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