]>
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; | |
0b7f569e KH |
298 | |
299 | if (!mm) | |
300 | return NULL; | |
54595fe2 KH |
301 | /* |
302 | * Because we have no locks, mm->owner's may be being moved to other | |
303 | * cgroup. We use css_tryget() here even if this looks | |
304 | * pessimistic (rather than adding locks here). | |
305 | */ | |
306 | rcu_read_lock(); | |
307 | do { | |
308 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
309 | if (unlikely(!mem)) | |
310 | break; | |
311 | } while (!css_tryget(&mem->css)); | |
312 | rcu_read_unlock(); | |
313 | return mem; | |
314 | } | |
315 | ||
316 | static bool mem_cgroup_is_obsolete(struct mem_cgroup *mem) | |
317 | { | |
318 | if (!mem) | |
319 | return true; | |
320 | return css_is_removed(&mem->css); | |
321 | } | |
322 | ||
14067bb3 KH |
323 | |
324 | /* | |
325 | * Call callback function against all cgroup under hierarchy tree. | |
326 | */ | |
327 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
328 | int (*func)(struct mem_cgroup *, void *)) | |
329 | { | |
330 | int found, ret, nextid; | |
331 | struct cgroup_subsys_state *css; | |
332 | struct mem_cgroup *mem; | |
333 | ||
334 | if (!root->use_hierarchy) | |
335 | return (*func)(root, data); | |
336 | ||
337 | nextid = 1; | |
338 | do { | |
339 | ret = 0; | |
340 | mem = NULL; | |
341 | ||
342 | rcu_read_lock(); | |
343 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
344 | &found); | |
345 | if (css && css_tryget(css)) | |
346 | mem = container_of(css, struct mem_cgroup, css); | |
347 | rcu_read_unlock(); | |
348 | ||
349 | if (mem) { | |
350 | ret = (*func)(mem, data); | |
351 | css_put(&mem->css); | |
352 | } | |
353 | nextid = found + 1; | |
354 | } while (!ret && css); | |
355 | ||
356 | return ret; | |
357 | } | |
358 | ||
08e552c6 KH |
359 | /* |
360 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
361 | * Operations are called by routine of global LRU independently from memcg. | |
362 | * What we have to take care of here is validness of pc->mem_cgroup. | |
363 | * | |
364 | * Changes to pc->mem_cgroup happens when | |
365 | * 1. charge | |
366 | * 2. moving account | |
367 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
368 | * It is added to LRU before charge. | |
369 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
370 | * When moving account, the page is not on LRU. It's isolated. | |
371 | */ | |
4f98a2fe | 372 | |
08e552c6 KH |
373 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
374 | { | |
375 | struct page_cgroup *pc; | |
376 | struct mem_cgroup *mem; | |
377 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 378 | |
f8d66542 | 379 | if (mem_cgroup_disabled()) |
08e552c6 KH |
380 | return; |
381 | pc = lookup_page_cgroup(page); | |
382 | /* can happen while we handle swapcache. */ | |
544122e5 | 383 | if (list_empty(&pc->lru) || !pc->mem_cgroup) |
08e552c6 | 384 | return; |
544122e5 KH |
385 | /* |
386 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
387 | * removed from global LRU. | |
388 | */ | |
08e552c6 KH |
389 | mz = page_cgroup_zoneinfo(pc); |
390 | mem = pc->mem_cgroup; | |
b69408e8 | 391 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
08e552c6 KH |
392 | list_del_init(&pc->lru); |
393 | return; | |
6d12e2d8 KH |
394 | } |
395 | ||
08e552c6 | 396 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 397 | { |
08e552c6 KH |
398 | mem_cgroup_del_lru_list(page, page_lru(page)); |
399 | } | |
b69408e8 | 400 | |
08e552c6 KH |
401 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
402 | { | |
403 | struct mem_cgroup_per_zone *mz; | |
404 | struct page_cgroup *pc; | |
b69408e8 | 405 | |
f8d66542 | 406 | if (mem_cgroup_disabled()) |
08e552c6 | 407 | return; |
6d12e2d8 | 408 | |
08e552c6 | 409 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
410 | /* |
411 | * Used bit is set without atomic ops but after smp_wmb(). | |
412 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
413 | */ | |
08e552c6 KH |
414 | smp_rmb(); |
415 | /* unused page is not rotated. */ | |
416 | if (!PageCgroupUsed(pc)) | |
417 | return; | |
418 | mz = page_cgroup_zoneinfo(pc); | |
419 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
420 | } |
421 | ||
08e552c6 | 422 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 423 | { |
08e552c6 KH |
424 | struct page_cgroup *pc; |
425 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 426 | |
f8d66542 | 427 | if (mem_cgroup_disabled()) |
08e552c6 KH |
428 | return; |
429 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
430 | /* |
431 | * Used bit is set without atomic ops but after smp_wmb(). | |
432 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
433 | */ | |
08e552c6 KH |
434 | smp_rmb(); |
435 | if (!PageCgroupUsed(pc)) | |
894bc310 | 436 | return; |
b69408e8 | 437 | |
08e552c6 | 438 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 439 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
08e552c6 KH |
440 | list_add(&pc->lru, &mz->lists[lru]); |
441 | } | |
544122e5 | 442 | |
08e552c6 | 443 | /* |
544122e5 KH |
444 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
445 | * lru because the page may.be reused after it's fully uncharged (because of | |
446 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
447 | * it again. This function is only used to charge SwapCache. It's done under | |
448 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 449 | */ |
544122e5 | 450 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 451 | { |
544122e5 KH |
452 | unsigned long flags; |
453 | struct zone *zone = page_zone(page); | |
454 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
455 | ||
456 | spin_lock_irqsave(&zone->lru_lock, flags); | |
457 | /* | |
458 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
459 | * is guarded by lock_page() because the page is SwapCache. | |
460 | */ | |
461 | if (!PageCgroupUsed(pc)) | |
462 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
463 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
464 | } |
465 | ||
544122e5 KH |
466 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
467 | { | |
468 | unsigned long flags; | |
469 | struct zone *zone = page_zone(page); | |
470 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
471 | ||
472 | spin_lock_irqsave(&zone->lru_lock, flags); | |
473 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
474 | if (PageLRU(page) && list_empty(&pc->lru)) | |
475 | mem_cgroup_add_lru_list(page, page_lru(page)); | |
476 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
477 | } | |
478 | ||
479 | ||
08e552c6 KH |
480 | void mem_cgroup_move_lists(struct page *page, |
481 | enum lru_list from, enum lru_list to) | |
482 | { | |
f8d66542 | 483 | if (mem_cgroup_disabled()) |
08e552c6 KH |
484 | return; |
485 | mem_cgroup_del_lru_list(page, from); | |
486 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
487 | } |
488 | ||
4c4a2214 DR |
489 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
490 | { | |
491 | int ret; | |
0b7f569e | 492 | struct mem_cgroup *curr = NULL; |
4c4a2214 DR |
493 | |
494 | task_lock(task); | |
0b7f569e KH |
495 | rcu_read_lock(); |
496 | curr = try_get_mem_cgroup_from_mm(task->mm); | |
497 | rcu_read_unlock(); | |
4c4a2214 | 498 | task_unlock(task); |
0b7f569e KH |
499 | if (!curr) |
500 | return 0; | |
501 | if (curr->use_hierarchy) | |
502 | ret = css_is_ancestor(&curr->css, &mem->css); | |
503 | else | |
504 | ret = (curr == mem); | |
505 | css_put(&curr->css); | |
4c4a2214 DR |
506 | return ret; |
507 | } | |
508 | ||
58ae83db KH |
509 | /* |
510 | * Calculate mapped_ratio under memory controller. This will be used in | |
511 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
512 | */ | |
513 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
514 | { | |
515 | long total, rss; | |
516 | ||
517 | /* | |
518 | * usage is recorded in bytes. But, here, we assume the number of | |
519 | * physical pages can be represented by "long" on any arch. | |
520 | */ | |
521 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
522 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
523 | return (int)((rss * 100L) / total); | |
524 | } | |
8869b8f6 | 525 | |
6c48a1d0 KH |
526 | /* |
527 | * prev_priority control...this will be used in memory reclaim path. | |
528 | */ | |
529 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
530 | { | |
2733c06a KM |
531 | int prev_priority; |
532 | ||
533 | spin_lock(&mem->reclaim_param_lock); | |
534 | prev_priority = mem->prev_priority; | |
535 | spin_unlock(&mem->reclaim_param_lock); | |
536 | ||
537 | return prev_priority; | |
6c48a1d0 KH |
538 | } |
539 | ||
540 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
541 | { | |
2733c06a | 542 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
543 | if (priority < mem->prev_priority) |
544 | mem->prev_priority = priority; | |
2733c06a | 545 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
546 | } |
547 | ||
548 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
549 | { | |
2733c06a | 550 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 | 551 | mem->prev_priority = priority; |
2733c06a | 552 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
553 | } |
554 | ||
c772be93 | 555 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
556 | { |
557 | unsigned long active; | |
558 | unsigned long inactive; | |
c772be93 KM |
559 | unsigned long gb; |
560 | unsigned long inactive_ratio; | |
14797e23 | 561 | |
14067bb3 KH |
562 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
563 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 564 | |
c772be93 KM |
565 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
566 | if (gb) | |
567 | inactive_ratio = int_sqrt(10 * gb); | |
568 | else | |
569 | inactive_ratio = 1; | |
570 | ||
571 | if (present_pages) { | |
572 | present_pages[0] = inactive; | |
573 | present_pages[1] = active; | |
574 | } | |
575 | ||
576 | return inactive_ratio; | |
577 | } | |
578 | ||
579 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
580 | { | |
581 | unsigned long active; | |
582 | unsigned long inactive; | |
583 | unsigned long present_pages[2]; | |
584 | unsigned long inactive_ratio; | |
585 | ||
586 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
587 | ||
588 | inactive = present_pages[0]; | |
589 | active = present_pages[1]; | |
590 | ||
591 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
592 | return 1; |
593 | ||
594 | return 0; | |
595 | } | |
596 | ||
a3d8e054 KM |
597 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
598 | struct zone *zone, | |
599 | enum lru_list lru) | |
600 | { | |
601 | int nid = zone->zone_pgdat->node_id; | |
602 | int zid = zone_idx(zone); | |
603 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
604 | ||
605 | return MEM_CGROUP_ZSTAT(mz, lru); | |
606 | } | |
607 | ||
3e2f41f1 KM |
608 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
609 | struct zone *zone) | |
610 | { | |
611 | int nid = zone->zone_pgdat->node_id; | |
612 | int zid = zone_idx(zone); | |
613 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
614 | ||
615 | return &mz->reclaim_stat; | |
616 | } | |
617 | ||
618 | struct zone_reclaim_stat * | |
619 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
620 | { | |
621 | struct page_cgroup *pc; | |
622 | struct mem_cgroup_per_zone *mz; | |
623 | ||
624 | if (mem_cgroup_disabled()) | |
625 | return NULL; | |
626 | ||
627 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
628 | /* |
629 | * Used bit is set without atomic ops but after smp_wmb(). | |
630 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
631 | */ | |
632 | smp_rmb(); | |
633 | if (!PageCgroupUsed(pc)) | |
634 | return NULL; | |
635 | ||
3e2f41f1 KM |
636 | mz = page_cgroup_zoneinfo(pc); |
637 | if (!mz) | |
638 | return NULL; | |
639 | ||
640 | return &mz->reclaim_stat; | |
641 | } | |
642 | ||
66e1707b BS |
643 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
644 | struct list_head *dst, | |
645 | unsigned long *scanned, int order, | |
646 | int mode, struct zone *z, | |
647 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 648 | int active, int file) |
66e1707b BS |
649 | { |
650 | unsigned long nr_taken = 0; | |
651 | struct page *page; | |
652 | unsigned long scan; | |
653 | LIST_HEAD(pc_list); | |
654 | struct list_head *src; | |
ff7283fa | 655 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
656 | int nid = z->zone_pgdat->node_id; |
657 | int zid = zone_idx(z); | |
658 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 659 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 660 | |
cf475ad2 | 661 | BUG_ON(!mem_cont); |
1ecaab2b | 662 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 663 | src = &mz->lists[lru]; |
66e1707b | 664 | |
ff7283fa KH |
665 | scan = 0; |
666 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 667 | if (scan >= nr_to_scan) |
ff7283fa | 668 | break; |
08e552c6 KH |
669 | |
670 | page = pc->page; | |
52d4b9ac KH |
671 | if (unlikely(!PageCgroupUsed(pc))) |
672 | continue; | |
436c6541 | 673 | if (unlikely(!PageLRU(page))) |
ff7283fa | 674 | continue; |
ff7283fa | 675 | |
436c6541 | 676 | scan++; |
4f98a2fe | 677 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
678 | list_move(&page->lru, dst); |
679 | nr_taken++; | |
680 | } | |
681 | } | |
682 | ||
66e1707b BS |
683 | *scanned = scan; |
684 | return nr_taken; | |
685 | } | |
686 | ||
6d61ef40 BS |
687 | #define mem_cgroup_from_res_counter(counter, member) \ |
688 | container_of(counter, struct mem_cgroup, member) | |
689 | ||
b85a96c0 DN |
690 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
691 | { | |
692 | if (do_swap_account) { | |
693 | if (res_counter_check_under_limit(&mem->res) && | |
694 | res_counter_check_under_limit(&mem->memsw)) | |
695 | return true; | |
696 | } else | |
697 | if (res_counter_check_under_limit(&mem->res)) | |
698 | return true; | |
699 | return false; | |
700 | } | |
701 | ||
a7885eb8 KM |
702 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
703 | { | |
704 | struct cgroup *cgrp = memcg->css.cgroup; | |
705 | unsigned int swappiness; | |
706 | ||
707 | /* root ? */ | |
708 | if (cgrp->parent == NULL) | |
709 | return vm_swappiness; | |
710 | ||
711 | spin_lock(&memcg->reclaim_param_lock); | |
712 | swappiness = memcg->swappiness; | |
713 | spin_unlock(&memcg->reclaim_param_lock); | |
714 | ||
715 | return swappiness; | |
716 | } | |
717 | ||
81d39c20 KH |
718 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
719 | { | |
720 | int *val = data; | |
721 | (*val)++; | |
722 | return 0; | |
723 | } | |
724 | /* | |
725 | * This function returns the number of memcg under hierarchy tree. Returns | |
726 | * 1(self count) if no children. | |
727 | */ | |
728 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
729 | { | |
730 | int num = 0; | |
731 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
732 | return num; | |
733 | } | |
734 | ||
6d61ef40 | 735 | /* |
04046e1a KH |
736 | * Visit the first child (need not be the first child as per the ordering |
737 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
738 | * that to reclaim free pages from. | |
739 | */ | |
740 | static struct mem_cgroup * | |
741 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
742 | { | |
743 | struct mem_cgroup *ret = NULL; | |
744 | struct cgroup_subsys_state *css; | |
745 | int nextid, found; | |
746 | ||
747 | if (!root_mem->use_hierarchy) { | |
748 | css_get(&root_mem->css); | |
749 | ret = root_mem; | |
750 | } | |
751 | ||
752 | while (!ret) { | |
753 | rcu_read_lock(); | |
754 | nextid = root_mem->last_scanned_child + 1; | |
755 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
756 | &found); | |
757 | if (css && css_tryget(css)) | |
758 | ret = container_of(css, struct mem_cgroup, css); | |
759 | ||
760 | rcu_read_unlock(); | |
761 | /* Updates scanning parameter */ | |
762 | spin_lock(&root_mem->reclaim_param_lock); | |
763 | if (!css) { | |
764 | /* this means start scan from ID:1 */ | |
765 | root_mem->last_scanned_child = 0; | |
766 | } else | |
767 | root_mem->last_scanned_child = found; | |
768 | spin_unlock(&root_mem->reclaim_param_lock); | |
769 | } | |
770 | ||
771 | return ret; | |
772 | } | |
773 | ||
774 | /* | |
775 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
776 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
777 | * based on its position in the children list. | |
6d61ef40 BS |
778 | * |
779 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
780 | * |
781 | * We give up and return to the caller when we visit root_mem twice. | |
782 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
783 | * |
784 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
785 | */ |
786 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
81d39c20 | 787 | gfp_t gfp_mask, bool noswap, bool shrink) |
6d61ef40 | 788 | { |
04046e1a KH |
789 | struct mem_cgroup *victim; |
790 | int ret, total = 0; | |
791 | int loop = 0; | |
792 | ||
793 | while (loop < 2) { | |
794 | victim = mem_cgroup_select_victim(root_mem); | |
795 | if (victim == root_mem) | |
796 | loop++; | |
797 | if (!mem_cgroup_local_usage(&victim->stat)) { | |
798 | /* this cgroup's local usage == 0 */ | |
799 | css_put(&victim->css); | |
6d61ef40 BS |
800 | continue; |
801 | } | |
04046e1a KH |
802 | /* we use swappiness of local cgroup */ |
803 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, noswap, | |
804 | get_swappiness(victim)); | |
805 | css_put(&victim->css); | |
81d39c20 KH |
806 | /* |
807 | * At shrinking usage, we can't check we should stop here or | |
808 | * reclaim more. It's depends on callers. last_scanned_child | |
809 | * will work enough for keeping fairness under tree. | |
810 | */ | |
811 | if (shrink) | |
812 | return ret; | |
04046e1a | 813 | total += ret; |
b85a96c0 | 814 | if (mem_cgroup_check_under_limit(root_mem)) |
04046e1a | 815 | return 1 + total; |
6d61ef40 | 816 | } |
04046e1a | 817 | return total; |
6d61ef40 BS |
818 | } |
819 | ||
a636b327 KH |
820 | bool mem_cgroup_oom_called(struct task_struct *task) |
821 | { | |
822 | bool ret = false; | |
823 | struct mem_cgroup *mem; | |
824 | struct mm_struct *mm; | |
825 | ||
826 | rcu_read_lock(); | |
827 | mm = task->mm; | |
828 | if (!mm) | |
829 | mm = &init_mm; | |
830 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
831 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
832 | ret = true; | |
833 | rcu_read_unlock(); | |
834 | return ret; | |
835 | } | |
0b7f569e KH |
836 | |
837 | static int record_last_oom_cb(struct mem_cgroup *mem, void *data) | |
838 | { | |
839 | mem->last_oom_jiffies = jiffies; | |
840 | return 0; | |
841 | } | |
842 | ||
843 | static void record_last_oom(struct mem_cgroup *mem) | |
844 | { | |
845 | mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb); | |
846 | } | |
847 | ||
848 | ||
f817ed48 KH |
849 | /* |
850 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
851 | * oom-killer can be invoked. | |
8a9f3ccd | 852 | */ |
f817ed48 | 853 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
8c7c6e34 KH |
854 | gfp_t gfp_mask, struct mem_cgroup **memcg, |
855 | bool oom) | |
8a9f3ccd | 856 | { |
6d61ef40 | 857 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 858 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
28dbc4b6 | 859 | struct res_counter *fail_res; |
a636b327 KH |
860 | |
861 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
862 | /* Don't account this! */ | |
863 | *memcg = NULL; | |
864 | return 0; | |
865 | } | |
866 | ||
8a9f3ccd | 867 | /* |
3be91277 HD |
868 | * We always charge the cgroup the mm_struct belongs to. |
869 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
870 | * thread group leader migrates. It's possible that mm is not |
871 | * set, if so charge the init_mm (happens for pagecache usage). | |
872 | */ | |
54595fe2 KH |
873 | mem = *memcg; |
874 | if (likely(!mem)) { | |
875 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 876 | *memcg = mem; |
e8589cc1 | 877 | } else { |
7a81b88c | 878 | css_get(&mem->css); |
e8589cc1 | 879 | } |
54595fe2 KH |
880 | if (unlikely(!mem)) |
881 | return 0; | |
882 | ||
883 | VM_BUG_ON(mem_cgroup_is_obsolete(mem)); | |
8a9f3ccd | 884 | |
8c7c6e34 KH |
885 | while (1) { |
886 | int ret; | |
887 | bool noswap = false; | |
7a81b88c | 888 | |
28dbc4b6 | 889 | ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res); |
8c7c6e34 KH |
890 | if (likely(!ret)) { |
891 | if (!do_swap_account) | |
892 | break; | |
28dbc4b6 BS |
893 | ret = res_counter_charge(&mem->memsw, PAGE_SIZE, |
894 | &fail_res); | |
8c7c6e34 KH |
895 | if (likely(!ret)) |
896 | break; | |
897 | /* mem+swap counter fails */ | |
898 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
899 | noswap = true; | |
6d61ef40 BS |
900 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
901 | memsw); | |
902 | } else | |
903 | /* mem counter fails */ | |
904 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
905 | res); | |
906 | ||
3be91277 | 907 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 908 | goto nomem; |
e1a1cd59 | 909 | |
6d61ef40 | 910 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask, |
81d39c20 | 911 | noswap, false); |
4d1c6273 DN |
912 | if (ret) |
913 | continue; | |
66e1707b BS |
914 | |
915 | /* | |
8869b8f6 HD |
916 | * try_to_free_mem_cgroup_pages() might not give us a full |
917 | * picture of reclaim. Some pages are reclaimed and might be | |
918 | * moved to swap cache or just unmapped from the cgroup. | |
919 | * Check the limit again to see if the reclaim reduced the | |
920 | * current usage of the cgroup before giving up | |
8c7c6e34 | 921 | * |
8869b8f6 | 922 | */ |
b85a96c0 DN |
923 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
924 | continue; | |
3be91277 HD |
925 | |
926 | if (!nr_retries--) { | |
a636b327 | 927 | if (oom) { |
7f4d454d | 928 | mutex_lock(&memcg_tasklist); |
88700756 | 929 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); |
7f4d454d | 930 | mutex_unlock(&memcg_tasklist); |
0b7f569e | 931 | record_last_oom(mem_over_limit); |
a636b327 | 932 | } |
7a81b88c | 933 | goto nomem; |
66e1707b | 934 | } |
8a9f3ccd | 935 | } |
7a81b88c KH |
936 | return 0; |
937 | nomem: | |
938 | css_put(&mem->css); | |
939 | return -ENOMEM; | |
940 | } | |
8a9f3ccd | 941 | |
b5a84319 KH |
942 | static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page) |
943 | { | |
944 | struct mem_cgroup *mem; | |
945 | swp_entry_t ent; | |
946 | ||
947 | if (!PageSwapCache(page)) | |
948 | return NULL; | |
949 | ||
950 | ent.val = page_private(page); | |
951 | mem = lookup_swap_cgroup(ent); | |
952 | if (!mem) | |
953 | return NULL; | |
954 | if (!css_tryget(&mem->css)) | |
955 | return NULL; | |
956 | return mem; | |
957 | } | |
958 | ||
7a81b88c | 959 | /* |
a5e924f5 | 960 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
961 | * USED state. If already USED, uncharge and return. |
962 | */ | |
963 | ||
964 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
965 | struct page_cgroup *pc, | |
966 | enum charge_type ctype) | |
967 | { | |
7a81b88c KH |
968 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
969 | if (!mem) | |
970 | return; | |
52d4b9ac KH |
971 | |
972 | lock_page_cgroup(pc); | |
973 | if (unlikely(PageCgroupUsed(pc))) { | |
974 | unlock_page_cgroup(pc); | |
975 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
976 | if (do_swap_account) |
977 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
52d4b9ac | 978 | css_put(&mem->css); |
7a81b88c | 979 | return; |
52d4b9ac | 980 | } |
8a9f3ccd | 981 | pc->mem_cgroup = mem; |
08e552c6 | 982 | smp_wmb(); |
c05555b5 | 983 | pc->flags = pcg_default_flags[ctype]; |
3be91277 | 984 | |
08e552c6 | 985 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 986 | |
52d4b9ac | 987 | unlock_page_cgroup(pc); |
7a81b88c | 988 | } |
66e1707b | 989 | |
f817ed48 KH |
990 | /** |
991 | * mem_cgroup_move_account - move account of the page | |
992 | * @pc: page_cgroup of the page. | |
993 | * @from: mem_cgroup which the page is moved from. | |
994 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
995 | * | |
996 | * The caller must confirm following. | |
08e552c6 | 997 | * - page is not on LRU (isolate_page() is useful.) |
f817ed48 KH |
998 | * |
999 | * returns 0 at success, | |
1000 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
1001 | * | |
1002 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
1003 | * new cgroup. It should be done by a caller. | |
1004 | */ | |
1005 | ||
1006 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
1007 | struct mem_cgroup *from, struct mem_cgroup *to) | |
1008 | { | |
1009 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
1010 | int nid, zid; | |
1011 | int ret = -EBUSY; | |
1012 | ||
f817ed48 | 1013 | VM_BUG_ON(from == to); |
08e552c6 | 1014 | VM_BUG_ON(PageLRU(pc->page)); |
f817ed48 KH |
1015 | |
1016 | nid = page_cgroup_nid(pc); | |
1017 | zid = page_cgroup_zid(pc); | |
1018 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
1019 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
1020 | ||
f817ed48 KH |
1021 | if (!trylock_page_cgroup(pc)) |
1022 | return ret; | |
1023 | ||
1024 | if (!PageCgroupUsed(pc)) | |
1025 | goto out; | |
1026 | ||
1027 | if (pc->mem_cgroup != from) | |
1028 | goto out; | |
1029 | ||
08e552c6 KH |
1030 | res_counter_uncharge(&from->res, PAGE_SIZE); |
1031 | mem_cgroup_charge_statistics(from, pc, false); | |
1032 | if (do_swap_account) | |
1033 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
40d58138 DN |
1034 | css_put(&from->css); |
1035 | ||
1036 | css_get(&to->css); | |
08e552c6 KH |
1037 | pc->mem_cgroup = to; |
1038 | mem_cgroup_charge_statistics(to, pc, true); | |
08e552c6 | 1039 | ret = 0; |
f817ed48 KH |
1040 | out: |
1041 | unlock_page_cgroup(pc); | |
1042 | return ret; | |
1043 | } | |
1044 | ||
1045 | /* | |
1046 | * move charges to its parent. | |
1047 | */ | |
1048 | ||
1049 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1050 | struct mem_cgroup *child, | |
1051 | gfp_t gfp_mask) | |
1052 | { | |
08e552c6 | 1053 | struct page *page = pc->page; |
f817ed48 KH |
1054 | struct cgroup *cg = child->css.cgroup; |
1055 | struct cgroup *pcg = cg->parent; | |
1056 | struct mem_cgroup *parent; | |
f817ed48 KH |
1057 | int ret; |
1058 | ||
1059 | /* Is ROOT ? */ | |
1060 | if (!pcg) | |
1061 | return -EINVAL; | |
1062 | ||
08e552c6 | 1063 | |
f817ed48 KH |
1064 | parent = mem_cgroup_from_cont(pcg); |
1065 | ||
08e552c6 | 1066 | |
f817ed48 | 1067 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); |
a636b327 | 1068 | if (ret || !parent) |
f817ed48 KH |
1069 | return ret; |
1070 | ||
40d58138 DN |
1071 | if (!get_page_unless_zero(page)) { |
1072 | ret = -EBUSY; | |
1073 | goto uncharge; | |
1074 | } | |
08e552c6 KH |
1075 | |
1076 | ret = isolate_lru_page(page); | |
1077 | ||
1078 | if (ret) | |
1079 | goto cancel; | |
f817ed48 | 1080 | |
f817ed48 | 1081 | ret = mem_cgroup_move_account(pc, child, parent); |
f817ed48 | 1082 | |
08e552c6 KH |
1083 | putback_lru_page(page); |
1084 | if (!ret) { | |
1085 | put_page(page); | |
40d58138 DN |
1086 | /* drop extra refcnt by try_charge() */ |
1087 | css_put(&parent->css); | |
08e552c6 | 1088 | return 0; |
8c7c6e34 | 1089 | } |
40d58138 | 1090 | |
08e552c6 | 1091 | cancel: |
40d58138 DN |
1092 | put_page(page); |
1093 | uncharge: | |
1094 | /* drop extra refcnt by try_charge() */ | |
1095 | css_put(&parent->css); | |
1096 | /* uncharge if move fails */ | |
08e552c6 KH |
1097 | res_counter_uncharge(&parent->res, PAGE_SIZE); |
1098 | if (do_swap_account) | |
1099 | res_counter_uncharge(&parent->memsw, PAGE_SIZE); | |
f817ed48 KH |
1100 | return ret; |
1101 | } | |
1102 | ||
7a81b88c KH |
1103 | /* |
1104 | * Charge the memory controller for page usage. | |
1105 | * Return | |
1106 | * 0 if the charge was successful | |
1107 | * < 0 if the cgroup is over its limit | |
1108 | */ | |
1109 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1110 | gfp_t gfp_mask, enum charge_type ctype, | |
1111 | struct mem_cgroup *memcg) | |
1112 | { | |
1113 | struct mem_cgroup *mem; | |
1114 | struct page_cgroup *pc; | |
1115 | int ret; | |
1116 | ||
1117 | pc = lookup_page_cgroup(page); | |
1118 | /* can happen at boot */ | |
1119 | if (unlikely(!pc)) | |
1120 | return 0; | |
1121 | prefetchw(pc); | |
1122 | ||
1123 | mem = memcg; | |
f817ed48 | 1124 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
a636b327 | 1125 | if (ret || !mem) |
7a81b88c KH |
1126 | return ret; |
1127 | ||
1128 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1129 | return 0; |
8a9f3ccd BS |
1130 | } |
1131 | ||
7a81b88c KH |
1132 | int mem_cgroup_newpage_charge(struct page *page, |
1133 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1134 | { |
f8d66542 | 1135 | if (mem_cgroup_disabled()) |
cede86ac | 1136 | return 0; |
52d4b9ac KH |
1137 | if (PageCompound(page)) |
1138 | return 0; | |
69029cd5 KH |
1139 | /* |
1140 | * If already mapped, we don't have to account. | |
1141 | * If page cache, page->mapping has address_space. | |
1142 | * But page->mapping may have out-of-use anon_vma pointer, | |
1143 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1144 | * is NULL. | |
1145 | */ | |
1146 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
1147 | return 0; | |
1148 | if (unlikely(!mm)) | |
1149 | mm = &init_mm; | |
217bc319 | 1150 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 1151 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
1152 | } |
1153 | ||
e1a1cd59 BS |
1154 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
1155 | gfp_t gfp_mask) | |
8697d331 | 1156 | { |
b5a84319 KH |
1157 | struct mem_cgroup *mem = NULL; |
1158 | int ret; | |
1159 | ||
f8d66542 | 1160 | if (mem_cgroup_disabled()) |
cede86ac | 1161 | return 0; |
52d4b9ac KH |
1162 | if (PageCompound(page)) |
1163 | return 0; | |
accf163e KH |
1164 | /* |
1165 | * Corner case handling. This is called from add_to_page_cache() | |
1166 | * in usual. But some FS (shmem) precharges this page before calling it | |
1167 | * and call add_to_page_cache() with GFP_NOWAIT. | |
1168 | * | |
1169 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
1170 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
1171 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
1172 | * And when the page is SwapCache, it should take swap information |
1173 | * into account. This is under lock_page() now. | |
accf163e KH |
1174 | */ |
1175 | if (!(gfp_mask & __GFP_WAIT)) { | |
1176 | struct page_cgroup *pc; | |
1177 | ||
52d4b9ac KH |
1178 | |
1179 | pc = lookup_page_cgroup(page); | |
1180 | if (!pc) | |
1181 | return 0; | |
1182 | lock_page_cgroup(pc); | |
1183 | if (PageCgroupUsed(pc)) { | |
1184 | unlock_page_cgroup(pc); | |
accf163e KH |
1185 | return 0; |
1186 | } | |
52d4b9ac | 1187 | unlock_page_cgroup(pc); |
accf163e KH |
1188 | } |
1189 | ||
b5a84319 KH |
1190 | if (do_swap_account && PageSwapCache(page)) { |
1191 | mem = try_get_mem_cgroup_from_swapcache(page); | |
1192 | if (mem) | |
1193 | mm = NULL; | |
1194 | else | |
1195 | mem = NULL; | |
1196 | /* SwapCache may be still linked to LRU now. */ | |
1197 | mem_cgroup_lru_del_before_commit_swapcache(page); | |
1198 | } | |
1199 | ||
1200 | if (unlikely(!mm && !mem)) | |
8697d331 | 1201 | mm = &init_mm; |
accf163e | 1202 | |
c05555b5 KH |
1203 | if (page_is_file_cache(page)) |
1204 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 1205 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 KH |
1206 | |
1207 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
1208 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
1209 | if (mem) | |
1210 | css_put(&mem->css); | |
1211 | if (PageSwapCache(page)) | |
1212 | mem_cgroup_lru_add_after_commit_swapcache(page); | |
1213 | ||
1214 | if (do_swap_account && !ret && PageSwapCache(page)) { | |
1215 | swp_entry_t ent = {.val = page_private(page)}; | |
1216 | /* avoid double counting */ | |
1217 | mem = swap_cgroup_record(ent, NULL); | |
1218 | if (mem) { | |
1219 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1220 | mem_cgroup_put(mem); | |
1221 | } | |
1222 | } | |
1223 | return ret; | |
e8589cc1 KH |
1224 | } |
1225 | ||
54595fe2 KH |
1226 | /* |
1227 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
1228 | * And when try_charge() successfully returns, one refcnt to memcg without | |
1229 | * struct page_cgroup is aquired. This refcnt will be cumsumed by | |
1230 | * "commit()" or removed by "cancel()" | |
1231 | */ | |
8c7c6e34 KH |
1232 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
1233 | struct page *page, | |
1234 | gfp_t mask, struct mem_cgroup **ptr) | |
1235 | { | |
1236 | struct mem_cgroup *mem; | |
54595fe2 | 1237 | int ret; |
8c7c6e34 | 1238 | |
f8d66542 | 1239 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
1240 | return 0; |
1241 | ||
1242 | if (!do_swap_account) | |
1243 | goto charge_cur_mm; | |
8c7c6e34 KH |
1244 | /* |
1245 | * A racing thread's fault, or swapoff, may have already updated | |
1246 | * the pte, and even removed page from swap cache: return success | |
1247 | * to go on to do_swap_page()'s pte_same() test, which should fail. | |
1248 | */ | |
1249 | if (!PageSwapCache(page)) | |
1250 | return 0; | |
b5a84319 | 1251 | mem = try_get_mem_cgroup_from_swapcache(page); |
54595fe2 KH |
1252 | if (!mem) |
1253 | goto charge_cur_mm; | |
8c7c6e34 | 1254 | *ptr = mem; |
54595fe2 KH |
1255 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true); |
1256 | /* drop extra refcnt from tryget */ | |
1257 | css_put(&mem->css); | |
1258 | return ret; | |
8c7c6e34 KH |
1259 | charge_cur_mm: |
1260 | if (unlikely(!mm)) | |
1261 | mm = &init_mm; | |
1262 | return __mem_cgroup_try_charge(mm, mask, ptr, true); | |
1263 | } | |
1264 | ||
7a81b88c KH |
1265 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
1266 | { | |
1267 | struct page_cgroup *pc; | |
1268 | ||
f8d66542 | 1269 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1270 | return; |
1271 | if (!ptr) | |
1272 | return; | |
1273 | pc = lookup_page_cgroup(page); | |
544122e5 | 1274 | mem_cgroup_lru_del_before_commit_swapcache(page); |
7a81b88c | 1275 | __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
544122e5 | 1276 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
1277 | /* |
1278 | * Now swap is on-memory. This means this page may be | |
1279 | * counted both as mem and swap....double count. | |
03f3c433 KH |
1280 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
1281 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
1282 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 1283 | */ |
03f3c433 | 1284 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 KH |
1285 | swp_entry_t ent = {.val = page_private(page)}; |
1286 | struct mem_cgroup *memcg; | |
1287 | memcg = swap_cgroup_record(ent, NULL); | |
1288 | if (memcg) { | |
8c7c6e34 KH |
1289 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
1290 | mem_cgroup_put(memcg); | |
1291 | } | |
1292 | ||
1293 | } | |
08e552c6 | 1294 | /* add this page(page_cgroup) to the LRU we want. */ |
544122e5 | 1295 | |
7a81b88c KH |
1296 | } |
1297 | ||
1298 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
1299 | { | |
f8d66542 | 1300 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1301 | return; |
1302 | if (!mem) | |
1303 | return; | |
1304 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
1305 | if (do_swap_account) |
1306 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
7a81b88c KH |
1307 | css_put(&mem->css); |
1308 | } | |
1309 | ||
1310 | ||
8a9f3ccd | 1311 | /* |
69029cd5 | 1312 | * uncharge if !page_mapped(page) |
8a9f3ccd | 1313 | */ |
8c7c6e34 | 1314 | static struct mem_cgroup * |
69029cd5 | 1315 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 1316 | { |
8289546e | 1317 | struct page_cgroup *pc; |
8c7c6e34 | 1318 | struct mem_cgroup *mem = NULL; |
072c56c1 | 1319 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 1320 | |
f8d66542 | 1321 | if (mem_cgroup_disabled()) |
8c7c6e34 | 1322 | return NULL; |
4077960e | 1323 | |
d13d1443 | 1324 | if (PageSwapCache(page)) |
8c7c6e34 | 1325 | return NULL; |
d13d1443 | 1326 | |
8697d331 | 1327 | /* |
3c541e14 | 1328 | * Check if our page_cgroup is valid |
8697d331 | 1329 | */ |
52d4b9ac KH |
1330 | pc = lookup_page_cgroup(page); |
1331 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 1332 | return NULL; |
b9c565d5 | 1333 | |
52d4b9ac | 1334 | lock_page_cgroup(pc); |
d13d1443 | 1335 | |
8c7c6e34 KH |
1336 | mem = pc->mem_cgroup; |
1337 | ||
d13d1443 KH |
1338 | if (!PageCgroupUsed(pc)) |
1339 | goto unlock_out; | |
1340 | ||
1341 | switch (ctype) { | |
1342 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1343 | if (page_mapped(page)) | |
1344 | goto unlock_out; | |
1345 | break; | |
1346 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
1347 | if (!PageAnon(page)) { /* Shared memory */ | |
1348 | if (page->mapping && !page_is_file_cache(page)) | |
1349 | goto unlock_out; | |
1350 | } else if (page_mapped(page)) /* Anon */ | |
1351 | goto unlock_out; | |
1352 | break; | |
1353 | default: | |
1354 | break; | |
52d4b9ac | 1355 | } |
d13d1443 | 1356 | |
8c7c6e34 KH |
1357 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1358 | if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)) | |
1359 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
08e552c6 | 1360 | mem_cgroup_charge_statistics(mem, pc, false); |
04046e1a | 1361 | |
52d4b9ac | 1362 | ClearPageCgroupUsed(pc); |
544122e5 KH |
1363 | /* |
1364 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
1365 | * freed from LRU. This is safe because uncharged page is expected not | |
1366 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
1367 | * special functions. | |
1368 | */ | |
b9c565d5 | 1369 | |
69029cd5 | 1370 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 1371 | unlock_page_cgroup(pc); |
fb59e9f1 | 1372 | |
a7fe942e KH |
1373 | /* at swapout, this memcg will be accessed to record to swap */ |
1374 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
1375 | css_put(&mem->css); | |
6d12e2d8 | 1376 | |
8c7c6e34 | 1377 | return mem; |
d13d1443 KH |
1378 | |
1379 | unlock_out: | |
1380 | unlock_page_cgroup(pc); | |
8c7c6e34 | 1381 | return NULL; |
3c541e14 BS |
1382 | } |
1383 | ||
69029cd5 KH |
1384 | void mem_cgroup_uncharge_page(struct page *page) |
1385 | { | |
52d4b9ac KH |
1386 | /* early check. */ |
1387 | if (page_mapped(page)) | |
1388 | return; | |
1389 | if (page->mapping && !PageAnon(page)) | |
1390 | return; | |
69029cd5 KH |
1391 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
1392 | } | |
1393 | ||
1394 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
1395 | { | |
1396 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 1397 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
1398 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
1399 | } | |
1400 | ||
8c7c6e34 KH |
1401 | /* |
1402 | * called from __delete_from_swap_cache() and drop "page" account. | |
1403 | * memcg information is recorded to swap_cgroup of "ent" | |
1404 | */ | |
1405 | void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent) | |
1406 | { | |
1407 | struct mem_cgroup *memcg; | |
1408 | ||
1409 | memcg = __mem_cgroup_uncharge_common(page, | |
1410 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT); | |
1411 | /* record memcg information */ | |
1412 | if (do_swap_account && memcg) { | |
1413 | swap_cgroup_record(ent, memcg); | |
1414 | mem_cgroup_get(memcg); | |
1415 | } | |
a7fe942e KH |
1416 | if (memcg) |
1417 | css_put(&memcg->css); | |
8c7c6e34 KH |
1418 | } |
1419 | ||
1420 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1421 | /* | |
1422 | * called from swap_entry_free(). remove record in swap_cgroup and | |
1423 | * uncharge "memsw" account. | |
1424 | */ | |
1425 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 1426 | { |
8c7c6e34 KH |
1427 | struct mem_cgroup *memcg; |
1428 | ||
1429 | if (!do_swap_account) | |
1430 | return; | |
1431 | ||
1432 | memcg = swap_cgroup_record(ent, NULL); | |
1433 | if (memcg) { | |
1434 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
1435 | mem_cgroup_put(memcg); | |
1436 | } | |
d13d1443 | 1437 | } |
8c7c6e34 | 1438 | #endif |
d13d1443 | 1439 | |
ae41be37 | 1440 | /* |
01b1ae63 KH |
1441 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
1442 | * page belongs to. | |
ae41be37 | 1443 | */ |
01b1ae63 | 1444 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
1445 | { |
1446 | struct page_cgroup *pc; | |
e8589cc1 | 1447 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 1448 | int ret = 0; |
8869b8f6 | 1449 | |
f8d66542 | 1450 | if (mem_cgroup_disabled()) |
4077960e BS |
1451 | return 0; |
1452 | ||
52d4b9ac KH |
1453 | pc = lookup_page_cgroup(page); |
1454 | lock_page_cgroup(pc); | |
1455 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
1456 | mem = pc->mem_cgroup; |
1457 | css_get(&mem->css); | |
e8589cc1 | 1458 | } |
52d4b9ac | 1459 | unlock_page_cgroup(pc); |
01b1ae63 | 1460 | |
e8589cc1 | 1461 | if (mem) { |
3bb4edf2 | 1462 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); |
e8589cc1 KH |
1463 | css_put(&mem->css); |
1464 | } | |
01b1ae63 | 1465 | *ptr = mem; |
e8589cc1 | 1466 | return ret; |
ae41be37 | 1467 | } |
8869b8f6 | 1468 | |
69029cd5 | 1469 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
1470 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
1471 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 1472 | { |
01b1ae63 KH |
1473 | struct page *target, *unused; |
1474 | struct page_cgroup *pc; | |
1475 | enum charge_type ctype; | |
1476 | ||
1477 | if (!mem) | |
1478 | return; | |
1479 | ||
1480 | /* at migration success, oldpage->mapping is NULL. */ | |
1481 | if (oldpage->mapping) { | |
1482 | target = oldpage; | |
1483 | unused = NULL; | |
1484 | } else { | |
1485 | target = newpage; | |
1486 | unused = oldpage; | |
1487 | } | |
1488 | ||
1489 | if (PageAnon(target)) | |
1490 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
1491 | else if (page_is_file_cache(target)) | |
1492 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
1493 | else | |
1494 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
1495 | ||
1496 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 1497 | if (unused) |
01b1ae63 KH |
1498 | __mem_cgroup_uncharge_common(unused, ctype); |
1499 | ||
1500 | pc = lookup_page_cgroup(target); | |
69029cd5 | 1501 | /* |
01b1ae63 KH |
1502 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
1503 | * So, double-counting is effectively avoided. | |
1504 | */ | |
1505 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
1506 | ||
1507 | /* | |
1508 | * Both of oldpage and newpage are still under lock_page(). | |
1509 | * Then, we don't have to care about race in radix-tree. | |
1510 | * But we have to be careful that this page is unmapped or not. | |
1511 | * | |
1512 | * There is a case for !page_mapped(). At the start of | |
1513 | * migration, oldpage was mapped. But now, it's zapped. | |
1514 | * But we know *target* page is not freed/reused under us. | |
1515 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 1516 | */ |
01b1ae63 KH |
1517 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
1518 | mem_cgroup_uncharge_page(target); | |
ae41be37 | 1519 | } |
78fb7466 | 1520 | |
c9b0ed51 KH |
1521 | /* |
1522 | * A call to try to shrink memory usage under specified resource controller. | |
1523 | * This is typically used for page reclaiming for shmem for reducing side | |
1524 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
1525 | */ | |
b5a84319 KH |
1526 | int mem_cgroup_shrink_usage(struct page *page, |
1527 | struct mm_struct *mm, | |
1528 | gfp_t gfp_mask) | |
c9b0ed51 | 1529 | { |
b5a84319 | 1530 | struct mem_cgroup *mem = NULL; |
c9b0ed51 KH |
1531 | int progress = 0; |
1532 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
1533 | ||
f8d66542 | 1534 | if (mem_cgroup_disabled()) |
cede86ac | 1535 | return 0; |
b5a84319 KH |
1536 | if (page) |
1537 | mem = try_get_mem_cgroup_from_swapcache(page); | |
1538 | if (!mem && mm) | |
1539 | mem = try_get_mem_cgroup_from_mm(mm); | |
54595fe2 | 1540 | if (unlikely(!mem)) |
31a78f23 | 1541 | return 0; |
c9b0ed51 KH |
1542 | |
1543 | do { | |
81d39c20 KH |
1544 | progress = mem_cgroup_hierarchical_reclaim(mem, |
1545 | gfp_mask, true, false); | |
b85a96c0 | 1546 | progress += mem_cgroup_check_under_limit(mem); |
c9b0ed51 KH |
1547 | } while (!progress && --retry); |
1548 | ||
1549 | css_put(&mem->css); | |
1550 | if (!retry) | |
1551 | return -ENOMEM; | |
1552 | return 0; | |
1553 | } | |
1554 | ||
8c7c6e34 KH |
1555 | static DEFINE_MUTEX(set_limit_mutex); |
1556 | ||
d38d2a75 | 1557 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 1558 | unsigned long long val) |
628f4235 | 1559 | { |
81d39c20 | 1560 | int retry_count; |
628f4235 | 1561 | int progress; |
8c7c6e34 | 1562 | u64 memswlimit; |
628f4235 | 1563 | int ret = 0; |
81d39c20 KH |
1564 | int children = mem_cgroup_count_children(memcg); |
1565 | u64 curusage, oldusage; | |
1566 | ||
1567 | /* | |
1568 | * For keeping hierarchical_reclaim simple, how long we should retry | |
1569 | * is depends on callers. We set our retry-count to be function | |
1570 | * of # of children which we should visit in this loop. | |
1571 | */ | |
1572 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
1573 | ||
1574 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 1575 | |
8c7c6e34 | 1576 | while (retry_count) { |
628f4235 KH |
1577 | if (signal_pending(current)) { |
1578 | ret = -EINTR; | |
1579 | break; | |
1580 | } | |
8c7c6e34 KH |
1581 | /* |
1582 | * Rather than hide all in some function, I do this in | |
1583 | * open coded manner. You see what this really does. | |
1584 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1585 | */ | |
1586 | mutex_lock(&set_limit_mutex); | |
1587 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1588 | if (memswlimit < val) { | |
1589 | ret = -EINVAL; | |
1590 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
1591 | break; |
1592 | } | |
8c7c6e34 KH |
1593 | ret = res_counter_set_limit(&memcg->res, val); |
1594 | mutex_unlock(&set_limit_mutex); | |
1595 | ||
1596 | if (!ret) | |
1597 | break; | |
1598 | ||
42e9abb6 | 1599 | progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, |
81d39c20 KH |
1600 | false, true); |
1601 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
1602 | /* Usage is reduced ? */ | |
1603 | if (curusage >= oldusage) | |
1604 | retry_count--; | |
1605 | else | |
1606 | oldusage = curusage; | |
8c7c6e34 | 1607 | } |
14797e23 | 1608 | |
8c7c6e34 KH |
1609 | return ret; |
1610 | } | |
1611 | ||
1612 | int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, | |
1613 | unsigned long long val) | |
1614 | { | |
81d39c20 | 1615 | int retry_count; |
8c7c6e34 | 1616 | u64 memlimit, oldusage, curusage; |
81d39c20 KH |
1617 | int children = mem_cgroup_count_children(memcg); |
1618 | int ret = -EBUSY; | |
8c7c6e34 KH |
1619 | |
1620 | if (!do_swap_account) | |
1621 | return -EINVAL; | |
81d39c20 KH |
1622 | /* see mem_cgroup_resize_res_limit */ |
1623 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
1624 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
1625 | while (retry_count) { |
1626 | if (signal_pending(current)) { | |
1627 | ret = -EINTR; | |
1628 | break; | |
1629 | } | |
1630 | /* | |
1631 | * Rather than hide all in some function, I do this in | |
1632 | * open coded manner. You see what this really does. | |
1633 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1634 | */ | |
1635 | mutex_lock(&set_limit_mutex); | |
1636 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1637 | if (memlimit > val) { | |
1638 | ret = -EINVAL; | |
1639 | mutex_unlock(&set_limit_mutex); | |
1640 | break; | |
1641 | } | |
1642 | ret = res_counter_set_limit(&memcg->memsw, val); | |
1643 | mutex_unlock(&set_limit_mutex); | |
1644 | ||
1645 | if (!ret) | |
1646 | break; | |
1647 | ||
81d39c20 | 1648 | mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true, true); |
8c7c6e34 | 1649 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 1650 | /* Usage is reduced ? */ |
8c7c6e34 | 1651 | if (curusage >= oldusage) |
628f4235 | 1652 | retry_count--; |
81d39c20 KH |
1653 | else |
1654 | oldusage = curusage; | |
628f4235 KH |
1655 | } |
1656 | return ret; | |
1657 | } | |
1658 | ||
cc847582 KH |
1659 | /* |
1660 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
1661 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
1662 | */ | |
f817ed48 | 1663 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 1664 | int node, int zid, enum lru_list lru) |
cc847582 | 1665 | { |
08e552c6 KH |
1666 | struct zone *zone; |
1667 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 1668 | struct page_cgroup *pc, *busy; |
08e552c6 | 1669 | unsigned long flags, loop; |
072c56c1 | 1670 | struct list_head *list; |
f817ed48 | 1671 | int ret = 0; |
072c56c1 | 1672 | |
08e552c6 KH |
1673 | zone = &NODE_DATA(node)->node_zones[zid]; |
1674 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 1675 | list = &mz->lists[lru]; |
cc847582 | 1676 | |
f817ed48 KH |
1677 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
1678 | /* give some margin against EBUSY etc...*/ | |
1679 | loop += 256; | |
1680 | busy = NULL; | |
1681 | while (loop--) { | |
1682 | ret = 0; | |
08e552c6 | 1683 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 1684 | if (list_empty(list)) { |
08e552c6 | 1685 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 1686 | break; |
f817ed48 KH |
1687 | } |
1688 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
1689 | if (busy == pc) { | |
1690 | list_move(&pc->lru, list); | |
1691 | busy = 0; | |
08e552c6 | 1692 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
1693 | continue; |
1694 | } | |
08e552c6 | 1695 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 1696 | |
2c26fdd7 | 1697 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 1698 | if (ret == -ENOMEM) |
52d4b9ac | 1699 | break; |
f817ed48 KH |
1700 | |
1701 | if (ret == -EBUSY || ret == -EINVAL) { | |
1702 | /* found lock contention or "pc" is obsolete. */ | |
1703 | busy = pc; | |
1704 | cond_resched(); | |
1705 | } else | |
1706 | busy = NULL; | |
cc847582 | 1707 | } |
08e552c6 | 1708 | |
f817ed48 KH |
1709 | if (!ret && !list_empty(list)) |
1710 | return -EBUSY; | |
1711 | return ret; | |
cc847582 KH |
1712 | } |
1713 | ||
1714 | /* | |
1715 | * make mem_cgroup's charge to be 0 if there is no task. | |
1716 | * This enables deleting this mem_cgroup. | |
1717 | */ | |
c1e862c1 | 1718 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 1719 | { |
f817ed48 KH |
1720 | int ret; |
1721 | int node, zid, shrink; | |
1722 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 1723 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 1724 | |
cc847582 | 1725 | css_get(&mem->css); |
f817ed48 KH |
1726 | |
1727 | shrink = 0; | |
c1e862c1 KH |
1728 | /* should free all ? */ |
1729 | if (free_all) | |
1730 | goto try_to_free; | |
f817ed48 | 1731 | move_account: |
1ecaab2b | 1732 | while (mem->res.usage > 0) { |
f817ed48 | 1733 | ret = -EBUSY; |
c1e862c1 KH |
1734 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
1735 | goto out; | |
1736 | ret = -EINTR; | |
1737 | if (signal_pending(current)) | |
cc847582 | 1738 | goto out; |
52d4b9ac KH |
1739 | /* This is for making all *used* pages to be on LRU. */ |
1740 | lru_add_drain_all(); | |
f817ed48 | 1741 | ret = 0; |
299b4eaa | 1742 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 1743 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 1744 | enum lru_list l; |
f817ed48 KH |
1745 | for_each_lru(l) { |
1746 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 1747 | node, zid, l); |
f817ed48 KH |
1748 | if (ret) |
1749 | break; | |
1750 | } | |
1ecaab2b | 1751 | } |
f817ed48 KH |
1752 | if (ret) |
1753 | break; | |
1754 | } | |
1755 | /* it seems parent cgroup doesn't have enough mem */ | |
1756 | if (ret == -ENOMEM) | |
1757 | goto try_to_free; | |
52d4b9ac | 1758 | cond_resched(); |
cc847582 KH |
1759 | } |
1760 | ret = 0; | |
1761 | out: | |
1762 | css_put(&mem->css); | |
1763 | return ret; | |
f817ed48 KH |
1764 | |
1765 | try_to_free: | |
c1e862c1 KH |
1766 | /* returns EBUSY if there is a task or if we come here twice. */ |
1767 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
1768 | ret = -EBUSY; |
1769 | goto out; | |
1770 | } | |
c1e862c1 KH |
1771 | /* we call try-to-free pages for make this cgroup empty */ |
1772 | lru_add_drain_all(); | |
f817ed48 KH |
1773 | /* try to free all pages in this cgroup */ |
1774 | shrink = 1; | |
1775 | while (nr_retries && mem->res.usage > 0) { | |
1776 | int progress; | |
c1e862c1 KH |
1777 | |
1778 | if (signal_pending(current)) { | |
1779 | ret = -EINTR; | |
1780 | goto out; | |
1781 | } | |
a7885eb8 KM |
1782 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
1783 | false, get_swappiness(mem)); | |
c1e862c1 | 1784 | if (!progress) { |
f817ed48 | 1785 | nr_retries--; |
c1e862c1 KH |
1786 | /* maybe some writeback is necessary */ |
1787 | congestion_wait(WRITE, HZ/10); | |
1788 | } | |
f817ed48 KH |
1789 | |
1790 | } | |
08e552c6 | 1791 | lru_add_drain(); |
f817ed48 KH |
1792 | /* try move_account...there may be some *locked* pages. */ |
1793 | if (mem->res.usage) | |
1794 | goto move_account; | |
1795 | ret = 0; | |
1796 | goto out; | |
cc847582 KH |
1797 | } |
1798 | ||
c1e862c1 KH |
1799 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
1800 | { | |
1801 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
1802 | } | |
1803 | ||
1804 | ||
18f59ea7 BS |
1805 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
1806 | { | |
1807 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
1808 | } | |
1809 | ||
1810 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
1811 | u64 val) | |
1812 | { | |
1813 | int retval = 0; | |
1814 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1815 | struct cgroup *parent = cont->parent; | |
1816 | struct mem_cgroup *parent_mem = NULL; | |
1817 | ||
1818 | if (parent) | |
1819 | parent_mem = mem_cgroup_from_cont(parent); | |
1820 | ||
1821 | cgroup_lock(); | |
1822 | /* | |
1823 | * If parent's use_hiearchy is set, we can't make any modifications | |
1824 | * in the child subtrees. If it is unset, then the change can | |
1825 | * occur, provided the current cgroup has no children. | |
1826 | * | |
1827 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
1828 | * set if there are no children. | |
1829 | */ | |
1830 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
1831 | (val == 1 || val == 0)) { | |
1832 | if (list_empty(&cont->children)) | |
1833 | mem->use_hierarchy = val; | |
1834 | else | |
1835 | retval = -EBUSY; | |
1836 | } else | |
1837 | retval = -EINVAL; | |
1838 | cgroup_unlock(); | |
1839 | ||
1840 | return retval; | |
1841 | } | |
1842 | ||
2c3daa72 | 1843 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 1844 | { |
8c7c6e34 KH |
1845 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
1846 | u64 val = 0; | |
1847 | int type, name; | |
1848 | ||
1849 | type = MEMFILE_TYPE(cft->private); | |
1850 | name = MEMFILE_ATTR(cft->private); | |
1851 | switch (type) { | |
1852 | case _MEM: | |
1853 | val = res_counter_read_u64(&mem->res, name); | |
1854 | break; | |
1855 | case _MEMSWAP: | |
1856 | if (do_swap_account) | |
1857 | val = res_counter_read_u64(&mem->memsw, name); | |
1858 | break; | |
1859 | default: | |
1860 | BUG(); | |
1861 | break; | |
1862 | } | |
1863 | return val; | |
8cdea7c0 | 1864 | } |
628f4235 KH |
1865 | /* |
1866 | * The user of this function is... | |
1867 | * RES_LIMIT. | |
1868 | */ | |
856c13aa PM |
1869 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
1870 | const char *buffer) | |
8cdea7c0 | 1871 | { |
628f4235 | 1872 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 1873 | int type, name; |
628f4235 KH |
1874 | unsigned long long val; |
1875 | int ret; | |
1876 | ||
8c7c6e34 KH |
1877 | type = MEMFILE_TYPE(cft->private); |
1878 | name = MEMFILE_ATTR(cft->private); | |
1879 | switch (name) { | |
628f4235 KH |
1880 | case RES_LIMIT: |
1881 | /* This function does all necessary parse...reuse it */ | |
1882 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
1883 | if (ret) |
1884 | break; | |
1885 | if (type == _MEM) | |
628f4235 | 1886 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
1887 | else |
1888 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 KH |
1889 | break; |
1890 | default: | |
1891 | ret = -EINVAL; /* should be BUG() ? */ | |
1892 | break; | |
1893 | } | |
1894 | return ret; | |
8cdea7c0 BS |
1895 | } |
1896 | ||
fee7b548 KH |
1897 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
1898 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
1899 | { | |
1900 | struct cgroup *cgroup; | |
1901 | unsigned long long min_limit, min_memsw_limit, tmp; | |
1902 | ||
1903 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1904 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1905 | cgroup = memcg->css.cgroup; | |
1906 | if (!memcg->use_hierarchy) | |
1907 | goto out; | |
1908 | ||
1909 | while (cgroup->parent) { | |
1910 | cgroup = cgroup->parent; | |
1911 | memcg = mem_cgroup_from_cont(cgroup); | |
1912 | if (!memcg->use_hierarchy) | |
1913 | break; | |
1914 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1915 | min_limit = min(min_limit, tmp); | |
1916 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1917 | min_memsw_limit = min(min_memsw_limit, tmp); | |
1918 | } | |
1919 | out: | |
1920 | *mem_limit = min_limit; | |
1921 | *memsw_limit = min_memsw_limit; | |
1922 | return; | |
1923 | } | |
1924 | ||
29f2a4da | 1925 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
1926 | { |
1927 | struct mem_cgroup *mem; | |
8c7c6e34 | 1928 | int type, name; |
c84872e1 PE |
1929 | |
1930 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
1931 | type = MEMFILE_TYPE(event); |
1932 | name = MEMFILE_ATTR(event); | |
1933 | switch (name) { | |
29f2a4da | 1934 | case RES_MAX_USAGE: |
8c7c6e34 KH |
1935 | if (type == _MEM) |
1936 | res_counter_reset_max(&mem->res); | |
1937 | else | |
1938 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
1939 | break; |
1940 | case RES_FAILCNT: | |
8c7c6e34 KH |
1941 | if (type == _MEM) |
1942 | res_counter_reset_failcnt(&mem->res); | |
1943 | else | |
1944 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
1945 | break; |
1946 | } | |
85cc59db | 1947 | return 0; |
c84872e1 PE |
1948 | } |
1949 | ||
14067bb3 KH |
1950 | |
1951 | /* For read statistics */ | |
1952 | enum { | |
1953 | MCS_CACHE, | |
1954 | MCS_RSS, | |
1955 | MCS_PGPGIN, | |
1956 | MCS_PGPGOUT, | |
1957 | MCS_INACTIVE_ANON, | |
1958 | MCS_ACTIVE_ANON, | |
1959 | MCS_INACTIVE_FILE, | |
1960 | MCS_ACTIVE_FILE, | |
1961 | MCS_UNEVICTABLE, | |
1962 | NR_MCS_STAT, | |
1963 | }; | |
1964 | ||
1965 | struct mcs_total_stat { | |
1966 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
1967 | }; |
1968 | ||
14067bb3 KH |
1969 | struct { |
1970 | char *local_name; | |
1971 | char *total_name; | |
1972 | } memcg_stat_strings[NR_MCS_STAT] = { | |
1973 | {"cache", "total_cache"}, | |
1974 | {"rss", "total_rss"}, | |
1975 | {"pgpgin", "total_pgpgin"}, | |
1976 | {"pgpgout", "total_pgpgout"}, | |
1977 | {"inactive_anon", "total_inactive_anon"}, | |
1978 | {"active_anon", "total_active_anon"}, | |
1979 | {"inactive_file", "total_inactive_file"}, | |
1980 | {"active_file", "total_active_file"}, | |
1981 | {"unevictable", "total_unevictable"} | |
1982 | }; | |
1983 | ||
1984 | ||
1985 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
1986 | { | |
1987 | struct mcs_total_stat *s = data; | |
1988 | s64 val; | |
1989 | ||
1990 | /* per cpu stat */ | |
1991 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE); | |
1992 | s->stat[MCS_CACHE] += val * PAGE_SIZE; | |
1993 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
1994 | s->stat[MCS_RSS] += val * PAGE_SIZE; | |
1995 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT); | |
1996 | s->stat[MCS_PGPGIN] += val; | |
1997 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT); | |
1998 | s->stat[MCS_PGPGOUT] += val; | |
1999 | ||
2000 | /* per zone stat */ | |
2001 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
2002 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
2003 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
2004 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
2005 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
2006 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
2007 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
2008 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
2009 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
2010 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
2011 | return 0; | |
2012 | } | |
2013 | ||
2014 | static void | |
2015 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
2016 | { | |
2017 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
2018 | } | |
2019 | ||
c64745cf PM |
2020 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
2021 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 2022 | { |
d2ceb9b7 | 2023 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 2024 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
2025 | int i; |
2026 | ||
14067bb3 KH |
2027 | memset(&mystat, 0, sizeof(mystat)); |
2028 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 2029 | |
14067bb3 KH |
2030 | for (i = 0; i < NR_MCS_STAT; i++) |
2031 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); | |
7b854121 | 2032 | |
14067bb3 | 2033 | /* Hierarchical information */ |
fee7b548 KH |
2034 | { |
2035 | unsigned long long limit, memsw_limit; | |
2036 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
2037 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
2038 | if (do_swap_account) | |
2039 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
2040 | } | |
7f016ee8 | 2041 | |
14067bb3 KH |
2042 | memset(&mystat, 0, sizeof(mystat)); |
2043 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
2044 | for (i = 0; i < NR_MCS_STAT; i++) | |
2045 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); | |
2046 | ||
2047 | ||
7f016ee8 | 2048 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 2049 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
2050 | |
2051 | { | |
2052 | int nid, zid; | |
2053 | struct mem_cgroup_per_zone *mz; | |
2054 | unsigned long recent_rotated[2] = {0, 0}; | |
2055 | unsigned long recent_scanned[2] = {0, 0}; | |
2056 | ||
2057 | for_each_online_node(nid) | |
2058 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
2059 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
2060 | ||
2061 | recent_rotated[0] += | |
2062 | mz->reclaim_stat.recent_rotated[0]; | |
2063 | recent_rotated[1] += | |
2064 | mz->reclaim_stat.recent_rotated[1]; | |
2065 | recent_scanned[0] += | |
2066 | mz->reclaim_stat.recent_scanned[0]; | |
2067 | recent_scanned[1] += | |
2068 | mz->reclaim_stat.recent_scanned[1]; | |
2069 | } | |
2070 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
2071 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
2072 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
2073 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
2074 | } | |
2075 | #endif | |
2076 | ||
d2ceb9b7 KH |
2077 | return 0; |
2078 | } | |
2079 | ||
a7885eb8 KM |
2080 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
2081 | { | |
2082 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2083 | ||
2084 | return get_swappiness(memcg); | |
2085 | } | |
2086 | ||
2087 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
2088 | u64 val) | |
2089 | { | |
2090 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2091 | struct mem_cgroup *parent; | |
068b38c1 | 2092 | |
a7885eb8 KM |
2093 | if (val > 100) |
2094 | return -EINVAL; | |
2095 | ||
2096 | if (cgrp->parent == NULL) | |
2097 | return -EINVAL; | |
2098 | ||
2099 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
2100 | |
2101 | cgroup_lock(); | |
2102 | ||
a7885eb8 KM |
2103 | /* If under hierarchy, only empty-root can set this value */ |
2104 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
2105 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
2106 | cgroup_unlock(); | |
a7885eb8 | 2107 | return -EINVAL; |
068b38c1 | 2108 | } |
a7885eb8 KM |
2109 | |
2110 | spin_lock(&memcg->reclaim_param_lock); | |
2111 | memcg->swappiness = val; | |
2112 | spin_unlock(&memcg->reclaim_param_lock); | |
2113 | ||
068b38c1 LZ |
2114 | cgroup_unlock(); |
2115 | ||
a7885eb8 KM |
2116 | return 0; |
2117 | } | |
2118 | ||
c1e862c1 | 2119 | |
8cdea7c0 BS |
2120 | static struct cftype mem_cgroup_files[] = { |
2121 | { | |
0eea1030 | 2122 | .name = "usage_in_bytes", |
8c7c6e34 | 2123 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 2124 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 2125 | }, |
c84872e1 PE |
2126 | { |
2127 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 2128 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 2129 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
2130 | .read_u64 = mem_cgroup_read, |
2131 | }, | |
8cdea7c0 | 2132 | { |
0eea1030 | 2133 | .name = "limit_in_bytes", |
8c7c6e34 | 2134 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 2135 | .write_string = mem_cgroup_write, |
2c3daa72 | 2136 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
2137 | }, |
2138 | { | |
2139 | .name = "failcnt", | |
8c7c6e34 | 2140 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 2141 | .trigger = mem_cgroup_reset, |
2c3daa72 | 2142 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 2143 | }, |
d2ceb9b7 KH |
2144 | { |
2145 | .name = "stat", | |
c64745cf | 2146 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 2147 | }, |
c1e862c1 KH |
2148 | { |
2149 | .name = "force_empty", | |
2150 | .trigger = mem_cgroup_force_empty_write, | |
2151 | }, | |
18f59ea7 BS |
2152 | { |
2153 | .name = "use_hierarchy", | |
2154 | .write_u64 = mem_cgroup_hierarchy_write, | |
2155 | .read_u64 = mem_cgroup_hierarchy_read, | |
2156 | }, | |
a7885eb8 KM |
2157 | { |
2158 | .name = "swappiness", | |
2159 | .read_u64 = mem_cgroup_swappiness_read, | |
2160 | .write_u64 = mem_cgroup_swappiness_write, | |
2161 | }, | |
8cdea7c0 BS |
2162 | }; |
2163 | ||
8c7c6e34 KH |
2164 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
2165 | static struct cftype memsw_cgroup_files[] = { | |
2166 | { | |
2167 | .name = "memsw.usage_in_bytes", | |
2168 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
2169 | .read_u64 = mem_cgroup_read, | |
2170 | }, | |
2171 | { | |
2172 | .name = "memsw.max_usage_in_bytes", | |
2173 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
2174 | .trigger = mem_cgroup_reset, | |
2175 | .read_u64 = mem_cgroup_read, | |
2176 | }, | |
2177 | { | |
2178 | .name = "memsw.limit_in_bytes", | |
2179 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
2180 | .write_string = mem_cgroup_write, | |
2181 | .read_u64 = mem_cgroup_read, | |
2182 | }, | |
2183 | { | |
2184 | .name = "memsw.failcnt", | |
2185 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
2186 | .trigger = mem_cgroup_reset, | |
2187 | .read_u64 = mem_cgroup_read, | |
2188 | }, | |
2189 | }; | |
2190 | ||
2191 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
2192 | { | |
2193 | if (!do_swap_account) | |
2194 | return 0; | |
2195 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
2196 | ARRAY_SIZE(memsw_cgroup_files)); | |
2197 | }; | |
2198 | #else | |
2199 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
2200 | { | |
2201 | return 0; | |
2202 | } | |
2203 | #endif | |
2204 | ||
6d12e2d8 KH |
2205 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2206 | { | |
2207 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 2208 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 2209 | enum lru_list l; |
41e3355d | 2210 | int zone, tmp = node; |
1ecaab2b KH |
2211 | /* |
2212 | * This routine is called against possible nodes. | |
2213 | * But it's BUG to call kmalloc() against offline node. | |
2214 | * | |
2215 | * TODO: this routine can waste much memory for nodes which will | |
2216 | * never be onlined. It's better to use memory hotplug callback | |
2217 | * function. | |
2218 | */ | |
41e3355d KH |
2219 | if (!node_state(node, N_NORMAL_MEMORY)) |
2220 | tmp = -1; | |
2221 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
2222 | if (!pn) |
2223 | return 1; | |
1ecaab2b | 2224 | |
6d12e2d8 KH |
2225 | mem->info.nodeinfo[node] = pn; |
2226 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
2227 | |
2228 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
2229 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
2230 | for_each_lru(l) |
2231 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 2232 | } |
6d12e2d8 KH |
2233 | return 0; |
2234 | } | |
2235 | ||
1ecaab2b KH |
2236 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2237 | { | |
2238 | kfree(mem->info.nodeinfo[node]); | |
2239 | } | |
2240 | ||
c8dad2bb JB |
2241 | static int mem_cgroup_size(void) |
2242 | { | |
2243 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
2244 | return sizeof(struct mem_cgroup) + cpustat_size; | |
2245 | } | |
2246 | ||
33327948 KH |
2247 | static struct mem_cgroup *mem_cgroup_alloc(void) |
2248 | { | |
2249 | struct mem_cgroup *mem; | |
c8dad2bb | 2250 | int size = mem_cgroup_size(); |
33327948 | 2251 | |
c8dad2bb JB |
2252 | if (size < PAGE_SIZE) |
2253 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 2254 | else |
c8dad2bb | 2255 | mem = vmalloc(size); |
33327948 KH |
2256 | |
2257 | if (mem) | |
c8dad2bb | 2258 | memset(mem, 0, size); |
33327948 KH |
2259 | return mem; |
2260 | } | |
2261 | ||
8c7c6e34 KH |
2262 | /* |
2263 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
2264 | * (scanning all at force_empty is too costly...) | |
2265 | * | |
2266 | * Instead of clearing all references at force_empty, we remember | |
2267 | * the number of reference from swap_cgroup and free mem_cgroup when | |
2268 | * it goes down to 0. | |
2269 | * | |
8c7c6e34 KH |
2270 | * Removal of cgroup itself succeeds regardless of refs from swap. |
2271 | */ | |
2272 | ||
a7ba0eef | 2273 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 2274 | { |
08e552c6 KH |
2275 | int node; |
2276 | ||
04046e1a KH |
2277 | free_css_id(&mem_cgroup_subsys, &mem->css); |
2278 | ||
08e552c6 KH |
2279 | for_each_node_state(node, N_POSSIBLE) |
2280 | free_mem_cgroup_per_zone_info(mem, node); | |
2281 | ||
c8dad2bb | 2282 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
2283 | kfree(mem); |
2284 | else | |
2285 | vfree(mem); | |
2286 | } | |
2287 | ||
8c7c6e34 KH |
2288 | static void mem_cgroup_get(struct mem_cgroup *mem) |
2289 | { | |
2290 | atomic_inc(&mem->refcnt); | |
2291 | } | |
2292 | ||
2293 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
2294 | { | |
7bcc1bb1 DN |
2295 | if (atomic_dec_and_test(&mem->refcnt)) { |
2296 | struct mem_cgroup *parent = parent_mem_cgroup(mem); | |
a7ba0eef | 2297 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
2298 | if (parent) |
2299 | mem_cgroup_put(parent); | |
2300 | } | |
8c7c6e34 KH |
2301 | } |
2302 | ||
7bcc1bb1 DN |
2303 | /* |
2304 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
2305 | */ | |
2306 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
2307 | { | |
2308 | if (!mem->res.parent) | |
2309 | return NULL; | |
2310 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
2311 | } | |
33327948 | 2312 | |
c077719b KH |
2313 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
2314 | static void __init enable_swap_cgroup(void) | |
2315 | { | |
f8d66542 | 2316 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
2317 | do_swap_account = 1; |
2318 | } | |
2319 | #else | |
2320 | static void __init enable_swap_cgroup(void) | |
2321 | { | |
2322 | } | |
2323 | #endif | |
2324 | ||
0eb253e2 | 2325 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
2326 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
2327 | { | |
28dbc4b6 | 2328 | struct mem_cgroup *mem, *parent; |
04046e1a | 2329 | long error = -ENOMEM; |
6d12e2d8 | 2330 | int node; |
8cdea7c0 | 2331 | |
c8dad2bb JB |
2332 | mem = mem_cgroup_alloc(); |
2333 | if (!mem) | |
04046e1a | 2334 | return ERR_PTR(error); |
78fb7466 | 2335 | |
6d12e2d8 KH |
2336 | for_each_node_state(node, N_POSSIBLE) |
2337 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
2338 | goto free_out; | |
c077719b | 2339 | /* root ? */ |
28dbc4b6 | 2340 | if (cont->parent == NULL) { |
c077719b | 2341 | enable_swap_cgroup(); |
28dbc4b6 | 2342 | parent = NULL; |
18f59ea7 | 2343 | } else { |
28dbc4b6 | 2344 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 BS |
2345 | mem->use_hierarchy = parent->use_hierarchy; |
2346 | } | |
28dbc4b6 | 2347 | |
18f59ea7 BS |
2348 | if (parent && parent->use_hierarchy) { |
2349 | res_counter_init(&mem->res, &parent->res); | |
2350 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
2351 | /* |
2352 | * We increment refcnt of the parent to ensure that we can | |
2353 | * safely access it on res_counter_charge/uncharge. | |
2354 | * This refcnt will be decremented when freeing this | |
2355 | * mem_cgroup(see mem_cgroup_put). | |
2356 | */ | |
2357 | mem_cgroup_get(parent); | |
18f59ea7 BS |
2358 | } else { |
2359 | res_counter_init(&mem->res, NULL); | |
2360 | res_counter_init(&mem->memsw, NULL); | |
2361 | } | |
04046e1a | 2362 | mem->last_scanned_child = 0; |
2733c06a | 2363 | spin_lock_init(&mem->reclaim_param_lock); |
6d61ef40 | 2364 | |
a7885eb8 KM |
2365 | if (parent) |
2366 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 2367 | atomic_set(&mem->refcnt, 1); |
8cdea7c0 | 2368 | return &mem->css; |
6d12e2d8 | 2369 | free_out: |
a7ba0eef | 2370 | __mem_cgroup_free(mem); |
04046e1a | 2371 | return ERR_PTR(error); |
8cdea7c0 BS |
2372 | } |
2373 | ||
ec64f515 | 2374 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
2375 | struct cgroup *cont) |
2376 | { | |
2377 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
2378 | |
2379 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
2380 | } |
2381 | ||
8cdea7c0 BS |
2382 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
2383 | struct cgroup *cont) | |
2384 | { | |
c268e994 | 2385 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 2386 | |
c268e994 | 2387 | mem_cgroup_put(mem); |
8cdea7c0 BS |
2388 | } |
2389 | ||
2390 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
2391 | struct cgroup *cont) | |
2392 | { | |
8c7c6e34 KH |
2393 | int ret; |
2394 | ||
2395 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
2396 | ARRAY_SIZE(mem_cgroup_files)); | |
2397 | ||
2398 | if (!ret) | |
2399 | ret = register_memsw_files(cont, ss); | |
2400 | return ret; | |
8cdea7c0 BS |
2401 | } |
2402 | ||
67e465a7 BS |
2403 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
2404 | struct cgroup *cont, | |
2405 | struct cgroup *old_cont, | |
2406 | struct task_struct *p) | |
2407 | { | |
7f4d454d | 2408 | mutex_lock(&memcg_tasklist); |
67e465a7 | 2409 | /* |
f9717d28 NK |
2410 | * FIXME: It's better to move charges of this process from old |
2411 | * memcg to new memcg. But it's just on TODO-List now. | |
67e465a7 | 2412 | */ |
7f4d454d | 2413 | mutex_unlock(&memcg_tasklist); |
67e465a7 BS |
2414 | } |
2415 | ||
8cdea7c0 BS |
2416 | struct cgroup_subsys mem_cgroup_subsys = { |
2417 | .name = "memory", | |
2418 | .subsys_id = mem_cgroup_subsys_id, | |
2419 | .create = mem_cgroup_create, | |
df878fb0 | 2420 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
2421 | .destroy = mem_cgroup_destroy, |
2422 | .populate = mem_cgroup_populate, | |
67e465a7 | 2423 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 2424 | .early_init = 0, |
04046e1a | 2425 | .use_id = 1, |
8cdea7c0 | 2426 | }; |
c077719b KH |
2427 | |
2428 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2429 | ||
2430 | static int __init disable_swap_account(char *s) | |
2431 | { | |
2432 | really_do_swap_account = 0; | |
2433 | return 1; | |
2434 | } | |
2435 | __setup("noswapaccount", disable_swap_account); | |
2436 | #endif |