]>
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> | |
e222432b | 30 | #include <linux/limits.h> |
8c7c6e34 | 31 | #include <linux/mutex.h> |
b6ac57d5 | 32 | #include <linux/slab.h> |
66e1707b BS |
33 | #include <linux/swap.h> |
34 | #include <linux/spinlock.h> | |
35 | #include <linux/fs.h> | |
d2ceb9b7 | 36 | #include <linux/seq_file.h> |
33327948 | 37 | #include <linux/vmalloc.h> |
b69408e8 | 38 | #include <linux/mm_inline.h> |
52d4b9ac | 39 | #include <linux/page_cgroup.h> |
08e552c6 | 40 | #include "internal.h" |
8cdea7c0 | 41 | |
8697d331 BS |
42 | #include <asm/uaccess.h> |
43 | ||
a181b0e8 | 44 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 45 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 46 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 47 | |
c077719b | 48 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 49 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b KH |
50 | int do_swap_account __read_mostly; |
51 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
52 | #else | |
53 | #define do_swap_account (0) | |
54 | #endif | |
55 | ||
7f4d454d | 56 | static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */ |
c077719b | 57 | |
d52aa412 KH |
58 | /* |
59 | * Statistics for memory cgroup. | |
60 | */ | |
61 | enum mem_cgroup_stat_index { | |
62 | /* | |
63 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
64 | */ | |
65 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f BS |
66 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
67 | MEM_CGROUP_STAT_MAPPED_FILE, /* # of pages charged as file rss */ | |
55e462b0 BR |
68 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
69 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
70 | |
71 | MEM_CGROUP_STAT_NSTATS, | |
72 | }; | |
73 | ||
74 | struct mem_cgroup_stat_cpu { | |
75 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
76 | } ____cacheline_aligned_in_smp; | |
77 | ||
78 | struct mem_cgroup_stat { | |
c8dad2bb | 79 | struct mem_cgroup_stat_cpu cpustat[0]; |
d52aa412 KH |
80 | }; |
81 | ||
82 | /* | |
83 | * For accounting under irq disable, no need for increment preempt count. | |
84 | */ | |
addb9efe | 85 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
86 | enum mem_cgroup_stat_index idx, int val) |
87 | { | |
addb9efe | 88 | stat->count[idx] += val; |
d52aa412 KH |
89 | } |
90 | ||
91 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
92 | enum mem_cgroup_stat_index idx) | |
93 | { | |
94 | int cpu; | |
95 | s64 ret = 0; | |
96 | for_each_possible_cpu(cpu) | |
97 | ret += stat->cpustat[cpu].count[idx]; | |
98 | return ret; | |
99 | } | |
100 | ||
04046e1a KH |
101 | static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat) |
102 | { | |
103 | s64 ret; | |
104 | ||
105 | ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE); | |
106 | ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS); | |
107 | return ret; | |
108 | } | |
109 | ||
6d12e2d8 KH |
110 | /* |
111 | * per-zone information in memory controller. | |
112 | */ | |
6d12e2d8 | 113 | struct mem_cgroup_per_zone { |
072c56c1 KH |
114 | /* |
115 | * spin_lock to protect the per cgroup LRU | |
116 | */ | |
b69408e8 CL |
117 | struct list_head lists[NR_LRU_LISTS]; |
118 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
119 | |
120 | struct zone_reclaim_stat reclaim_stat; | |
6d12e2d8 KH |
121 | }; |
122 | /* Macro for accessing counter */ | |
123 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
124 | ||
125 | struct mem_cgroup_per_node { | |
126 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
127 | }; | |
128 | ||
129 | struct mem_cgroup_lru_info { | |
130 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
131 | }; | |
132 | ||
8cdea7c0 BS |
133 | /* |
134 | * The memory controller data structure. The memory controller controls both | |
135 | * page cache and RSS per cgroup. We would eventually like to provide | |
136 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
137 | * to help the administrator determine what knobs to tune. | |
138 | * | |
139 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
140 | * we hit the water mark. May be even add a low water mark, such that |
141 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
142 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
143 | */ |
144 | struct mem_cgroup { | |
145 | struct cgroup_subsys_state css; | |
146 | /* | |
147 | * the counter to account for memory usage | |
148 | */ | |
149 | struct res_counter res; | |
8c7c6e34 KH |
150 | /* |
151 | * the counter to account for mem+swap usage. | |
152 | */ | |
153 | struct res_counter memsw; | |
78fb7466 PE |
154 | /* |
155 | * Per cgroup active and inactive list, similar to the | |
156 | * per zone LRU lists. | |
78fb7466 | 157 | */ |
6d12e2d8 | 158 | struct mem_cgroup_lru_info info; |
072c56c1 | 159 | |
2733c06a KM |
160 | /* |
161 | protect against reclaim related member. | |
162 | */ | |
163 | spinlock_t reclaim_param_lock; | |
164 | ||
6c48a1d0 | 165 | int prev_priority; /* for recording reclaim priority */ |
6d61ef40 BS |
166 | |
167 | /* | |
168 | * While reclaiming in a hiearchy, we cache the last child we | |
04046e1a | 169 | * reclaimed from. |
6d61ef40 | 170 | */ |
04046e1a | 171 | int last_scanned_child; |
18f59ea7 BS |
172 | /* |
173 | * Should the accounting and control be hierarchical, per subtree? | |
174 | */ | |
175 | bool use_hierarchy; | |
a636b327 | 176 | unsigned long last_oom_jiffies; |
8c7c6e34 | 177 | atomic_t refcnt; |
14797e23 | 178 | |
a7885eb8 KM |
179 | unsigned int swappiness; |
180 | ||
22a668d7 KH |
181 | /* set when res.limit == memsw.limit */ |
182 | bool memsw_is_minimum; | |
183 | ||
d52aa412 | 184 | /* |
c8dad2bb | 185 | * statistics. This must be placed at the end of memcg. |
d52aa412 KH |
186 | */ |
187 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
188 | }; |
189 | ||
217bc319 KH |
190 | enum charge_type { |
191 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
192 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 193 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 194 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 195 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 196 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
197 | NR_CHARGE_TYPE, |
198 | }; | |
199 | ||
52d4b9ac KH |
200 | /* only for here (for easy reading.) */ |
201 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
202 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 203 | #define PCGF_LOCK (1UL << PCG_LOCK) |
4b3bde4c BS |
204 | /* Not used, but added here for completeness */ |
205 | #define PCGF_ACCT (1UL << PCG_ACCT) | |
217bc319 | 206 | |
8c7c6e34 KH |
207 | /* for encoding cft->private value on file */ |
208 | #define _MEM (0) | |
209 | #define _MEMSWAP (1) | |
210 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
211 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
212 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
213 | ||
214 | static void mem_cgroup_get(struct mem_cgroup *mem); | |
215 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 216 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
8c7c6e34 | 217 | |
c05555b5 KH |
218 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
219 | struct page_cgroup *pc, | |
220 | bool charge) | |
d52aa412 KH |
221 | { |
222 | int val = (charge)? 1 : -1; | |
223 | struct mem_cgroup_stat *stat = &mem->stat; | |
addb9efe | 224 | struct mem_cgroup_stat_cpu *cpustat; |
08e552c6 | 225 | int cpu = get_cpu(); |
d52aa412 | 226 | |
08e552c6 | 227 | cpustat = &stat->cpustat[cpu]; |
c05555b5 | 228 | if (PageCgroupCache(pc)) |
addb9efe | 229 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 230 | else |
addb9efe | 231 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
232 | |
233 | if (charge) | |
addb9efe | 234 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
235 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
236 | else | |
addb9efe | 237 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 238 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
08e552c6 | 239 | put_cpu(); |
6d12e2d8 KH |
240 | } |
241 | ||
d5b69e38 | 242 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
243 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
244 | { | |
6d12e2d8 KH |
245 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
246 | } | |
247 | ||
d5b69e38 | 248 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
249 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
250 | { | |
251 | struct mem_cgroup *mem = pc->mem_cgroup; | |
252 | int nid = page_cgroup_nid(pc); | |
253 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 254 | |
54992762 KM |
255 | if (!mem) |
256 | return NULL; | |
257 | ||
6d12e2d8 KH |
258 | return mem_cgroup_zoneinfo(mem, nid, zid); |
259 | } | |
260 | ||
14067bb3 | 261 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 262 | enum lru_list idx) |
6d12e2d8 KH |
263 | { |
264 | int nid, zid; | |
265 | struct mem_cgroup_per_zone *mz; | |
266 | u64 total = 0; | |
267 | ||
268 | for_each_online_node(nid) | |
269 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
270 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
271 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
272 | } | |
273 | return total; | |
d52aa412 KH |
274 | } |
275 | ||
d5b69e38 | 276 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
277 | { |
278 | return container_of(cgroup_subsys_state(cont, | |
279 | mem_cgroup_subsys_id), struct mem_cgroup, | |
280 | css); | |
281 | } | |
282 | ||
cf475ad2 | 283 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 284 | { |
31a78f23 BS |
285 | /* |
286 | * mm_update_next_owner() may clear mm->owner to NULL | |
287 | * if it races with swapoff, page migration, etc. | |
288 | * So this can be called with p == NULL. | |
289 | */ | |
290 | if (unlikely(!p)) | |
291 | return NULL; | |
292 | ||
78fb7466 PE |
293 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
294 | struct mem_cgroup, css); | |
295 | } | |
296 | ||
54595fe2 KH |
297 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
298 | { | |
299 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
300 | |
301 | if (!mm) | |
302 | return NULL; | |
54595fe2 KH |
303 | /* |
304 | * Because we have no locks, mm->owner's may be being moved to other | |
305 | * cgroup. We use css_tryget() here even if this looks | |
306 | * pessimistic (rather than adding locks here). | |
307 | */ | |
308 | rcu_read_lock(); | |
309 | do { | |
310 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
311 | if (unlikely(!mem)) | |
312 | break; | |
313 | } while (!css_tryget(&mem->css)); | |
314 | rcu_read_unlock(); | |
315 | return mem; | |
316 | } | |
317 | ||
14067bb3 KH |
318 | /* |
319 | * Call callback function against all cgroup under hierarchy tree. | |
320 | */ | |
321 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
322 | int (*func)(struct mem_cgroup *, void *)) | |
323 | { | |
324 | int found, ret, nextid; | |
325 | struct cgroup_subsys_state *css; | |
326 | struct mem_cgroup *mem; | |
327 | ||
328 | if (!root->use_hierarchy) | |
329 | return (*func)(root, data); | |
330 | ||
331 | nextid = 1; | |
332 | do { | |
333 | ret = 0; | |
334 | mem = NULL; | |
335 | ||
336 | rcu_read_lock(); | |
337 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
338 | &found); | |
339 | if (css && css_tryget(css)) | |
340 | mem = container_of(css, struct mem_cgroup, css); | |
341 | rcu_read_unlock(); | |
342 | ||
343 | if (mem) { | |
344 | ret = (*func)(mem, data); | |
345 | css_put(&mem->css); | |
346 | } | |
347 | nextid = found + 1; | |
348 | } while (!ret && css); | |
349 | ||
350 | return ret; | |
351 | } | |
352 | ||
4b3bde4c BS |
353 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
354 | { | |
355 | return (mem == root_mem_cgroup); | |
356 | } | |
357 | ||
08e552c6 KH |
358 | /* |
359 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
360 | * Operations are called by routine of global LRU independently from memcg. | |
361 | * What we have to take care of here is validness of pc->mem_cgroup. | |
362 | * | |
363 | * Changes to pc->mem_cgroup happens when | |
364 | * 1. charge | |
365 | * 2. moving account | |
366 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
367 | * It is added to LRU before charge. | |
368 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
369 | * When moving account, the page is not on LRU. It's isolated. | |
370 | */ | |
4f98a2fe | 371 | |
08e552c6 KH |
372 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
373 | { | |
374 | struct page_cgroup *pc; | |
08e552c6 | 375 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 376 | |
f8d66542 | 377 | if (mem_cgroup_disabled()) |
08e552c6 KH |
378 | return; |
379 | pc = lookup_page_cgroup(page); | |
380 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 381 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 382 | return; |
4b3bde4c | 383 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
384 | /* |
385 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
386 | * removed from global LRU. | |
387 | */ | |
08e552c6 | 388 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 389 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
4b3bde4c BS |
390 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
391 | return; | |
392 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 KH |
393 | list_del_init(&pc->lru); |
394 | return; | |
6d12e2d8 KH |
395 | } |
396 | ||
08e552c6 | 397 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 398 | { |
08e552c6 KH |
399 | mem_cgroup_del_lru_list(page, page_lru(page)); |
400 | } | |
b69408e8 | 401 | |
08e552c6 KH |
402 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
403 | { | |
404 | struct mem_cgroup_per_zone *mz; | |
405 | struct page_cgroup *pc; | |
b69408e8 | 406 | |
f8d66542 | 407 | if (mem_cgroup_disabled()) |
08e552c6 | 408 | return; |
6d12e2d8 | 409 | |
08e552c6 | 410 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
411 | /* |
412 | * Used bit is set without atomic ops but after smp_wmb(). | |
413 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
414 | */ | |
08e552c6 | 415 | smp_rmb(); |
4b3bde4c BS |
416 | /* unused or root page is not rotated. */ |
417 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 KH |
418 | return; |
419 | mz = page_cgroup_zoneinfo(pc); | |
420 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
421 | } |
422 | ||
08e552c6 | 423 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 424 | { |
08e552c6 KH |
425 | struct page_cgroup *pc; |
426 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 427 | |
f8d66542 | 428 | if (mem_cgroup_disabled()) |
08e552c6 KH |
429 | return; |
430 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 431 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
bd112db8 DN |
432 | /* |
433 | * Used bit is set without atomic ops but after smp_wmb(). | |
434 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
435 | */ | |
08e552c6 KH |
436 | smp_rmb(); |
437 | if (!PageCgroupUsed(pc)) | |
894bc310 | 438 | return; |
b69408e8 | 439 | |
08e552c6 | 440 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 441 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
4b3bde4c BS |
442 | SetPageCgroupAcctLRU(pc); |
443 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
444 | return; | |
08e552c6 KH |
445 | list_add(&pc->lru, &mz->lists[lru]); |
446 | } | |
544122e5 | 447 | |
08e552c6 | 448 | /* |
544122e5 KH |
449 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
450 | * lru because the page may.be reused after it's fully uncharged (because of | |
451 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
452 | * it again. This function is only used to charge SwapCache. It's done under | |
453 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 454 | */ |
544122e5 | 455 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 456 | { |
544122e5 KH |
457 | unsigned long flags; |
458 | struct zone *zone = page_zone(page); | |
459 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
460 | ||
461 | spin_lock_irqsave(&zone->lru_lock, flags); | |
462 | /* | |
463 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
464 | * is guarded by lock_page() because the page is SwapCache. | |
465 | */ | |
466 | if (!PageCgroupUsed(pc)) | |
467 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
468 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
469 | } |
470 | ||
544122e5 KH |
471 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
472 | { | |
473 | unsigned long flags; | |
474 | struct zone *zone = page_zone(page); | |
475 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
476 | ||
477 | spin_lock_irqsave(&zone->lru_lock, flags); | |
478 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 479 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
480 | mem_cgroup_add_lru_list(page, page_lru(page)); |
481 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
482 | } | |
483 | ||
484 | ||
08e552c6 KH |
485 | void mem_cgroup_move_lists(struct page *page, |
486 | enum lru_list from, enum lru_list to) | |
487 | { | |
f8d66542 | 488 | if (mem_cgroup_disabled()) |
08e552c6 KH |
489 | return; |
490 | mem_cgroup_del_lru_list(page, from); | |
491 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
492 | } |
493 | ||
4c4a2214 DR |
494 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
495 | { | |
496 | int ret; | |
0b7f569e | 497 | struct mem_cgroup *curr = NULL; |
4c4a2214 DR |
498 | |
499 | task_lock(task); | |
0b7f569e KH |
500 | rcu_read_lock(); |
501 | curr = try_get_mem_cgroup_from_mm(task->mm); | |
502 | rcu_read_unlock(); | |
4c4a2214 | 503 | task_unlock(task); |
0b7f569e KH |
504 | if (!curr) |
505 | return 0; | |
506 | if (curr->use_hierarchy) | |
507 | ret = css_is_ancestor(&curr->css, &mem->css); | |
508 | else | |
509 | ret = (curr == mem); | |
510 | css_put(&curr->css); | |
4c4a2214 DR |
511 | return ret; |
512 | } | |
513 | ||
6c48a1d0 KH |
514 | /* |
515 | * prev_priority control...this will be used in memory reclaim path. | |
516 | */ | |
517 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
518 | { | |
2733c06a KM |
519 | int prev_priority; |
520 | ||
521 | spin_lock(&mem->reclaim_param_lock); | |
522 | prev_priority = mem->prev_priority; | |
523 | spin_unlock(&mem->reclaim_param_lock); | |
524 | ||
525 | return prev_priority; | |
6c48a1d0 KH |
526 | } |
527 | ||
528 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
529 | { | |
2733c06a | 530 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
531 | if (priority < mem->prev_priority) |
532 | mem->prev_priority = priority; | |
2733c06a | 533 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
534 | } |
535 | ||
536 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
537 | { | |
2733c06a | 538 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 | 539 | mem->prev_priority = priority; |
2733c06a | 540 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
541 | } |
542 | ||
c772be93 | 543 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
544 | { |
545 | unsigned long active; | |
546 | unsigned long inactive; | |
c772be93 KM |
547 | unsigned long gb; |
548 | unsigned long inactive_ratio; | |
14797e23 | 549 | |
14067bb3 KH |
550 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
551 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 552 | |
c772be93 KM |
553 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
554 | if (gb) | |
555 | inactive_ratio = int_sqrt(10 * gb); | |
556 | else | |
557 | inactive_ratio = 1; | |
558 | ||
559 | if (present_pages) { | |
560 | present_pages[0] = inactive; | |
561 | present_pages[1] = active; | |
562 | } | |
563 | ||
564 | return inactive_ratio; | |
565 | } | |
566 | ||
567 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
568 | { | |
569 | unsigned long active; | |
570 | unsigned long inactive; | |
571 | unsigned long present_pages[2]; | |
572 | unsigned long inactive_ratio; | |
573 | ||
574 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
575 | ||
576 | inactive = present_pages[0]; | |
577 | active = present_pages[1]; | |
578 | ||
579 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
580 | return 1; |
581 | ||
582 | return 0; | |
583 | } | |
584 | ||
56e49d21 RR |
585 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
586 | { | |
587 | unsigned long active; | |
588 | unsigned long inactive; | |
589 | ||
590 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
591 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
592 | ||
593 | return (active > inactive); | |
594 | } | |
595 | ||
a3d8e054 KM |
596 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
597 | struct zone *zone, | |
598 | enum lru_list lru) | |
599 | { | |
600 | int nid = zone->zone_pgdat->node_id; | |
601 | int zid = zone_idx(zone); | |
602 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
603 | ||
604 | return MEM_CGROUP_ZSTAT(mz, lru); | |
605 | } | |
606 | ||
3e2f41f1 KM |
607 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
608 | struct zone *zone) | |
609 | { | |
610 | int nid = zone->zone_pgdat->node_id; | |
611 | int zid = zone_idx(zone); | |
612 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
613 | ||
614 | return &mz->reclaim_stat; | |
615 | } | |
616 | ||
617 | struct zone_reclaim_stat * | |
618 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
619 | { | |
620 | struct page_cgroup *pc; | |
621 | struct mem_cgroup_per_zone *mz; | |
622 | ||
623 | if (mem_cgroup_disabled()) | |
624 | return NULL; | |
625 | ||
626 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
627 | /* |
628 | * Used bit is set without atomic ops but after smp_wmb(). | |
629 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
630 | */ | |
631 | smp_rmb(); | |
632 | if (!PageCgroupUsed(pc)) | |
633 | return NULL; | |
634 | ||
3e2f41f1 KM |
635 | mz = page_cgroup_zoneinfo(pc); |
636 | if (!mz) | |
637 | return NULL; | |
638 | ||
639 | return &mz->reclaim_stat; | |
640 | } | |
641 | ||
66e1707b BS |
642 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
643 | struct list_head *dst, | |
644 | unsigned long *scanned, int order, | |
645 | int mode, struct zone *z, | |
646 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 647 | int active, int file) |
66e1707b BS |
648 | { |
649 | unsigned long nr_taken = 0; | |
650 | struct page *page; | |
651 | unsigned long scan; | |
652 | LIST_HEAD(pc_list); | |
653 | struct list_head *src; | |
ff7283fa | 654 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
655 | int nid = z->zone_pgdat->node_id; |
656 | int zid = zone_idx(z); | |
657 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 658 | int lru = LRU_FILE * file + active; |
2ffebca6 | 659 | int ret; |
66e1707b | 660 | |
cf475ad2 | 661 | BUG_ON(!mem_cont); |
1ecaab2b | 662 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 663 | src = &mz->lists[lru]; |
66e1707b | 664 | |
ff7283fa KH |
665 | scan = 0; |
666 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 667 | if (scan >= nr_to_scan) |
ff7283fa | 668 | break; |
08e552c6 KH |
669 | |
670 | page = pc->page; | |
52d4b9ac KH |
671 | if (unlikely(!PageCgroupUsed(pc))) |
672 | continue; | |
436c6541 | 673 | if (unlikely(!PageLRU(page))) |
ff7283fa | 674 | continue; |
ff7283fa | 675 | |
436c6541 | 676 | scan++; |
2ffebca6 KH |
677 | ret = __isolate_lru_page(page, mode, file); |
678 | switch (ret) { | |
679 | case 0: | |
66e1707b | 680 | list_move(&page->lru, dst); |
2ffebca6 | 681 | mem_cgroup_del_lru(page); |
66e1707b | 682 | nr_taken++; |
2ffebca6 KH |
683 | break; |
684 | case -EBUSY: | |
685 | /* we don't affect global LRU but rotate in our LRU */ | |
686 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
687 | break; | |
688 | default: | |
689 | break; | |
66e1707b BS |
690 | } |
691 | } | |
692 | ||
66e1707b BS |
693 | *scanned = scan; |
694 | return nr_taken; | |
695 | } | |
696 | ||
6d61ef40 BS |
697 | #define mem_cgroup_from_res_counter(counter, member) \ |
698 | container_of(counter, struct mem_cgroup, member) | |
699 | ||
b85a96c0 DN |
700 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
701 | { | |
702 | if (do_swap_account) { | |
703 | if (res_counter_check_under_limit(&mem->res) && | |
704 | res_counter_check_under_limit(&mem->memsw)) | |
705 | return true; | |
706 | } else | |
707 | if (res_counter_check_under_limit(&mem->res)) | |
708 | return true; | |
709 | return false; | |
710 | } | |
711 | ||
a7885eb8 KM |
712 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
713 | { | |
714 | struct cgroup *cgrp = memcg->css.cgroup; | |
715 | unsigned int swappiness; | |
716 | ||
717 | /* root ? */ | |
718 | if (cgrp->parent == NULL) | |
719 | return vm_swappiness; | |
720 | ||
721 | spin_lock(&memcg->reclaim_param_lock); | |
722 | swappiness = memcg->swappiness; | |
723 | spin_unlock(&memcg->reclaim_param_lock); | |
724 | ||
725 | return swappiness; | |
726 | } | |
727 | ||
81d39c20 KH |
728 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
729 | { | |
730 | int *val = data; | |
731 | (*val)++; | |
732 | return 0; | |
733 | } | |
e222432b BS |
734 | |
735 | /** | |
736 | * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode. | |
737 | * @memcg: The memory cgroup that went over limit | |
738 | * @p: Task that is going to be killed | |
739 | * | |
740 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
741 | * enabled | |
742 | */ | |
743 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
744 | { | |
745 | struct cgroup *task_cgrp; | |
746 | struct cgroup *mem_cgrp; | |
747 | /* | |
748 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
749 | * on the assumption that OOM is serialized for memory controller. | |
750 | * If this assumption is broken, revisit this code. | |
751 | */ | |
752 | static char memcg_name[PATH_MAX]; | |
753 | int ret; | |
754 | ||
755 | if (!memcg) | |
756 | return; | |
757 | ||
758 | ||
759 | rcu_read_lock(); | |
760 | ||
761 | mem_cgrp = memcg->css.cgroup; | |
762 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
763 | ||
764 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
765 | if (ret < 0) { | |
766 | /* | |
767 | * Unfortunately, we are unable to convert to a useful name | |
768 | * But we'll still print out the usage information | |
769 | */ | |
770 | rcu_read_unlock(); | |
771 | goto done; | |
772 | } | |
773 | rcu_read_unlock(); | |
774 | ||
775 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
776 | ||
777 | rcu_read_lock(); | |
778 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
779 | if (ret < 0) { | |
780 | rcu_read_unlock(); | |
781 | goto done; | |
782 | } | |
783 | rcu_read_unlock(); | |
784 | ||
785 | /* | |
786 | * Continues from above, so we don't need an KERN_ level | |
787 | */ | |
788 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
789 | done: | |
790 | ||
791 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
792 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
793 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
794 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
795 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
796 | "failcnt %llu\n", | |
797 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
798 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
799 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
800 | } | |
801 | ||
81d39c20 KH |
802 | /* |
803 | * This function returns the number of memcg under hierarchy tree. Returns | |
804 | * 1(self count) if no children. | |
805 | */ | |
806 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
807 | { | |
808 | int num = 0; | |
809 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
810 | return num; | |
811 | } | |
812 | ||
6d61ef40 | 813 | /* |
04046e1a KH |
814 | * Visit the first child (need not be the first child as per the ordering |
815 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
816 | * that to reclaim free pages from. | |
817 | */ | |
818 | static struct mem_cgroup * | |
819 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
820 | { | |
821 | struct mem_cgroup *ret = NULL; | |
822 | struct cgroup_subsys_state *css; | |
823 | int nextid, found; | |
824 | ||
825 | if (!root_mem->use_hierarchy) { | |
826 | css_get(&root_mem->css); | |
827 | ret = root_mem; | |
828 | } | |
829 | ||
830 | while (!ret) { | |
831 | rcu_read_lock(); | |
832 | nextid = root_mem->last_scanned_child + 1; | |
833 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
834 | &found); | |
835 | if (css && css_tryget(css)) | |
836 | ret = container_of(css, struct mem_cgroup, css); | |
837 | ||
838 | rcu_read_unlock(); | |
839 | /* Updates scanning parameter */ | |
840 | spin_lock(&root_mem->reclaim_param_lock); | |
841 | if (!css) { | |
842 | /* this means start scan from ID:1 */ | |
843 | root_mem->last_scanned_child = 0; | |
844 | } else | |
845 | root_mem->last_scanned_child = found; | |
846 | spin_unlock(&root_mem->reclaim_param_lock); | |
847 | } | |
848 | ||
849 | return ret; | |
850 | } | |
851 | ||
852 | /* | |
853 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
854 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
855 | * based on its position in the children list. | |
6d61ef40 BS |
856 | * |
857 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
858 | * |
859 | * We give up and return to the caller when we visit root_mem twice. | |
860 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
861 | * |
862 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
863 | */ |
864 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
81d39c20 | 865 | gfp_t gfp_mask, bool noswap, bool shrink) |
6d61ef40 | 866 | { |
04046e1a KH |
867 | struct mem_cgroup *victim; |
868 | int ret, total = 0; | |
869 | int loop = 0; | |
870 | ||
22a668d7 KH |
871 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
872 | if (root_mem->memsw_is_minimum) | |
873 | noswap = true; | |
874 | ||
04046e1a KH |
875 | while (loop < 2) { |
876 | victim = mem_cgroup_select_victim(root_mem); | |
877 | if (victim == root_mem) | |
878 | loop++; | |
879 | if (!mem_cgroup_local_usage(&victim->stat)) { | |
880 | /* this cgroup's local usage == 0 */ | |
881 | css_put(&victim->css); | |
6d61ef40 BS |
882 | continue; |
883 | } | |
04046e1a KH |
884 | /* we use swappiness of local cgroup */ |
885 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap, | |
886 | get_swappiness(victim)); | |
887 | css_put(&victim->css); | |
81d39c20 KH |
888 | /* |
889 | * At shrinking usage, we can't check we should stop here or | |
890 | * reclaim more. It's depends on callers. last_scanned_child | |
891 | * will work enough for keeping fairness under tree. | |
892 | */ | |
893 | if (shrink) | |
894 | return ret; | |
04046e1a | 895 | total += ret; |
b85a96c0 | 896 | if (mem_cgroup_check_under_limit(root_mem)) |
04046e1a | 897 | return 1 + total; |
6d61ef40 | 898 | } |
04046e1a | 899 | return total; |
6d61ef40 BS |
900 | } |
901 | ||
a636b327 KH |
902 | bool mem_cgroup_oom_called(struct task_struct *task) |
903 | { | |
904 | bool ret = false; | |
905 | struct mem_cgroup *mem; | |
906 | struct mm_struct *mm; | |
907 | ||
908 | rcu_read_lock(); | |
909 | mm = task->mm; | |
910 | if (!mm) | |
911 | mm = &init_mm; | |
912 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
913 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
914 | ret = true; | |
915 | rcu_read_unlock(); | |
916 | return ret; | |
917 | } | |
0b7f569e KH |
918 | |
919 | static int record_last_oom_cb(struct mem_cgroup *mem, void *data) | |
920 | { | |
921 | mem->last_oom_jiffies = jiffies; | |
922 | return 0; | |
923 | } | |
924 | ||
925 | static void record_last_oom(struct mem_cgroup *mem) | |
926 | { | |
927 | mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb); | |
928 | } | |
929 | ||
d69b042f BS |
930 | /* |
931 | * Currently used to update mapped file statistics, but the routine can be | |
932 | * generalized to update other statistics as well. | |
933 | */ | |
934 | void mem_cgroup_update_mapped_file_stat(struct page *page, int val) | |
935 | { | |
936 | struct mem_cgroup *mem; | |
937 | struct mem_cgroup_stat *stat; | |
938 | struct mem_cgroup_stat_cpu *cpustat; | |
939 | int cpu; | |
940 | struct page_cgroup *pc; | |
941 | ||
942 | if (!page_is_file_cache(page)) | |
943 | return; | |
944 | ||
945 | pc = lookup_page_cgroup(page); | |
946 | if (unlikely(!pc)) | |
947 | return; | |
948 | ||
949 | lock_page_cgroup(pc); | |
950 | mem = pc->mem_cgroup; | |
951 | if (!mem) | |
952 | goto done; | |
953 | ||
954 | if (!PageCgroupUsed(pc)) | |
955 | goto done; | |
956 | ||
957 | /* | |
958 | * Preemption is already disabled, we don't need get_cpu() | |
959 | */ | |
960 | cpu = smp_processor_id(); | |
961 | stat = &mem->stat; | |
962 | cpustat = &stat->cpustat[cpu]; | |
963 | ||
964 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, val); | |
965 | done: | |
966 | unlock_page_cgroup(pc); | |
967 | } | |
0b7f569e | 968 | |
f817ed48 KH |
969 | /* |
970 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
971 | * oom-killer can be invoked. | |
8a9f3ccd | 972 | */ |
f817ed48 | 973 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
8c7c6e34 KH |
974 | gfp_t gfp_mask, struct mem_cgroup **memcg, |
975 | bool oom) | |
8a9f3ccd | 976 | { |
6d61ef40 | 977 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 978 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
28dbc4b6 | 979 | struct res_counter *fail_res; |
a636b327 KH |
980 | |
981 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
982 | /* Don't account this! */ | |
983 | *memcg = NULL; | |
984 | return 0; | |
985 | } | |
986 | ||
8a9f3ccd | 987 | /* |
3be91277 HD |
988 | * We always charge the cgroup the mm_struct belongs to. |
989 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
990 | * thread group leader migrates. It's possible that mm is not |
991 | * set, if so charge the init_mm (happens for pagecache usage). | |
992 | */ | |
54595fe2 KH |
993 | mem = *memcg; |
994 | if (likely(!mem)) { | |
995 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 996 | *memcg = mem; |
e8589cc1 | 997 | } else { |
7a81b88c | 998 | css_get(&mem->css); |
e8589cc1 | 999 | } |
54595fe2 KH |
1000 | if (unlikely(!mem)) |
1001 | return 0; | |
1002 | ||
46f7e602 | 1003 | VM_BUG_ON(css_is_removed(&mem->css)); |
8a9f3ccd | 1004 | |
8c7c6e34 KH |
1005 | while (1) { |
1006 | int ret; | |
1007 | bool noswap = false; | |
7a81b88c | 1008 | |
28dbc4b6 | 1009 | ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res); |
8c7c6e34 KH |
1010 | if (likely(!ret)) { |
1011 | if (!do_swap_account) | |
1012 | break; | |
28dbc4b6 BS |
1013 | ret = res_counter_charge(&mem->memsw, PAGE_SIZE, |
1014 | &fail_res); | |
8c7c6e34 KH |
1015 | if (likely(!ret)) |
1016 | break; | |
1017 | /* mem+swap counter fails */ | |
1018 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
1019 | noswap = true; | |
6d61ef40 BS |
1020 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
1021 | memsw); | |
1022 | } else | |
1023 | /* mem counter fails */ | |
1024 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
1025 | res); | |
1026 | ||
3be91277 | 1027 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 1028 | goto nomem; |
e1a1cd59 | 1029 | |
6d61ef40 | 1030 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask, |
81d39c20 | 1031 | noswap, false); |
4d1c6273 DN |
1032 | if (ret) |
1033 | continue; | |
66e1707b BS |
1034 | |
1035 | /* | |
8869b8f6 HD |
1036 | * try_to_free_mem_cgroup_pages() might not give us a full |
1037 | * picture of reclaim. Some pages are reclaimed and might be | |
1038 | * moved to swap cache or just unmapped from the cgroup. | |
1039 | * Check the limit again to see if the reclaim reduced the | |
1040 | * current usage of the cgroup before giving up | |
8c7c6e34 | 1041 | * |
8869b8f6 | 1042 | */ |
b85a96c0 DN |
1043 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
1044 | continue; | |
3be91277 HD |
1045 | |
1046 | if (!nr_retries--) { | |
a636b327 | 1047 | if (oom) { |
7f4d454d | 1048 | mutex_lock(&memcg_tasklist); |
88700756 | 1049 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); |
7f4d454d | 1050 | mutex_unlock(&memcg_tasklist); |
0b7f569e | 1051 | record_last_oom(mem_over_limit); |
a636b327 | 1052 | } |
7a81b88c | 1053 | goto nomem; |
66e1707b | 1054 | } |
8a9f3ccd | 1055 | } |
7a81b88c KH |
1056 | return 0; |
1057 | nomem: | |
1058 | css_put(&mem->css); | |
1059 | return -ENOMEM; | |
1060 | } | |
8a9f3ccd | 1061 | |
a3b2d692 KH |
1062 | |
1063 | /* | |
1064 | * A helper function to get mem_cgroup from ID. must be called under | |
1065 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
1066 | * it's concern. (dropping refcnt from swap can be called against removed | |
1067 | * memcg.) | |
1068 | */ | |
1069 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
1070 | { | |
1071 | struct cgroup_subsys_state *css; | |
1072 | ||
1073 | /* ID 0 is unused ID */ | |
1074 | if (!id) | |
1075 | return NULL; | |
1076 | css = css_lookup(&mem_cgroup_subsys, id); | |
1077 | if (!css) | |
1078 | return NULL; | |
1079 | return container_of(css, struct mem_cgroup, css); | |
1080 | } | |
1081 | ||
b5a84319 KH |
1082 | static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page) |
1083 | { | |
1084 | struct mem_cgroup *mem; | |
3c776e64 | 1085 | struct page_cgroup *pc; |
a3b2d692 | 1086 | unsigned short id; |
b5a84319 KH |
1087 | swp_entry_t ent; |
1088 | ||
3c776e64 DN |
1089 | VM_BUG_ON(!PageLocked(page)); |
1090 | ||
b5a84319 KH |
1091 | if (!PageSwapCache(page)) |
1092 | return NULL; | |
1093 | ||
3c776e64 | 1094 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 1095 | lock_page_cgroup(pc); |
a3b2d692 | 1096 | if (PageCgroupUsed(pc)) { |
3c776e64 | 1097 | mem = pc->mem_cgroup; |
a3b2d692 KH |
1098 | if (mem && !css_tryget(&mem->css)) |
1099 | mem = NULL; | |
1100 | } else { | |
3c776e64 | 1101 | ent.val = page_private(page); |
a3b2d692 KH |
1102 | id = lookup_swap_cgroup(ent); |
1103 | rcu_read_lock(); | |
1104 | mem = mem_cgroup_lookup(id); | |
1105 | if (mem && !css_tryget(&mem->css)) | |
1106 | mem = NULL; | |
1107 | rcu_read_unlock(); | |
3c776e64 | 1108 | } |
c0bd3f63 | 1109 | unlock_page_cgroup(pc); |
b5a84319 KH |
1110 | return mem; |
1111 | } | |
1112 | ||
7a81b88c | 1113 | /* |
a5e924f5 | 1114 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
1115 | * USED state. If already USED, uncharge and return. |
1116 | */ | |
1117 | ||
1118 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
1119 | struct page_cgroup *pc, | |
1120 | enum charge_type ctype) | |
1121 | { | |
7a81b88c KH |
1122 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
1123 | if (!mem) | |
1124 | return; | |
52d4b9ac KH |
1125 | |
1126 | lock_page_cgroup(pc); | |
1127 | if (unlikely(PageCgroupUsed(pc))) { | |
1128 | unlock_page_cgroup(pc); | |
1129 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
1130 | if (do_swap_account) |
1131 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
52d4b9ac | 1132 | css_put(&mem->css); |
7a81b88c | 1133 | return; |
52d4b9ac | 1134 | } |
4b3bde4c | 1135 | |
8a9f3ccd | 1136 | pc->mem_cgroup = mem; |
08e552c6 | 1137 | smp_wmb(); |
4b3bde4c BS |
1138 | switch (ctype) { |
1139 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
1140 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
1141 | SetPageCgroupCache(pc); | |
1142 | SetPageCgroupUsed(pc); | |
1143 | break; | |
1144 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1145 | ClearPageCgroupCache(pc); | |
1146 | SetPageCgroupUsed(pc); | |
1147 | break; | |
1148 | default: | |
1149 | break; | |
1150 | } | |
3be91277 | 1151 | |
08e552c6 | 1152 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 1153 | |
52d4b9ac | 1154 | unlock_page_cgroup(pc); |
7a81b88c | 1155 | } |
66e1707b | 1156 | |
f817ed48 KH |
1157 | /** |
1158 | * mem_cgroup_move_account - move account of the page | |
1159 | * @pc: page_cgroup of the page. | |
1160 | * @from: mem_cgroup which the page is moved from. | |
1161 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
1162 | * | |
1163 | * The caller must confirm following. | |
08e552c6 | 1164 | * - page is not on LRU (isolate_page() is useful.) |
f817ed48 KH |
1165 | * |
1166 | * returns 0 at success, | |
1167 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
1168 | * | |
1169 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
1170 | * new cgroup. It should be done by a caller. | |
1171 | */ | |
1172 | ||
1173 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
1174 | struct mem_cgroup *from, struct mem_cgroup *to) | |
1175 | { | |
1176 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
1177 | int nid, zid; | |
1178 | int ret = -EBUSY; | |
d69b042f BS |
1179 | struct page *page; |
1180 | int cpu; | |
1181 | struct mem_cgroup_stat *stat; | |
1182 | struct mem_cgroup_stat_cpu *cpustat; | |
f817ed48 | 1183 | |
f817ed48 | 1184 | VM_BUG_ON(from == to); |
08e552c6 | 1185 | VM_BUG_ON(PageLRU(pc->page)); |
f817ed48 KH |
1186 | |
1187 | nid = page_cgroup_nid(pc); | |
1188 | zid = page_cgroup_zid(pc); | |
1189 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
1190 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
1191 | ||
f817ed48 KH |
1192 | if (!trylock_page_cgroup(pc)) |
1193 | return ret; | |
1194 | ||
1195 | if (!PageCgroupUsed(pc)) | |
1196 | goto out; | |
1197 | ||
1198 | if (pc->mem_cgroup != from) | |
1199 | goto out; | |
1200 | ||
08e552c6 KH |
1201 | res_counter_uncharge(&from->res, PAGE_SIZE); |
1202 | mem_cgroup_charge_statistics(from, pc, false); | |
d69b042f BS |
1203 | |
1204 | page = pc->page; | |
1205 | if (page_is_file_cache(page) && page_mapped(page)) { | |
1206 | cpu = smp_processor_id(); | |
1207 | /* Update mapped_file data for mem_cgroup "from" */ | |
1208 | stat = &from->stat; | |
1209 | cpustat = &stat->cpustat[cpu]; | |
1210 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, | |
1211 | -1); | |
1212 | ||
1213 | /* Update mapped_file data for mem_cgroup "to" */ | |
1214 | stat = &to->stat; | |
1215 | cpustat = &stat->cpustat[cpu]; | |
1216 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, | |
1217 | 1); | |
1218 | } | |
1219 | ||
08e552c6 KH |
1220 | if (do_swap_account) |
1221 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
40d58138 DN |
1222 | css_put(&from->css); |
1223 | ||
1224 | css_get(&to->css); | |
08e552c6 KH |
1225 | pc->mem_cgroup = to; |
1226 | mem_cgroup_charge_statistics(to, pc, true); | |
08e552c6 | 1227 | ret = 0; |
f817ed48 KH |
1228 | out: |
1229 | unlock_page_cgroup(pc); | |
88703267 KH |
1230 | /* |
1231 | * We charges against "to" which may not have any tasks. Then, "to" | |
1232 | * can be under rmdir(). But in current implementation, caller of | |
1233 | * this function is just force_empty() and it's garanteed that | |
1234 | * "to" is never removed. So, we don't check rmdir status here. | |
1235 | */ | |
f817ed48 KH |
1236 | return ret; |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * move charges to its parent. | |
1241 | */ | |
1242 | ||
1243 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1244 | struct mem_cgroup *child, | |
1245 | gfp_t gfp_mask) | |
1246 | { | |
08e552c6 | 1247 | struct page *page = pc->page; |
f817ed48 KH |
1248 | struct cgroup *cg = child->css.cgroup; |
1249 | struct cgroup *pcg = cg->parent; | |
1250 | struct mem_cgroup *parent; | |
f817ed48 KH |
1251 | int ret; |
1252 | ||
1253 | /* Is ROOT ? */ | |
1254 | if (!pcg) | |
1255 | return -EINVAL; | |
1256 | ||
08e552c6 | 1257 | |
f817ed48 KH |
1258 | parent = mem_cgroup_from_cont(pcg); |
1259 | ||
08e552c6 | 1260 | |
f817ed48 | 1261 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); |
a636b327 | 1262 | if (ret || !parent) |
f817ed48 KH |
1263 | return ret; |
1264 | ||
40d58138 DN |
1265 | if (!get_page_unless_zero(page)) { |
1266 | ret = -EBUSY; | |
1267 | goto uncharge; | |
1268 | } | |
08e552c6 KH |
1269 | |
1270 | ret = isolate_lru_page(page); | |
1271 | ||
1272 | if (ret) | |
1273 | goto cancel; | |
f817ed48 | 1274 | |
f817ed48 | 1275 | ret = mem_cgroup_move_account(pc, child, parent); |
f817ed48 | 1276 | |
08e552c6 KH |
1277 | putback_lru_page(page); |
1278 | if (!ret) { | |
1279 | put_page(page); | |
40d58138 DN |
1280 | /* drop extra refcnt by try_charge() */ |
1281 | css_put(&parent->css); | |
08e552c6 | 1282 | return 0; |
8c7c6e34 | 1283 | } |
40d58138 | 1284 | |
08e552c6 | 1285 | cancel: |
40d58138 DN |
1286 | put_page(page); |
1287 | uncharge: | |
1288 | /* drop extra refcnt by try_charge() */ | |
1289 | css_put(&parent->css); | |
1290 | /* uncharge if move fails */ | |
08e552c6 KH |
1291 | res_counter_uncharge(&parent->res, PAGE_SIZE); |
1292 | if (do_swap_account) | |
1293 | res_counter_uncharge(&parent->memsw, PAGE_SIZE); | |
f817ed48 KH |
1294 | return ret; |
1295 | } | |
1296 | ||
7a81b88c KH |
1297 | /* |
1298 | * Charge the memory controller for page usage. | |
1299 | * Return | |
1300 | * 0 if the charge was successful | |
1301 | * < 0 if the cgroup is over its limit | |
1302 | */ | |
1303 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1304 | gfp_t gfp_mask, enum charge_type ctype, | |
1305 | struct mem_cgroup *memcg) | |
1306 | { | |
1307 | struct mem_cgroup *mem; | |
1308 | struct page_cgroup *pc; | |
1309 | int ret; | |
1310 | ||
1311 | pc = lookup_page_cgroup(page); | |
1312 | /* can happen at boot */ | |
1313 | if (unlikely(!pc)) | |
1314 | return 0; | |
1315 | prefetchw(pc); | |
1316 | ||
1317 | mem = memcg; | |
f817ed48 | 1318 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
a636b327 | 1319 | if (ret || !mem) |
7a81b88c KH |
1320 | return ret; |
1321 | ||
1322 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1323 | return 0; |
8a9f3ccd BS |
1324 | } |
1325 | ||
7a81b88c KH |
1326 | int mem_cgroup_newpage_charge(struct page *page, |
1327 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1328 | { |
f8d66542 | 1329 | if (mem_cgroup_disabled()) |
cede86ac | 1330 | return 0; |
52d4b9ac KH |
1331 | if (PageCompound(page)) |
1332 | return 0; | |
69029cd5 KH |
1333 | /* |
1334 | * If already mapped, we don't have to account. | |
1335 | * If page cache, page->mapping has address_space. | |
1336 | * But page->mapping may have out-of-use anon_vma pointer, | |
1337 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1338 | * is NULL. | |
1339 | */ | |
1340 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
1341 | return 0; | |
1342 | if (unlikely(!mm)) | |
1343 | mm = &init_mm; | |
217bc319 | 1344 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 1345 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
1346 | } |
1347 | ||
83aae4c7 DN |
1348 | static void |
1349 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1350 | enum charge_type ctype); | |
1351 | ||
e1a1cd59 BS |
1352 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
1353 | gfp_t gfp_mask) | |
8697d331 | 1354 | { |
b5a84319 KH |
1355 | struct mem_cgroup *mem = NULL; |
1356 | int ret; | |
1357 | ||
f8d66542 | 1358 | if (mem_cgroup_disabled()) |
cede86ac | 1359 | return 0; |
52d4b9ac KH |
1360 | if (PageCompound(page)) |
1361 | return 0; | |
accf163e KH |
1362 | /* |
1363 | * Corner case handling. This is called from add_to_page_cache() | |
1364 | * in usual. But some FS (shmem) precharges this page before calling it | |
1365 | * and call add_to_page_cache() with GFP_NOWAIT. | |
1366 | * | |
1367 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
1368 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
1369 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
1370 | * And when the page is SwapCache, it should take swap information |
1371 | * into account. This is under lock_page() now. | |
accf163e KH |
1372 | */ |
1373 | if (!(gfp_mask & __GFP_WAIT)) { | |
1374 | struct page_cgroup *pc; | |
1375 | ||
52d4b9ac KH |
1376 | |
1377 | pc = lookup_page_cgroup(page); | |
1378 | if (!pc) | |
1379 | return 0; | |
1380 | lock_page_cgroup(pc); | |
1381 | if (PageCgroupUsed(pc)) { | |
1382 | unlock_page_cgroup(pc); | |
accf163e KH |
1383 | return 0; |
1384 | } | |
52d4b9ac | 1385 | unlock_page_cgroup(pc); |
accf163e KH |
1386 | } |
1387 | ||
b5a84319 | 1388 | if (unlikely(!mm && !mem)) |
8697d331 | 1389 | mm = &init_mm; |
accf163e | 1390 | |
c05555b5 KH |
1391 | if (page_is_file_cache(page)) |
1392 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 1393 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 | 1394 | |
83aae4c7 DN |
1395 | /* shmem */ |
1396 | if (PageSwapCache(page)) { | |
1397 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
1398 | if (!ret) | |
1399 | __mem_cgroup_commit_charge_swapin(page, mem, | |
1400 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
1401 | } else | |
1402 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
1403 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
b5a84319 | 1404 | |
b5a84319 | 1405 | return ret; |
e8589cc1 KH |
1406 | } |
1407 | ||
54595fe2 KH |
1408 | /* |
1409 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
1410 | * And when try_charge() successfully returns, one refcnt to memcg without | |
1411 | * struct page_cgroup is aquired. This refcnt will be cumsumed by | |
1412 | * "commit()" or removed by "cancel()" | |
1413 | */ | |
8c7c6e34 KH |
1414 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
1415 | struct page *page, | |
1416 | gfp_t mask, struct mem_cgroup **ptr) | |
1417 | { | |
1418 | struct mem_cgroup *mem; | |
54595fe2 | 1419 | int ret; |
8c7c6e34 | 1420 | |
f8d66542 | 1421 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
1422 | return 0; |
1423 | ||
1424 | if (!do_swap_account) | |
1425 | goto charge_cur_mm; | |
8c7c6e34 KH |
1426 | /* |
1427 | * A racing thread's fault, or swapoff, may have already updated | |
1428 | * the pte, and even removed page from swap cache: return success | |
1429 | * to go on to do_swap_page()'s pte_same() test, which should fail. | |
1430 | */ | |
1431 | if (!PageSwapCache(page)) | |
1432 | return 0; | |
b5a84319 | 1433 | mem = try_get_mem_cgroup_from_swapcache(page); |
54595fe2 KH |
1434 | if (!mem) |
1435 | goto charge_cur_mm; | |
8c7c6e34 | 1436 | *ptr = mem; |
54595fe2 KH |
1437 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true); |
1438 | /* drop extra refcnt from tryget */ | |
1439 | css_put(&mem->css); | |
1440 | return ret; | |
8c7c6e34 KH |
1441 | charge_cur_mm: |
1442 | if (unlikely(!mm)) | |
1443 | mm = &init_mm; | |
1444 | return __mem_cgroup_try_charge(mm, mask, ptr, true); | |
1445 | } | |
1446 | ||
83aae4c7 DN |
1447 | static void |
1448 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1449 | enum charge_type ctype) | |
7a81b88c KH |
1450 | { |
1451 | struct page_cgroup *pc; | |
1452 | ||
f8d66542 | 1453 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1454 | return; |
1455 | if (!ptr) | |
1456 | return; | |
88703267 | 1457 | cgroup_exclude_rmdir(&ptr->css); |
7a81b88c | 1458 | pc = lookup_page_cgroup(page); |
544122e5 | 1459 | mem_cgroup_lru_del_before_commit_swapcache(page); |
83aae4c7 | 1460 | __mem_cgroup_commit_charge(ptr, pc, ctype); |
544122e5 | 1461 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
1462 | /* |
1463 | * Now swap is on-memory. This means this page may be | |
1464 | * counted both as mem and swap....double count. | |
03f3c433 KH |
1465 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
1466 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
1467 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 1468 | */ |
03f3c433 | 1469 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 1470 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 1471 | unsigned short id; |
8c7c6e34 | 1472 | struct mem_cgroup *memcg; |
a3b2d692 KH |
1473 | |
1474 | id = swap_cgroup_record(ent, 0); | |
1475 | rcu_read_lock(); | |
1476 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 1477 | if (memcg) { |
a3b2d692 KH |
1478 | /* |
1479 | * This recorded memcg can be obsolete one. So, avoid | |
1480 | * calling css_tryget | |
1481 | */ | |
8c7c6e34 KH |
1482 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
1483 | mem_cgroup_put(memcg); | |
1484 | } | |
a3b2d692 | 1485 | rcu_read_unlock(); |
8c7c6e34 | 1486 | } |
88703267 KH |
1487 | /* |
1488 | * At swapin, we may charge account against cgroup which has no tasks. | |
1489 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
1490 | * In that case, we need to call pre_destroy() again. check it here. | |
1491 | */ | |
1492 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
1493 | } |
1494 | ||
83aae4c7 DN |
1495 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
1496 | { | |
1497 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
1498 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
1499 | } | |
1500 | ||
7a81b88c KH |
1501 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
1502 | { | |
f8d66542 | 1503 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1504 | return; |
1505 | if (!mem) | |
1506 | return; | |
1507 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
1508 | if (do_swap_account) |
1509 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
7a81b88c KH |
1510 | css_put(&mem->css); |
1511 | } | |
1512 | ||
1513 | ||
8a9f3ccd | 1514 | /* |
69029cd5 | 1515 | * uncharge if !page_mapped(page) |
8a9f3ccd | 1516 | */ |
8c7c6e34 | 1517 | static struct mem_cgroup * |
69029cd5 | 1518 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 1519 | { |
8289546e | 1520 | struct page_cgroup *pc; |
8c7c6e34 | 1521 | struct mem_cgroup *mem = NULL; |
072c56c1 | 1522 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 1523 | |
f8d66542 | 1524 | if (mem_cgroup_disabled()) |
8c7c6e34 | 1525 | return NULL; |
4077960e | 1526 | |
d13d1443 | 1527 | if (PageSwapCache(page)) |
8c7c6e34 | 1528 | return NULL; |
d13d1443 | 1529 | |
8697d331 | 1530 | /* |
3c541e14 | 1531 | * Check if our page_cgroup is valid |
8697d331 | 1532 | */ |
52d4b9ac KH |
1533 | pc = lookup_page_cgroup(page); |
1534 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 1535 | return NULL; |
b9c565d5 | 1536 | |
52d4b9ac | 1537 | lock_page_cgroup(pc); |
d13d1443 | 1538 | |
8c7c6e34 KH |
1539 | mem = pc->mem_cgroup; |
1540 | ||
d13d1443 KH |
1541 | if (!PageCgroupUsed(pc)) |
1542 | goto unlock_out; | |
1543 | ||
1544 | switch (ctype) { | |
1545 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 1546 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
d13d1443 KH |
1547 | if (page_mapped(page)) |
1548 | goto unlock_out; | |
1549 | break; | |
1550 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
1551 | if (!PageAnon(page)) { /* Shared memory */ | |
1552 | if (page->mapping && !page_is_file_cache(page)) | |
1553 | goto unlock_out; | |
1554 | } else if (page_mapped(page)) /* Anon */ | |
1555 | goto unlock_out; | |
1556 | break; | |
1557 | default: | |
1558 | break; | |
52d4b9ac | 1559 | } |
d13d1443 | 1560 | |
8c7c6e34 KH |
1561 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1562 | if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)) | |
1563 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
08e552c6 | 1564 | mem_cgroup_charge_statistics(mem, pc, false); |
04046e1a | 1565 | |
52d4b9ac | 1566 | ClearPageCgroupUsed(pc); |
544122e5 KH |
1567 | /* |
1568 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
1569 | * freed from LRU. This is safe because uncharged page is expected not | |
1570 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
1571 | * special functions. | |
1572 | */ | |
b9c565d5 | 1573 | |
69029cd5 | 1574 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 1575 | unlock_page_cgroup(pc); |
fb59e9f1 | 1576 | |
a7fe942e KH |
1577 | /* at swapout, this memcg will be accessed to record to swap */ |
1578 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
1579 | css_put(&mem->css); | |
6d12e2d8 | 1580 | |
8c7c6e34 | 1581 | return mem; |
d13d1443 KH |
1582 | |
1583 | unlock_out: | |
1584 | unlock_page_cgroup(pc); | |
8c7c6e34 | 1585 | return NULL; |
3c541e14 BS |
1586 | } |
1587 | ||
69029cd5 KH |
1588 | void mem_cgroup_uncharge_page(struct page *page) |
1589 | { | |
52d4b9ac KH |
1590 | /* early check. */ |
1591 | if (page_mapped(page)) | |
1592 | return; | |
1593 | if (page->mapping && !PageAnon(page)) | |
1594 | return; | |
69029cd5 KH |
1595 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
1596 | } | |
1597 | ||
1598 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
1599 | { | |
1600 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 1601 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
1602 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
1603 | } | |
1604 | ||
e767e056 | 1605 | #ifdef CONFIG_SWAP |
8c7c6e34 | 1606 | /* |
e767e056 | 1607 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
1608 | * memcg information is recorded to swap_cgroup of "ent" |
1609 | */ | |
8a9478ca KH |
1610 | void |
1611 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
1612 | { |
1613 | struct mem_cgroup *memcg; | |
8a9478ca KH |
1614 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
1615 | ||
1616 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
1617 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
1618 | ||
1619 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 1620 | |
8c7c6e34 | 1621 | /* record memcg information */ |
8a9478ca | 1622 | if (do_swap_account && swapout && memcg) { |
a3b2d692 | 1623 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 KH |
1624 | mem_cgroup_get(memcg); |
1625 | } | |
8a9478ca | 1626 | if (swapout && memcg) |
a7fe942e | 1627 | css_put(&memcg->css); |
8c7c6e34 | 1628 | } |
e767e056 | 1629 | #endif |
8c7c6e34 KH |
1630 | |
1631 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1632 | /* | |
1633 | * called from swap_entry_free(). remove record in swap_cgroup and | |
1634 | * uncharge "memsw" account. | |
1635 | */ | |
1636 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 1637 | { |
8c7c6e34 | 1638 | struct mem_cgroup *memcg; |
a3b2d692 | 1639 | unsigned short id; |
8c7c6e34 KH |
1640 | |
1641 | if (!do_swap_account) | |
1642 | return; | |
1643 | ||
a3b2d692 KH |
1644 | id = swap_cgroup_record(ent, 0); |
1645 | rcu_read_lock(); | |
1646 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 1647 | if (memcg) { |
a3b2d692 KH |
1648 | /* |
1649 | * We uncharge this because swap is freed. | |
1650 | * This memcg can be obsolete one. We avoid calling css_tryget | |
1651 | */ | |
8c7c6e34 KH |
1652 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
1653 | mem_cgroup_put(memcg); | |
1654 | } | |
a3b2d692 | 1655 | rcu_read_unlock(); |
d13d1443 | 1656 | } |
8c7c6e34 | 1657 | #endif |
d13d1443 | 1658 | |
ae41be37 | 1659 | /* |
01b1ae63 KH |
1660 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
1661 | * page belongs to. | |
ae41be37 | 1662 | */ |
01b1ae63 | 1663 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
1664 | { |
1665 | struct page_cgroup *pc; | |
e8589cc1 | 1666 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 1667 | int ret = 0; |
8869b8f6 | 1668 | |
f8d66542 | 1669 | if (mem_cgroup_disabled()) |
4077960e BS |
1670 | return 0; |
1671 | ||
52d4b9ac KH |
1672 | pc = lookup_page_cgroup(page); |
1673 | lock_page_cgroup(pc); | |
1674 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
1675 | mem = pc->mem_cgroup; |
1676 | css_get(&mem->css); | |
e8589cc1 | 1677 | } |
52d4b9ac | 1678 | unlock_page_cgroup(pc); |
01b1ae63 | 1679 | |
e8589cc1 | 1680 | if (mem) { |
3bb4edf2 | 1681 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); |
e8589cc1 KH |
1682 | css_put(&mem->css); |
1683 | } | |
01b1ae63 | 1684 | *ptr = mem; |
e8589cc1 | 1685 | return ret; |
ae41be37 | 1686 | } |
8869b8f6 | 1687 | |
69029cd5 | 1688 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
1689 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
1690 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 1691 | { |
01b1ae63 KH |
1692 | struct page *target, *unused; |
1693 | struct page_cgroup *pc; | |
1694 | enum charge_type ctype; | |
1695 | ||
1696 | if (!mem) | |
1697 | return; | |
88703267 | 1698 | cgroup_exclude_rmdir(&mem->css); |
01b1ae63 KH |
1699 | /* at migration success, oldpage->mapping is NULL. */ |
1700 | if (oldpage->mapping) { | |
1701 | target = oldpage; | |
1702 | unused = NULL; | |
1703 | } else { | |
1704 | target = newpage; | |
1705 | unused = oldpage; | |
1706 | } | |
1707 | ||
1708 | if (PageAnon(target)) | |
1709 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
1710 | else if (page_is_file_cache(target)) | |
1711 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
1712 | else | |
1713 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
1714 | ||
1715 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 1716 | if (unused) |
01b1ae63 KH |
1717 | __mem_cgroup_uncharge_common(unused, ctype); |
1718 | ||
1719 | pc = lookup_page_cgroup(target); | |
69029cd5 | 1720 | /* |
01b1ae63 KH |
1721 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
1722 | * So, double-counting is effectively avoided. | |
1723 | */ | |
1724 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
1725 | ||
1726 | /* | |
1727 | * Both of oldpage and newpage are still under lock_page(). | |
1728 | * Then, we don't have to care about race in radix-tree. | |
1729 | * But we have to be careful that this page is unmapped or not. | |
1730 | * | |
1731 | * There is a case for !page_mapped(). At the start of | |
1732 | * migration, oldpage was mapped. But now, it's zapped. | |
1733 | * But we know *target* page is not freed/reused under us. | |
1734 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 1735 | */ |
01b1ae63 KH |
1736 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
1737 | mem_cgroup_uncharge_page(target); | |
88703267 KH |
1738 | /* |
1739 | * At migration, we may charge account against cgroup which has no tasks | |
1740 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
1741 | * In that case, we need to call pre_destroy() again. check it here. | |
1742 | */ | |
1743 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 1744 | } |
78fb7466 | 1745 | |
c9b0ed51 | 1746 | /* |
ae3abae6 DN |
1747 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
1748 | * Calling hierarchical_reclaim is not enough because we should update | |
1749 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
1750 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
1751 | * not from the memcg which this page would be charged to. | |
1752 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 1753 | */ |
ae3abae6 | 1754 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
1755 | struct mm_struct *mm, |
1756 | gfp_t gfp_mask) | |
c9b0ed51 | 1757 | { |
b5a84319 | 1758 | struct mem_cgroup *mem = NULL; |
ae3abae6 | 1759 | int ret; |
c9b0ed51 | 1760 | |
f8d66542 | 1761 | if (mem_cgroup_disabled()) |
cede86ac | 1762 | return 0; |
c9b0ed51 | 1763 | |
ae3abae6 DN |
1764 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
1765 | if (!ret) | |
1766 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 1767 | |
ae3abae6 | 1768 | return ret; |
c9b0ed51 KH |
1769 | } |
1770 | ||
8c7c6e34 KH |
1771 | static DEFINE_MUTEX(set_limit_mutex); |
1772 | ||
d38d2a75 | 1773 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 1774 | unsigned long long val) |
628f4235 | 1775 | { |
81d39c20 | 1776 | int retry_count; |
628f4235 | 1777 | int progress; |
8c7c6e34 | 1778 | u64 memswlimit; |
628f4235 | 1779 | int ret = 0; |
81d39c20 KH |
1780 | int children = mem_cgroup_count_children(memcg); |
1781 | u64 curusage, oldusage; | |
1782 | ||
1783 | /* | |
1784 | * For keeping hierarchical_reclaim simple, how long we should retry | |
1785 | * is depends on callers. We set our retry-count to be function | |
1786 | * of # of children which we should visit in this loop. | |
1787 | */ | |
1788 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
1789 | ||
1790 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 1791 | |
8c7c6e34 | 1792 | while (retry_count) { |
628f4235 KH |
1793 | if (signal_pending(current)) { |
1794 | ret = -EINTR; | |
1795 | break; | |
1796 | } | |
8c7c6e34 KH |
1797 | /* |
1798 | * Rather than hide all in some function, I do this in | |
1799 | * open coded manner. You see what this really does. | |
1800 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1801 | */ | |
1802 | mutex_lock(&set_limit_mutex); | |
1803 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1804 | if (memswlimit < val) { | |
1805 | ret = -EINVAL; | |
1806 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
1807 | break; |
1808 | } | |
8c7c6e34 | 1809 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
1810 | if (!ret) { |
1811 | if (memswlimit == val) | |
1812 | memcg->memsw_is_minimum = true; | |
1813 | else | |
1814 | memcg->memsw_is_minimum = false; | |
1815 | } | |
8c7c6e34 KH |
1816 | mutex_unlock(&set_limit_mutex); |
1817 | ||
1818 | if (!ret) | |
1819 | break; | |
1820 | ||
42e9abb6 | 1821 | progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, |
81d39c20 KH |
1822 | false, true); |
1823 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
1824 | /* Usage is reduced ? */ | |
1825 | if (curusage >= oldusage) | |
1826 | retry_count--; | |
1827 | else | |
1828 | oldusage = curusage; | |
8c7c6e34 | 1829 | } |
14797e23 | 1830 | |
8c7c6e34 KH |
1831 | return ret; |
1832 | } | |
1833 | ||
338c8431 LZ |
1834 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
1835 | unsigned long long val) | |
8c7c6e34 | 1836 | { |
81d39c20 | 1837 | int retry_count; |
8c7c6e34 | 1838 | u64 memlimit, oldusage, curusage; |
81d39c20 KH |
1839 | int children = mem_cgroup_count_children(memcg); |
1840 | int ret = -EBUSY; | |
8c7c6e34 | 1841 | |
81d39c20 KH |
1842 | /* see mem_cgroup_resize_res_limit */ |
1843 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
1844 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
1845 | while (retry_count) { |
1846 | if (signal_pending(current)) { | |
1847 | ret = -EINTR; | |
1848 | break; | |
1849 | } | |
1850 | /* | |
1851 | * Rather than hide all in some function, I do this in | |
1852 | * open coded manner. You see what this really does. | |
1853 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1854 | */ | |
1855 | mutex_lock(&set_limit_mutex); | |
1856 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1857 | if (memlimit > val) { | |
1858 | ret = -EINVAL; | |
1859 | mutex_unlock(&set_limit_mutex); | |
1860 | break; | |
1861 | } | |
1862 | ret = res_counter_set_limit(&memcg->memsw, val); | |
22a668d7 KH |
1863 | if (!ret) { |
1864 | if (memlimit == val) | |
1865 | memcg->memsw_is_minimum = true; | |
1866 | else | |
1867 | memcg->memsw_is_minimum = false; | |
1868 | } | |
8c7c6e34 KH |
1869 | mutex_unlock(&set_limit_mutex); |
1870 | ||
1871 | if (!ret) | |
1872 | break; | |
1873 | ||
81d39c20 | 1874 | mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true); |
8c7c6e34 | 1875 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 1876 | /* Usage is reduced ? */ |
8c7c6e34 | 1877 | if (curusage >= oldusage) |
628f4235 | 1878 | retry_count--; |
81d39c20 KH |
1879 | else |
1880 | oldusage = curusage; | |
628f4235 KH |
1881 | } |
1882 | return ret; | |
1883 | } | |
1884 | ||
cc847582 KH |
1885 | /* |
1886 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
1887 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
1888 | */ | |
f817ed48 | 1889 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 1890 | int node, int zid, enum lru_list lru) |
cc847582 | 1891 | { |
08e552c6 KH |
1892 | struct zone *zone; |
1893 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 1894 | struct page_cgroup *pc, *busy; |
08e552c6 | 1895 | unsigned long flags, loop; |
072c56c1 | 1896 | struct list_head *list; |
f817ed48 | 1897 | int ret = 0; |
072c56c1 | 1898 | |
08e552c6 KH |
1899 | zone = &NODE_DATA(node)->node_zones[zid]; |
1900 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 1901 | list = &mz->lists[lru]; |
cc847582 | 1902 | |
f817ed48 KH |
1903 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
1904 | /* give some margin against EBUSY etc...*/ | |
1905 | loop += 256; | |
1906 | busy = NULL; | |
1907 | while (loop--) { | |
1908 | ret = 0; | |
08e552c6 | 1909 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 1910 | if (list_empty(list)) { |
08e552c6 | 1911 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 1912 | break; |
f817ed48 KH |
1913 | } |
1914 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
1915 | if (busy == pc) { | |
1916 | list_move(&pc->lru, list); | |
1917 | busy = 0; | |
08e552c6 | 1918 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
1919 | continue; |
1920 | } | |
08e552c6 | 1921 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 1922 | |
2c26fdd7 | 1923 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 1924 | if (ret == -ENOMEM) |
52d4b9ac | 1925 | break; |
f817ed48 KH |
1926 | |
1927 | if (ret == -EBUSY || ret == -EINVAL) { | |
1928 | /* found lock contention or "pc" is obsolete. */ | |
1929 | busy = pc; | |
1930 | cond_resched(); | |
1931 | } else | |
1932 | busy = NULL; | |
cc847582 | 1933 | } |
08e552c6 | 1934 | |
f817ed48 KH |
1935 | if (!ret && !list_empty(list)) |
1936 | return -EBUSY; | |
1937 | return ret; | |
cc847582 KH |
1938 | } |
1939 | ||
1940 | /* | |
1941 | * make mem_cgroup's charge to be 0 if there is no task. | |
1942 | * This enables deleting this mem_cgroup. | |
1943 | */ | |
c1e862c1 | 1944 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 1945 | { |
f817ed48 KH |
1946 | int ret; |
1947 | int node, zid, shrink; | |
1948 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 1949 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 1950 | |
cc847582 | 1951 | css_get(&mem->css); |
f817ed48 KH |
1952 | |
1953 | shrink = 0; | |
c1e862c1 KH |
1954 | /* should free all ? */ |
1955 | if (free_all) | |
1956 | goto try_to_free; | |
f817ed48 | 1957 | move_account: |
1ecaab2b | 1958 | while (mem->res.usage > 0) { |
f817ed48 | 1959 | ret = -EBUSY; |
c1e862c1 KH |
1960 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
1961 | goto out; | |
1962 | ret = -EINTR; | |
1963 | if (signal_pending(current)) | |
cc847582 | 1964 | goto out; |
52d4b9ac KH |
1965 | /* This is for making all *used* pages to be on LRU. */ |
1966 | lru_add_drain_all(); | |
f817ed48 | 1967 | ret = 0; |
299b4eaa | 1968 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 1969 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 1970 | enum lru_list l; |
f817ed48 KH |
1971 | for_each_lru(l) { |
1972 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 1973 | node, zid, l); |
f817ed48 KH |
1974 | if (ret) |
1975 | break; | |
1976 | } | |
1ecaab2b | 1977 | } |
f817ed48 KH |
1978 | if (ret) |
1979 | break; | |
1980 | } | |
1981 | /* it seems parent cgroup doesn't have enough mem */ | |
1982 | if (ret == -ENOMEM) | |
1983 | goto try_to_free; | |
52d4b9ac | 1984 | cond_resched(); |
cc847582 KH |
1985 | } |
1986 | ret = 0; | |
1987 | out: | |
1988 | css_put(&mem->css); | |
1989 | return ret; | |
f817ed48 KH |
1990 | |
1991 | try_to_free: | |
c1e862c1 KH |
1992 | /* returns EBUSY if there is a task or if we come here twice. */ |
1993 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
1994 | ret = -EBUSY; |
1995 | goto out; | |
1996 | } | |
c1e862c1 KH |
1997 | /* we call try-to-free pages for make this cgroup empty */ |
1998 | lru_add_drain_all(); | |
f817ed48 KH |
1999 | /* try to free all pages in this cgroup */ |
2000 | shrink = 1; | |
2001 | while (nr_retries && mem->res.usage > 0) { | |
2002 | int progress; | |
c1e862c1 KH |
2003 | |
2004 | if (signal_pending(current)) { | |
2005 | ret = -EINTR; | |
2006 | goto out; | |
2007 | } | |
a7885eb8 KM |
2008 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
2009 | false, get_swappiness(mem)); | |
c1e862c1 | 2010 | if (!progress) { |
f817ed48 | 2011 | nr_retries--; |
c1e862c1 | 2012 | /* maybe some writeback is necessary */ |
8aa7e847 | 2013 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2014 | } |
f817ed48 KH |
2015 | |
2016 | } | |
08e552c6 | 2017 | lru_add_drain(); |
f817ed48 KH |
2018 | /* try move_account...there may be some *locked* pages. */ |
2019 | if (mem->res.usage) | |
2020 | goto move_account; | |
2021 | ret = 0; | |
2022 | goto out; | |
cc847582 KH |
2023 | } |
2024 | ||
c1e862c1 KH |
2025 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
2026 | { | |
2027 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
2028 | } | |
2029 | ||
2030 | ||
18f59ea7 BS |
2031 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
2032 | { | |
2033 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
2034 | } | |
2035 | ||
2036 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
2037 | u64 val) | |
2038 | { | |
2039 | int retval = 0; | |
2040 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
2041 | struct cgroup *parent = cont->parent; | |
2042 | struct mem_cgroup *parent_mem = NULL; | |
2043 | ||
2044 | if (parent) | |
2045 | parent_mem = mem_cgroup_from_cont(parent); | |
2046 | ||
2047 | cgroup_lock(); | |
2048 | /* | |
2049 | * If parent's use_hiearchy is set, we can't make any modifications | |
2050 | * in the child subtrees. If it is unset, then the change can | |
2051 | * occur, provided the current cgroup has no children. | |
2052 | * | |
2053 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2054 | * set if there are no children. | |
2055 | */ | |
2056 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
2057 | (val == 1 || val == 0)) { | |
2058 | if (list_empty(&cont->children)) | |
2059 | mem->use_hierarchy = val; | |
2060 | else | |
2061 | retval = -EBUSY; | |
2062 | } else | |
2063 | retval = -EINVAL; | |
2064 | cgroup_unlock(); | |
2065 | ||
2066 | return retval; | |
2067 | } | |
2068 | ||
2c3daa72 | 2069 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 2070 | { |
8c7c6e34 KH |
2071 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
2072 | u64 val = 0; | |
2073 | int type, name; | |
2074 | ||
2075 | type = MEMFILE_TYPE(cft->private); | |
2076 | name = MEMFILE_ATTR(cft->private); | |
2077 | switch (type) { | |
2078 | case _MEM: | |
2079 | val = res_counter_read_u64(&mem->res, name); | |
2080 | break; | |
2081 | case _MEMSWAP: | |
338c8431 | 2082 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
2083 | break; |
2084 | default: | |
2085 | BUG(); | |
2086 | break; | |
2087 | } | |
2088 | return val; | |
8cdea7c0 | 2089 | } |
628f4235 KH |
2090 | /* |
2091 | * The user of this function is... | |
2092 | * RES_LIMIT. | |
2093 | */ | |
856c13aa PM |
2094 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
2095 | const char *buffer) | |
8cdea7c0 | 2096 | { |
628f4235 | 2097 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 2098 | int type, name; |
628f4235 KH |
2099 | unsigned long long val; |
2100 | int ret; | |
2101 | ||
8c7c6e34 KH |
2102 | type = MEMFILE_TYPE(cft->private); |
2103 | name = MEMFILE_ATTR(cft->private); | |
2104 | switch (name) { | |
628f4235 | 2105 | case RES_LIMIT: |
4b3bde4c BS |
2106 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2107 | ret = -EINVAL; | |
2108 | break; | |
2109 | } | |
628f4235 KH |
2110 | /* This function does all necessary parse...reuse it */ |
2111 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
2112 | if (ret) |
2113 | break; | |
2114 | if (type == _MEM) | |
628f4235 | 2115 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
2116 | else |
2117 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 KH |
2118 | break; |
2119 | default: | |
2120 | ret = -EINVAL; /* should be BUG() ? */ | |
2121 | break; | |
2122 | } | |
2123 | return ret; | |
8cdea7c0 BS |
2124 | } |
2125 | ||
fee7b548 KH |
2126 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
2127 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
2128 | { | |
2129 | struct cgroup *cgroup; | |
2130 | unsigned long long min_limit, min_memsw_limit, tmp; | |
2131 | ||
2132 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2133 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2134 | cgroup = memcg->css.cgroup; | |
2135 | if (!memcg->use_hierarchy) | |
2136 | goto out; | |
2137 | ||
2138 | while (cgroup->parent) { | |
2139 | cgroup = cgroup->parent; | |
2140 | memcg = mem_cgroup_from_cont(cgroup); | |
2141 | if (!memcg->use_hierarchy) | |
2142 | break; | |
2143 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2144 | min_limit = min(min_limit, tmp); | |
2145 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2146 | min_memsw_limit = min(min_memsw_limit, tmp); | |
2147 | } | |
2148 | out: | |
2149 | *mem_limit = min_limit; | |
2150 | *memsw_limit = min_memsw_limit; | |
2151 | return; | |
2152 | } | |
2153 | ||
29f2a4da | 2154 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
2155 | { |
2156 | struct mem_cgroup *mem; | |
8c7c6e34 | 2157 | int type, name; |
c84872e1 PE |
2158 | |
2159 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
2160 | type = MEMFILE_TYPE(event); |
2161 | name = MEMFILE_ATTR(event); | |
2162 | switch (name) { | |
29f2a4da | 2163 | case RES_MAX_USAGE: |
8c7c6e34 KH |
2164 | if (type == _MEM) |
2165 | res_counter_reset_max(&mem->res); | |
2166 | else | |
2167 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
2168 | break; |
2169 | case RES_FAILCNT: | |
8c7c6e34 KH |
2170 | if (type == _MEM) |
2171 | res_counter_reset_failcnt(&mem->res); | |
2172 | else | |
2173 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
2174 | break; |
2175 | } | |
85cc59db | 2176 | return 0; |
c84872e1 PE |
2177 | } |
2178 | ||
14067bb3 KH |
2179 | |
2180 | /* For read statistics */ | |
2181 | enum { | |
2182 | MCS_CACHE, | |
2183 | MCS_RSS, | |
d69b042f | 2184 | MCS_MAPPED_FILE, |
14067bb3 KH |
2185 | MCS_PGPGIN, |
2186 | MCS_PGPGOUT, | |
2187 | MCS_INACTIVE_ANON, | |
2188 | MCS_ACTIVE_ANON, | |
2189 | MCS_INACTIVE_FILE, | |
2190 | MCS_ACTIVE_FILE, | |
2191 | MCS_UNEVICTABLE, | |
2192 | NR_MCS_STAT, | |
2193 | }; | |
2194 | ||
2195 | struct mcs_total_stat { | |
2196 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
2197 | }; |
2198 | ||
14067bb3 KH |
2199 | struct { |
2200 | char *local_name; | |
2201 | char *total_name; | |
2202 | } memcg_stat_strings[NR_MCS_STAT] = { | |
2203 | {"cache", "total_cache"}, | |
2204 | {"rss", "total_rss"}, | |
d69b042f | 2205 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
2206 | {"pgpgin", "total_pgpgin"}, |
2207 | {"pgpgout", "total_pgpgout"}, | |
2208 | {"inactive_anon", "total_inactive_anon"}, | |
2209 | {"active_anon", "total_active_anon"}, | |
2210 | {"inactive_file", "total_inactive_file"}, | |
2211 | {"active_file", "total_active_file"}, | |
2212 | {"unevictable", "total_unevictable"} | |
2213 | }; | |
2214 | ||
2215 | ||
2216 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
2217 | { | |
2218 | struct mcs_total_stat *s = data; | |
2219 | s64 val; | |
2220 | ||
2221 | /* per cpu stat */ | |
2222 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE); | |
2223 | s->stat[MCS_CACHE] += val * PAGE_SIZE; | |
2224 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
2225 | s->stat[MCS_RSS] += val * PAGE_SIZE; | |
d69b042f BS |
2226 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_MAPPED_FILE); |
2227 | s->stat[MCS_MAPPED_FILE] += val * PAGE_SIZE; | |
14067bb3 KH |
2228 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT); |
2229 | s->stat[MCS_PGPGIN] += val; | |
2230 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT); | |
2231 | s->stat[MCS_PGPGOUT] += val; | |
2232 | ||
2233 | /* per zone stat */ | |
2234 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
2235 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
2236 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
2237 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
2238 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
2239 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
2240 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
2241 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
2242 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
2243 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
2244 | return 0; | |
2245 | } | |
2246 | ||
2247 | static void | |
2248 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
2249 | { | |
2250 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
2251 | } | |
2252 | ||
c64745cf PM |
2253 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
2254 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 2255 | { |
d2ceb9b7 | 2256 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 2257 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
2258 | int i; |
2259 | ||
14067bb3 KH |
2260 | memset(&mystat, 0, sizeof(mystat)); |
2261 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 2262 | |
14067bb3 KH |
2263 | for (i = 0; i < NR_MCS_STAT; i++) |
2264 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); | |
7b854121 | 2265 | |
14067bb3 | 2266 | /* Hierarchical information */ |
fee7b548 KH |
2267 | { |
2268 | unsigned long long limit, memsw_limit; | |
2269 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
2270 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
2271 | if (do_swap_account) | |
2272 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
2273 | } | |
7f016ee8 | 2274 | |
14067bb3 KH |
2275 | memset(&mystat, 0, sizeof(mystat)); |
2276 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
2277 | for (i = 0; i < NR_MCS_STAT; i++) | |
2278 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); | |
2279 | ||
2280 | ||
7f016ee8 | 2281 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 2282 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
2283 | |
2284 | { | |
2285 | int nid, zid; | |
2286 | struct mem_cgroup_per_zone *mz; | |
2287 | unsigned long recent_rotated[2] = {0, 0}; | |
2288 | unsigned long recent_scanned[2] = {0, 0}; | |
2289 | ||
2290 | for_each_online_node(nid) | |
2291 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
2292 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
2293 | ||
2294 | recent_rotated[0] += | |
2295 | mz->reclaim_stat.recent_rotated[0]; | |
2296 | recent_rotated[1] += | |
2297 | mz->reclaim_stat.recent_rotated[1]; | |
2298 | recent_scanned[0] += | |
2299 | mz->reclaim_stat.recent_scanned[0]; | |
2300 | recent_scanned[1] += | |
2301 | mz->reclaim_stat.recent_scanned[1]; | |
2302 | } | |
2303 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
2304 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
2305 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
2306 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
2307 | } | |
2308 | #endif | |
2309 | ||
d2ceb9b7 KH |
2310 | return 0; |
2311 | } | |
2312 | ||
a7885eb8 KM |
2313 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
2314 | { | |
2315 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2316 | ||
2317 | return get_swappiness(memcg); | |
2318 | } | |
2319 | ||
2320 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
2321 | u64 val) | |
2322 | { | |
2323 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2324 | struct mem_cgroup *parent; | |
068b38c1 | 2325 | |
a7885eb8 KM |
2326 | if (val > 100) |
2327 | return -EINVAL; | |
2328 | ||
2329 | if (cgrp->parent == NULL) | |
2330 | return -EINVAL; | |
2331 | ||
2332 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
2333 | |
2334 | cgroup_lock(); | |
2335 | ||
a7885eb8 KM |
2336 | /* If under hierarchy, only empty-root can set this value */ |
2337 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
2338 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
2339 | cgroup_unlock(); | |
a7885eb8 | 2340 | return -EINVAL; |
068b38c1 | 2341 | } |
a7885eb8 KM |
2342 | |
2343 | spin_lock(&memcg->reclaim_param_lock); | |
2344 | memcg->swappiness = val; | |
2345 | spin_unlock(&memcg->reclaim_param_lock); | |
2346 | ||
068b38c1 LZ |
2347 | cgroup_unlock(); |
2348 | ||
a7885eb8 KM |
2349 | return 0; |
2350 | } | |
2351 | ||
c1e862c1 | 2352 | |
8cdea7c0 BS |
2353 | static struct cftype mem_cgroup_files[] = { |
2354 | { | |
0eea1030 | 2355 | .name = "usage_in_bytes", |
8c7c6e34 | 2356 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 2357 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 2358 | }, |
c84872e1 PE |
2359 | { |
2360 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 2361 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 2362 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
2363 | .read_u64 = mem_cgroup_read, |
2364 | }, | |
8cdea7c0 | 2365 | { |
0eea1030 | 2366 | .name = "limit_in_bytes", |
8c7c6e34 | 2367 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 2368 | .write_string = mem_cgroup_write, |
2c3daa72 | 2369 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
2370 | }, |
2371 | { | |
2372 | .name = "failcnt", | |
8c7c6e34 | 2373 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 2374 | .trigger = mem_cgroup_reset, |
2c3daa72 | 2375 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 2376 | }, |
d2ceb9b7 KH |
2377 | { |
2378 | .name = "stat", | |
c64745cf | 2379 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 2380 | }, |
c1e862c1 KH |
2381 | { |
2382 | .name = "force_empty", | |
2383 | .trigger = mem_cgroup_force_empty_write, | |
2384 | }, | |
18f59ea7 BS |
2385 | { |
2386 | .name = "use_hierarchy", | |
2387 | .write_u64 = mem_cgroup_hierarchy_write, | |
2388 | .read_u64 = mem_cgroup_hierarchy_read, | |
2389 | }, | |
a7885eb8 KM |
2390 | { |
2391 | .name = "swappiness", | |
2392 | .read_u64 = mem_cgroup_swappiness_read, | |
2393 | .write_u64 = mem_cgroup_swappiness_write, | |
2394 | }, | |
8cdea7c0 BS |
2395 | }; |
2396 | ||
8c7c6e34 KH |
2397 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
2398 | static struct cftype memsw_cgroup_files[] = { | |
2399 | { | |
2400 | .name = "memsw.usage_in_bytes", | |
2401 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
2402 | .read_u64 = mem_cgroup_read, | |
2403 | }, | |
2404 | { | |
2405 | .name = "memsw.max_usage_in_bytes", | |
2406 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
2407 | .trigger = mem_cgroup_reset, | |
2408 | .read_u64 = mem_cgroup_read, | |
2409 | }, | |
2410 | { | |
2411 | .name = "memsw.limit_in_bytes", | |
2412 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
2413 | .write_string = mem_cgroup_write, | |
2414 | .read_u64 = mem_cgroup_read, | |
2415 | }, | |
2416 | { | |
2417 | .name = "memsw.failcnt", | |
2418 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
2419 | .trigger = mem_cgroup_reset, | |
2420 | .read_u64 = mem_cgroup_read, | |
2421 | }, | |
2422 | }; | |
2423 | ||
2424 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
2425 | { | |
2426 | if (!do_swap_account) | |
2427 | return 0; | |
2428 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
2429 | ARRAY_SIZE(memsw_cgroup_files)); | |
2430 | }; | |
2431 | #else | |
2432 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
2433 | { | |
2434 | return 0; | |
2435 | } | |
2436 | #endif | |
2437 | ||
6d12e2d8 KH |
2438 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2439 | { | |
2440 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 2441 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 2442 | enum lru_list l; |
41e3355d | 2443 | int zone, tmp = node; |
1ecaab2b KH |
2444 | /* |
2445 | * This routine is called against possible nodes. | |
2446 | * But it's BUG to call kmalloc() against offline node. | |
2447 | * | |
2448 | * TODO: this routine can waste much memory for nodes which will | |
2449 | * never be onlined. It's better to use memory hotplug callback | |
2450 | * function. | |
2451 | */ | |
41e3355d KH |
2452 | if (!node_state(node, N_NORMAL_MEMORY)) |
2453 | tmp = -1; | |
2454 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
2455 | if (!pn) |
2456 | return 1; | |
1ecaab2b | 2457 | |
6d12e2d8 KH |
2458 | mem->info.nodeinfo[node] = pn; |
2459 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
2460 | |
2461 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
2462 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
2463 | for_each_lru(l) |
2464 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 2465 | } |
6d12e2d8 KH |
2466 | return 0; |
2467 | } | |
2468 | ||
1ecaab2b KH |
2469 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2470 | { | |
2471 | kfree(mem->info.nodeinfo[node]); | |
2472 | } | |
2473 | ||
c8dad2bb JB |
2474 | static int mem_cgroup_size(void) |
2475 | { | |
2476 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
2477 | return sizeof(struct mem_cgroup) + cpustat_size; | |
2478 | } | |
2479 | ||
33327948 KH |
2480 | static struct mem_cgroup *mem_cgroup_alloc(void) |
2481 | { | |
2482 | struct mem_cgroup *mem; | |
c8dad2bb | 2483 | int size = mem_cgroup_size(); |
33327948 | 2484 | |
c8dad2bb JB |
2485 | if (size < PAGE_SIZE) |
2486 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 2487 | else |
c8dad2bb | 2488 | mem = vmalloc(size); |
33327948 KH |
2489 | |
2490 | if (mem) | |
c8dad2bb | 2491 | memset(mem, 0, size); |
33327948 KH |
2492 | return mem; |
2493 | } | |
2494 | ||
8c7c6e34 KH |
2495 | /* |
2496 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
2497 | * (scanning all at force_empty is too costly...) | |
2498 | * | |
2499 | * Instead of clearing all references at force_empty, we remember | |
2500 | * the number of reference from swap_cgroup and free mem_cgroup when | |
2501 | * it goes down to 0. | |
2502 | * | |
8c7c6e34 KH |
2503 | * Removal of cgroup itself succeeds regardless of refs from swap. |
2504 | */ | |
2505 | ||
a7ba0eef | 2506 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 2507 | { |
08e552c6 KH |
2508 | int node; |
2509 | ||
04046e1a KH |
2510 | free_css_id(&mem_cgroup_subsys, &mem->css); |
2511 | ||
08e552c6 KH |
2512 | for_each_node_state(node, N_POSSIBLE) |
2513 | free_mem_cgroup_per_zone_info(mem, node); | |
2514 | ||
c8dad2bb | 2515 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
2516 | kfree(mem); |
2517 | else | |
2518 | vfree(mem); | |
2519 | } | |
2520 | ||
8c7c6e34 KH |
2521 | static void mem_cgroup_get(struct mem_cgroup *mem) |
2522 | { | |
2523 | atomic_inc(&mem->refcnt); | |
2524 | } | |
2525 | ||
2526 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
2527 | { | |
7bcc1bb1 DN |
2528 | if (atomic_dec_and_test(&mem->refcnt)) { |
2529 | struct mem_cgroup *parent = parent_mem_cgroup(mem); | |
a7ba0eef | 2530 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
2531 | if (parent) |
2532 | mem_cgroup_put(parent); | |
2533 | } | |
8c7c6e34 KH |
2534 | } |
2535 | ||
7bcc1bb1 DN |
2536 | /* |
2537 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
2538 | */ | |
2539 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
2540 | { | |
2541 | if (!mem->res.parent) | |
2542 | return NULL; | |
2543 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
2544 | } | |
33327948 | 2545 | |
c077719b KH |
2546 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
2547 | static void __init enable_swap_cgroup(void) | |
2548 | { | |
f8d66542 | 2549 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
2550 | do_swap_account = 1; |
2551 | } | |
2552 | #else | |
2553 | static void __init enable_swap_cgroup(void) | |
2554 | { | |
2555 | } | |
2556 | #endif | |
2557 | ||
0eb253e2 | 2558 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
2559 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
2560 | { | |
28dbc4b6 | 2561 | struct mem_cgroup *mem, *parent; |
04046e1a | 2562 | long error = -ENOMEM; |
6d12e2d8 | 2563 | int node; |
8cdea7c0 | 2564 | |
c8dad2bb JB |
2565 | mem = mem_cgroup_alloc(); |
2566 | if (!mem) | |
04046e1a | 2567 | return ERR_PTR(error); |
78fb7466 | 2568 | |
6d12e2d8 KH |
2569 | for_each_node_state(node, N_POSSIBLE) |
2570 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
2571 | goto free_out; | |
c077719b | 2572 | /* root ? */ |
28dbc4b6 | 2573 | if (cont->parent == NULL) { |
c077719b | 2574 | enable_swap_cgroup(); |
28dbc4b6 | 2575 | parent = NULL; |
4b3bde4c | 2576 | root_mem_cgroup = mem; |
18f59ea7 | 2577 | } else { |
28dbc4b6 | 2578 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 BS |
2579 | mem->use_hierarchy = parent->use_hierarchy; |
2580 | } | |
28dbc4b6 | 2581 | |
18f59ea7 BS |
2582 | if (parent && parent->use_hierarchy) { |
2583 | res_counter_init(&mem->res, &parent->res); | |
2584 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
2585 | /* |
2586 | * We increment refcnt of the parent to ensure that we can | |
2587 | * safely access it on res_counter_charge/uncharge. | |
2588 | * This refcnt will be decremented when freeing this | |
2589 | * mem_cgroup(see mem_cgroup_put). | |
2590 | */ | |
2591 | mem_cgroup_get(parent); | |
18f59ea7 BS |
2592 | } else { |
2593 | res_counter_init(&mem->res, NULL); | |
2594 | res_counter_init(&mem->memsw, NULL); | |
2595 | } | |
04046e1a | 2596 | mem->last_scanned_child = 0; |
2733c06a | 2597 | spin_lock_init(&mem->reclaim_param_lock); |
6d61ef40 | 2598 | |
a7885eb8 KM |
2599 | if (parent) |
2600 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 2601 | atomic_set(&mem->refcnt, 1); |
8cdea7c0 | 2602 | return &mem->css; |
6d12e2d8 | 2603 | free_out: |
a7ba0eef | 2604 | __mem_cgroup_free(mem); |
4b3bde4c | 2605 | root_mem_cgroup = NULL; |
04046e1a | 2606 | return ERR_PTR(error); |
8cdea7c0 BS |
2607 | } |
2608 | ||
ec64f515 | 2609 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
2610 | struct cgroup *cont) |
2611 | { | |
2612 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
2613 | |
2614 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
2615 | } |
2616 | ||
8cdea7c0 BS |
2617 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
2618 | struct cgroup *cont) | |
2619 | { | |
c268e994 | 2620 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 2621 | |
c268e994 | 2622 | mem_cgroup_put(mem); |
8cdea7c0 BS |
2623 | } |
2624 | ||
2625 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
2626 | struct cgroup *cont) | |
2627 | { | |
8c7c6e34 KH |
2628 | int ret; |
2629 | ||
2630 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
2631 | ARRAY_SIZE(mem_cgroup_files)); | |
2632 | ||
2633 | if (!ret) | |
2634 | ret = register_memsw_files(cont, ss); | |
2635 | return ret; | |
8cdea7c0 BS |
2636 | } |
2637 | ||
67e465a7 BS |
2638 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
2639 | struct cgroup *cont, | |
2640 | struct cgroup *old_cont, | |
be367d09 BB |
2641 | struct task_struct *p, |
2642 | bool threadgroup) | |
67e465a7 | 2643 | { |
7f4d454d | 2644 | mutex_lock(&memcg_tasklist); |
67e465a7 | 2645 | /* |
f9717d28 NK |
2646 | * FIXME: It's better to move charges of this process from old |
2647 | * memcg to new memcg. But it's just on TODO-List now. | |
67e465a7 | 2648 | */ |
7f4d454d | 2649 | mutex_unlock(&memcg_tasklist); |
67e465a7 BS |
2650 | } |
2651 | ||
8cdea7c0 BS |
2652 | struct cgroup_subsys mem_cgroup_subsys = { |
2653 | .name = "memory", | |
2654 | .subsys_id = mem_cgroup_subsys_id, | |
2655 | .create = mem_cgroup_create, | |
df878fb0 | 2656 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
2657 | .destroy = mem_cgroup_destroy, |
2658 | .populate = mem_cgroup_populate, | |
67e465a7 | 2659 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 2660 | .early_init = 0, |
04046e1a | 2661 | .use_id = 1, |
8cdea7c0 | 2662 | }; |
c077719b KH |
2663 | |
2664 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2665 | ||
2666 | static int __init disable_swap_account(char *s) | |
2667 | { | |
2668 | really_do_swap_account = 0; | |
2669 | return 1; | |
2670 | } | |
2671 | __setup("noswapaccount", disable_swap_account); | |
2672 | #endif |