]>
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> | |
8c7c6e34 | 30 | #include <linux/mutex.h> |
b6ac57d5 | 31 | #include <linux/slab.h> |
66e1707b BS |
32 | #include <linux/swap.h> |
33 | #include <linux/spinlock.h> | |
34 | #include <linux/fs.h> | |
d2ceb9b7 | 35 | #include <linux/seq_file.h> |
33327948 | 36 | #include <linux/vmalloc.h> |
b69408e8 | 37 | #include <linux/mm_inline.h> |
52d4b9ac | 38 | #include <linux/page_cgroup.h> |
08e552c6 | 39 | #include "internal.h" |
8cdea7c0 | 40 | |
8697d331 BS |
41 | #include <asm/uaccess.h> |
42 | ||
a181b0e8 | 43 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 44 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 45 | |
c077719b KH |
46 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
47 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */ | |
48 | int do_swap_account __read_mostly; | |
49 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
50 | #else | |
51 | #define do_swap_account (0) | |
52 | #endif | |
53 | ||
7f4d454d | 54 | static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */ |
c077719b | 55 | |
d52aa412 KH |
56 | /* |
57 | * Statistics for memory cgroup. | |
58 | */ | |
59 | enum mem_cgroup_stat_index { | |
60 | /* | |
61 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
62 | */ | |
63 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
64 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
55e462b0 BR |
65 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
66 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
67 | |
68 | MEM_CGROUP_STAT_NSTATS, | |
69 | }; | |
70 | ||
71 | struct mem_cgroup_stat_cpu { | |
72 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
73 | } ____cacheline_aligned_in_smp; | |
74 | ||
75 | struct mem_cgroup_stat { | |
c8dad2bb | 76 | struct mem_cgroup_stat_cpu cpustat[0]; |
d52aa412 KH |
77 | }; |
78 | ||
79 | /* | |
80 | * For accounting under irq disable, no need for increment preempt count. | |
81 | */ | |
addb9efe | 82 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
83 | enum mem_cgroup_stat_index idx, int val) |
84 | { | |
addb9efe | 85 | stat->count[idx] += val; |
d52aa412 KH |
86 | } |
87 | ||
88 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
89 | enum mem_cgroup_stat_index idx) | |
90 | { | |
91 | int cpu; | |
92 | s64 ret = 0; | |
93 | for_each_possible_cpu(cpu) | |
94 | ret += stat->cpustat[cpu].count[idx]; | |
95 | return ret; | |
96 | } | |
97 | ||
04046e1a KH |
98 | static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat) |
99 | { | |
100 | s64 ret; | |
101 | ||
102 | ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE); | |
103 | ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS); | |
104 | return ret; | |
105 | } | |
106 | ||
6d12e2d8 KH |
107 | /* |
108 | * per-zone information in memory controller. | |
109 | */ | |
6d12e2d8 | 110 | struct mem_cgroup_per_zone { |
072c56c1 KH |
111 | /* |
112 | * spin_lock to protect the per cgroup LRU | |
113 | */ | |
b69408e8 CL |
114 | struct list_head lists[NR_LRU_LISTS]; |
115 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
116 | |
117 | struct zone_reclaim_stat reclaim_stat; | |
6d12e2d8 KH |
118 | }; |
119 | /* Macro for accessing counter */ | |
120 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
121 | ||
122 | struct mem_cgroup_per_node { | |
123 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
124 | }; | |
125 | ||
126 | struct mem_cgroup_lru_info { | |
127 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
128 | }; | |
129 | ||
8cdea7c0 BS |
130 | /* |
131 | * The memory controller data structure. The memory controller controls both | |
132 | * page cache and RSS per cgroup. We would eventually like to provide | |
133 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
134 | * to help the administrator determine what knobs to tune. | |
135 | * | |
136 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
137 | * we hit the water mark. May be even add a low water mark, such that |
138 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
139 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
140 | */ |
141 | struct mem_cgroup { | |
142 | struct cgroup_subsys_state css; | |
143 | /* | |
144 | * the counter to account for memory usage | |
145 | */ | |
146 | struct res_counter res; | |
8c7c6e34 KH |
147 | /* |
148 | * the counter to account for mem+swap usage. | |
149 | */ | |
150 | struct res_counter memsw; | |
78fb7466 PE |
151 | /* |
152 | * Per cgroup active and inactive list, similar to the | |
153 | * per zone LRU lists. | |
78fb7466 | 154 | */ |
6d12e2d8 | 155 | struct mem_cgroup_lru_info info; |
072c56c1 | 156 | |
2733c06a KM |
157 | /* |
158 | protect against reclaim related member. | |
159 | */ | |
160 | spinlock_t reclaim_param_lock; | |
161 | ||
6c48a1d0 | 162 | int prev_priority; /* for recording reclaim priority */ |
6d61ef40 BS |
163 | |
164 | /* | |
165 | * While reclaiming in a hiearchy, we cache the last child we | |
04046e1a | 166 | * reclaimed from. |
6d61ef40 | 167 | */ |
04046e1a | 168 | int last_scanned_child; |
18f59ea7 BS |
169 | /* |
170 | * Should the accounting and control be hierarchical, per subtree? | |
171 | */ | |
172 | bool use_hierarchy; | |
a636b327 | 173 | unsigned long last_oom_jiffies; |
8c7c6e34 | 174 | atomic_t refcnt; |
14797e23 | 175 | |
a7885eb8 KM |
176 | unsigned int swappiness; |
177 | ||
d52aa412 | 178 | /* |
c8dad2bb | 179 | * statistics. This must be placed at the end of memcg. |
d52aa412 KH |
180 | */ |
181 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
182 | }; |
183 | ||
217bc319 KH |
184 | enum charge_type { |
185 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
186 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 187 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 188 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 189 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
c05555b5 KH |
190 | NR_CHARGE_TYPE, |
191 | }; | |
192 | ||
52d4b9ac KH |
193 | /* only for here (for easy reading.) */ |
194 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
195 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 196 | #define PCGF_LOCK (1UL << PCG_LOCK) |
c05555b5 KH |
197 | static const unsigned long |
198 | pcg_default_flags[NR_CHARGE_TYPE] = { | |
08e552c6 KH |
199 | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */ |
200 | PCGF_USED | PCGF_LOCK, /* Anon */ | |
201 | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ | |
52d4b9ac | 202 | 0, /* FORCE */ |
217bc319 KH |
203 | }; |
204 | ||
8c7c6e34 KH |
205 | /* for encoding cft->private value on file */ |
206 | #define _MEM (0) | |
207 | #define _MEMSWAP (1) | |
208 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
209 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
210 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
211 | ||
212 | static void mem_cgroup_get(struct mem_cgroup *mem); | |
213 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 214 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
8c7c6e34 | 215 | |
c05555b5 KH |
216 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
217 | struct page_cgroup *pc, | |
218 | bool charge) | |
d52aa412 KH |
219 | { |
220 | int val = (charge)? 1 : -1; | |
221 | struct mem_cgroup_stat *stat = &mem->stat; | |
addb9efe | 222 | struct mem_cgroup_stat_cpu *cpustat; |
08e552c6 | 223 | int cpu = get_cpu(); |
d52aa412 | 224 | |
08e552c6 | 225 | cpustat = &stat->cpustat[cpu]; |
c05555b5 | 226 | if (PageCgroupCache(pc)) |
addb9efe | 227 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 228 | else |
addb9efe | 229 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
230 | |
231 | if (charge) | |
addb9efe | 232 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
233 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
234 | else | |
addb9efe | 235 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 236 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
08e552c6 | 237 | put_cpu(); |
6d12e2d8 KH |
238 | } |
239 | ||
d5b69e38 | 240 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
241 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
242 | { | |
6d12e2d8 KH |
243 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
244 | } | |
245 | ||
d5b69e38 | 246 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
247 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
248 | { | |
249 | struct mem_cgroup *mem = pc->mem_cgroup; | |
250 | int nid = page_cgroup_nid(pc); | |
251 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 252 | |
54992762 KM |
253 | if (!mem) |
254 | return NULL; | |
255 | ||
6d12e2d8 KH |
256 | return mem_cgroup_zoneinfo(mem, nid, zid); |
257 | } | |
258 | ||
14067bb3 | 259 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 260 | enum lru_list idx) |
6d12e2d8 KH |
261 | { |
262 | int nid, zid; | |
263 | struct mem_cgroup_per_zone *mz; | |
264 | u64 total = 0; | |
265 | ||
266 | for_each_online_node(nid) | |
267 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
268 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
269 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
270 | } | |
271 | return total; | |
d52aa412 KH |
272 | } |
273 | ||
d5b69e38 | 274 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
275 | { |
276 | return container_of(cgroup_subsys_state(cont, | |
277 | mem_cgroup_subsys_id), struct mem_cgroup, | |
278 | css); | |
279 | } | |
280 | ||
cf475ad2 | 281 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 282 | { |
31a78f23 BS |
283 | /* |
284 | * mm_update_next_owner() may clear mm->owner to NULL | |
285 | * if it races with swapoff, page migration, etc. | |
286 | * So this can be called with p == NULL. | |
287 | */ | |
288 | if (unlikely(!p)) | |
289 | return NULL; | |
290 | ||
78fb7466 PE |
291 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
292 | struct mem_cgroup, css); | |
293 | } | |
294 | ||
54595fe2 KH |
295 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
296 | { | |
297 | struct mem_cgroup *mem = NULL; | |
298 | /* | |
299 | * Because we have no locks, mm->owner's may be being moved to other | |
300 | * cgroup. We use css_tryget() here even if this looks | |
301 | * pessimistic (rather than adding locks here). | |
302 | */ | |
303 | rcu_read_lock(); | |
304 | do { | |
305 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
306 | if (unlikely(!mem)) | |
307 | break; | |
308 | } while (!css_tryget(&mem->css)); | |
309 | rcu_read_unlock(); | |
310 | return mem; | |
311 | } | |
312 | ||
313 | static bool mem_cgroup_is_obsolete(struct mem_cgroup *mem) | |
314 | { | |
315 | if (!mem) | |
316 | return true; | |
317 | return css_is_removed(&mem->css); | |
318 | } | |
319 | ||
14067bb3 KH |
320 | |
321 | /* | |
322 | * Call callback function against all cgroup under hierarchy tree. | |
323 | */ | |
324 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
325 | int (*func)(struct mem_cgroup *, void *)) | |
326 | { | |
327 | int found, ret, nextid; | |
328 | struct cgroup_subsys_state *css; | |
329 | struct mem_cgroup *mem; | |
330 | ||
331 | if (!root->use_hierarchy) | |
332 | return (*func)(root, data); | |
333 | ||
334 | nextid = 1; | |
335 | do { | |
336 | ret = 0; | |
337 | mem = NULL; | |
338 | ||
339 | rcu_read_lock(); | |
340 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
341 | &found); | |
342 | if (css && css_tryget(css)) | |
343 | mem = container_of(css, struct mem_cgroup, css); | |
344 | rcu_read_unlock(); | |
345 | ||
346 | if (mem) { | |
347 | ret = (*func)(mem, data); | |
348 | css_put(&mem->css); | |
349 | } | |
350 | nextid = found + 1; | |
351 | } while (!ret && css); | |
352 | ||
353 | return ret; | |
354 | } | |
355 | ||
08e552c6 KH |
356 | /* |
357 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
358 | * Operations are called by routine of global LRU independently from memcg. | |
359 | * What we have to take care of here is validness of pc->mem_cgroup. | |
360 | * | |
361 | * Changes to pc->mem_cgroup happens when | |
362 | * 1. charge | |
363 | * 2. moving account | |
364 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
365 | * It is added to LRU before charge. | |
366 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
367 | * When moving account, the page is not on LRU. It's isolated. | |
368 | */ | |
4f98a2fe | 369 | |
08e552c6 KH |
370 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
371 | { | |
372 | struct page_cgroup *pc; | |
373 | struct mem_cgroup *mem; | |
374 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 375 | |
f8d66542 | 376 | if (mem_cgroup_disabled()) |
08e552c6 KH |
377 | return; |
378 | pc = lookup_page_cgroup(page); | |
379 | /* can happen while we handle swapcache. */ | |
544122e5 | 380 | if (list_empty(&pc->lru) || !pc->mem_cgroup) |
08e552c6 | 381 | return; |
544122e5 KH |
382 | /* |
383 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
384 | * removed from global LRU. | |
385 | */ | |
08e552c6 KH |
386 | mz = page_cgroup_zoneinfo(pc); |
387 | mem = pc->mem_cgroup; | |
b69408e8 | 388 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
08e552c6 KH |
389 | list_del_init(&pc->lru); |
390 | return; | |
6d12e2d8 KH |
391 | } |
392 | ||
08e552c6 | 393 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 394 | { |
08e552c6 KH |
395 | mem_cgroup_del_lru_list(page, page_lru(page)); |
396 | } | |
b69408e8 | 397 | |
08e552c6 KH |
398 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
399 | { | |
400 | struct mem_cgroup_per_zone *mz; | |
401 | struct page_cgroup *pc; | |
b69408e8 | 402 | |
f8d66542 | 403 | if (mem_cgroup_disabled()) |
08e552c6 | 404 | return; |
6d12e2d8 | 405 | |
08e552c6 | 406 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
407 | /* |
408 | * Used bit is set without atomic ops but after smp_wmb(). | |
409 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
410 | */ | |
08e552c6 KH |
411 | smp_rmb(); |
412 | /* unused page is not rotated. */ | |
413 | if (!PageCgroupUsed(pc)) | |
414 | return; | |
415 | mz = page_cgroup_zoneinfo(pc); | |
416 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
417 | } |
418 | ||
08e552c6 | 419 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 420 | { |
08e552c6 KH |
421 | struct page_cgroup *pc; |
422 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 423 | |
f8d66542 | 424 | if (mem_cgroup_disabled()) |
08e552c6 KH |
425 | return; |
426 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
427 | /* |
428 | * Used bit is set without atomic ops but after smp_wmb(). | |
429 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
430 | */ | |
08e552c6 KH |
431 | smp_rmb(); |
432 | if (!PageCgroupUsed(pc)) | |
894bc310 | 433 | return; |
b69408e8 | 434 | |
08e552c6 | 435 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 436 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
08e552c6 KH |
437 | list_add(&pc->lru, &mz->lists[lru]); |
438 | } | |
544122e5 | 439 | |
08e552c6 | 440 | /* |
544122e5 KH |
441 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
442 | * lru because the page may.be reused after it's fully uncharged (because of | |
443 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
444 | * it again. This function is only used to charge SwapCache. It's done under | |
445 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 446 | */ |
544122e5 | 447 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 448 | { |
544122e5 KH |
449 | unsigned long flags; |
450 | struct zone *zone = page_zone(page); | |
451 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
452 | ||
453 | spin_lock_irqsave(&zone->lru_lock, flags); | |
454 | /* | |
455 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
456 | * is guarded by lock_page() because the page is SwapCache. | |
457 | */ | |
458 | if (!PageCgroupUsed(pc)) | |
459 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
460 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
461 | } |
462 | ||
544122e5 KH |
463 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
464 | { | |
465 | unsigned long flags; | |
466 | struct zone *zone = page_zone(page); | |
467 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
468 | ||
469 | spin_lock_irqsave(&zone->lru_lock, flags); | |
470 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
471 | if (PageLRU(page) && list_empty(&pc->lru)) | |
472 | mem_cgroup_add_lru_list(page, page_lru(page)); | |
473 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
474 | } | |
475 | ||
476 | ||
08e552c6 KH |
477 | void mem_cgroup_move_lists(struct page *page, |
478 | enum lru_list from, enum lru_list to) | |
479 | { | |
f8d66542 | 480 | if (mem_cgroup_disabled()) |
08e552c6 KH |
481 | return; |
482 | mem_cgroup_del_lru_list(page, from); | |
483 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
484 | } |
485 | ||
4c4a2214 DR |
486 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
487 | { | |
488 | int ret; | |
489 | ||
490 | task_lock(task); | |
bd845e38 | 491 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
492 | task_unlock(task); |
493 | return ret; | |
494 | } | |
495 | ||
58ae83db KH |
496 | /* |
497 | * Calculate mapped_ratio under memory controller. This will be used in | |
498 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
499 | */ | |
500 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
501 | { | |
502 | long total, rss; | |
503 | ||
504 | /* | |
505 | * usage is recorded in bytes. But, here, we assume the number of | |
506 | * physical pages can be represented by "long" on any arch. | |
507 | */ | |
508 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
509 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
510 | return (int)((rss * 100L) / total); | |
511 | } | |
8869b8f6 | 512 | |
6c48a1d0 KH |
513 | /* |
514 | * prev_priority control...this will be used in memory reclaim path. | |
515 | */ | |
516 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
517 | { | |
2733c06a KM |
518 | int prev_priority; |
519 | ||
520 | spin_lock(&mem->reclaim_param_lock); | |
521 | prev_priority = mem->prev_priority; | |
522 | spin_unlock(&mem->reclaim_param_lock); | |
523 | ||
524 | return prev_priority; | |
6c48a1d0 KH |
525 | } |
526 | ||
527 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
528 | { | |
2733c06a | 529 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
530 | if (priority < mem->prev_priority) |
531 | mem->prev_priority = priority; | |
2733c06a | 532 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
533 | } |
534 | ||
535 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
536 | { | |
2733c06a | 537 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 | 538 | mem->prev_priority = priority; |
2733c06a | 539 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
540 | } |
541 | ||
c772be93 | 542 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
543 | { |
544 | unsigned long active; | |
545 | unsigned long inactive; | |
c772be93 KM |
546 | unsigned long gb; |
547 | unsigned long inactive_ratio; | |
14797e23 | 548 | |
14067bb3 KH |
549 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
550 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 551 | |
c772be93 KM |
552 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
553 | if (gb) | |
554 | inactive_ratio = int_sqrt(10 * gb); | |
555 | else | |
556 | inactive_ratio = 1; | |
557 | ||
558 | if (present_pages) { | |
559 | present_pages[0] = inactive; | |
560 | present_pages[1] = active; | |
561 | } | |
562 | ||
563 | return inactive_ratio; | |
564 | } | |
565 | ||
566 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
567 | { | |
568 | unsigned long active; | |
569 | unsigned long inactive; | |
570 | unsigned long present_pages[2]; | |
571 | unsigned long inactive_ratio; | |
572 | ||
573 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
574 | ||
575 | inactive = present_pages[0]; | |
576 | active = present_pages[1]; | |
577 | ||
578 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
579 | return 1; |
580 | ||
581 | return 0; | |
582 | } | |
583 | ||
a3d8e054 KM |
584 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
585 | struct zone *zone, | |
586 | enum lru_list lru) | |
587 | { | |
588 | int nid = zone->zone_pgdat->node_id; | |
589 | int zid = zone_idx(zone); | |
590 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
591 | ||
592 | return MEM_CGROUP_ZSTAT(mz, lru); | |
593 | } | |
594 | ||
3e2f41f1 KM |
595 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
596 | struct zone *zone) | |
597 | { | |
598 | int nid = zone->zone_pgdat->node_id; | |
599 | int zid = zone_idx(zone); | |
600 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
601 | ||
602 | return &mz->reclaim_stat; | |
603 | } | |
604 | ||
605 | struct zone_reclaim_stat * | |
606 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
607 | { | |
608 | struct page_cgroup *pc; | |
609 | struct mem_cgroup_per_zone *mz; | |
610 | ||
611 | if (mem_cgroup_disabled()) | |
612 | return NULL; | |
613 | ||
614 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
615 | /* |
616 | * Used bit is set without atomic ops but after smp_wmb(). | |
617 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
618 | */ | |
619 | smp_rmb(); | |
620 | if (!PageCgroupUsed(pc)) | |
621 | return NULL; | |
622 | ||
3e2f41f1 KM |
623 | mz = page_cgroup_zoneinfo(pc); |
624 | if (!mz) | |
625 | return NULL; | |
626 | ||
627 | return &mz->reclaim_stat; | |
628 | } | |
629 | ||
66e1707b BS |
630 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
631 | struct list_head *dst, | |
632 | unsigned long *scanned, int order, | |
633 | int mode, struct zone *z, | |
634 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 635 | int active, int file) |
66e1707b BS |
636 | { |
637 | unsigned long nr_taken = 0; | |
638 | struct page *page; | |
639 | unsigned long scan; | |
640 | LIST_HEAD(pc_list); | |
641 | struct list_head *src; | |
ff7283fa | 642 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
643 | int nid = z->zone_pgdat->node_id; |
644 | int zid = zone_idx(z); | |
645 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 646 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 647 | |
cf475ad2 | 648 | BUG_ON(!mem_cont); |
1ecaab2b | 649 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 650 | src = &mz->lists[lru]; |
66e1707b | 651 | |
ff7283fa KH |
652 | scan = 0; |
653 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 654 | if (scan >= nr_to_scan) |
ff7283fa | 655 | break; |
08e552c6 KH |
656 | |
657 | page = pc->page; | |
52d4b9ac KH |
658 | if (unlikely(!PageCgroupUsed(pc))) |
659 | continue; | |
436c6541 | 660 | if (unlikely(!PageLRU(page))) |
ff7283fa | 661 | continue; |
ff7283fa | 662 | |
436c6541 | 663 | scan++; |
4f98a2fe | 664 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
665 | list_move(&page->lru, dst); |
666 | nr_taken++; | |
667 | } | |
668 | } | |
669 | ||
66e1707b BS |
670 | *scanned = scan; |
671 | return nr_taken; | |
672 | } | |
673 | ||
6d61ef40 BS |
674 | #define mem_cgroup_from_res_counter(counter, member) \ |
675 | container_of(counter, struct mem_cgroup, member) | |
676 | ||
b85a96c0 DN |
677 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
678 | { | |
679 | if (do_swap_account) { | |
680 | if (res_counter_check_under_limit(&mem->res) && | |
681 | res_counter_check_under_limit(&mem->memsw)) | |
682 | return true; | |
683 | } else | |
684 | if (res_counter_check_under_limit(&mem->res)) | |
685 | return true; | |
686 | return false; | |
687 | } | |
688 | ||
a7885eb8 KM |
689 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
690 | { | |
691 | struct cgroup *cgrp = memcg->css.cgroup; | |
692 | unsigned int swappiness; | |
693 | ||
694 | /* root ? */ | |
695 | if (cgrp->parent == NULL) | |
696 | return vm_swappiness; | |
697 | ||
698 | spin_lock(&memcg->reclaim_param_lock); | |
699 | swappiness = memcg->swappiness; | |
700 | spin_unlock(&memcg->reclaim_param_lock); | |
701 | ||
702 | return swappiness; | |
703 | } | |
704 | ||
81d39c20 KH |
705 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
706 | { | |
707 | int *val = data; | |
708 | (*val)++; | |
709 | return 0; | |
710 | } | |
711 | /* | |
712 | * This function returns the number of memcg under hierarchy tree. Returns | |
713 | * 1(self count) if no children. | |
714 | */ | |
715 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
716 | { | |
717 | int num = 0; | |
718 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
719 | return num; | |
720 | } | |
721 | ||
6d61ef40 | 722 | /* |
04046e1a KH |
723 | * Visit the first child (need not be the first child as per the ordering |
724 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
725 | * that to reclaim free pages from. | |
726 | */ | |
727 | static struct mem_cgroup * | |
728 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
729 | { | |
730 | struct mem_cgroup *ret = NULL; | |
731 | struct cgroup_subsys_state *css; | |
732 | int nextid, found; | |
733 | ||
734 | if (!root_mem->use_hierarchy) { | |
735 | css_get(&root_mem->css); | |
736 | ret = root_mem; | |
737 | } | |
738 | ||
739 | while (!ret) { | |
740 | rcu_read_lock(); | |
741 | nextid = root_mem->last_scanned_child + 1; | |
742 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
743 | &found); | |
744 | if (css && css_tryget(css)) | |
745 | ret = container_of(css, struct mem_cgroup, css); | |
746 | ||
747 | rcu_read_unlock(); | |
748 | /* Updates scanning parameter */ | |
749 | spin_lock(&root_mem->reclaim_param_lock); | |
750 | if (!css) { | |
751 | /* this means start scan from ID:1 */ | |
752 | root_mem->last_scanned_child = 0; | |
753 | } else | |
754 | root_mem->last_scanned_child = found; | |
755 | spin_unlock(&root_mem->reclaim_param_lock); | |
756 | } | |
757 | ||
758 | return ret; | |
759 | } | |
760 | ||
761 | /* | |
762 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
763 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
764 | * based on its position in the children list. | |
6d61ef40 BS |
765 | * |
766 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
767 | * |
768 | * We give up and return to the caller when we visit root_mem twice. | |
769 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
770 | * |
771 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
772 | */ |
773 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
81d39c20 | 774 | gfp_t gfp_mask, bool noswap, bool shrink) |
6d61ef40 | 775 | { |
04046e1a KH |
776 | struct mem_cgroup *victim; |
777 | int ret, total = 0; | |
778 | int loop = 0; | |
779 | ||
780 | while (loop < 2) { | |
781 | victim = mem_cgroup_select_victim(root_mem); | |
782 | if (victim == root_mem) | |
783 | loop++; | |
784 | if (!mem_cgroup_local_usage(&victim->stat)) { | |
785 | /* this cgroup's local usage == 0 */ | |
786 | css_put(&victim->css); | |
6d61ef40 BS |
787 | continue; |
788 | } | |
04046e1a KH |
789 | /* we use swappiness of local cgroup */ |
790 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap, | |
791 | get_swappiness(victim)); | |
792 | css_put(&victim->css); | |
81d39c20 KH |
793 | /* |
794 | * At shrinking usage, we can't check we should stop here or | |
795 | * reclaim more. It's depends on callers. last_scanned_child | |
796 | * will work enough for keeping fairness under tree. | |
797 | */ | |
798 | if (shrink) | |
799 | return ret; | |
04046e1a | 800 | total += ret; |
b85a96c0 | 801 | if (mem_cgroup_check_under_limit(root_mem)) |
04046e1a | 802 | return 1 + total; |
6d61ef40 | 803 | } |
04046e1a | 804 | return total; |
6d61ef40 BS |
805 | } |
806 | ||
a636b327 KH |
807 | bool mem_cgroup_oom_called(struct task_struct *task) |
808 | { | |
809 | bool ret = false; | |
810 | struct mem_cgroup *mem; | |
811 | struct mm_struct *mm; | |
812 | ||
813 | rcu_read_lock(); | |
814 | mm = task->mm; | |
815 | if (!mm) | |
816 | mm = &init_mm; | |
817 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
818 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
819 | ret = true; | |
820 | rcu_read_unlock(); | |
821 | return ret; | |
822 | } | |
f817ed48 KH |
823 | /* |
824 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
825 | * oom-killer can be invoked. | |
8a9f3ccd | 826 | */ |
f817ed48 | 827 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
8c7c6e34 KH |
828 | gfp_t gfp_mask, struct mem_cgroup **memcg, |
829 | bool oom) | |
8a9f3ccd | 830 | { |
6d61ef40 | 831 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 832 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
28dbc4b6 | 833 | struct res_counter *fail_res; |
a636b327 KH |
834 | |
835 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
836 | /* Don't account this! */ | |
837 | *memcg = NULL; | |
838 | return 0; | |
839 | } | |
840 | ||
8a9f3ccd | 841 | /* |
3be91277 HD |
842 | * We always charge the cgroup the mm_struct belongs to. |
843 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
844 | * thread group leader migrates. It's possible that mm is not |
845 | * set, if so charge the init_mm (happens for pagecache usage). | |
846 | */ | |
54595fe2 KH |
847 | mem = *memcg; |
848 | if (likely(!mem)) { | |
849 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 850 | *memcg = mem; |
e8589cc1 | 851 | } else { |
7a81b88c | 852 | css_get(&mem->css); |
e8589cc1 | 853 | } |
54595fe2 KH |
854 | if (unlikely(!mem)) |
855 | return 0; | |
856 | ||
857 | VM_BUG_ON(mem_cgroup_is_obsolete(mem)); | |
8a9f3ccd | 858 | |
8c7c6e34 KH |
859 | while (1) { |
860 | int ret; | |
861 | bool noswap = false; | |
7a81b88c | 862 | |
28dbc4b6 | 863 | ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res); |
8c7c6e34 KH |
864 | if (likely(!ret)) { |
865 | if (!do_swap_account) | |
866 | break; | |
28dbc4b6 BS |
867 | ret = res_counter_charge(&mem->memsw, PAGE_SIZE, |
868 | &fail_res); | |
8c7c6e34 KH |
869 | if (likely(!ret)) |
870 | break; | |
871 | /* mem+swap counter fails */ | |
872 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
873 | noswap = true; | |
6d61ef40 BS |
874 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
875 | memsw); | |
876 | } else | |
877 | /* mem counter fails */ | |
878 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
879 | res); | |
880 | ||
3be91277 | 881 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 882 | goto nomem; |
e1a1cd59 | 883 | |
6d61ef40 | 884 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask, |
81d39c20 | 885 | noswap, false); |
4d1c6273 DN |
886 | if (ret) |
887 | continue; | |
66e1707b BS |
888 | |
889 | /* | |
8869b8f6 HD |
890 | * try_to_free_mem_cgroup_pages() might not give us a full |
891 | * picture of reclaim. Some pages are reclaimed and might be | |
892 | * moved to swap cache or just unmapped from the cgroup. | |
893 | * Check the limit again to see if the reclaim reduced the | |
894 | * current usage of the cgroup before giving up | |
8c7c6e34 | 895 | * |
8869b8f6 | 896 | */ |
b85a96c0 DN |
897 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
898 | continue; | |
3be91277 HD |
899 | |
900 | if (!nr_retries--) { | |
a636b327 | 901 | if (oom) { |
7f4d454d | 902 | mutex_lock(&memcg_tasklist); |
88700756 | 903 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); |
7f4d454d | 904 | mutex_unlock(&memcg_tasklist); |
88700756 | 905 | mem_over_limit->last_oom_jiffies = jiffies; |
a636b327 | 906 | } |
7a81b88c | 907 | goto nomem; |
66e1707b | 908 | } |
8a9f3ccd | 909 | } |
7a81b88c KH |
910 | return 0; |
911 | nomem: | |
912 | css_put(&mem->css); | |
913 | return -ENOMEM; | |
914 | } | |
8a9f3ccd | 915 | |
b5a84319 KH |
916 | static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page) |
917 | { | |
918 | struct mem_cgroup *mem; | |
919 | swp_entry_t ent; | |
920 | ||
921 | if (!PageSwapCache(page)) | |
922 | return NULL; | |
923 | ||
924 | ent.val = page_private(page); | |
925 | mem = lookup_swap_cgroup(ent); | |
926 | if (!mem) | |
927 | return NULL; | |
928 | if (!css_tryget(&mem->css)) | |
929 | return NULL; | |
930 | return mem; | |
931 | } | |
932 | ||
7a81b88c | 933 | /* |
a5e924f5 | 934 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
935 | * USED state. If already USED, uncharge and return. |
936 | */ | |
937 | ||
938 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
939 | struct page_cgroup *pc, | |
940 | enum charge_type ctype) | |
941 | { | |
7a81b88c KH |
942 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
943 | if (!mem) | |
944 | return; | |
52d4b9ac KH |
945 | |
946 | lock_page_cgroup(pc); | |
947 | if (unlikely(PageCgroupUsed(pc))) { | |
948 | unlock_page_cgroup(pc); | |
949 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
950 | if (do_swap_account) |
951 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
52d4b9ac | 952 | css_put(&mem->css); |
7a81b88c | 953 | return; |
52d4b9ac | 954 | } |
8a9f3ccd | 955 | pc->mem_cgroup = mem; |
08e552c6 | 956 | smp_wmb(); |
c05555b5 | 957 | pc->flags = pcg_default_flags[ctype]; |
3be91277 | 958 | |
08e552c6 | 959 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 960 | |
52d4b9ac | 961 | unlock_page_cgroup(pc); |
7a81b88c | 962 | } |
66e1707b | 963 | |
f817ed48 KH |
964 | /** |
965 | * mem_cgroup_move_account - move account of the page | |
966 | * @pc: page_cgroup of the page. | |
967 | * @from: mem_cgroup which the page is moved from. | |
968 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
969 | * | |
970 | * The caller must confirm following. | |
08e552c6 | 971 | * - page is not on LRU (isolate_page() is useful.) |
f817ed48 KH |
972 | * |
973 | * returns 0 at success, | |
974 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
975 | * | |
976 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
977 | * new cgroup. It should be done by a caller. | |
978 | */ | |
979 | ||
980 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
981 | struct mem_cgroup *from, struct mem_cgroup *to) | |
982 | { | |
983 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
984 | int nid, zid; | |
985 | int ret = -EBUSY; | |
986 | ||
f817ed48 | 987 | VM_BUG_ON(from == to); |
08e552c6 | 988 | VM_BUG_ON(PageLRU(pc->page)); |
f817ed48 KH |
989 | |
990 | nid = page_cgroup_nid(pc); | |
991 | zid = page_cgroup_zid(pc); | |
992 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
993 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
994 | ||
f817ed48 KH |
995 | if (!trylock_page_cgroup(pc)) |
996 | return ret; | |
997 | ||
998 | if (!PageCgroupUsed(pc)) | |
999 | goto out; | |
1000 | ||
1001 | if (pc->mem_cgroup != from) | |
1002 | goto out; | |
1003 | ||
08e552c6 KH |
1004 | res_counter_uncharge(&from->res, PAGE_SIZE); |
1005 | mem_cgroup_charge_statistics(from, pc, false); | |
1006 | if (do_swap_account) | |
1007 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
40d58138 DN |
1008 | css_put(&from->css); |
1009 | ||
1010 | css_get(&to->css); | |
08e552c6 KH |
1011 | pc->mem_cgroup = to; |
1012 | mem_cgroup_charge_statistics(to, pc, true); | |
08e552c6 | 1013 | ret = 0; |
f817ed48 KH |
1014 | out: |
1015 | unlock_page_cgroup(pc); | |
1016 | return ret; | |
1017 | } | |
1018 | ||
1019 | /* | |
1020 | * move charges to its parent. | |
1021 | */ | |
1022 | ||
1023 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1024 | struct mem_cgroup *child, | |
1025 | gfp_t gfp_mask) | |
1026 | { | |
08e552c6 | 1027 | struct page *page = pc->page; |
f817ed48 KH |
1028 | struct cgroup *cg = child->css.cgroup; |
1029 | struct cgroup *pcg = cg->parent; | |
1030 | struct mem_cgroup *parent; | |
f817ed48 KH |
1031 | int ret; |
1032 | ||
1033 | /* Is ROOT ? */ | |
1034 | if (!pcg) | |
1035 | return -EINVAL; | |
1036 | ||
08e552c6 | 1037 | |
f817ed48 KH |
1038 | parent = mem_cgroup_from_cont(pcg); |
1039 | ||
08e552c6 | 1040 | |
f817ed48 | 1041 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); |
a636b327 | 1042 | if (ret || !parent) |
f817ed48 KH |
1043 | return ret; |
1044 | ||
40d58138 DN |
1045 | if (!get_page_unless_zero(page)) { |
1046 | ret = -EBUSY; | |
1047 | goto uncharge; | |
1048 | } | |
08e552c6 KH |
1049 | |
1050 | ret = isolate_lru_page(page); | |
1051 | ||
1052 | if (ret) | |
1053 | goto cancel; | |
f817ed48 | 1054 | |
f817ed48 | 1055 | ret = mem_cgroup_move_account(pc, child, parent); |
f817ed48 | 1056 | |
08e552c6 KH |
1057 | putback_lru_page(page); |
1058 | if (!ret) { | |
1059 | put_page(page); | |
40d58138 DN |
1060 | /* drop extra refcnt by try_charge() */ |
1061 | css_put(&parent->css); | |
08e552c6 | 1062 | return 0; |
8c7c6e34 | 1063 | } |
40d58138 | 1064 | |
08e552c6 | 1065 | cancel: |
40d58138 DN |
1066 | put_page(page); |
1067 | uncharge: | |
1068 | /* drop extra refcnt by try_charge() */ | |
1069 | css_put(&parent->css); | |
1070 | /* uncharge if move fails */ | |
08e552c6 KH |
1071 | res_counter_uncharge(&parent->res, PAGE_SIZE); |
1072 | if (do_swap_account) | |
1073 | res_counter_uncharge(&parent->memsw, PAGE_SIZE); | |
f817ed48 KH |
1074 | return ret; |
1075 | } | |
1076 | ||
7a81b88c KH |
1077 | /* |
1078 | * Charge the memory controller for page usage. | |
1079 | * Return | |
1080 | * 0 if the charge was successful | |
1081 | * < 0 if the cgroup is over its limit | |
1082 | */ | |
1083 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1084 | gfp_t gfp_mask, enum charge_type ctype, | |
1085 | struct mem_cgroup *memcg) | |
1086 | { | |
1087 | struct mem_cgroup *mem; | |
1088 | struct page_cgroup *pc; | |
1089 | int ret; | |
1090 | ||
1091 | pc = lookup_page_cgroup(page); | |
1092 | /* can happen at boot */ | |
1093 | if (unlikely(!pc)) | |
1094 | return 0; | |
1095 | prefetchw(pc); | |
1096 | ||
1097 | mem = memcg; | |
f817ed48 | 1098 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
a636b327 | 1099 | if (ret || !mem) |
7a81b88c KH |
1100 | return ret; |
1101 | ||
1102 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1103 | return 0; |
8a9f3ccd BS |
1104 | } |
1105 | ||
7a81b88c KH |
1106 | int mem_cgroup_newpage_charge(struct page *page, |
1107 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1108 | { |
f8d66542 | 1109 | if (mem_cgroup_disabled()) |
cede86ac | 1110 | return 0; |
52d4b9ac KH |
1111 | if (PageCompound(page)) |
1112 | return 0; | |
69029cd5 KH |
1113 | /* |
1114 | * If already mapped, we don't have to account. | |
1115 | * If page cache, page->mapping has address_space. | |
1116 | * But page->mapping may have out-of-use anon_vma pointer, | |
1117 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1118 | * is NULL. | |
1119 | */ | |
1120 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
1121 | return 0; | |
1122 | if (unlikely(!mm)) | |
1123 | mm = &init_mm; | |
217bc319 | 1124 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 1125 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
1126 | } |
1127 | ||
e1a1cd59 BS |
1128 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
1129 | gfp_t gfp_mask) | |
8697d331 | 1130 | { |
b5a84319 KH |
1131 | struct mem_cgroup *mem = NULL; |
1132 | int ret; | |
1133 | ||
f8d66542 | 1134 | if (mem_cgroup_disabled()) |
cede86ac | 1135 | return 0; |
52d4b9ac KH |
1136 | if (PageCompound(page)) |
1137 | return 0; | |
accf163e KH |
1138 | /* |
1139 | * Corner case handling. This is called from add_to_page_cache() | |
1140 | * in usual. But some FS (shmem) precharges this page before calling it | |
1141 | * and call add_to_page_cache() with GFP_NOWAIT. | |
1142 | * | |
1143 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
1144 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
1145 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
1146 | * And when the page is SwapCache, it should take swap information |
1147 | * into account. This is under lock_page() now. | |
accf163e KH |
1148 | */ |
1149 | if (!(gfp_mask & __GFP_WAIT)) { | |
1150 | struct page_cgroup *pc; | |
1151 | ||
52d4b9ac KH |
1152 | |
1153 | pc = lookup_page_cgroup(page); | |
1154 | if (!pc) | |
1155 | return 0; | |
1156 | lock_page_cgroup(pc); | |
1157 | if (PageCgroupUsed(pc)) { | |
1158 | unlock_page_cgroup(pc); | |
accf163e KH |
1159 | return 0; |
1160 | } | |
52d4b9ac | 1161 | unlock_page_cgroup(pc); |
accf163e KH |
1162 | } |
1163 | ||
b5a84319 KH |
1164 | if (do_swap_account && PageSwapCache(page)) { |
1165 | mem = try_get_mem_cgroup_from_swapcache(page); | |
1166 | if (mem) | |
1167 | mm = NULL; | |
1168 | else | |
1169 | mem = NULL; | |
1170 | /* SwapCache may be still linked to LRU now. */ | |
1171 | mem_cgroup_lru_del_before_commit_swapcache(page); | |
1172 | } | |
1173 | ||
1174 | if (unlikely(!mm && !mem)) | |
8697d331 | 1175 | mm = &init_mm; |
accf163e | 1176 | |
c05555b5 KH |
1177 | if (page_is_file_cache(page)) |
1178 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 1179 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 KH |
1180 | |
1181 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
1182 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
1183 | if (mem) | |
1184 | css_put(&mem->css); | |
1185 | if (PageSwapCache(page)) | |
1186 | mem_cgroup_lru_add_after_commit_swapcache(page); | |
1187 | ||
1188 | if (do_swap_account && !ret && PageSwapCache(page)) { | |
1189 | swp_entry_t ent = {.val = page_private(page)}; | |
1190 | /* avoid double counting */ | |
1191 | mem = swap_cgroup_record(ent, NULL); | |
1192 | if (mem) { | |
1193 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1194 | mem_cgroup_put(mem); | |
1195 | } | |
1196 | } | |
1197 | return ret; | |
e8589cc1 KH |
1198 | } |
1199 | ||
54595fe2 KH |
1200 | /* |
1201 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
1202 | * And when try_charge() successfully returns, one refcnt to memcg without | |
1203 | * struct page_cgroup is aquired. This refcnt will be cumsumed by | |
1204 | * "commit()" or removed by "cancel()" | |
1205 | */ | |
8c7c6e34 KH |
1206 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
1207 | struct page *page, | |
1208 | gfp_t mask, struct mem_cgroup **ptr) | |
1209 | { | |
1210 | struct mem_cgroup *mem; | |
54595fe2 | 1211 | int ret; |
8c7c6e34 | 1212 | |
f8d66542 | 1213 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
1214 | return 0; |
1215 | ||
1216 | if (!do_swap_account) | |
1217 | goto charge_cur_mm; | |
8c7c6e34 KH |
1218 | /* |
1219 | * A racing thread's fault, or swapoff, may have already updated | |
1220 | * the pte, and even removed page from swap cache: return success | |
1221 | * to go on to do_swap_page()'s pte_same() test, which should fail. | |
1222 | */ | |
1223 | if (!PageSwapCache(page)) | |
1224 | return 0; | |
b5a84319 | 1225 | mem = try_get_mem_cgroup_from_swapcache(page); |
54595fe2 KH |
1226 | if (!mem) |
1227 | goto charge_cur_mm; | |
8c7c6e34 | 1228 | *ptr = mem; |
54595fe2 KH |
1229 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true); |
1230 | /* drop extra refcnt from tryget */ | |
1231 | css_put(&mem->css); | |
1232 | return ret; | |
8c7c6e34 KH |
1233 | charge_cur_mm: |
1234 | if (unlikely(!mm)) | |
1235 | mm = &init_mm; | |
1236 | return __mem_cgroup_try_charge(mm, mask, ptr, true); | |
1237 | } | |
1238 | ||
7a81b88c KH |
1239 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
1240 | { | |
1241 | struct page_cgroup *pc; | |
1242 | ||
f8d66542 | 1243 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1244 | return; |
1245 | if (!ptr) | |
1246 | return; | |
1247 | pc = lookup_page_cgroup(page); | |
544122e5 | 1248 | mem_cgroup_lru_del_before_commit_swapcache(page); |
7a81b88c | 1249 | __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
544122e5 | 1250 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
1251 | /* |
1252 | * Now swap is on-memory. This means this page may be | |
1253 | * counted both as mem and swap....double count. | |
03f3c433 KH |
1254 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
1255 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
1256 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 1257 | */ |
03f3c433 | 1258 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 KH |
1259 | swp_entry_t ent = {.val = page_private(page)}; |
1260 | struct mem_cgroup *memcg; | |
1261 | memcg = swap_cgroup_record(ent, NULL); | |
1262 | if (memcg) { | |
8c7c6e34 KH |
1263 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
1264 | mem_cgroup_put(memcg); | |
1265 | } | |
1266 | ||
1267 | } | |
08e552c6 | 1268 | /* add this page(page_cgroup) to the LRU we want. */ |
544122e5 | 1269 | |
7a81b88c KH |
1270 | } |
1271 | ||
1272 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
1273 | { | |
f8d66542 | 1274 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1275 | return; |
1276 | if (!mem) | |
1277 | return; | |
1278 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
1279 | if (do_swap_account) |
1280 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
7a81b88c KH |
1281 | css_put(&mem->css); |
1282 | } | |
1283 | ||
1284 | ||
8a9f3ccd | 1285 | /* |
69029cd5 | 1286 | * uncharge if !page_mapped(page) |
8a9f3ccd | 1287 | */ |
8c7c6e34 | 1288 | static struct mem_cgroup * |
69029cd5 | 1289 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 1290 | { |
8289546e | 1291 | struct page_cgroup *pc; |
8c7c6e34 | 1292 | struct mem_cgroup *mem = NULL; |
072c56c1 | 1293 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 1294 | |
f8d66542 | 1295 | if (mem_cgroup_disabled()) |
8c7c6e34 | 1296 | return NULL; |
4077960e | 1297 | |
d13d1443 | 1298 | if (PageSwapCache(page)) |
8c7c6e34 | 1299 | return NULL; |
d13d1443 | 1300 | |
8697d331 | 1301 | /* |
3c541e14 | 1302 | * Check if our page_cgroup is valid |
8697d331 | 1303 | */ |
52d4b9ac KH |
1304 | pc = lookup_page_cgroup(page); |
1305 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 1306 | return NULL; |
b9c565d5 | 1307 | |
52d4b9ac | 1308 | lock_page_cgroup(pc); |
d13d1443 | 1309 | |
8c7c6e34 KH |
1310 | mem = pc->mem_cgroup; |
1311 | ||
d13d1443 KH |
1312 | if (!PageCgroupUsed(pc)) |
1313 | goto unlock_out; | |
1314 | ||
1315 | switch (ctype) { | |
1316 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1317 | if (page_mapped(page)) | |
1318 | goto unlock_out; | |
1319 | break; | |
1320 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
1321 | if (!PageAnon(page)) { /* Shared memory */ | |
1322 | if (page->mapping && !page_is_file_cache(page)) | |
1323 | goto unlock_out; | |
1324 | } else if (page_mapped(page)) /* Anon */ | |
1325 | goto unlock_out; | |
1326 | break; | |
1327 | default: | |
1328 | break; | |
52d4b9ac | 1329 | } |
d13d1443 | 1330 | |
8c7c6e34 KH |
1331 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1332 | if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)) | |
1333 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
08e552c6 | 1334 | mem_cgroup_charge_statistics(mem, pc, false); |
04046e1a | 1335 | |
52d4b9ac | 1336 | ClearPageCgroupUsed(pc); |
544122e5 KH |
1337 | /* |
1338 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
1339 | * freed from LRU. This is safe because uncharged page is expected not | |
1340 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
1341 | * special functions. | |
1342 | */ | |
b9c565d5 | 1343 | |
69029cd5 | 1344 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 1345 | unlock_page_cgroup(pc); |
fb59e9f1 | 1346 | |
a7fe942e KH |
1347 | /* at swapout, this memcg will be accessed to record to swap */ |
1348 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
1349 | css_put(&mem->css); | |
6d12e2d8 | 1350 | |
8c7c6e34 | 1351 | return mem; |
d13d1443 KH |
1352 | |
1353 | unlock_out: | |
1354 | unlock_page_cgroup(pc); | |
8c7c6e34 | 1355 | return NULL; |
3c541e14 BS |
1356 | } |
1357 | ||
69029cd5 KH |
1358 | void mem_cgroup_uncharge_page(struct page *page) |
1359 | { | |
52d4b9ac KH |
1360 | /* early check. */ |
1361 | if (page_mapped(page)) | |
1362 | return; | |
1363 | if (page->mapping && !PageAnon(page)) | |
1364 | return; | |
69029cd5 KH |
1365 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
1366 | } | |
1367 | ||
1368 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
1369 | { | |
1370 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 1371 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
1372 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
1373 | } | |
1374 | ||
8c7c6e34 KH |
1375 | /* |
1376 | * called from __delete_from_swap_cache() and drop "page" account. | |
1377 | * memcg information is recorded to swap_cgroup of "ent" | |
1378 | */ | |
1379 | void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent) | |
1380 | { | |
1381 | struct mem_cgroup *memcg; | |
1382 | ||
1383 | memcg = __mem_cgroup_uncharge_common(page, | |
1384 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT); | |
1385 | /* record memcg information */ | |
1386 | if (do_swap_account && memcg) { | |
1387 | swap_cgroup_record(ent, memcg); | |
1388 | mem_cgroup_get(memcg); | |
1389 | } | |
a7fe942e KH |
1390 | if (memcg) |
1391 | css_put(&memcg->css); | |
8c7c6e34 KH |
1392 | } |
1393 | ||
1394 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1395 | /* | |
1396 | * called from swap_entry_free(). remove record in swap_cgroup and | |
1397 | * uncharge "memsw" account. | |
1398 | */ | |
1399 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 1400 | { |
8c7c6e34 KH |
1401 | struct mem_cgroup *memcg; |
1402 | ||
1403 | if (!do_swap_account) | |
1404 | return; | |
1405 | ||
1406 | memcg = swap_cgroup_record(ent, NULL); | |
1407 | if (memcg) { | |
1408 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
1409 | mem_cgroup_put(memcg); | |
1410 | } | |
d13d1443 | 1411 | } |
8c7c6e34 | 1412 | #endif |
d13d1443 | 1413 | |
ae41be37 | 1414 | /* |
01b1ae63 KH |
1415 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
1416 | * page belongs to. | |
ae41be37 | 1417 | */ |
01b1ae63 | 1418 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
1419 | { |
1420 | struct page_cgroup *pc; | |
e8589cc1 | 1421 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 1422 | int ret = 0; |
8869b8f6 | 1423 | |
f8d66542 | 1424 | if (mem_cgroup_disabled()) |
4077960e BS |
1425 | return 0; |
1426 | ||
52d4b9ac KH |
1427 | pc = lookup_page_cgroup(page); |
1428 | lock_page_cgroup(pc); | |
1429 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
1430 | mem = pc->mem_cgroup; |
1431 | css_get(&mem->css); | |
e8589cc1 | 1432 | } |
52d4b9ac | 1433 | unlock_page_cgroup(pc); |
01b1ae63 | 1434 | |
e8589cc1 | 1435 | if (mem) { |
3bb4edf2 | 1436 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); |
e8589cc1 KH |
1437 | css_put(&mem->css); |
1438 | } | |
01b1ae63 | 1439 | *ptr = mem; |
e8589cc1 | 1440 | return ret; |
ae41be37 | 1441 | } |
8869b8f6 | 1442 | |
69029cd5 | 1443 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
1444 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
1445 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 1446 | { |
01b1ae63 KH |
1447 | struct page *target, *unused; |
1448 | struct page_cgroup *pc; | |
1449 | enum charge_type ctype; | |
1450 | ||
1451 | if (!mem) | |
1452 | return; | |
1453 | ||
1454 | /* at migration success, oldpage->mapping is NULL. */ | |
1455 | if (oldpage->mapping) { | |
1456 | target = oldpage; | |
1457 | unused = NULL; | |
1458 | } else { | |
1459 | target = newpage; | |
1460 | unused = oldpage; | |
1461 | } | |
1462 | ||
1463 | if (PageAnon(target)) | |
1464 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
1465 | else if (page_is_file_cache(target)) | |
1466 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
1467 | else | |
1468 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
1469 | ||
1470 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 1471 | if (unused) |
01b1ae63 KH |
1472 | __mem_cgroup_uncharge_common(unused, ctype); |
1473 | ||
1474 | pc = lookup_page_cgroup(target); | |
69029cd5 | 1475 | /* |
01b1ae63 KH |
1476 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
1477 | * So, double-counting is effectively avoided. | |
1478 | */ | |
1479 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
1480 | ||
1481 | /* | |
1482 | * Both of oldpage and newpage are still under lock_page(). | |
1483 | * Then, we don't have to care about race in radix-tree. | |
1484 | * But we have to be careful that this page is unmapped or not. | |
1485 | * | |
1486 | * There is a case for !page_mapped(). At the start of | |
1487 | * migration, oldpage was mapped. But now, it's zapped. | |
1488 | * But we know *target* page is not freed/reused under us. | |
1489 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 1490 | */ |
01b1ae63 KH |
1491 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
1492 | mem_cgroup_uncharge_page(target); | |
ae41be37 | 1493 | } |
78fb7466 | 1494 | |
c9b0ed51 KH |
1495 | /* |
1496 | * A call to try to shrink memory usage under specified resource controller. | |
1497 | * This is typically used for page reclaiming for shmem for reducing side | |
1498 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
1499 | */ | |
b5a84319 KH |
1500 | int mem_cgroup_shrink_usage(struct page *page, |
1501 | struct mm_struct *mm, | |
1502 | gfp_t gfp_mask) | |
c9b0ed51 | 1503 | { |
b5a84319 | 1504 | struct mem_cgroup *mem = NULL; |
c9b0ed51 KH |
1505 | int progress = 0; |
1506 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
1507 | ||
f8d66542 | 1508 | if (mem_cgroup_disabled()) |
cede86ac | 1509 | return 0; |
b5a84319 KH |
1510 | if (page) |
1511 | mem = try_get_mem_cgroup_from_swapcache(page); | |
1512 | if (!mem && mm) | |
1513 | mem = try_get_mem_cgroup_from_mm(mm); | |
54595fe2 | 1514 | if (unlikely(!mem)) |
31a78f23 | 1515 | return 0; |
c9b0ed51 KH |
1516 | |
1517 | do { | |
81d39c20 KH |
1518 | progress = mem_cgroup_hierarchical_reclaim(mem, |
1519 | gfp_mask, true, false); | |
b85a96c0 | 1520 | progress += mem_cgroup_check_under_limit(mem); |
c9b0ed51 KH |
1521 | } while (!progress && --retry); |
1522 | ||
1523 | css_put(&mem->css); | |
1524 | if (!retry) | |
1525 | return -ENOMEM; | |
1526 | return 0; | |
1527 | } | |
1528 | ||
8c7c6e34 KH |
1529 | static DEFINE_MUTEX(set_limit_mutex); |
1530 | ||
d38d2a75 | 1531 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 1532 | unsigned long long val) |
628f4235 | 1533 | { |
81d39c20 | 1534 | int retry_count; |
628f4235 | 1535 | int progress; |
8c7c6e34 | 1536 | u64 memswlimit; |
628f4235 | 1537 | int ret = 0; |
81d39c20 KH |
1538 | int children = mem_cgroup_count_children(memcg); |
1539 | u64 curusage, oldusage; | |
1540 | ||
1541 | /* | |
1542 | * For keeping hierarchical_reclaim simple, how long we should retry | |
1543 | * is depends on callers. We set our retry-count to be function | |
1544 | * of # of children which we should visit in this loop. | |
1545 | */ | |
1546 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
1547 | ||
1548 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 1549 | |
8c7c6e34 | 1550 | while (retry_count) { |
628f4235 KH |
1551 | if (signal_pending(current)) { |
1552 | ret = -EINTR; | |
1553 | break; | |
1554 | } | |
8c7c6e34 KH |
1555 | /* |
1556 | * Rather than hide all in some function, I do this in | |
1557 | * open coded manner. You see what this really does. | |
1558 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1559 | */ | |
1560 | mutex_lock(&set_limit_mutex); | |
1561 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1562 | if (memswlimit < val) { | |
1563 | ret = -EINVAL; | |
1564 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
1565 | break; |
1566 | } | |
8c7c6e34 KH |
1567 | ret = res_counter_set_limit(&memcg->res, val); |
1568 | mutex_unlock(&set_limit_mutex); | |
1569 | ||
1570 | if (!ret) | |
1571 | break; | |
1572 | ||
42e9abb6 | 1573 | progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, |
81d39c20 KH |
1574 | false, true); |
1575 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
1576 | /* Usage is reduced ? */ | |
1577 | if (curusage >= oldusage) | |
1578 | retry_count--; | |
1579 | else | |
1580 | oldusage = curusage; | |
8c7c6e34 | 1581 | } |
14797e23 | 1582 | |
8c7c6e34 KH |
1583 | return ret; |
1584 | } | |
1585 | ||
1586 | int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, | |
1587 | unsigned long long val) | |
1588 | { | |
81d39c20 | 1589 | int retry_count; |
8c7c6e34 | 1590 | u64 memlimit, oldusage, curusage; |
81d39c20 KH |
1591 | int children = mem_cgroup_count_children(memcg); |
1592 | int ret = -EBUSY; | |
8c7c6e34 KH |
1593 | |
1594 | if (!do_swap_account) | |
1595 | return -EINVAL; | |
81d39c20 KH |
1596 | /* see mem_cgroup_resize_res_limit */ |
1597 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
1598 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
1599 | while (retry_count) { |
1600 | if (signal_pending(current)) { | |
1601 | ret = -EINTR; | |
1602 | break; | |
1603 | } | |
1604 | /* | |
1605 | * Rather than hide all in some function, I do this in | |
1606 | * open coded manner. You see what this really does. | |
1607 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1608 | */ | |
1609 | mutex_lock(&set_limit_mutex); | |
1610 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1611 | if (memlimit > val) { | |
1612 | ret = -EINVAL; | |
1613 | mutex_unlock(&set_limit_mutex); | |
1614 | break; | |
1615 | } | |
1616 | ret = res_counter_set_limit(&memcg->memsw, val); | |
1617 | mutex_unlock(&set_limit_mutex); | |
1618 | ||
1619 | if (!ret) | |
1620 | break; | |
1621 | ||
81d39c20 | 1622 | mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true); |
8c7c6e34 | 1623 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 1624 | /* Usage is reduced ? */ |
8c7c6e34 | 1625 | if (curusage >= oldusage) |
628f4235 | 1626 | retry_count--; |
81d39c20 KH |
1627 | else |
1628 | oldusage = curusage; | |
628f4235 KH |
1629 | } |
1630 | return ret; | |
1631 | } | |
1632 | ||
cc847582 KH |
1633 | /* |
1634 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
1635 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
1636 | */ | |
f817ed48 | 1637 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 1638 | int node, int zid, enum lru_list lru) |
cc847582 | 1639 | { |
08e552c6 KH |
1640 | struct zone *zone; |
1641 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 1642 | struct page_cgroup *pc, *busy; |
08e552c6 | 1643 | unsigned long flags, loop; |
072c56c1 | 1644 | struct list_head *list; |
f817ed48 | 1645 | int ret = 0; |
072c56c1 | 1646 | |
08e552c6 KH |
1647 | zone = &NODE_DATA(node)->node_zones[zid]; |
1648 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 1649 | list = &mz->lists[lru]; |
cc847582 | 1650 | |
f817ed48 KH |
1651 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
1652 | /* give some margin against EBUSY etc...*/ | |
1653 | loop += 256; | |
1654 | busy = NULL; | |
1655 | while (loop--) { | |
1656 | ret = 0; | |
08e552c6 | 1657 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 1658 | if (list_empty(list)) { |
08e552c6 | 1659 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 1660 | break; |
f817ed48 KH |
1661 | } |
1662 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
1663 | if (busy == pc) { | |
1664 | list_move(&pc->lru, list); | |
1665 | busy = 0; | |
08e552c6 | 1666 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
1667 | continue; |
1668 | } | |
08e552c6 | 1669 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 1670 | |
2c26fdd7 | 1671 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 1672 | if (ret == -ENOMEM) |
52d4b9ac | 1673 | break; |
f817ed48 KH |
1674 | |
1675 | if (ret == -EBUSY || ret == -EINVAL) { | |
1676 | /* found lock contention or "pc" is obsolete. */ | |
1677 | busy = pc; | |
1678 | cond_resched(); | |
1679 | } else | |
1680 | busy = NULL; | |
cc847582 | 1681 | } |
08e552c6 | 1682 | |
f817ed48 KH |
1683 | if (!ret && !list_empty(list)) |
1684 | return -EBUSY; | |
1685 | return ret; | |
cc847582 KH |
1686 | } |
1687 | ||
1688 | /* | |
1689 | * make mem_cgroup's charge to be 0 if there is no task. | |
1690 | * This enables deleting this mem_cgroup. | |
1691 | */ | |
c1e862c1 | 1692 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 1693 | { |
f817ed48 KH |
1694 | int ret; |
1695 | int node, zid, shrink; | |
1696 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 1697 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 1698 | |
cc847582 | 1699 | css_get(&mem->css); |
f817ed48 KH |
1700 | |
1701 | shrink = 0; | |
c1e862c1 KH |
1702 | /* should free all ? */ |
1703 | if (free_all) | |
1704 | goto try_to_free; | |
f817ed48 | 1705 | move_account: |
1ecaab2b | 1706 | while (mem->res.usage > 0) { |
f817ed48 | 1707 | ret = -EBUSY; |
c1e862c1 KH |
1708 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
1709 | goto out; | |
1710 | ret = -EINTR; | |
1711 | if (signal_pending(current)) | |
cc847582 | 1712 | goto out; |
52d4b9ac KH |
1713 | /* This is for making all *used* pages to be on LRU. */ |
1714 | lru_add_drain_all(); | |
f817ed48 | 1715 | ret = 0; |
299b4eaa | 1716 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 1717 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 1718 | enum lru_list l; |
f817ed48 KH |
1719 | for_each_lru(l) { |
1720 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 1721 | node, zid, l); |
f817ed48 KH |
1722 | if (ret) |
1723 | break; | |
1724 | } | |
1ecaab2b | 1725 | } |
f817ed48 KH |
1726 | if (ret) |
1727 | break; | |
1728 | } | |
1729 | /* it seems parent cgroup doesn't have enough mem */ | |
1730 | if (ret == -ENOMEM) | |
1731 | goto try_to_free; | |
52d4b9ac | 1732 | cond_resched(); |
cc847582 KH |
1733 | } |
1734 | ret = 0; | |
1735 | out: | |
1736 | css_put(&mem->css); | |
1737 | return ret; | |
f817ed48 KH |
1738 | |
1739 | try_to_free: | |
c1e862c1 KH |
1740 | /* returns EBUSY if there is a task or if we come here twice. */ |
1741 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
1742 | ret = -EBUSY; |
1743 | goto out; | |
1744 | } | |
c1e862c1 KH |
1745 | /* we call try-to-free pages for make this cgroup empty */ |
1746 | lru_add_drain_all(); | |
f817ed48 KH |
1747 | /* try to free all pages in this cgroup */ |
1748 | shrink = 1; | |
1749 | while (nr_retries && mem->res.usage > 0) { | |
1750 | int progress; | |
c1e862c1 KH |
1751 | |
1752 | if (signal_pending(current)) { | |
1753 | ret = -EINTR; | |
1754 | goto out; | |
1755 | } | |
a7885eb8 KM |
1756 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
1757 | false, get_swappiness(mem)); | |
c1e862c1 | 1758 | if (!progress) { |
f817ed48 | 1759 | nr_retries--; |
c1e862c1 KH |
1760 | /* maybe some writeback is necessary */ |
1761 | congestion_wait(WRITE, HZ/10); | |
1762 | } | |
f817ed48 KH |
1763 | |
1764 | } | |
08e552c6 | 1765 | lru_add_drain(); |
f817ed48 KH |
1766 | /* try move_account...there may be some *locked* pages. */ |
1767 | if (mem->res.usage) | |
1768 | goto move_account; | |
1769 | ret = 0; | |
1770 | goto out; | |
cc847582 KH |
1771 | } |
1772 | ||
c1e862c1 KH |
1773 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
1774 | { | |
1775 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
1776 | } | |
1777 | ||
1778 | ||
18f59ea7 BS |
1779 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
1780 | { | |
1781 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
1782 | } | |
1783 | ||
1784 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
1785 | u64 val) | |
1786 | { | |
1787 | int retval = 0; | |
1788 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1789 | struct cgroup *parent = cont->parent; | |
1790 | struct mem_cgroup *parent_mem = NULL; | |
1791 | ||
1792 | if (parent) | |
1793 | parent_mem = mem_cgroup_from_cont(parent); | |
1794 | ||
1795 | cgroup_lock(); | |
1796 | /* | |
1797 | * If parent's use_hiearchy is set, we can't make any modifications | |
1798 | * in the child subtrees. If it is unset, then the change can | |
1799 | * occur, provided the current cgroup has no children. | |
1800 | * | |
1801 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
1802 | * set if there are no children. | |
1803 | */ | |
1804 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
1805 | (val == 1 || val == 0)) { | |
1806 | if (list_empty(&cont->children)) | |
1807 | mem->use_hierarchy = val; | |
1808 | else | |
1809 | retval = -EBUSY; | |
1810 | } else | |
1811 | retval = -EINVAL; | |
1812 | cgroup_unlock(); | |
1813 | ||
1814 | return retval; | |
1815 | } | |
1816 | ||
2c3daa72 | 1817 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 1818 | { |
8c7c6e34 KH |
1819 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
1820 | u64 val = 0; | |
1821 | int type, name; | |
1822 | ||
1823 | type = MEMFILE_TYPE(cft->private); | |
1824 | name = MEMFILE_ATTR(cft->private); | |
1825 | switch (type) { | |
1826 | case _MEM: | |
1827 | val = res_counter_read_u64(&mem->res, name); | |
1828 | break; | |
1829 | case _MEMSWAP: | |
1830 | if (do_swap_account) | |
1831 | val = res_counter_read_u64(&mem->memsw, name); | |
1832 | break; | |
1833 | default: | |
1834 | BUG(); | |
1835 | break; | |
1836 | } | |
1837 | return val; | |
8cdea7c0 | 1838 | } |
628f4235 KH |
1839 | /* |
1840 | * The user of this function is... | |
1841 | * RES_LIMIT. | |
1842 | */ | |
856c13aa PM |
1843 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
1844 | const char *buffer) | |
8cdea7c0 | 1845 | { |
628f4235 | 1846 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 1847 | int type, name; |
628f4235 KH |
1848 | unsigned long long val; |
1849 | int ret; | |
1850 | ||
8c7c6e34 KH |
1851 | type = MEMFILE_TYPE(cft->private); |
1852 | name = MEMFILE_ATTR(cft->private); | |
1853 | switch (name) { | |
628f4235 KH |
1854 | case RES_LIMIT: |
1855 | /* This function does all necessary parse...reuse it */ | |
1856 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
1857 | if (ret) |
1858 | break; | |
1859 | if (type == _MEM) | |
628f4235 | 1860 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
1861 | else |
1862 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 KH |
1863 | break; |
1864 | default: | |
1865 | ret = -EINVAL; /* should be BUG() ? */ | |
1866 | break; | |
1867 | } | |
1868 | return ret; | |
8cdea7c0 BS |
1869 | } |
1870 | ||
fee7b548 KH |
1871 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
1872 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
1873 | { | |
1874 | struct cgroup *cgroup; | |
1875 | unsigned long long min_limit, min_memsw_limit, tmp; | |
1876 | ||
1877 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1878 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1879 | cgroup = memcg->css.cgroup; | |
1880 | if (!memcg->use_hierarchy) | |
1881 | goto out; | |
1882 | ||
1883 | while (cgroup->parent) { | |
1884 | cgroup = cgroup->parent; | |
1885 | memcg = mem_cgroup_from_cont(cgroup); | |
1886 | if (!memcg->use_hierarchy) | |
1887 | break; | |
1888 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1889 | min_limit = min(min_limit, tmp); | |
1890 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1891 | min_memsw_limit = min(min_memsw_limit, tmp); | |
1892 | } | |
1893 | out: | |
1894 | *mem_limit = min_limit; | |
1895 | *memsw_limit = min_memsw_limit; | |
1896 | return; | |
1897 | } | |
1898 | ||
29f2a4da | 1899 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
1900 | { |
1901 | struct mem_cgroup *mem; | |
8c7c6e34 | 1902 | int type, name; |
c84872e1 PE |
1903 | |
1904 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
1905 | type = MEMFILE_TYPE(event); |
1906 | name = MEMFILE_ATTR(event); | |
1907 | switch (name) { | |
29f2a4da | 1908 | case RES_MAX_USAGE: |
8c7c6e34 KH |
1909 | if (type == _MEM) |
1910 | res_counter_reset_max(&mem->res); | |
1911 | else | |
1912 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
1913 | break; |
1914 | case RES_FAILCNT: | |
8c7c6e34 KH |
1915 | if (type == _MEM) |
1916 | res_counter_reset_failcnt(&mem->res); | |
1917 | else | |
1918 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
1919 | break; |
1920 | } | |
85cc59db | 1921 | return 0; |
c84872e1 PE |
1922 | } |
1923 | ||
14067bb3 KH |
1924 | |
1925 | /* For read statistics */ | |
1926 | enum { | |
1927 | MCS_CACHE, | |
1928 | MCS_RSS, | |
1929 | MCS_PGPGIN, | |
1930 | MCS_PGPGOUT, | |
1931 | MCS_INACTIVE_ANON, | |
1932 | MCS_ACTIVE_ANON, | |
1933 | MCS_INACTIVE_FILE, | |
1934 | MCS_ACTIVE_FILE, | |
1935 | MCS_UNEVICTABLE, | |
1936 | NR_MCS_STAT, | |
1937 | }; | |
1938 | ||
1939 | struct mcs_total_stat { | |
1940 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
1941 | }; |
1942 | ||
14067bb3 KH |
1943 | struct { |
1944 | char *local_name; | |
1945 | char *total_name; | |
1946 | } memcg_stat_strings[NR_MCS_STAT] = { | |
1947 | {"cache", "total_cache"}, | |
1948 | {"rss", "total_rss"}, | |
1949 | {"pgpgin", "total_pgpgin"}, | |
1950 | {"pgpgout", "total_pgpgout"}, | |
1951 | {"inactive_anon", "total_inactive_anon"}, | |
1952 | {"active_anon", "total_active_anon"}, | |
1953 | {"inactive_file", "total_inactive_file"}, | |
1954 | {"active_file", "total_active_file"}, | |
1955 | {"unevictable", "total_unevictable"} | |
1956 | }; | |
1957 | ||
1958 | ||
1959 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
1960 | { | |
1961 | struct mcs_total_stat *s = data; | |
1962 | s64 val; | |
1963 | ||
1964 | /* per cpu stat */ | |
1965 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE); | |
1966 | s->stat[MCS_CACHE] += val * PAGE_SIZE; | |
1967 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
1968 | s->stat[MCS_RSS] += val * PAGE_SIZE; | |
1969 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT); | |
1970 | s->stat[MCS_PGPGIN] += val; | |
1971 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT); | |
1972 | s->stat[MCS_PGPGOUT] += val; | |
1973 | ||
1974 | /* per zone stat */ | |
1975 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
1976 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
1977 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
1978 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
1979 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
1980 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
1981 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
1982 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
1983 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
1984 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
1985 | return 0; | |
1986 | } | |
1987 | ||
1988 | static void | |
1989 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
1990 | { | |
1991 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
1992 | } | |
1993 | ||
c64745cf PM |
1994 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
1995 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 1996 | { |
d2ceb9b7 | 1997 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 1998 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
1999 | int i; |
2000 | ||
14067bb3 KH |
2001 | memset(&mystat, 0, sizeof(mystat)); |
2002 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 2003 | |
14067bb3 KH |
2004 | for (i = 0; i < NR_MCS_STAT; i++) |
2005 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); | |
7b854121 | 2006 | |
14067bb3 | 2007 | /* Hierarchical information */ |
fee7b548 KH |
2008 | { |
2009 | unsigned long long limit, memsw_limit; | |
2010 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
2011 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
2012 | if (do_swap_account) | |
2013 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
2014 | } | |
7f016ee8 | 2015 | |
14067bb3 KH |
2016 | memset(&mystat, 0, sizeof(mystat)); |
2017 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
2018 | for (i = 0; i < NR_MCS_STAT; i++) | |
2019 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); | |
2020 | ||
2021 | ||
7f016ee8 | 2022 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 2023 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
2024 | |
2025 | { | |
2026 | int nid, zid; | |
2027 | struct mem_cgroup_per_zone *mz; | |
2028 | unsigned long recent_rotated[2] = {0, 0}; | |
2029 | unsigned long recent_scanned[2] = {0, 0}; | |
2030 | ||
2031 | for_each_online_node(nid) | |
2032 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
2033 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
2034 | ||
2035 | recent_rotated[0] += | |
2036 | mz->reclaim_stat.recent_rotated[0]; | |
2037 | recent_rotated[1] += | |
2038 | mz->reclaim_stat.recent_rotated[1]; | |
2039 | recent_scanned[0] += | |
2040 | mz->reclaim_stat.recent_scanned[0]; | |
2041 | recent_scanned[1] += | |
2042 | mz->reclaim_stat.recent_scanned[1]; | |
2043 | } | |
2044 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
2045 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
2046 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
2047 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
2048 | } | |
2049 | #endif | |
2050 | ||
d2ceb9b7 KH |
2051 | return 0; |
2052 | } | |
2053 | ||
a7885eb8 KM |
2054 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
2055 | { | |
2056 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2057 | ||
2058 | return get_swappiness(memcg); | |
2059 | } | |
2060 | ||
2061 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
2062 | u64 val) | |
2063 | { | |
2064 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2065 | struct mem_cgroup *parent; | |
068b38c1 | 2066 | |
a7885eb8 KM |
2067 | if (val > 100) |
2068 | return -EINVAL; | |
2069 | ||
2070 | if (cgrp->parent == NULL) | |
2071 | return -EINVAL; | |
2072 | ||
2073 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
2074 | |
2075 | cgroup_lock(); | |
2076 | ||
a7885eb8 KM |
2077 | /* If under hierarchy, only empty-root can set this value */ |
2078 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
2079 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
2080 | cgroup_unlock(); | |
a7885eb8 | 2081 | return -EINVAL; |
068b38c1 | 2082 | } |
a7885eb8 KM |
2083 | |
2084 | spin_lock(&memcg->reclaim_param_lock); | |
2085 | memcg->swappiness = val; | |
2086 | spin_unlock(&memcg->reclaim_param_lock); | |
2087 | ||
068b38c1 LZ |
2088 | cgroup_unlock(); |
2089 | ||
a7885eb8 KM |
2090 | return 0; |
2091 | } | |
2092 | ||
c1e862c1 | 2093 | |
8cdea7c0 BS |
2094 | static struct cftype mem_cgroup_files[] = { |
2095 | { | |
0eea1030 | 2096 | .name = "usage_in_bytes", |
8c7c6e34 | 2097 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 2098 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 2099 | }, |
c84872e1 PE |
2100 | { |
2101 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 2102 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 2103 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
2104 | .read_u64 = mem_cgroup_read, |
2105 | }, | |
8cdea7c0 | 2106 | { |
0eea1030 | 2107 | .name = "limit_in_bytes", |
8c7c6e34 | 2108 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 2109 | .write_string = mem_cgroup_write, |
2c3daa72 | 2110 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
2111 | }, |
2112 | { | |
2113 | .name = "failcnt", | |
8c7c6e34 | 2114 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 2115 | .trigger = mem_cgroup_reset, |
2c3daa72 | 2116 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 2117 | }, |
d2ceb9b7 KH |
2118 | { |
2119 | .name = "stat", | |
c64745cf | 2120 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 2121 | }, |
c1e862c1 KH |
2122 | { |
2123 | .name = "force_empty", | |
2124 | .trigger = mem_cgroup_force_empty_write, | |
2125 | }, | |
18f59ea7 BS |
2126 | { |
2127 | .name = "use_hierarchy", | |
2128 | .write_u64 = mem_cgroup_hierarchy_write, | |
2129 | .read_u64 = mem_cgroup_hierarchy_read, | |
2130 | }, | |
a7885eb8 KM |
2131 | { |
2132 | .name = "swappiness", | |
2133 | .read_u64 = mem_cgroup_swappiness_read, | |
2134 | .write_u64 = mem_cgroup_swappiness_write, | |
2135 | }, | |
8cdea7c0 BS |
2136 | }; |
2137 | ||
8c7c6e34 KH |
2138 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
2139 | static struct cftype memsw_cgroup_files[] = { | |
2140 | { | |
2141 | .name = "memsw.usage_in_bytes", | |
2142 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
2143 | .read_u64 = mem_cgroup_read, | |
2144 | }, | |
2145 | { | |
2146 | .name = "memsw.max_usage_in_bytes", | |
2147 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
2148 | .trigger = mem_cgroup_reset, | |
2149 | .read_u64 = mem_cgroup_read, | |
2150 | }, | |
2151 | { | |
2152 | .name = "memsw.limit_in_bytes", | |
2153 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
2154 | .write_string = mem_cgroup_write, | |
2155 | .read_u64 = mem_cgroup_read, | |
2156 | }, | |
2157 | { | |
2158 | .name = "memsw.failcnt", | |
2159 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
2160 | .trigger = mem_cgroup_reset, | |
2161 | .read_u64 = mem_cgroup_read, | |
2162 | }, | |
2163 | }; | |
2164 | ||
2165 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
2166 | { | |
2167 | if (!do_swap_account) | |
2168 | return 0; | |
2169 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
2170 | ARRAY_SIZE(memsw_cgroup_files)); | |
2171 | }; | |
2172 | #else | |
2173 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
2174 | { | |
2175 | return 0; | |
2176 | } | |
2177 | #endif | |
2178 | ||
6d12e2d8 KH |
2179 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2180 | { | |
2181 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 2182 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 2183 | enum lru_list l; |
41e3355d | 2184 | int zone, tmp = node; |
1ecaab2b KH |
2185 | /* |
2186 | * This routine is called against possible nodes. | |
2187 | * But it's BUG to call kmalloc() against offline node. | |
2188 | * | |
2189 | * TODO: this routine can waste much memory for nodes which will | |
2190 | * never be onlined. It's better to use memory hotplug callback | |
2191 | * function. | |
2192 | */ | |
41e3355d KH |
2193 | if (!node_state(node, N_NORMAL_MEMORY)) |
2194 | tmp = -1; | |
2195 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
2196 | if (!pn) |
2197 | return 1; | |
1ecaab2b | 2198 | |
6d12e2d8 KH |
2199 | mem->info.nodeinfo[node] = pn; |
2200 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
2201 | |
2202 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
2203 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
2204 | for_each_lru(l) |
2205 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 2206 | } |
6d12e2d8 KH |
2207 | return 0; |
2208 | } | |
2209 | ||
1ecaab2b KH |
2210 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2211 | { | |
2212 | kfree(mem->info.nodeinfo[node]); | |
2213 | } | |
2214 | ||
c8dad2bb JB |
2215 | static int mem_cgroup_size(void) |
2216 | { | |
2217 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
2218 | return sizeof(struct mem_cgroup) + cpustat_size; | |
2219 | } | |
2220 | ||
33327948 KH |
2221 | static struct mem_cgroup *mem_cgroup_alloc(void) |
2222 | { | |
2223 | struct mem_cgroup *mem; | |
c8dad2bb | 2224 | int size = mem_cgroup_size(); |
33327948 | 2225 | |
c8dad2bb JB |
2226 | if (size < PAGE_SIZE) |
2227 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 2228 | else |
c8dad2bb | 2229 | mem = vmalloc(size); |
33327948 KH |
2230 | |
2231 | if (mem) | |
c8dad2bb | 2232 | memset(mem, 0, size); |
33327948 KH |
2233 | return mem; |
2234 | } | |
2235 | ||
8c7c6e34 KH |
2236 | /* |
2237 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
2238 | * (scanning all at force_empty is too costly...) | |
2239 | * | |
2240 | * Instead of clearing all references at force_empty, we remember | |
2241 | * the number of reference from swap_cgroup and free mem_cgroup when | |
2242 | * it goes down to 0. | |
2243 | * | |
8c7c6e34 KH |
2244 | * Removal of cgroup itself succeeds regardless of refs from swap. |
2245 | */ | |
2246 | ||
a7ba0eef | 2247 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 2248 | { |
08e552c6 KH |
2249 | int node; |
2250 | ||
04046e1a KH |
2251 | free_css_id(&mem_cgroup_subsys, &mem->css); |
2252 | ||
08e552c6 KH |
2253 | for_each_node_state(node, N_POSSIBLE) |
2254 | free_mem_cgroup_per_zone_info(mem, node); | |
2255 | ||
c8dad2bb | 2256 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
2257 | kfree(mem); |
2258 | else | |
2259 | vfree(mem); | |
2260 | } | |
2261 | ||
8c7c6e34 KH |
2262 | static void mem_cgroup_get(struct mem_cgroup *mem) |
2263 | { | |
2264 | atomic_inc(&mem->refcnt); | |
2265 | } | |
2266 | ||
2267 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
2268 | { | |
7bcc1bb1 DN |
2269 | if (atomic_dec_and_test(&mem->refcnt)) { |
2270 | struct mem_cgroup *parent = parent_mem_cgroup(mem); | |
a7ba0eef | 2271 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
2272 | if (parent) |
2273 | mem_cgroup_put(parent); | |
2274 | } | |
8c7c6e34 KH |
2275 | } |
2276 | ||
7bcc1bb1 DN |
2277 | /* |
2278 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
2279 | */ | |
2280 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
2281 | { | |
2282 | if (!mem->res.parent) | |
2283 | return NULL; | |
2284 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
2285 | } | |
33327948 | 2286 | |
c077719b KH |
2287 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
2288 | static void __init enable_swap_cgroup(void) | |
2289 | { | |
f8d66542 | 2290 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
2291 | do_swap_account = 1; |
2292 | } | |
2293 | #else | |
2294 | static void __init enable_swap_cgroup(void) | |
2295 | { | |
2296 | } | |
2297 | #endif | |
2298 | ||
0eb253e2 | 2299 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
2300 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
2301 | { | |
28dbc4b6 | 2302 | struct mem_cgroup *mem, *parent; |
04046e1a | 2303 | long error = -ENOMEM; |
6d12e2d8 | 2304 | int node; |
8cdea7c0 | 2305 | |
c8dad2bb JB |
2306 | mem = mem_cgroup_alloc(); |
2307 | if (!mem) | |
04046e1a | 2308 | return ERR_PTR(error); |
78fb7466 | 2309 | |
6d12e2d8 KH |
2310 | for_each_node_state(node, N_POSSIBLE) |
2311 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
2312 | goto free_out; | |
c077719b | 2313 | /* root ? */ |
28dbc4b6 | 2314 | if (cont->parent == NULL) { |
c077719b | 2315 | enable_swap_cgroup(); |
28dbc4b6 | 2316 | parent = NULL; |
18f59ea7 | 2317 | } else { |
28dbc4b6 | 2318 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 BS |
2319 | mem->use_hierarchy = parent->use_hierarchy; |
2320 | } | |
28dbc4b6 | 2321 | |
18f59ea7 BS |
2322 | if (parent && parent->use_hierarchy) { |
2323 | res_counter_init(&mem->res, &parent->res); | |
2324 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
2325 | /* |
2326 | * We increment refcnt of the parent to ensure that we can | |
2327 | * safely access it on res_counter_charge/uncharge. | |
2328 | * This refcnt will be decremented when freeing this | |
2329 | * mem_cgroup(see mem_cgroup_put). | |
2330 | */ | |
2331 | mem_cgroup_get(parent); | |
18f59ea7 BS |
2332 | } else { |
2333 | res_counter_init(&mem->res, NULL); | |
2334 | res_counter_init(&mem->memsw, NULL); | |
2335 | } | |
04046e1a | 2336 | mem->last_scanned_child = 0; |
2733c06a | 2337 | spin_lock_init(&mem->reclaim_param_lock); |
6d61ef40 | 2338 | |
a7885eb8 KM |
2339 | if (parent) |
2340 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 2341 | atomic_set(&mem->refcnt, 1); |
8cdea7c0 | 2342 | return &mem->css; |
6d12e2d8 | 2343 | free_out: |
a7ba0eef | 2344 | __mem_cgroup_free(mem); |
04046e1a | 2345 | return ERR_PTR(error); |
8cdea7c0 BS |
2346 | } |
2347 | ||
ec64f515 | 2348 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
2349 | struct cgroup *cont) |
2350 | { | |
2351 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
2352 | |
2353 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
2354 | } |
2355 | ||
8cdea7c0 BS |
2356 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
2357 | struct cgroup *cont) | |
2358 | { | |
c268e994 | 2359 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 2360 | |
c268e994 | 2361 | mem_cgroup_put(mem); |
8cdea7c0 BS |
2362 | } |
2363 | ||
2364 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
2365 | struct cgroup *cont) | |
2366 | { | |
8c7c6e34 KH |
2367 | int ret; |
2368 | ||
2369 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
2370 | ARRAY_SIZE(mem_cgroup_files)); | |
2371 | ||
2372 | if (!ret) | |
2373 | ret = register_memsw_files(cont, ss); | |
2374 | return ret; | |
8cdea7c0 BS |
2375 | } |
2376 | ||
67e465a7 BS |
2377 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
2378 | struct cgroup *cont, | |
2379 | struct cgroup *old_cont, | |
2380 | struct task_struct *p) | |
2381 | { | |
7f4d454d | 2382 | mutex_lock(&memcg_tasklist); |
67e465a7 | 2383 | /* |
f9717d28 NK |
2384 | * FIXME: It's better to move charges of this process from old |
2385 | * memcg to new memcg. But it's just on TODO-List now. | |
67e465a7 | 2386 | */ |
7f4d454d | 2387 | mutex_unlock(&memcg_tasklist); |
67e465a7 BS |
2388 | } |
2389 | ||
8cdea7c0 BS |
2390 | struct cgroup_subsys mem_cgroup_subsys = { |
2391 | .name = "memory", | |
2392 | .subsys_id = mem_cgroup_subsys_id, | |
2393 | .create = mem_cgroup_create, | |
df878fb0 | 2394 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
2395 | .destroy = mem_cgroup_destroy, |
2396 | .populate = mem_cgroup_populate, | |
67e465a7 | 2397 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 2398 | .early_init = 0, |
04046e1a | 2399 | .use_id = 1, |
8cdea7c0 | 2400 | }; |
c077719b KH |
2401 | |
2402 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2403 | ||
2404 | static int __init disable_swap_account(char *s) | |
2405 | { | |
2406 | really_do_swap_account = 0; | |
2407 | return 1; | |
2408 | } | |
2409 | __setup("noswapaccount", disable_swap_account); | |
2410 | #endif |