]>
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 | ||
54 | ||
d52aa412 KH |
55 | /* |
56 | * Statistics for memory cgroup. | |
57 | */ | |
58 | enum mem_cgroup_stat_index { | |
59 | /* | |
60 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
61 | */ | |
62 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
63 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
55e462b0 BR |
64 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
65 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
66 | |
67 | MEM_CGROUP_STAT_NSTATS, | |
68 | }; | |
69 | ||
70 | struct mem_cgroup_stat_cpu { | |
71 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
72 | } ____cacheline_aligned_in_smp; | |
73 | ||
74 | struct mem_cgroup_stat { | |
c8dad2bb | 75 | struct mem_cgroup_stat_cpu cpustat[0]; |
d52aa412 KH |
76 | }; |
77 | ||
78 | /* | |
79 | * For accounting under irq disable, no need for increment preempt count. | |
80 | */ | |
addb9efe | 81 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
82 | enum mem_cgroup_stat_index idx, int val) |
83 | { | |
addb9efe | 84 | stat->count[idx] += val; |
d52aa412 KH |
85 | } |
86 | ||
87 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
88 | enum mem_cgroup_stat_index idx) | |
89 | { | |
90 | int cpu; | |
91 | s64 ret = 0; | |
92 | for_each_possible_cpu(cpu) | |
93 | ret += stat->cpustat[cpu].count[idx]; | |
94 | return ret; | |
95 | } | |
96 | ||
6d12e2d8 KH |
97 | /* |
98 | * per-zone information in memory controller. | |
99 | */ | |
6d12e2d8 | 100 | struct mem_cgroup_per_zone { |
072c56c1 KH |
101 | /* |
102 | * spin_lock to protect the per cgroup LRU | |
103 | */ | |
b69408e8 CL |
104 | struct list_head lists[NR_LRU_LISTS]; |
105 | unsigned long count[NR_LRU_LISTS]; | |
6d12e2d8 KH |
106 | }; |
107 | /* Macro for accessing counter */ | |
108 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
109 | ||
110 | struct mem_cgroup_per_node { | |
111 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
112 | }; | |
113 | ||
114 | struct mem_cgroup_lru_info { | |
115 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
116 | }; | |
117 | ||
8cdea7c0 BS |
118 | /* |
119 | * The memory controller data structure. The memory controller controls both | |
120 | * page cache and RSS per cgroup. We would eventually like to provide | |
121 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
122 | * to help the administrator determine what knobs to tune. | |
123 | * | |
124 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
125 | * we hit the water mark. May be even add a low water mark, such that |
126 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
127 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
128 | */ |
129 | struct mem_cgroup { | |
130 | struct cgroup_subsys_state css; | |
131 | /* | |
132 | * the counter to account for memory usage | |
133 | */ | |
134 | struct res_counter res; | |
8c7c6e34 KH |
135 | /* |
136 | * the counter to account for mem+swap usage. | |
137 | */ | |
138 | struct res_counter memsw; | |
78fb7466 PE |
139 | /* |
140 | * Per cgroup active and inactive list, similar to the | |
141 | * per zone LRU lists. | |
78fb7466 | 142 | */ |
6d12e2d8 | 143 | struct mem_cgroup_lru_info info; |
072c56c1 | 144 | |
6c48a1d0 | 145 | int prev_priority; /* for recording reclaim priority */ |
6d61ef40 BS |
146 | |
147 | /* | |
148 | * While reclaiming in a hiearchy, we cache the last child we | |
149 | * reclaimed from. Protected by cgroup_lock() | |
150 | */ | |
151 | struct mem_cgroup *last_scanned_child; | |
18f59ea7 BS |
152 | /* |
153 | * Should the accounting and control be hierarchical, per subtree? | |
154 | */ | |
155 | bool use_hierarchy; | |
a636b327 | 156 | unsigned long last_oom_jiffies; |
8c7c6e34 KH |
157 | int obsolete; |
158 | atomic_t refcnt; | |
14797e23 KM |
159 | |
160 | unsigned int inactive_ratio; | |
161 | ||
d52aa412 | 162 | /* |
c8dad2bb | 163 | * statistics. This must be placed at the end of memcg. |
d52aa412 KH |
164 | */ |
165 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
166 | }; |
167 | ||
217bc319 KH |
168 | enum charge_type { |
169 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
170 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 171 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 172 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 173 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
c05555b5 KH |
174 | NR_CHARGE_TYPE, |
175 | }; | |
176 | ||
52d4b9ac KH |
177 | /* only for here (for easy reading.) */ |
178 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
179 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 180 | #define PCGF_LOCK (1UL << PCG_LOCK) |
c05555b5 KH |
181 | static const unsigned long |
182 | pcg_default_flags[NR_CHARGE_TYPE] = { | |
08e552c6 KH |
183 | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */ |
184 | PCGF_USED | PCGF_LOCK, /* Anon */ | |
185 | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ | |
52d4b9ac | 186 | 0, /* FORCE */ |
217bc319 KH |
187 | }; |
188 | ||
8c7c6e34 KH |
189 | |
190 | /* for encoding cft->private value on file */ | |
191 | #define _MEM (0) | |
192 | #define _MEMSWAP (1) | |
193 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
194 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
195 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
196 | ||
197 | static void mem_cgroup_get(struct mem_cgroup *mem); | |
198 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
199 | ||
c05555b5 KH |
200 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
201 | struct page_cgroup *pc, | |
202 | bool charge) | |
d52aa412 KH |
203 | { |
204 | int val = (charge)? 1 : -1; | |
205 | struct mem_cgroup_stat *stat = &mem->stat; | |
addb9efe | 206 | struct mem_cgroup_stat_cpu *cpustat; |
08e552c6 | 207 | int cpu = get_cpu(); |
d52aa412 | 208 | |
08e552c6 | 209 | cpustat = &stat->cpustat[cpu]; |
c05555b5 | 210 | if (PageCgroupCache(pc)) |
addb9efe | 211 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 212 | else |
addb9efe | 213 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
214 | |
215 | if (charge) | |
addb9efe | 216 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
217 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
218 | else | |
addb9efe | 219 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 220 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
08e552c6 | 221 | put_cpu(); |
6d12e2d8 KH |
222 | } |
223 | ||
d5b69e38 | 224 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
225 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
226 | { | |
6d12e2d8 KH |
227 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
228 | } | |
229 | ||
d5b69e38 | 230 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
231 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
232 | { | |
233 | struct mem_cgroup *mem = pc->mem_cgroup; | |
234 | int nid = page_cgroup_nid(pc); | |
235 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 236 | |
54992762 KM |
237 | if (!mem) |
238 | return NULL; | |
239 | ||
6d12e2d8 KH |
240 | return mem_cgroup_zoneinfo(mem, nid, zid); |
241 | } | |
242 | ||
243 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
b69408e8 | 244 | enum lru_list idx) |
6d12e2d8 KH |
245 | { |
246 | int nid, zid; | |
247 | struct mem_cgroup_per_zone *mz; | |
248 | u64 total = 0; | |
249 | ||
250 | for_each_online_node(nid) | |
251 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
252 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
253 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
254 | } | |
255 | return total; | |
d52aa412 KH |
256 | } |
257 | ||
d5b69e38 | 258 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
259 | { |
260 | return container_of(cgroup_subsys_state(cont, | |
261 | mem_cgroup_subsys_id), struct mem_cgroup, | |
262 | css); | |
263 | } | |
264 | ||
cf475ad2 | 265 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 266 | { |
31a78f23 BS |
267 | /* |
268 | * mm_update_next_owner() may clear mm->owner to NULL | |
269 | * if it races with swapoff, page migration, etc. | |
270 | * So this can be called with p == NULL. | |
271 | */ | |
272 | if (unlikely(!p)) | |
273 | return NULL; | |
274 | ||
78fb7466 PE |
275 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
276 | struct mem_cgroup, css); | |
277 | } | |
278 | ||
08e552c6 KH |
279 | /* |
280 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
281 | * Operations are called by routine of global LRU independently from memcg. | |
282 | * What we have to take care of here is validness of pc->mem_cgroup. | |
283 | * | |
284 | * Changes to pc->mem_cgroup happens when | |
285 | * 1. charge | |
286 | * 2. moving account | |
287 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
288 | * It is added to LRU before charge. | |
289 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
290 | * When moving account, the page is not on LRU. It's isolated. | |
291 | */ | |
4f98a2fe | 292 | |
08e552c6 KH |
293 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
294 | { | |
295 | struct page_cgroup *pc; | |
296 | struct mem_cgroup *mem; | |
297 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 298 | |
f8d66542 | 299 | if (mem_cgroup_disabled()) |
08e552c6 KH |
300 | return; |
301 | pc = lookup_page_cgroup(page); | |
302 | /* can happen while we handle swapcache. */ | |
303 | if (list_empty(&pc->lru)) | |
304 | return; | |
305 | mz = page_cgroup_zoneinfo(pc); | |
306 | mem = pc->mem_cgroup; | |
b69408e8 | 307 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
08e552c6 KH |
308 | list_del_init(&pc->lru); |
309 | return; | |
6d12e2d8 KH |
310 | } |
311 | ||
08e552c6 | 312 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 313 | { |
08e552c6 KH |
314 | mem_cgroup_del_lru_list(page, page_lru(page)); |
315 | } | |
b69408e8 | 316 | |
08e552c6 KH |
317 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
318 | { | |
319 | struct mem_cgroup_per_zone *mz; | |
320 | struct page_cgroup *pc; | |
b69408e8 | 321 | |
f8d66542 | 322 | if (mem_cgroup_disabled()) |
08e552c6 | 323 | return; |
6d12e2d8 | 324 | |
08e552c6 KH |
325 | pc = lookup_page_cgroup(page); |
326 | smp_rmb(); | |
327 | /* unused page is not rotated. */ | |
328 | if (!PageCgroupUsed(pc)) | |
329 | return; | |
330 | mz = page_cgroup_zoneinfo(pc); | |
331 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
332 | } |
333 | ||
08e552c6 | 334 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 335 | { |
08e552c6 KH |
336 | struct page_cgroup *pc; |
337 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 338 | |
f8d66542 | 339 | if (mem_cgroup_disabled()) |
08e552c6 KH |
340 | return; |
341 | pc = lookup_page_cgroup(page); | |
342 | /* barrier to sync with "charge" */ | |
343 | smp_rmb(); | |
344 | if (!PageCgroupUsed(pc)) | |
894bc310 | 345 | return; |
b69408e8 | 346 | |
08e552c6 | 347 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 348 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
08e552c6 KH |
349 | list_add(&pc->lru, &mz->lists[lru]); |
350 | } | |
351 | /* | |
352 | * To add swapcache into LRU. Be careful to all this function. | |
353 | * zone->lru_lock shouldn't be held and irq must not be disabled. | |
354 | */ | |
355 | static void mem_cgroup_lru_fixup(struct page *page) | |
356 | { | |
357 | if (!isolate_lru_page(page)) | |
358 | putback_lru_page(page); | |
359 | } | |
360 | ||
361 | void mem_cgroup_move_lists(struct page *page, | |
362 | enum lru_list from, enum lru_list to) | |
363 | { | |
f8d66542 | 364 | if (mem_cgroup_disabled()) |
08e552c6 KH |
365 | return; |
366 | mem_cgroup_del_lru_list(page, from); | |
367 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
368 | } |
369 | ||
4c4a2214 DR |
370 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
371 | { | |
372 | int ret; | |
373 | ||
374 | task_lock(task); | |
bd845e38 | 375 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
376 | task_unlock(task); |
377 | return ret; | |
378 | } | |
379 | ||
58ae83db KH |
380 | /* |
381 | * Calculate mapped_ratio under memory controller. This will be used in | |
382 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
383 | */ | |
384 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
385 | { | |
386 | long total, rss; | |
387 | ||
388 | /* | |
389 | * usage is recorded in bytes. But, here, we assume the number of | |
390 | * physical pages can be represented by "long" on any arch. | |
391 | */ | |
392 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
393 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
394 | return (int)((rss * 100L) / total); | |
395 | } | |
8869b8f6 | 396 | |
6c48a1d0 KH |
397 | /* |
398 | * prev_priority control...this will be used in memory reclaim path. | |
399 | */ | |
400 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
401 | { | |
402 | return mem->prev_priority; | |
403 | } | |
404 | ||
405 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
406 | { | |
407 | if (priority < mem->prev_priority) | |
408 | mem->prev_priority = priority; | |
409 | } | |
410 | ||
411 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
412 | { | |
413 | mem->prev_priority = priority; | |
414 | } | |
415 | ||
cc38108e KH |
416 | /* |
417 | * Calculate # of pages to be scanned in this priority/zone. | |
418 | * See also vmscan.c | |
419 | * | |
420 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
421 | * (see include/linux/mmzone.h) | |
422 | */ | |
423 | ||
b69408e8 CL |
424 | long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, |
425 | int priority, enum lru_list lru) | |
cc38108e | 426 | { |
b69408e8 | 427 | long nr_pages; |
cc38108e KH |
428 | int nid = zone->zone_pgdat->node_id; |
429 | int zid = zone_idx(zone); | |
430 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
431 | ||
b69408e8 | 432 | nr_pages = MEM_CGROUP_ZSTAT(mz, lru); |
cc38108e | 433 | |
b69408e8 | 434 | return (nr_pages >> priority); |
cc38108e KH |
435 | } |
436 | ||
14797e23 KM |
437 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone) |
438 | { | |
439 | unsigned long active; | |
440 | unsigned long inactive; | |
441 | ||
442 | inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON); | |
443 | active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON); | |
444 | ||
445 | if (inactive * memcg->inactive_ratio < active) | |
446 | return 1; | |
447 | ||
448 | return 0; | |
449 | } | |
450 | ||
66e1707b BS |
451 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
452 | struct list_head *dst, | |
453 | unsigned long *scanned, int order, | |
454 | int mode, struct zone *z, | |
455 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 456 | int active, int file) |
66e1707b BS |
457 | { |
458 | unsigned long nr_taken = 0; | |
459 | struct page *page; | |
460 | unsigned long scan; | |
461 | LIST_HEAD(pc_list); | |
462 | struct list_head *src; | |
ff7283fa | 463 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
464 | int nid = z->zone_pgdat->node_id; |
465 | int zid = zone_idx(z); | |
466 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 467 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 468 | |
cf475ad2 | 469 | BUG_ON(!mem_cont); |
1ecaab2b | 470 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 471 | src = &mz->lists[lru]; |
66e1707b | 472 | |
ff7283fa KH |
473 | scan = 0; |
474 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 475 | if (scan >= nr_to_scan) |
ff7283fa | 476 | break; |
08e552c6 KH |
477 | |
478 | page = pc->page; | |
52d4b9ac KH |
479 | if (unlikely(!PageCgroupUsed(pc))) |
480 | continue; | |
436c6541 | 481 | if (unlikely(!PageLRU(page))) |
ff7283fa | 482 | continue; |
ff7283fa | 483 | |
436c6541 | 484 | scan++; |
4f98a2fe | 485 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
486 | list_move(&page->lru, dst); |
487 | nr_taken++; | |
488 | } | |
489 | } | |
490 | ||
66e1707b BS |
491 | *scanned = scan; |
492 | return nr_taken; | |
493 | } | |
494 | ||
6d61ef40 BS |
495 | #define mem_cgroup_from_res_counter(counter, member) \ |
496 | container_of(counter, struct mem_cgroup, member) | |
497 | ||
498 | /* | |
499 | * This routine finds the DFS walk successor. This routine should be | |
500 | * called with cgroup_mutex held | |
501 | */ | |
502 | static struct mem_cgroup * | |
503 | mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem) | |
504 | { | |
505 | struct cgroup *cgroup, *curr_cgroup, *root_cgroup; | |
506 | ||
507 | curr_cgroup = curr->css.cgroup; | |
508 | root_cgroup = root_mem->css.cgroup; | |
509 | ||
510 | if (!list_empty(&curr_cgroup->children)) { | |
511 | /* | |
512 | * Walk down to children | |
513 | */ | |
514 | mem_cgroup_put(curr); | |
515 | cgroup = list_entry(curr_cgroup->children.next, | |
516 | struct cgroup, sibling); | |
517 | curr = mem_cgroup_from_cont(cgroup); | |
518 | mem_cgroup_get(curr); | |
519 | goto done; | |
520 | } | |
521 | ||
522 | visit_parent: | |
523 | if (curr_cgroup == root_cgroup) { | |
524 | mem_cgroup_put(curr); | |
525 | curr = root_mem; | |
526 | mem_cgroup_get(curr); | |
527 | goto done; | |
528 | } | |
529 | ||
530 | /* | |
531 | * Goto next sibling | |
532 | */ | |
533 | if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) { | |
534 | mem_cgroup_put(curr); | |
535 | cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup, | |
536 | sibling); | |
537 | curr = mem_cgroup_from_cont(cgroup); | |
538 | mem_cgroup_get(curr); | |
539 | goto done; | |
540 | } | |
541 | ||
542 | /* | |
543 | * Go up to next parent and next parent's sibling if need be | |
544 | */ | |
545 | curr_cgroup = curr_cgroup->parent; | |
546 | goto visit_parent; | |
547 | ||
548 | done: | |
549 | root_mem->last_scanned_child = curr; | |
550 | return curr; | |
551 | } | |
552 | ||
553 | /* | |
554 | * Visit the first child (need not be the first child as per the ordering | |
555 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
556 | * that to reclaim free pages from. | |
557 | */ | |
558 | static struct mem_cgroup * | |
559 | mem_cgroup_get_first_node(struct mem_cgroup *root_mem) | |
560 | { | |
561 | struct cgroup *cgroup; | |
562 | struct mem_cgroup *ret; | |
563 | bool obsolete = (root_mem->last_scanned_child && | |
564 | root_mem->last_scanned_child->obsolete); | |
565 | ||
566 | /* | |
567 | * Scan all children under the mem_cgroup mem | |
568 | */ | |
569 | cgroup_lock(); | |
570 | if (list_empty(&root_mem->css.cgroup->children)) { | |
571 | ret = root_mem; | |
572 | goto done; | |
573 | } | |
574 | ||
575 | if (!root_mem->last_scanned_child || obsolete) { | |
576 | ||
577 | if (obsolete) | |
578 | mem_cgroup_put(root_mem->last_scanned_child); | |
579 | ||
580 | cgroup = list_first_entry(&root_mem->css.cgroup->children, | |
581 | struct cgroup, sibling); | |
582 | ret = mem_cgroup_from_cont(cgroup); | |
583 | mem_cgroup_get(ret); | |
584 | } else | |
585 | ret = mem_cgroup_get_next_node(root_mem->last_scanned_child, | |
586 | root_mem); | |
587 | ||
588 | done: | |
589 | root_mem->last_scanned_child = ret; | |
590 | cgroup_unlock(); | |
591 | return ret; | |
592 | } | |
593 | ||
b85a96c0 DN |
594 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
595 | { | |
596 | if (do_swap_account) { | |
597 | if (res_counter_check_under_limit(&mem->res) && | |
598 | res_counter_check_under_limit(&mem->memsw)) | |
599 | return true; | |
600 | } else | |
601 | if (res_counter_check_under_limit(&mem->res)) | |
602 | return true; | |
603 | return false; | |
604 | } | |
605 | ||
6d61ef40 BS |
606 | /* |
607 | * Dance down the hierarchy if needed to reclaim memory. We remember the | |
608 | * last child we reclaimed from, so that we don't end up penalizing | |
609 | * one child extensively based on its position in the children list. | |
610 | * | |
611 | * root_mem is the original ancestor that we've been reclaim from. | |
612 | */ | |
613 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
614 | gfp_t gfp_mask, bool noswap) | |
615 | { | |
616 | struct mem_cgroup *next_mem; | |
617 | int ret = 0; | |
618 | ||
619 | /* | |
620 | * Reclaim unconditionally and don't check for return value. | |
621 | * We need to reclaim in the current group and down the tree. | |
622 | * One might think about checking for children before reclaiming, | |
623 | * but there might be left over accounting, even after children | |
624 | * have left. | |
625 | */ | |
626 | ret = try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap); | |
b85a96c0 | 627 | if (mem_cgroup_check_under_limit(root_mem)) |
6d61ef40 | 628 | return 0; |
670ec2f1 DN |
629 | if (!root_mem->use_hierarchy) |
630 | return ret; | |
6d61ef40 BS |
631 | |
632 | next_mem = mem_cgroup_get_first_node(root_mem); | |
633 | ||
634 | while (next_mem != root_mem) { | |
635 | if (next_mem->obsolete) { | |
636 | mem_cgroup_put(next_mem); | |
637 | cgroup_lock(); | |
638 | next_mem = mem_cgroup_get_first_node(root_mem); | |
639 | cgroup_unlock(); | |
640 | continue; | |
641 | } | |
642 | ret = try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap); | |
b85a96c0 | 643 | if (mem_cgroup_check_under_limit(root_mem)) |
6d61ef40 BS |
644 | return 0; |
645 | cgroup_lock(); | |
646 | next_mem = mem_cgroup_get_next_node(next_mem, root_mem); | |
647 | cgroup_unlock(); | |
648 | } | |
649 | return ret; | |
650 | } | |
651 | ||
a636b327 KH |
652 | bool mem_cgroup_oom_called(struct task_struct *task) |
653 | { | |
654 | bool ret = false; | |
655 | struct mem_cgroup *mem; | |
656 | struct mm_struct *mm; | |
657 | ||
658 | rcu_read_lock(); | |
659 | mm = task->mm; | |
660 | if (!mm) | |
661 | mm = &init_mm; | |
662 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
663 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
664 | ret = true; | |
665 | rcu_read_unlock(); | |
666 | return ret; | |
667 | } | |
f817ed48 KH |
668 | /* |
669 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
670 | * oom-killer can be invoked. | |
8a9f3ccd | 671 | */ |
f817ed48 | 672 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
8c7c6e34 KH |
673 | gfp_t gfp_mask, struct mem_cgroup **memcg, |
674 | bool oom) | |
8a9f3ccd | 675 | { |
6d61ef40 | 676 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 677 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
28dbc4b6 | 678 | struct res_counter *fail_res; |
a636b327 KH |
679 | |
680 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
681 | /* Don't account this! */ | |
682 | *memcg = NULL; | |
683 | return 0; | |
684 | } | |
685 | ||
8a9f3ccd | 686 | /* |
3be91277 HD |
687 | * We always charge the cgroup the mm_struct belongs to. |
688 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
689 | * thread group leader migrates. It's possible that mm is not |
690 | * set, if so charge the init_mm (happens for pagecache usage). | |
691 | */ | |
7a81b88c | 692 | if (likely(!*memcg)) { |
e8589cc1 KH |
693 | rcu_read_lock(); |
694 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
695 | if (unlikely(!mem)) { |
696 | rcu_read_unlock(); | |
31a78f23 BS |
697 | return 0; |
698 | } | |
e8589cc1 KH |
699 | /* |
700 | * For every charge from the cgroup, increment reference count | |
701 | */ | |
702 | css_get(&mem->css); | |
7a81b88c | 703 | *memcg = mem; |
e8589cc1 KH |
704 | rcu_read_unlock(); |
705 | } else { | |
7a81b88c KH |
706 | mem = *memcg; |
707 | css_get(&mem->css); | |
e8589cc1 | 708 | } |
8a9f3ccd | 709 | |
8c7c6e34 KH |
710 | while (1) { |
711 | int ret; | |
712 | bool noswap = false; | |
7a81b88c | 713 | |
28dbc4b6 | 714 | ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res); |
8c7c6e34 KH |
715 | if (likely(!ret)) { |
716 | if (!do_swap_account) | |
717 | break; | |
28dbc4b6 BS |
718 | ret = res_counter_charge(&mem->memsw, PAGE_SIZE, |
719 | &fail_res); | |
8c7c6e34 KH |
720 | if (likely(!ret)) |
721 | break; | |
722 | /* mem+swap counter fails */ | |
723 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
724 | noswap = true; | |
6d61ef40 BS |
725 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
726 | memsw); | |
727 | } else | |
728 | /* mem counter fails */ | |
729 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
730 | res); | |
731 | ||
3be91277 | 732 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 733 | goto nomem; |
e1a1cd59 | 734 | |
6d61ef40 BS |
735 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask, |
736 | noswap); | |
66e1707b BS |
737 | |
738 | /* | |
8869b8f6 HD |
739 | * try_to_free_mem_cgroup_pages() might not give us a full |
740 | * picture of reclaim. Some pages are reclaimed and might be | |
741 | * moved to swap cache or just unmapped from the cgroup. | |
742 | * Check the limit again to see if the reclaim reduced the | |
743 | * current usage of the cgroup before giving up | |
8c7c6e34 | 744 | * |
8869b8f6 | 745 | */ |
b85a96c0 DN |
746 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
747 | continue; | |
3be91277 HD |
748 | |
749 | if (!nr_retries--) { | |
a636b327 | 750 | if (oom) { |
88700756 KH |
751 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); |
752 | mem_over_limit->last_oom_jiffies = jiffies; | |
a636b327 | 753 | } |
7a81b88c | 754 | goto nomem; |
66e1707b | 755 | } |
8a9f3ccd | 756 | } |
7a81b88c KH |
757 | return 0; |
758 | nomem: | |
759 | css_put(&mem->css); | |
760 | return -ENOMEM; | |
761 | } | |
8a9f3ccd | 762 | |
f817ed48 KH |
763 | /** |
764 | * mem_cgroup_try_charge - get charge of PAGE_SIZE. | |
765 | * @mm: an mm_struct which is charged against. (when *memcg is NULL) | |
766 | * @gfp_mask: gfp_mask for reclaim. | |
767 | * @memcg: a pointer to memory cgroup which is charged against. | |
768 | * | |
769 | * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated | |
770 | * memory cgroup from @mm is got and stored in *memcg. | |
771 | * | |
772 | * Returns 0 if success. -ENOMEM at failure. | |
773 | * This call can invoke OOM-Killer. | |
774 | */ | |
775 | ||
776 | int mem_cgroup_try_charge(struct mm_struct *mm, | |
777 | gfp_t mask, struct mem_cgroup **memcg) | |
778 | { | |
779 | return __mem_cgroup_try_charge(mm, mask, memcg, true); | |
780 | } | |
781 | ||
7a81b88c KH |
782 | /* |
783 | * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be | |
784 | * USED state. If already USED, uncharge and return. | |
785 | */ | |
786 | ||
787 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
788 | struct page_cgroup *pc, | |
789 | enum charge_type ctype) | |
790 | { | |
7a81b88c KH |
791 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
792 | if (!mem) | |
793 | return; | |
52d4b9ac KH |
794 | |
795 | lock_page_cgroup(pc); | |
796 | if (unlikely(PageCgroupUsed(pc))) { | |
797 | unlock_page_cgroup(pc); | |
798 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
799 | if (do_swap_account) |
800 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
52d4b9ac | 801 | css_put(&mem->css); |
7a81b88c | 802 | return; |
52d4b9ac | 803 | } |
8a9f3ccd | 804 | pc->mem_cgroup = mem; |
08e552c6 | 805 | smp_wmb(); |
c05555b5 | 806 | pc->flags = pcg_default_flags[ctype]; |
3be91277 | 807 | |
08e552c6 | 808 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 809 | |
52d4b9ac | 810 | unlock_page_cgroup(pc); |
7a81b88c | 811 | } |
66e1707b | 812 | |
f817ed48 KH |
813 | /** |
814 | * mem_cgroup_move_account - move account of the page | |
815 | * @pc: page_cgroup of the page. | |
816 | * @from: mem_cgroup which the page is moved from. | |
817 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
818 | * | |
819 | * The caller must confirm following. | |
08e552c6 | 820 | * - page is not on LRU (isolate_page() is useful.) |
f817ed48 KH |
821 | * |
822 | * returns 0 at success, | |
823 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
824 | * | |
825 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
826 | * new cgroup. It should be done by a caller. | |
827 | */ | |
828 | ||
829 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
830 | struct mem_cgroup *from, struct mem_cgroup *to) | |
831 | { | |
832 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
833 | int nid, zid; | |
834 | int ret = -EBUSY; | |
835 | ||
f817ed48 | 836 | VM_BUG_ON(from == to); |
08e552c6 | 837 | VM_BUG_ON(PageLRU(pc->page)); |
f817ed48 KH |
838 | |
839 | nid = page_cgroup_nid(pc); | |
840 | zid = page_cgroup_zid(pc); | |
841 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
842 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
843 | ||
f817ed48 KH |
844 | if (!trylock_page_cgroup(pc)) |
845 | return ret; | |
846 | ||
847 | if (!PageCgroupUsed(pc)) | |
848 | goto out; | |
849 | ||
850 | if (pc->mem_cgroup != from) | |
851 | goto out; | |
852 | ||
08e552c6 KH |
853 | css_put(&from->css); |
854 | res_counter_uncharge(&from->res, PAGE_SIZE); | |
855 | mem_cgroup_charge_statistics(from, pc, false); | |
856 | if (do_swap_account) | |
857 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
858 | pc->mem_cgroup = to; | |
859 | mem_cgroup_charge_statistics(to, pc, true); | |
860 | css_get(&to->css); | |
861 | ret = 0; | |
f817ed48 KH |
862 | out: |
863 | unlock_page_cgroup(pc); | |
864 | return ret; | |
865 | } | |
866 | ||
867 | /* | |
868 | * move charges to its parent. | |
869 | */ | |
870 | ||
871 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
872 | struct mem_cgroup *child, | |
873 | gfp_t gfp_mask) | |
874 | { | |
08e552c6 | 875 | struct page *page = pc->page; |
f817ed48 KH |
876 | struct cgroup *cg = child->css.cgroup; |
877 | struct cgroup *pcg = cg->parent; | |
878 | struct mem_cgroup *parent; | |
f817ed48 KH |
879 | int ret; |
880 | ||
881 | /* Is ROOT ? */ | |
882 | if (!pcg) | |
883 | return -EINVAL; | |
884 | ||
08e552c6 | 885 | |
f817ed48 KH |
886 | parent = mem_cgroup_from_cont(pcg); |
887 | ||
08e552c6 | 888 | |
f817ed48 | 889 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); |
a636b327 | 890 | if (ret || !parent) |
f817ed48 KH |
891 | return ret; |
892 | ||
08e552c6 KH |
893 | if (!get_page_unless_zero(page)) |
894 | return -EBUSY; | |
895 | ||
896 | ret = isolate_lru_page(page); | |
897 | ||
898 | if (ret) | |
899 | goto cancel; | |
f817ed48 | 900 | |
f817ed48 | 901 | ret = mem_cgroup_move_account(pc, child, parent); |
f817ed48 | 902 | |
08e552c6 | 903 | /* drop extra refcnt by try_charge() (move_account increment one) */ |
f817ed48 | 904 | css_put(&parent->css); |
08e552c6 KH |
905 | putback_lru_page(page); |
906 | if (!ret) { | |
907 | put_page(page); | |
908 | return 0; | |
8c7c6e34 | 909 | } |
08e552c6 KH |
910 | /* uncharge if move fails */ |
911 | cancel: | |
912 | res_counter_uncharge(&parent->res, PAGE_SIZE); | |
913 | if (do_swap_account) | |
914 | res_counter_uncharge(&parent->memsw, PAGE_SIZE); | |
915 | put_page(page); | |
f817ed48 KH |
916 | return ret; |
917 | } | |
918 | ||
7a81b88c KH |
919 | /* |
920 | * Charge the memory controller for page usage. | |
921 | * Return | |
922 | * 0 if the charge was successful | |
923 | * < 0 if the cgroup is over its limit | |
924 | */ | |
925 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
926 | gfp_t gfp_mask, enum charge_type ctype, | |
927 | struct mem_cgroup *memcg) | |
928 | { | |
929 | struct mem_cgroup *mem; | |
930 | struct page_cgroup *pc; | |
931 | int ret; | |
932 | ||
933 | pc = lookup_page_cgroup(page); | |
934 | /* can happen at boot */ | |
935 | if (unlikely(!pc)) | |
936 | return 0; | |
937 | prefetchw(pc); | |
938 | ||
939 | mem = memcg; | |
f817ed48 | 940 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
a636b327 | 941 | if (ret || !mem) |
7a81b88c KH |
942 | return ret; |
943 | ||
944 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 945 | return 0; |
8a9f3ccd BS |
946 | } |
947 | ||
7a81b88c KH |
948 | int mem_cgroup_newpage_charge(struct page *page, |
949 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 950 | { |
f8d66542 | 951 | if (mem_cgroup_disabled()) |
cede86ac | 952 | return 0; |
52d4b9ac KH |
953 | if (PageCompound(page)) |
954 | return 0; | |
69029cd5 KH |
955 | /* |
956 | * If already mapped, we don't have to account. | |
957 | * If page cache, page->mapping has address_space. | |
958 | * But page->mapping may have out-of-use anon_vma pointer, | |
959 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
960 | * is NULL. | |
961 | */ | |
962 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
963 | return 0; | |
964 | if (unlikely(!mm)) | |
965 | mm = &init_mm; | |
217bc319 | 966 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 967 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
968 | } |
969 | ||
e1a1cd59 BS |
970 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
971 | gfp_t gfp_mask) | |
8697d331 | 972 | { |
f8d66542 | 973 | if (mem_cgroup_disabled()) |
cede86ac | 974 | return 0; |
52d4b9ac KH |
975 | if (PageCompound(page)) |
976 | return 0; | |
accf163e KH |
977 | /* |
978 | * Corner case handling. This is called from add_to_page_cache() | |
979 | * in usual. But some FS (shmem) precharges this page before calling it | |
980 | * and call add_to_page_cache() with GFP_NOWAIT. | |
981 | * | |
982 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
983 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
984 | * charge twice. (It works but has to pay a bit larger cost.) | |
985 | */ | |
986 | if (!(gfp_mask & __GFP_WAIT)) { | |
987 | struct page_cgroup *pc; | |
988 | ||
52d4b9ac KH |
989 | |
990 | pc = lookup_page_cgroup(page); | |
991 | if (!pc) | |
992 | return 0; | |
993 | lock_page_cgroup(pc); | |
994 | if (PageCgroupUsed(pc)) { | |
995 | unlock_page_cgroup(pc); | |
accf163e KH |
996 | return 0; |
997 | } | |
52d4b9ac | 998 | unlock_page_cgroup(pc); |
accf163e KH |
999 | } |
1000 | ||
69029cd5 | 1001 | if (unlikely(!mm)) |
8697d331 | 1002 | mm = &init_mm; |
accf163e | 1003 | |
c05555b5 KH |
1004 | if (page_is_file_cache(page)) |
1005 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 1006 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
c05555b5 KH |
1007 | else |
1008 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
1009 | MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL); | |
e8589cc1 KH |
1010 | } |
1011 | ||
8c7c6e34 KH |
1012 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
1013 | struct page *page, | |
1014 | gfp_t mask, struct mem_cgroup **ptr) | |
1015 | { | |
1016 | struct mem_cgroup *mem; | |
1017 | swp_entry_t ent; | |
1018 | ||
f8d66542 | 1019 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
1020 | return 0; |
1021 | ||
1022 | if (!do_swap_account) | |
1023 | goto charge_cur_mm; | |
1024 | ||
1025 | /* | |
1026 | * A racing thread's fault, or swapoff, may have already updated | |
1027 | * the pte, and even removed page from swap cache: return success | |
1028 | * to go on to do_swap_page()'s pte_same() test, which should fail. | |
1029 | */ | |
1030 | if (!PageSwapCache(page)) | |
1031 | return 0; | |
1032 | ||
1033 | ent.val = page_private(page); | |
1034 | ||
1035 | mem = lookup_swap_cgroup(ent); | |
1036 | if (!mem || mem->obsolete) | |
1037 | goto charge_cur_mm; | |
1038 | *ptr = mem; | |
1039 | return __mem_cgroup_try_charge(NULL, mask, ptr, true); | |
1040 | charge_cur_mm: | |
1041 | if (unlikely(!mm)) | |
1042 | mm = &init_mm; | |
1043 | return __mem_cgroup_try_charge(mm, mask, ptr, true); | |
1044 | } | |
1045 | ||
d13d1443 | 1046 | #ifdef CONFIG_SWAP |
8c7c6e34 | 1047 | |
d13d1443 KH |
1048 | int mem_cgroup_cache_charge_swapin(struct page *page, |
1049 | struct mm_struct *mm, gfp_t mask, bool locked) | |
1050 | { | |
1051 | int ret = 0; | |
1052 | ||
f8d66542 | 1053 | if (mem_cgroup_disabled()) |
d13d1443 KH |
1054 | return 0; |
1055 | if (unlikely(!mm)) | |
1056 | mm = &init_mm; | |
1057 | if (!locked) | |
1058 | lock_page(page); | |
1059 | /* | |
1060 | * If not locked, the page can be dropped from SwapCache until | |
1061 | * we reach here. | |
1062 | */ | |
1063 | if (PageSwapCache(page)) { | |
8c7c6e34 KH |
1064 | struct mem_cgroup *mem = NULL; |
1065 | swp_entry_t ent; | |
1066 | ||
1067 | ent.val = page_private(page); | |
1068 | if (do_swap_account) { | |
1069 | mem = lookup_swap_cgroup(ent); | |
1070 | if (mem && mem->obsolete) | |
1071 | mem = NULL; | |
1072 | if (mem) | |
1073 | mm = NULL; | |
1074 | } | |
d13d1443 | 1075 | ret = mem_cgroup_charge_common(page, mm, mask, |
8c7c6e34 KH |
1076 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); |
1077 | ||
1078 | if (!ret && do_swap_account) { | |
1079 | /* avoid double counting */ | |
1080 | mem = swap_cgroup_record(ent, NULL); | |
1081 | if (mem) { | |
1082 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1083 | mem_cgroup_put(mem); | |
1084 | } | |
1085 | } | |
d13d1443 KH |
1086 | } |
1087 | if (!locked) | |
1088 | unlock_page(page); | |
08e552c6 KH |
1089 | /* add this page(page_cgroup) to the LRU we want. */ |
1090 | mem_cgroup_lru_fixup(page); | |
d13d1443 KH |
1091 | |
1092 | return ret; | |
1093 | } | |
1094 | #endif | |
1095 | ||
7a81b88c KH |
1096 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
1097 | { | |
1098 | struct page_cgroup *pc; | |
1099 | ||
f8d66542 | 1100 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1101 | return; |
1102 | if (!ptr) | |
1103 | return; | |
1104 | pc = lookup_page_cgroup(page); | |
1105 | __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
8c7c6e34 KH |
1106 | /* |
1107 | * Now swap is on-memory. This means this page may be | |
1108 | * counted both as mem and swap....double count. | |
1109 | * Fix it by uncharging from memsw. This SwapCache is stable | |
1110 | * because we're still under lock_page(). | |
1111 | */ | |
1112 | if (do_swap_account) { | |
1113 | swp_entry_t ent = {.val = page_private(page)}; | |
1114 | struct mem_cgroup *memcg; | |
1115 | memcg = swap_cgroup_record(ent, NULL); | |
1116 | if (memcg) { | |
1117 | /* If memcg is obsolete, memcg can be != ptr */ | |
1118 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
1119 | mem_cgroup_put(memcg); | |
1120 | } | |
1121 | ||
1122 | } | |
08e552c6 KH |
1123 | /* add this page(page_cgroup) to the LRU we want. */ |
1124 | mem_cgroup_lru_fixup(page); | |
7a81b88c KH |
1125 | } |
1126 | ||
1127 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
1128 | { | |
f8d66542 | 1129 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1130 | return; |
1131 | if (!mem) | |
1132 | return; | |
1133 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
1134 | if (do_swap_account) |
1135 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
7a81b88c KH |
1136 | css_put(&mem->css); |
1137 | } | |
1138 | ||
1139 | ||
8a9f3ccd | 1140 | /* |
69029cd5 | 1141 | * uncharge if !page_mapped(page) |
8a9f3ccd | 1142 | */ |
8c7c6e34 | 1143 | static struct mem_cgroup * |
69029cd5 | 1144 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 1145 | { |
8289546e | 1146 | struct page_cgroup *pc; |
8c7c6e34 | 1147 | struct mem_cgroup *mem = NULL; |
072c56c1 | 1148 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 1149 | |
f8d66542 | 1150 | if (mem_cgroup_disabled()) |
8c7c6e34 | 1151 | return NULL; |
4077960e | 1152 | |
d13d1443 | 1153 | if (PageSwapCache(page)) |
8c7c6e34 | 1154 | return NULL; |
d13d1443 | 1155 | |
8697d331 | 1156 | /* |
3c541e14 | 1157 | * Check if our page_cgroup is valid |
8697d331 | 1158 | */ |
52d4b9ac KH |
1159 | pc = lookup_page_cgroup(page); |
1160 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 1161 | return NULL; |
b9c565d5 | 1162 | |
52d4b9ac | 1163 | lock_page_cgroup(pc); |
d13d1443 | 1164 | |
8c7c6e34 KH |
1165 | mem = pc->mem_cgroup; |
1166 | ||
d13d1443 KH |
1167 | if (!PageCgroupUsed(pc)) |
1168 | goto unlock_out; | |
1169 | ||
1170 | switch (ctype) { | |
1171 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1172 | if (page_mapped(page)) | |
1173 | goto unlock_out; | |
1174 | break; | |
1175 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
1176 | if (!PageAnon(page)) { /* Shared memory */ | |
1177 | if (page->mapping && !page_is_file_cache(page)) | |
1178 | goto unlock_out; | |
1179 | } else if (page_mapped(page)) /* Anon */ | |
1180 | goto unlock_out; | |
1181 | break; | |
1182 | default: | |
1183 | break; | |
52d4b9ac | 1184 | } |
d13d1443 | 1185 | |
8c7c6e34 KH |
1186 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1187 | if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)) | |
1188 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1189 | ||
08e552c6 | 1190 | mem_cgroup_charge_statistics(mem, pc, false); |
52d4b9ac | 1191 | ClearPageCgroupUsed(pc); |
b9c565d5 | 1192 | |
69029cd5 | 1193 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 1194 | unlock_page_cgroup(pc); |
fb59e9f1 | 1195 | |
a7fe942e KH |
1196 | /* at swapout, this memcg will be accessed to record to swap */ |
1197 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
1198 | css_put(&mem->css); | |
6d12e2d8 | 1199 | |
8c7c6e34 | 1200 | return mem; |
d13d1443 KH |
1201 | |
1202 | unlock_out: | |
1203 | unlock_page_cgroup(pc); | |
8c7c6e34 | 1204 | return NULL; |
3c541e14 BS |
1205 | } |
1206 | ||
69029cd5 KH |
1207 | void mem_cgroup_uncharge_page(struct page *page) |
1208 | { | |
52d4b9ac KH |
1209 | /* early check. */ |
1210 | if (page_mapped(page)) | |
1211 | return; | |
1212 | if (page->mapping && !PageAnon(page)) | |
1213 | return; | |
69029cd5 KH |
1214 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
1215 | } | |
1216 | ||
1217 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
1218 | { | |
1219 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 1220 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
1221 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
1222 | } | |
1223 | ||
8c7c6e34 KH |
1224 | /* |
1225 | * called from __delete_from_swap_cache() and drop "page" account. | |
1226 | * memcg information is recorded to swap_cgroup of "ent" | |
1227 | */ | |
1228 | void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent) | |
1229 | { | |
1230 | struct mem_cgroup *memcg; | |
1231 | ||
1232 | memcg = __mem_cgroup_uncharge_common(page, | |
1233 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT); | |
1234 | /* record memcg information */ | |
1235 | if (do_swap_account && memcg) { | |
1236 | swap_cgroup_record(ent, memcg); | |
1237 | mem_cgroup_get(memcg); | |
1238 | } | |
a7fe942e KH |
1239 | if (memcg) |
1240 | css_put(&memcg->css); | |
8c7c6e34 KH |
1241 | } |
1242 | ||
1243 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1244 | /* | |
1245 | * called from swap_entry_free(). remove record in swap_cgroup and | |
1246 | * uncharge "memsw" account. | |
1247 | */ | |
1248 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 1249 | { |
8c7c6e34 KH |
1250 | struct mem_cgroup *memcg; |
1251 | ||
1252 | if (!do_swap_account) | |
1253 | return; | |
1254 | ||
1255 | memcg = swap_cgroup_record(ent, NULL); | |
1256 | if (memcg) { | |
1257 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
1258 | mem_cgroup_put(memcg); | |
1259 | } | |
d13d1443 | 1260 | } |
8c7c6e34 | 1261 | #endif |
d13d1443 | 1262 | |
ae41be37 | 1263 | /* |
01b1ae63 KH |
1264 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
1265 | * page belongs to. | |
ae41be37 | 1266 | */ |
01b1ae63 | 1267 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
1268 | { |
1269 | struct page_cgroup *pc; | |
e8589cc1 | 1270 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 1271 | int ret = 0; |
8869b8f6 | 1272 | |
f8d66542 | 1273 | if (mem_cgroup_disabled()) |
4077960e BS |
1274 | return 0; |
1275 | ||
52d4b9ac KH |
1276 | pc = lookup_page_cgroup(page); |
1277 | lock_page_cgroup(pc); | |
1278 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
1279 | mem = pc->mem_cgroup; |
1280 | css_get(&mem->css); | |
e8589cc1 | 1281 | } |
52d4b9ac | 1282 | unlock_page_cgroup(pc); |
01b1ae63 | 1283 | |
e8589cc1 | 1284 | if (mem) { |
2c26fdd7 | 1285 | ret = mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem); |
e8589cc1 KH |
1286 | css_put(&mem->css); |
1287 | } | |
01b1ae63 | 1288 | *ptr = mem; |
e8589cc1 | 1289 | return ret; |
ae41be37 | 1290 | } |
8869b8f6 | 1291 | |
69029cd5 | 1292 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
1293 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
1294 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 1295 | { |
01b1ae63 KH |
1296 | struct page *target, *unused; |
1297 | struct page_cgroup *pc; | |
1298 | enum charge_type ctype; | |
1299 | ||
1300 | if (!mem) | |
1301 | return; | |
1302 | ||
1303 | /* at migration success, oldpage->mapping is NULL. */ | |
1304 | if (oldpage->mapping) { | |
1305 | target = oldpage; | |
1306 | unused = NULL; | |
1307 | } else { | |
1308 | target = newpage; | |
1309 | unused = oldpage; | |
1310 | } | |
1311 | ||
1312 | if (PageAnon(target)) | |
1313 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
1314 | else if (page_is_file_cache(target)) | |
1315 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
1316 | else | |
1317 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
1318 | ||
1319 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 1320 | if (unused) |
01b1ae63 KH |
1321 | __mem_cgroup_uncharge_common(unused, ctype); |
1322 | ||
1323 | pc = lookup_page_cgroup(target); | |
69029cd5 | 1324 | /* |
01b1ae63 KH |
1325 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
1326 | * So, double-counting is effectively avoided. | |
1327 | */ | |
1328 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
1329 | ||
1330 | /* | |
1331 | * Both of oldpage and newpage are still under lock_page(). | |
1332 | * Then, we don't have to care about race in radix-tree. | |
1333 | * But we have to be careful that this page is unmapped or not. | |
1334 | * | |
1335 | * There is a case for !page_mapped(). At the start of | |
1336 | * migration, oldpage was mapped. But now, it's zapped. | |
1337 | * But we know *target* page is not freed/reused under us. | |
1338 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 1339 | */ |
01b1ae63 KH |
1340 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
1341 | mem_cgroup_uncharge_page(target); | |
ae41be37 | 1342 | } |
78fb7466 | 1343 | |
c9b0ed51 KH |
1344 | /* |
1345 | * A call to try to shrink memory usage under specified resource controller. | |
1346 | * This is typically used for page reclaiming for shmem for reducing side | |
1347 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
1348 | */ | |
1349 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) | |
1350 | { | |
1351 | struct mem_cgroup *mem; | |
1352 | int progress = 0; | |
1353 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
1354 | ||
f8d66542 | 1355 | if (mem_cgroup_disabled()) |
cede86ac | 1356 | return 0; |
9623e078 HD |
1357 | if (!mm) |
1358 | return 0; | |
cede86ac | 1359 | |
c9b0ed51 KH |
1360 | rcu_read_lock(); |
1361 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
1362 | if (unlikely(!mem)) { |
1363 | rcu_read_unlock(); | |
1364 | return 0; | |
1365 | } | |
c9b0ed51 KH |
1366 | css_get(&mem->css); |
1367 | rcu_read_unlock(); | |
1368 | ||
1369 | do { | |
8c7c6e34 | 1370 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask, true); |
b85a96c0 | 1371 | progress += mem_cgroup_check_under_limit(mem); |
c9b0ed51 KH |
1372 | } while (!progress && --retry); |
1373 | ||
1374 | css_put(&mem->css); | |
1375 | if (!retry) | |
1376 | return -ENOMEM; | |
1377 | return 0; | |
1378 | } | |
1379 | ||
14797e23 KM |
1380 | /* |
1381 | * The inactive anon list should be small enough that the VM never has to | |
1382 | * do too much work, but large enough that each inactive page has a chance | |
1383 | * to be referenced again before it is swapped out. | |
1384 | * | |
1385 | * this calculation is straightforward porting from | |
1386 | * page_alloc.c::setup_per_zone_inactive_ratio(). | |
1387 | * it describe more detail. | |
1388 | */ | |
1389 | static void mem_cgroup_set_inactive_ratio(struct mem_cgroup *memcg) | |
1390 | { | |
1391 | unsigned int gb, ratio; | |
1392 | ||
1393 | gb = res_counter_read_u64(&memcg->res, RES_LIMIT) >> 30; | |
1394 | if (gb) | |
1395 | ratio = int_sqrt(10 * gb); | |
1396 | else | |
1397 | ratio = 1; | |
1398 | ||
1399 | memcg->inactive_ratio = ratio; | |
1400 | ||
1401 | } | |
1402 | ||
8c7c6e34 KH |
1403 | static DEFINE_MUTEX(set_limit_mutex); |
1404 | ||
d38d2a75 | 1405 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 1406 | unsigned long long val) |
628f4235 KH |
1407 | { |
1408 | ||
1409 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
1410 | int progress; | |
8c7c6e34 | 1411 | u64 memswlimit; |
628f4235 KH |
1412 | int ret = 0; |
1413 | ||
8c7c6e34 | 1414 | while (retry_count) { |
628f4235 KH |
1415 | if (signal_pending(current)) { |
1416 | ret = -EINTR; | |
1417 | break; | |
1418 | } | |
8c7c6e34 KH |
1419 | /* |
1420 | * Rather than hide all in some function, I do this in | |
1421 | * open coded manner. You see what this really does. | |
1422 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1423 | */ | |
1424 | mutex_lock(&set_limit_mutex); | |
1425 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1426 | if (memswlimit < val) { | |
1427 | ret = -EINVAL; | |
1428 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
1429 | break; |
1430 | } | |
8c7c6e34 KH |
1431 | ret = res_counter_set_limit(&memcg->res, val); |
1432 | mutex_unlock(&set_limit_mutex); | |
1433 | ||
1434 | if (!ret) | |
1435 | break; | |
1436 | ||
bced0520 | 1437 | progress = try_to_free_mem_cgroup_pages(memcg, |
2c26fdd7 | 1438 | GFP_KERNEL, false); |
8c7c6e34 KH |
1439 | if (!progress) retry_count--; |
1440 | } | |
14797e23 KM |
1441 | |
1442 | if (!ret) | |
1443 | mem_cgroup_set_inactive_ratio(memcg); | |
1444 | ||
8c7c6e34 KH |
1445 | return ret; |
1446 | } | |
1447 | ||
1448 | int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, | |
1449 | unsigned long long val) | |
1450 | { | |
1451 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
1452 | u64 memlimit, oldusage, curusage; | |
1453 | int ret; | |
1454 | ||
1455 | if (!do_swap_account) | |
1456 | return -EINVAL; | |
1457 | ||
1458 | while (retry_count) { | |
1459 | if (signal_pending(current)) { | |
1460 | ret = -EINTR; | |
1461 | break; | |
1462 | } | |
1463 | /* | |
1464 | * Rather than hide all in some function, I do this in | |
1465 | * open coded manner. You see what this really does. | |
1466 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1467 | */ | |
1468 | mutex_lock(&set_limit_mutex); | |
1469 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1470 | if (memlimit > val) { | |
1471 | ret = -EINVAL; | |
1472 | mutex_unlock(&set_limit_mutex); | |
1473 | break; | |
1474 | } | |
1475 | ret = res_counter_set_limit(&memcg->memsw, val); | |
1476 | mutex_unlock(&set_limit_mutex); | |
1477 | ||
1478 | if (!ret) | |
1479 | break; | |
1480 | ||
1481 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
2c26fdd7 | 1482 | try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, true); |
8c7c6e34 KH |
1483 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
1484 | if (curusage >= oldusage) | |
628f4235 KH |
1485 | retry_count--; |
1486 | } | |
1487 | return ret; | |
1488 | } | |
1489 | ||
cc847582 KH |
1490 | /* |
1491 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
1492 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
1493 | */ | |
f817ed48 | 1494 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 1495 | int node, int zid, enum lru_list lru) |
cc847582 | 1496 | { |
08e552c6 KH |
1497 | struct zone *zone; |
1498 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 1499 | struct page_cgroup *pc, *busy; |
08e552c6 | 1500 | unsigned long flags, loop; |
072c56c1 | 1501 | struct list_head *list; |
f817ed48 | 1502 | int ret = 0; |
072c56c1 | 1503 | |
08e552c6 KH |
1504 | zone = &NODE_DATA(node)->node_zones[zid]; |
1505 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 1506 | list = &mz->lists[lru]; |
cc847582 | 1507 | |
f817ed48 KH |
1508 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
1509 | /* give some margin against EBUSY etc...*/ | |
1510 | loop += 256; | |
1511 | busy = NULL; | |
1512 | while (loop--) { | |
1513 | ret = 0; | |
08e552c6 | 1514 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 1515 | if (list_empty(list)) { |
08e552c6 | 1516 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 1517 | break; |
f817ed48 KH |
1518 | } |
1519 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
1520 | if (busy == pc) { | |
1521 | list_move(&pc->lru, list); | |
1522 | busy = 0; | |
08e552c6 | 1523 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
1524 | continue; |
1525 | } | |
08e552c6 | 1526 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 1527 | |
2c26fdd7 | 1528 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 1529 | if (ret == -ENOMEM) |
52d4b9ac | 1530 | break; |
f817ed48 KH |
1531 | |
1532 | if (ret == -EBUSY || ret == -EINVAL) { | |
1533 | /* found lock contention or "pc" is obsolete. */ | |
1534 | busy = pc; | |
1535 | cond_resched(); | |
1536 | } else | |
1537 | busy = NULL; | |
cc847582 | 1538 | } |
08e552c6 | 1539 | |
f817ed48 KH |
1540 | if (!ret && !list_empty(list)) |
1541 | return -EBUSY; | |
1542 | return ret; | |
cc847582 KH |
1543 | } |
1544 | ||
1545 | /* | |
1546 | * make mem_cgroup's charge to be 0 if there is no task. | |
1547 | * This enables deleting this mem_cgroup. | |
1548 | */ | |
c1e862c1 | 1549 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 1550 | { |
f817ed48 KH |
1551 | int ret; |
1552 | int node, zid, shrink; | |
1553 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 1554 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 1555 | |
cc847582 | 1556 | css_get(&mem->css); |
f817ed48 KH |
1557 | |
1558 | shrink = 0; | |
c1e862c1 KH |
1559 | /* should free all ? */ |
1560 | if (free_all) | |
1561 | goto try_to_free; | |
f817ed48 | 1562 | move_account: |
1ecaab2b | 1563 | while (mem->res.usage > 0) { |
f817ed48 | 1564 | ret = -EBUSY; |
c1e862c1 KH |
1565 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
1566 | goto out; | |
1567 | ret = -EINTR; | |
1568 | if (signal_pending(current)) | |
cc847582 | 1569 | goto out; |
52d4b9ac KH |
1570 | /* This is for making all *used* pages to be on LRU. */ |
1571 | lru_add_drain_all(); | |
f817ed48 KH |
1572 | ret = 0; |
1573 | for_each_node_state(node, N_POSSIBLE) { | |
1574 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { | |
b69408e8 | 1575 | enum lru_list l; |
f817ed48 KH |
1576 | for_each_lru(l) { |
1577 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 1578 | node, zid, l); |
f817ed48 KH |
1579 | if (ret) |
1580 | break; | |
1581 | } | |
1ecaab2b | 1582 | } |
f817ed48 KH |
1583 | if (ret) |
1584 | break; | |
1585 | } | |
1586 | /* it seems parent cgroup doesn't have enough mem */ | |
1587 | if (ret == -ENOMEM) | |
1588 | goto try_to_free; | |
52d4b9ac | 1589 | cond_resched(); |
cc847582 KH |
1590 | } |
1591 | ret = 0; | |
1592 | out: | |
1593 | css_put(&mem->css); | |
1594 | return ret; | |
f817ed48 KH |
1595 | |
1596 | try_to_free: | |
c1e862c1 KH |
1597 | /* returns EBUSY if there is a task or if we come here twice. */ |
1598 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
1599 | ret = -EBUSY; |
1600 | goto out; | |
1601 | } | |
c1e862c1 KH |
1602 | /* we call try-to-free pages for make this cgroup empty */ |
1603 | lru_add_drain_all(); | |
f817ed48 KH |
1604 | /* try to free all pages in this cgroup */ |
1605 | shrink = 1; | |
1606 | while (nr_retries && mem->res.usage > 0) { | |
1607 | int progress; | |
c1e862c1 KH |
1608 | |
1609 | if (signal_pending(current)) { | |
1610 | ret = -EINTR; | |
1611 | goto out; | |
1612 | } | |
f817ed48 | 1613 | progress = try_to_free_mem_cgroup_pages(mem, |
2c26fdd7 | 1614 | GFP_KERNEL, false); |
c1e862c1 | 1615 | if (!progress) { |
f817ed48 | 1616 | nr_retries--; |
c1e862c1 KH |
1617 | /* maybe some writeback is necessary */ |
1618 | congestion_wait(WRITE, HZ/10); | |
1619 | } | |
f817ed48 KH |
1620 | |
1621 | } | |
08e552c6 | 1622 | lru_add_drain(); |
f817ed48 KH |
1623 | /* try move_account...there may be some *locked* pages. */ |
1624 | if (mem->res.usage) | |
1625 | goto move_account; | |
1626 | ret = 0; | |
1627 | goto out; | |
cc847582 KH |
1628 | } |
1629 | ||
c1e862c1 KH |
1630 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
1631 | { | |
1632 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
1633 | } | |
1634 | ||
1635 | ||
18f59ea7 BS |
1636 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
1637 | { | |
1638 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
1639 | } | |
1640 | ||
1641 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
1642 | u64 val) | |
1643 | { | |
1644 | int retval = 0; | |
1645 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1646 | struct cgroup *parent = cont->parent; | |
1647 | struct mem_cgroup *parent_mem = NULL; | |
1648 | ||
1649 | if (parent) | |
1650 | parent_mem = mem_cgroup_from_cont(parent); | |
1651 | ||
1652 | cgroup_lock(); | |
1653 | /* | |
1654 | * If parent's use_hiearchy is set, we can't make any modifications | |
1655 | * in the child subtrees. If it is unset, then the change can | |
1656 | * occur, provided the current cgroup has no children. | |
1657 | * | |
1658 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
1659 | * set if there are no children. | |
1660 | */ | |
1661 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
1662 | (val == 1 || val == 0)) { | |
1663 | if (list_empty(&cont->children)) | |
1664 | mem->use_hierarchy = val; | |
1665 | else | |
1666 | retval = -EBUSY; | |
1667 | } else | |
1668 | retval = -EINVAL; | |
1669 | cgroup_unlock(); | |
1670 | ||
1671 | return retval; | |
1672 | } | |
1673 | ||
2c3daa72 | 1674 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 1675 | { |
8c7c6e34 KH |
1676 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
1677 | u64 val = 0; | |
1678 | int type, name; | |
1679 | ||
1680 | type = MEMFILE_TYPE(cft->private); | |
1681 | name = MEMFILE_ATTR(cft->private); | |
1682 | switch (type) { | |
1683 | case _MEM: | |
1684 | val = res_counter_read_u64(&mem->res, name); | |
1685 | break; | |
1686 | case _MEMSWAP: | |
1687 | if (do_swap_account) | |
1688 | val = res_counter_read_u64(&mem->memsw, name); | |
1689 | break; | |
1690 | default: | |
1691 | BUG(); | |
1692 | break; | |
1693 | } | |
1694 | return val; | |
8cdea7c0 | 1695 | } |
628f4235 KH |
1696 | /* |
1697 | * The user of this function is... | |
1698 | * RES_LIMIT. | |
1699 | */ | |
856c13aa PM |
1700 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
1701 | const char *buffer) | |
8cdea7c0 | 1702 | { |
628f4235 | 1703 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 1704 | int type, name; |
628f4235 KH |
1705 | unsigned long long val; |
1706 | int ret; | |
1707 | ||
8c7c6e34 KH |
1708 | type = MEMFILE_TYPE(cft->private); |
1709 | name = MEMFILE_ATTR(cft->private); | |
1710 | switch (name) { | |
628f4235 KH |
1711 | case RES_LIMIT: |
1712 | /* This function does all necessary parse...reuse it */ | |
1713 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
1714 | if (ret) |
1715 | break; | |
1716 | if (type == _MEM) | |
628f4235 | 1717 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
1718 | else |
1719 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 KH |
1720 | break; |
1721 | default: | |
1722 | ret = -EINVAL; /* should be BUG() ? */ | |
1723 | break; | |
1724 | } | |
1725 | return ret; | |
8cdea7c0 BS |
1726 | } |
1727 | ||
29f2a4da | 1728 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
1729 | { |
1730 | struct mem_cgroup *mem; | |
8c7c6e34 | 1731 | int type, name; |
c84872e1 PE |
1732 | |
1733 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
1734 | type = MEMFILE_TYPE(event); |
1735 | name = MEMFILE_ATTR(event); | |
1736 | switch (name) { | |
29f2a4da | 1737 | case RES_MAX_USAGE: |
8c7c6e34 KH |
1738 | if (type == _MEM) |
1739 | res_counter_reset_max(&mem->res); | |
1740 | else | |
1741 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
1742 | break; |
1743 | case RES_FAILCNT: | |
8c7c6e34 KH |
1744 | if (type == _MEM) |
1745 | res_counter_reset_failcnt(&mem->res); | |
1746 | else | |
1747 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
1748 | break; |
1749 | } | |
85cc59db | 1750 | return 0; |
c84872e1 PE |
1751 | } |
1752 | ||
d2ceb9b7 KH |
1753 | static const struct mem_cgroup_stat_desc { |
1754 | const char *msg; | |
1755 | u64 unit; | |
1756 | } mem_cgroup_stat_desc[] = { | |
1757 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
1758 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
1759 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
1760 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
1761 | }; |
1762 | ||
c64745cf PM |
1763 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
1764 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 1765 | { |
d2ceb9b7 KH |
1766 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
1767 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
1768 | int i; | |
1769 | ||
1770 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1771 | s64 val; | |
1772 | ||
1773 | val = mem_cgroup_read_stat(stat, i); | |
1774 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 1775 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 1776 | } |
6d12e2d8 KH |
1777 | /* showing # of active pages */ |
1778 | { | |
4f98a2fe RR |
1779 | unsigned long active_anon, inactive_anon; |
1780 | unsigned long active_file, inactive_file; | |
7b854121 | 1781 | unsigned long unevictable; |
4f98a2fe RR |
1782 | |
1783 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1784 | LRU_INACTIVE_ANON); | |
1785 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1786 | LRU_ACTIVE_ANON); | |
1787 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1788 | LRU_INACTIVE_FILE); | |
1789 | active_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1790 | LRU_ACTIVE_FILE); | |
7b854121 LS |
1791 | unevictable = mem_cgroup_get_all_zonestat(mem_cont, |
1792 | LRU_UNEVICTABLE); | |
1793 | ||
4f98a2fe RR |
1794 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); |
1795 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); | |
1796 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); | |
1797 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); | |
7b854121 LS |
1798 | cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); |
1799 | ||
6d12e2d8 | 1800 | } |
d2ceb9b7 KH |
1801 | return 0; |
1802 | } | |
1803 | ||
c1e862c1 | 1804 | |
8cdea7c0 BS |
1805 | static struct cftype mem_cgroup_files[] = { |
1806 | { | |
0eea1030 | 1807 | .name = "usage_in_bytes", |
8c7c6e34 | 1808 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 1809 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1810 | }, |
c84872e1 PE |
1811 | { |
1812 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 1813 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 1814 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
1815 | .read_u64 = mem_cgroup_read, |
1816 | }, | |
8cdea7c0 | 1817 | { |
0eea1030 | 1818 | .name = "limit_in_bytes", |
8c7c6e34 | 1819 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 1820 | .write_string = mem_cgroup_write, |
2c3daa72 | 1821 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
1822 | }, |
1823 | { | |
1824 | .name = "failcnt", | |
8c7c6e34 | 1825 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 1826 | .trigger = mem_cgroup_reset, |
2c3daa72 | 1827 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1828 | }, |
d2ceb9b7 KH |
1829 | { |
1830 | .name = "stat", | |
c64745cf | 1831 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 1832 | }, |
c1e862c1 KH |
1833 | { |
1834 | .name = "force_empty", | |
1835 | .trigger = mem_cgroup_force_empty_write, | |
1836 | }, | |
18f59ea7 BS |
1837 | { |
1838 | .name = "use_hierarchy", | |
1839 | .write_u64 = mem_cgroup_hierarchy_write, | |
1840 | .read_u64 = mem_cgroup_hierarchy_read, | |
1841 | }, | |
8cdea7c0 BS |
1842 | }; |
1843 | ||
8c7c6e34 KH |
1844 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
1845 | static struct cftype memsw_cgroup_files[] = { | |
1846 | { | |
1847 | .name = "memsw.usage_in_bytes", | |
1848 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
1849 | .read_u64 = mem_cgroup_read, | |
1850 | }, | |
1851 | { | |
1852 | .name = "memsw.max_usage_in_bytes", | |
1853 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
1854 | .trigger = mem_cgroup_reset, | |
1855 | .read_u64 = mem_cgroup_read, | |
1856 | }, | |
1857 | { | |
1858 | .name = "memsw.limit_in_bytes", | |
1859 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
1860 | .write_string = mem_cgroup_write, | |
1861 | .read_u64 = mem_cgroup_read, | |
1862 | }, | |
1863 | { | |
1864 | .name = "memsw.failcnt", | |
1865 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
1866 | .trigger = mem_cgroup_reset, | |
1867 | .read_u64 = mem_cgroup_read, | |
1868 | }, | |
1869 | }; | |
1870 | ||
1871 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
1872 | { | |
1873 | if (!do_swap_account) | |
1874 | return 0; | |
1875 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
1876 | ARRAY_SIZE(memsw_cgroup_files)); | |
1877 | }; | |
1878 | #else | |
1879 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
1880 | { | |
1881 | return 0; | |
1882 | } | |
1883 | #endif | |
1884 | ||
6d12e2d8 KH |
1885 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1886 | { | |
1887 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 1888 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1889 | enum lru_list l; |
41e3355d | 1890 | int zone, tmp = node; |
1ecaab2b KH |
1891 | /* |
1892 | * This routine is called against possible nodes. | |
1893 | * But it's BUG to call kmalloc() against offline node. | |
1894 | * | |
1895 | * TODO: this routine can waste much memory for nodes which will | |
1896 | * never be onlined. It's better to use memory hotplug callback | |
1897 | * function. | |
1898 | */ | |
41e3355d KH |
1899 | if (!node_state(node, N_NORMAL_MEMORY)) |
1900 | tmp = -1; | |
1901 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
1902 | if (!pn) |
1903 | return 1; | |
1ecaab2b | 1904 | |
6d12e2d8 KH |
1905 | mem->info.nodeinfo[node] = pn; |
1906 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1907 | |
1908 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1909 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
1910 | for_each_lru(l) |
1911 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 1912 | } |
6d12e2d8 KH |
1913 | return 0; |
1914 | } | |
1915 | ||
1ecaab2b KH |
1916 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1917 | { | |
1918 | kfree(mem->info.nodeinfo[node]); | |
1919 | } | |
1920 | ||
c8dad2bb JB |
1921 | static int mem_cgroup_size(void) |
1922 | { | |
1923 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
1924 | return sizeof(struct mem_cgroup) + cpustat_size; | |
1925 | } | |
1926 | ||
33327948 KH |
1927 | static struct mem_cgroup *mem_cgroup_alloc(void) |
1928 | { | |
1929 | struct mem_cgroup *mem; | |
c8dad2bb | 1930 | int size = mem_cgroup_size(); |
33327948 | 1931 | |
c8dad2bb JB |
1932 | if (size < PAGE_SIZE) |
1933 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 1934 | else |
c8dad2bb | 1935 | mem = vmalloc(size); |
33327948 KH |
1936 | |
1937 | if (mem) | |
c8dad2bb | 1938 | memset(mem, 0, size); |
33327948 KH |
1939 | return mem; |
1940 | } | |
1941 | ||
8c7c6e34 KH |
1942 | /* |
1943 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
1944 | * (scanning all at force_empty is too costly...) | |
1945 | * | |
1946 | * Instead of clearing all references at force_empty, we remember | |
1947 | * the number of reference from swap_cgroup and free mem_cgroup when | |
1948 | * it goes down to 0. | |
1949 | * | |
1950 | * When mem_cgroup is destroyed, mem->obsolete will be set to 0 and | |
1951 | * entry which points to this memcg will be ignore at swapin. | |
1952 | * | |
1953 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
1954 | */ | |
1955 | ||
33327948 KH |
1956 | static void mem_cgroup_free(struct mem_cgroup *mem) |
1957 | { | |
08e552c6 KH |
1958 | int node; |
1959 | ||
8c7c6e34 KH |
1960 | if (atomic_read(&mem->refcnt) > 0) |
1961 | return; | |
08e552c6 KH |
1962 | |
1963 | ||
1964 | for_each_node_state(node, N_POSSIBLE) | |
1965 | free_mem_cgroup_per_zone_info(mem, node); | |
1966 | ||
c8dad2bb | 1967 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
1968 | kfree(mem); |
1969 | else | |
1970 | vfree(mem); | |
1971 | } | |
1972 | ||
8c7c6e34 KH |
1973 | static void mem_cgroup_get(struct mem_cgroup *mem) |
1974 | { | |
1975 | atomic_inc(&mem->refcnt); | |
1976 | } | |
1977 | ||
1978 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
1979 | { | |
1980 | if (atomic_dec_and_test(&mem->refcnt)) { | |
1981 | if (!mem->obsolete) | |
1982 | return; | |
1983 | mem_cgroup_free(mem); | |
1984 | } | |
1985 | } | |
1986 | ||
33327948 | 1987 | |
c077719b KH |
1988 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
1989 | static void __init enable_swap_cgroup(void) | |
1990 | { | |
f8d66542 | 1991 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
1992 | do_swap_account = 1; |
1993 | } | |
1994 | #else | |
1995 | static void __init enable_swap_cgroup(void) | |
1996 | { | |
1997 | } | |
1998 | #endif | |
1999 | ||
8cdea7c0 BS |
2000 | static struct cgroup_subsys_state * |
2001 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
2002 | { | |
28dbc4b6 | 2003 | struct mem_cgroup *mem, *parent; |
6d12e2d8 | 2004 | int node; |
8cdea7c0 | 2005 | |
c8dad2bb JB |
2006 | mem = mem_cgroup_alloc(); |
2007 | if (!mem) | |
2008 | return ERR_PTR(-ENOMEM); | |
78fb7466 | 2009 | |
6d12e2d8 KH |
2010 | for_each_node_state(node, N_POSSIBLE) |
2011 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
2012 | goto free_out; | |
c077719b | 2013 | /* root ? */ |
28dbc4b6 | 2014 | if (cont->parent == NULL) { |
c077719b | 2015 | enable_swap_cgroup(); |
28dbc4b6 | 2016 | parent = NULL; |
18f59ea7 | 2017 | } else { |
28dbc4b6 | 2018 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 BS |
2019 | mem->use_hierarchy = parent->use_hierarchy; |
2020 | } | |
28dbc4b6 | 2021 | |
18f59ea7 BS |
2022 | if (parent && parent->use_hierarchy) { |
2023 | res_counter_init(&mem->res, &parent->res); | |
2024 | res_counter_init(&mem->memsw, &parent->memsw); | |
2025 | } else { | |
2026 | res_counter_init(&mem->res, NULL); | |
2027 | res_counter_init(&mem->memsw, NULL); | |
2028 | } | |
14797e23 | 2029 | mem_cgroup_set_inactive_ratio(mem); |
6d61ef40 BS |
2030 | mem->last_scanned_child = NULL; |
2031 | ||
8cdea7c0 | 2032 | return &mem->css; |
6d12e2d8 KH |
2033 | free_out: |
2034 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 2035 | free_mem_cgroup_per_zone_info(mem, node); |
c8dad2bb | 2036 | mem_cgroup_free(mem); |
2dda81ca | 2037 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
2038 | } |
2039 | ||
df878fb0 KH |
2040 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
2041 | struct cgroup *cont) | |
2042 | { | |
2043 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 | 2044 | mem->obsolete = 1; |
c1e862c1 | 2045 | mem_cgroup_force_empty(mem, false); |
df878fb0 KH |
2046 | } |
2047 | ||
8cdea7c0 BS |
2048 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
2049 | struct cgroup *cont) | |
2050 | { | |
33327948 | 2051 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
2052 | } |
2053 | ||
2054 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
2055 | struct cgroup *cont) | |
2056 | { | |
8c7c6e34 KH |
2057 | int ret; |
2058 | ||
2059 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
2060 | ARRAY_SIZE(mem_cgroup_files)); | |
2061 | ||
2062 | if (!ret) | |
2063 | ret = register_memsw_files(cont, ss); | |
2064 | return ret; | |
8cdea7c0 BS |
2065 | } |
2066 | ||
67e465a7 BS |
2067 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
2068 | struct cgroup *cont, | |
2069 | struct cgroup *old_cont, | |
2070 | struct task_struct *p) | |
2071 | { | |
67e465a7 | 2072 | /* |
f9717d28 NK |
2073 | * FIXME: It's better to move charges of this process from old |
2074 | * memcg to new memcg. But it's just on TODO-List now. | |
67e465a7 | 2075 | */ |
67e465a7 BS |
2076 | } |
2077 | ||
8cdea7c0 BS |
2078 | struct cgroup_subsys mem_cgroup_subsys = { |
2079 | .name = "memory", | |
2080 | .subsys_id = mem_cgroup_subsys_id, | |
2081 | .create = mem_cgroup_create, | |
df878fb0 | 2082 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
2083 | .destroy = mem_cgroup_destroy, |
2084 | .populate = mem_cgroup_populate, | |
67e465a7 | 2085 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 2086 | .early_init = 0, |
8cdea7c0 | 2087 | }; |
c077719b KH |
2088 | |
2089 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2090 | ||
2091 | static int __init disable_swap_account(char *s) | |
2092 | { | |
2093 | really_do_swap_account = 0; | |
2094 | return 1; | |
2095 | } | |
2096 | __setup("noswapaccount", disable_swap_account); | |
2097 | #endif |