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