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> |
d52aa412 | 24 | #include <linux/smp.h> |
8a9f3ccd | 25 | #include <linux/page-flags.h> |
66e1707b | 26 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
27 | #include <linux/bit_spinlock.h> |
28 | #include <linux/rcupdate.h> | |
b6ac57d5 | 29 | #include <linux/slab.h> |
66e1707b BS |
30 | #include <linux/swap.h> |
31 | #include <linux/spinlock.h> | |
32 | #include <linux/fs.h> | |
d2ceb9b7 | 33 | #include <linux/seq_file.h> |
33327948 | 34 | #include <linux/vmalloc.h> |
b69408e8 | 35 | #include <linux/mm_inline.h> |
52d4b9ac | 36 | #include <linux/page_cgroup.h> |
8cdea7c0 | 37 | |
8697d331 BS |
38 | #include <asm/uaccess.h> |
39 | ||
a181b0e8 | 40 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 41 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 42 | |
d52aa412 KH |
43 | /* |
44 | * Statistics for memory cgroup. | |
45 | */ | |
46 | enum mem_cgroup_stat_index { | |
47 | /* | |
48 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
49 | */ | |
50 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
51 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
55e462b0 BR |
52 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
53 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
54 | |
55 | MEM_CGROUP_STAT_NSTATS, | |
56 | }; | |
57 | ||
58 | struct mem_cgroup_stat_cpu { | |
59 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
60 | } ____cacheline_aligned_in_smp; | |
61 | ||
62 | struct mem_cgroup_stat { | |
63 | struct mem_cgroup_stat_cpu cpustat[NR_CPUS]; | |
64 | }; | |
65 | ||
66 | /* | |
67 | * For accounting under irq disable, no need for increment preempt count. | |
68 | */ | |
addb9efe | 69 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
70 | enum mem_cgroup_stat_index idx, int val) |
71 | { | |
addb9efe | 72 | stat->count[idx] += val; |
d52aa412 KH |
73 | } |
74 | ||
75 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
76 | enum mem_cgroup_stat_index idx) | |
77 | { | |
78 | int cpu; | |
79 | s64 ret = 0; | |
80 | for_each_possible_cpu(cpu) | |
81 | ret += stat->cpustat[cpu].count[idx]; | |
82 | return ret; | |
83 | } | |
84 | ||
6d12e2d8 KH |
85 | /* |
86 | * per-zone information in memory controller. | |
87 | */ | |
6d12e2d8 | 88 | struct mem_cgroup_per_zone { |
072c56c1 KH |
89 | /* |
90 | * spin_lock to protect the per cgroup LRU | |
91 | */ | |
92 | spinlock_t lru_lock; | |
b69408e8 CL |
93 | struct list_head lists[NR_LRU_LISTS]; |
94 | unsigned long count[NR_LRU_LISTS]; | |
6d12e2d8 KH |
95 | }; |
96 | /* Macro for accessing counter */ | |
97 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
98 | ||
99 | struct mem_cgroup_per_node { | |
100 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
101 | }; | |
102 | ||
103 | struct mem_cgroup_lru_info { | |
104 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
105 | }; | |
106 | ||
8cdea7c0 BS |
107 | /* |
108 | * The memory controller data structure. The memory controller controls both | |
109 | * page cache and RSS per cgroup. We would eventually like to provide | |
110 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
111 | * to help the administrator determine what knobs to tune. | |
112 | * | |
113 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
114 | * we hit the water mark. May be even add a low water mark, such that |
115 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
116 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
117 | */ |
118 | struct mem_cgroup { | |
119 | struct cgroup_subsys_state css; | |
120 | /* | |
121 | * the counter to account for memory usage | |
122 | */ | |
123 | struct res_counter res; | |
78fb7466 PE |
124 | /* |
125 | * Per cgroup active and inactive list, similar to the | |
126 | * per zone LRU lists. | |
78fb7466 | 127 | */ |
6d12e2d8 | 128 | struct mem_cgroup_lru_info info; |
072c56c1 | 129 | |
6c48a1d0 | 130 | int prev_priority; /* for recording reclaim priority */ |
d52aa412 KH |
131 | /* |
132 | * statistics. | |
133 | */ | |
134 | struct mem_cgroup_stat stat; | |
8cdea7c0 | 135 | }; |
8869b8f6 | 136 | static struct mem_cgroup init_mem_cgroup; |
8cdea7c0 | 137 | |
217bc319 KH |
138 | enum charge_type { |
139 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
140 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 141 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 KH |
142 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
143 | NR_CHARGE_TYPE, | |
144 | }; | |
145 | ||
52d4b9ac KH |
146 | /* only for here (for easy reading.) */ |
147 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
148 | #define PCGF_USED (1UL << PCG_USED) | |
149 | #define PCGF_ACTIVE (1UL << PCG_ACTIVE) | |
150 | #define PCGF_LOCK (1UL << PCG_LOCK) | |
151 | #define PCGF_FILE (1UL << PCG_FILE) | |
c05555b5 KH |
152 | static const unsigned long |
153 | pcg_default_flags[NR_CHARGE_TYPE] = { | |
52d4b9ac KH |
154 | PCGF_CACHE | PCGF_FILE | PCGF_USED | PCGF_LOCK, /* File Cache */ |
155 | PCGF_ACTIVE | PCGF_USED | PCGF_LOCK, /* Anon */ | |
156 | PCGF_ACTIVE | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ | |
157 | 0, /* FORCE */ | |
217bc319 KH |
158 | }; |
159 | ||
d52aa412 KH |
160 | /* |
161 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
162 | */ | |
c05555b5 KH |
163 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
164 | struct page_cgroup *pc, | |
165 | bool charge) | |
d52aa412 KH |
166 | { |
167 | int val = (charge)? 1 : -1; | |
168 | struct mem_cgroup_stat *stat = &mem->stat; | |
addb9efe | 169 | struct mem_cgroup_stat_cpu *cpustat; |
d52aa412 | 170 | |
8869b8f6 | 171 | VM_BUG_ON(!irqs_disabled()); |
addb9efe KH |
172 | |
173 | cpustat = &stat->cpustat[smp_processor_id()]; | |
c05555b5 | 174 | if (PageCgroupCache(pc)) |
addb9efe | 175 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 176 | else |
addb9efe | 177 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
178 | |
179 | if (charge) | |
addb9efe | 180 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
181 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
182 | else | |
addb9efe | 183 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 184 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
6d12e2d8 KH |
185 | } |
186 | ||
d5b69e38 | 187 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
188 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
189 | { | |
6d12e2d8 KH |
190 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
191 | } | |
192 | ||
d5b69e38 | 193 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
194 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
195 | { | |
196 | struct mem_cgroup *mem = pc->mem_cgroup; | |
197 | int nid = page_cgroup_nid(pc); | |
198 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 199 | |
6d12e2d8 KH |
200 | return mem_cgroup_zoneinfo(mem, nid, zid); |
201 | } | |
202 | ||
203 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
b69408e8 | 204 | enum lru_list idx) |
6d12e2d8 KH |
205 | { |
206 | int nid, zid; | |
207 | struct mem_cgroup_per_zone *mz; | |
208 | u64 total = 0; | |
209 | ||
210 | for_each_online_node(nid) | |
211 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
212 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
213 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
214 | } | |
215 | return total; | |
d52aa412 KH |
216 | } |
217 | ||
d5b69e38 | 218 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
219 | { |
220 | return container_of(cgroup_subsys_state(cont, | |
221 | mem_cgroup_subsys_id), struct mem_cgroup, | |
222 | css); | |
223 | } | |
224 | ||
cf475ad2 | 225 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 226 | { |
31a78f23 BS |
227 | /* |
228 | * mm_update_next_owner() may clear mm->owner to NULL | |
229 | * if it races with swapoff, page migration, etc. | |
230 | * So this can be called with p == NULL. | |
231 | */ | |
232 | if (unlikely(!p)) | |
233 | return NULL; | |
234 | ||
78fb7466 PE |
235 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
236 | struct mem_cgroup, css); | |
237 | } | |
238 | ||
3eae90c3 KH |
239 | static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, |
240 | struct page_cgroup *pc) | |
6d12e2d8 | 241 | { |
4f98a2fe RR |
242 | int lru = LRU_BASE; |
243 | ||
c05555b5 | 244 | if (PageCgroupUnevictable(pc)) |
894bc310 LS |
245 | lru = LRU_UNEVICTABLE; |
246 | else { | |
c05555b5 | 247 | if (PageCgroupActive(pc)) |
894bc310 | 248 | lru += LRU_ACTIVE; |
c05555b5 | 249 | if (PageCgroupFile(pc)) |
894bc310 LS |
250 | lru += LRU_FILE; |
251 | } | |
6d12e2d8 | 252 | |
b69408e8 | 253 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
6d12e2d8 | 254 | |
c05555b5 | 255 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, false); |
508b7be0 | 256 | list_del(&pc->lru); |
6d12e2d8 KH |
257 | } |
258 | ||
3eae90c3 KH |
259 | static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, |
260 | struct page_cgroup *pc) | |
6d12e2d8 | 261 | { |
4f98a2fe | 262 | int lru = LRU_BASE; |
b69408e8 | 263 | |
c05555b5 | 264 | if (PageCgroupUnevictable(pc)) |
894bc310 LS |
265 | lru = LRU_UNEVICTABLE; |
266 | else { | |
c05555b5 | 267 | if (PageCgroupActive(pc)) |
894bc310 | 268 | lru += LRU_ACTIVE; |
c05555b5 | 269 | if (PageCgroupFile(pc)) |
894bc310 LS |
270 | lru += LRU_FILE; |
271 | } | |
b69408e8 CL |
272 | |
273 | MEM_CGROUP_ZSTAT(mz, lru) += 1; | |
274 | list_add(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 | 275 | |
c05555b5 | 276 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true); |
6d12e2d8 KH |
277 | } |
278 | ||
894bc310 | 279 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru) |
66e1707b | 280 | { |
6d12e2d8 | 281 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); |
c05555b5 KH |
282 | int active = PageCgroupActive(pc); |
283 | int file = PageCgroupFile(pc); | |
284 | int unevictable = PageCgroupUnevictable(pc); | |
894bc310 LS |
285 | enum lru_list from = unevictable ? LRU_UNEVICTABLE : |
286 | (LRU_FILE * !!file + !!active); | |
6d12e2d8 | 287 | |
894bc310 LS |
288 | if (lru == from) |
289 | return; | |
b69408e8 | 290 | |
894bc310 | 291 | MEM_CGROUP_ZSTAT(mz, from) -= 1; |
c05555b5 KH |
292 | /* |
293 | * However this is done under mz->lru_lock, another flags, which | |
294 | * are not related to LRU, will be modified from out-of-lock. | |
295 | * We have to use atomic set/clear flags. | |
296 | */ | |
894bc310 | 297 | if (is_unevictable_lru(lru)) { |
c05555b5 KH |
298 | ClearPageCgroupActive(pc); |
299 | SetPageCgroupUnevictable(pc); | |
894bc310 LS |
300 | } else { |
301 | if (is_active_lru(lru)) | |
c05555b5 | 302 | SetPageCgroupActive(pc); |
894bc310 | 303 | else |
c05555b5 KH |
304 | ClearPageCgroupActive(pc); |
305 | ClearPageCgroupUnevictable(pc); | |
894bc310 | 306 | } |
b69408e8 | 307 | |
b69408e8 CL |
308 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
309 | list_move(&pc->lru, &mz->lists[lru]); | |
66e1707b BS |
310 | } |
311 | ||
4c4a2214 DR |
312 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
313 | { | |
314 | int ret; | |
315 | ||
316 | task_lock(task); | |
bd845e38 | 317 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
318 | task_unlock(task); |
319 | return ret; | |
320 | } | |
321 | ||
66e1707b BS |
322 | /* |
323 | * This routine assumes that the appropriate zone's lru lock is already held | |
324 | */ | |
894bc310 | 325 | void mem_cgroup_move_lists(struct page *page, enum lru_list lru) |
66e1707b | 326 | { |
427d5416 | 327 | struct page_cgroup *pc; |
072c56c1 KH |
328 | struct mem_cgroup_per_zone *mz; |
329 | unsigned long flags; | |
330 | ||
cede86ac LZ |
331 | if (mem_cgroup_subsys.disabled) |
332 | return; | |
333 | ||
2680eed7 HD |
334 | /* |
335 | * We cannot lock_page_cgroup while holding zone's lru_lock, | |
336 | * because other holders of lock_page_cgroup can be interrupted | |
337 | * with an attempt to rotate_reclaimable_page. But we cannot | |
338 | * safely get to page_cgroup without it, so just try_lock it: | |
339 | * mem_cgroup_isolate_pages allows for page left on wrong list. | |
340 | */ | |
52d4b9ac KH |
341 | pc = lookup_page_cgroup(page); |
342 | if (!trylock_page_cgroup(pc)) | |
66e1707b | 343 | return; |
52d4b9ac | 344 | if (pc && PageCgroupUsed(pc)) { |
2680eed7 | 345 | mz = page_cgroup_zoneinfo(pc); |
2680eed7 | 346 | spin_lock_irqsave(&mz->lru_lock, flags); |
894bc310 | 347 | __mem_cgroup_move_lists(pc, lru); |
2680eed7 | 348 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9b3c0a07 | 349 | } |
52d4b9ac | 350 | unlock_page_cgroup(pc); |
66e1707b BS |
351 | } |
352 | ||
58ae83db KH |
353 | /* |
354 | * Calculate mapped_ratio under memory controller. This will be used in | |
355 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
356 | */ | |
357 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
358 | { | |
359 | long total, rss; | |
360 | ||
361 | /* | |
362 | * usage is recorded in bytes. But, here, we assume the number of | |
363 | * physical pages can be represented by "long" on any arch. | |
364 | */ | |
365 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
366 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
367 | return (int)((rss * 100L) / total); | |
368 | } | |
8869b8f6 | 369 | |
6c48a1d0 KH |
370 | /* |
371 | * prev_priority control...this will be used in memory reclaim path. | |
372 | */ | |
373 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
374 | { | |
375 | return mem->prev_priority; | |
376 | } | |
377 | ||
378 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
379 | { | |
380 | if (priority < mem->prev_priority) | |
381 | mem->prev_priority = priority; | |
382 | } | |
383 | ||
384 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
385 | { | |
386 | mem->prev_priority = priority; | |
387 | } | |
388 | ||
cc38108e KH |
389 | /* |
390 | * Calculate # of pages to be scanned in this priority/zone. | |
391 | * See also vmscan.c | |
392 | * | |
393 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
394 | * (see include/linux/mmzone.h) | |
395 | */ | |
396 | ||
b69408e8 CL |
397 | long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, |
398 | int priority, enum lru_list lru) | |
cc38108e | 399 | { |
b69408e8 | 400 | long nr_pages; |
cc38108e KH |
401 | int nid = zone->zone_pgdat->node_id; |
402 | int zid = zone_idx(zone); | |
403 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
404 | ||
b69408e8 | 405 | nr_pages = MEM_CGROUP_ZSTAT(mz, lru); |
cc38108e | 406 | |
b69408e8 | 407 | return (nr_pages >> priority); |
cc38108e KH |
408 | } |
409 | ||
66e1707b BS |
410 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
411 | struct list_head *dst, | |
412 | unsigned long *scanned, int order, | |
413 | int mode, struct zone *z, | |
414 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 415 | int active, int file) |
66e1707b BS |
416 | { |
417 | unsigned long nr_taken = 0; | |
418 | struct page *page; | |
419 | unsigned long scan; | |
420 | LIST_HEAD(pc_list); | |
421 | struct list_head *src; | |
ff7283fa | 422 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
423 | int nid = z->zone_pgdat->node_id; |
424 | int zid = zone_idx(z); | |
425 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 426 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 427 | |
cf475ad2 | 428 | BUG_ON(!mem_cont); |
1ecaab2b | 429 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 430 | src = &mz->lists[lru]; |
66e1707b | 431 | |
072c56c1 | 432 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
433 | scan = 0; |
434 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 435 | if (scan >= nr_to_scan) |
ff7283fa | 436 | break; |
52d4b9ac KH |
437 | if (unlikely(!PageCgroupUsed(pc))) |
438 | continue; | |
66e1707b | 439 | page = pc->page; |
66e1707b | 440 | |
436c6541 | 441 | if (unlikely(!PageLRU(page))) |
ff7283fa | 442 | continue; |
ff7283fa | 443 | |
4f98a2fe RR |
444 | /* |
445 | * TODO: play better with lumpy reclaim, grabbing anything. | |
446 | */ | |
894bc310 LS |
447 | if (PageUnevictable(page) || |
448 | (PageActive(page) && !active) || | |
449 | (!PageActive(page) && active)) { | |
450 | __mem_cgroup_move_lists(pc, page_lru(page)); | |
66e1707b BS |
451 | continue; |
452 | } | |
453 | ||
436c6541 HD |
454 | scan++; |
455 | list_move(&pc->lru, &pc_list); | |
66e1707b | 456 | |
4f98a2fe | 457 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
458 | list_move(&page->lru, dst); |
459 | nr_taken++; | |
460 | } | |
461 | } | |
462 | ||
463 | list_splice(&pc_list, src); | |
072c56c1 | 464 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
465 | |
466 | *scanned = scan; | |
467 | return nr_taken; | |
468 | } | |
469 | ||
7a81b88c KH |
470 | |
471 | /** | |
472 | * mem_cgroup_try_charge - get charge of PAGE_SIZE. | |
473 | * @mm: an mm_struct which is charged against. (when *memcg is NULL) | |
474 | * @gfp_mask: gfp_mask for reclaim. | |
475 | * @memcg: a pointer to memory cgroup which is charged against. | |
476 | * | |
477 | * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated | |
478 | * memory cgroup from @mm is got and stored in *memcg. | |
479 | * | |
480 | * Returns 0 if success. -ENOMEM at failure. | |
8a9f3ccd | 481 | */ |
7a81b88c KH |
482 | |
483 | int mem_cgroup_try_charge(struct mm_struct *mm, | |
484 | gfp_t gfp_mask, struct mem_cgroup **memcg) | |
8a9f3ccd BS |
485 | { |
486 | struct mem_cgroup *mem; | |
7a81b88c | 487 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
8a9f3ccd | 488 | /* |
3be91277 HD |
489 | * We always charge the cgroup the mm_struct belongs to. |
490 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
491 | * thread group leader migrates. It's possible that mm is not |
492 | * set, if so charge the init_mm (happens for pagecache usage). | |
493 | */ | |
7a81b88c | 494 | if (likely(!*memcg)) { |
e8589cc1 KH |
495 | rcu_read_lock(); |
496 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
497 | if (unlikely(!mem)) { |
498 | rcu_read_unlock(); | |
31a78f23 BS |
499 | return 0; |
500 | } | |
e8589cc1 KH |
501 | /* |
502 | * For every charge from the cgroup, increment reference count | |
503 | */ | |
504 | css_get(&mem->css); | |
7a81b88c | 505 | *memcg = mem; |
e8589cc1 KH |
506 | rcu_read_unlock(); |
507 | } else { | |
7a81b88c KH |
508 | mem = *memcg; |
509 | css_get(&mem->css); | |
e8589cc1 | 510 | } |
8a9f3ccd | 511 | |
7a81b88c | 512 | |
addb9efe | 513 | while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) { |
3be91277 | 514 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 515 | goto nomem; |
e1a1cd59 BS |
516 | |
517 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
518 | continue; |
519 | ||
520 | /* | |
8869b8f6 HD |
521 | * try_to_free_mem_cgroup_pages() might not give us a full |
522 | * picture of reclaim. Some pages are reclaimed and might be | |
523 | * moved to swap cache or just unmapped from the cgroup. | |
524 | * Check the limit again to see if the reclaim reduced the | |
525 | * current usage of the cgroup before giving up | |
526 | */ | |
66e1707b BS |
527 | if (res_counter_check_under_limit(&mem->res)) |
528 | continue; | |
3be91277 HD |
529 | |
530 | if (!nr_retries--) { | |
531 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
7a81b88c | 532 | goto nomem; |
66e1707b | 533 | } |
8a9f3ccd | 534 | } |
7a81b88c KH |
535 | return 0; |
536 | nomem: | |
537 | css_put(&mem->css); | |
538 | return -ENOMEM; | |
539 | } | |
8a9f3ccd | 540 | |
7a81b88c KH |
541 | /* |
542 | * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be | |
543 | * USED state. If already USED, uncharge and return. | |
544 | */ | |
545 | ||
546 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
547 | struct page_cgroup *pc, | |
548 | enum charge_type ctype) | |
549 | { | |
550 | struct mem_cgroup_per_zone *mz; | |
551 | unsigned long flags; | |
552 | ||
553 | /* try_charge() can return NULL to *memcg, taking care of it. */ | |
554 | if (!mem) | |
555 | return; | |
52d4b9ac KH |
556 | |
557 | lock_page_cgroup(pc); | |
558 | if (unlikely(PageCgroupUsed(pc))) { | |
559 | unlock_page_cgroup(pc); | |
560 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
561 | css_put(&mem->css); | |
7a81b88c | 562 | return; |
52d4b9ac | 563 | } |
8a9f3ccd | 564 | pc->mem_cgroup = mem; |
508b7be0 KH |
565 | /* |
566 | * If a page is accounted as a page cache, insert to inactive list. | |
567 | * If anon, insert to active list. | |
568 | */ | |
c05555b5 | 569 | pc->flags = pcg_default_flags[ctype]; |
3be91277 | 570 | |
072c56c1 | 571 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 572 | |
072c56c1 | 573 | spin_lock_irqsave(&mz->lru_lock, flags); |
3eae90c3 | 574 | __mem_cgroup_add_list(mz, pc); |
072c56c1 | 575 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
52d4b9ac | 576 | unlock_page_cgroup(pc); |
7a81b88c | 577 | } |
66e1707b | 578 | |
7a81b88c KH |
579 | /* |
580 | * Charge the memory controller for page usage. | |
581 | * Return | |
582 | * 0 if the charge was successful | |
583 | * < 0 if the cgroup is over its limit | |
584 | */ | |
585 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
586 | gfp_t gfp_mask, enum charge_type ctype, | |
587 | struct mem_cgroup *memcg) | |
588 | { | |
589 | struct mem_cgroup *mem; | |
590 | struct page_cgroup *pc; | |
591 | int ret; | |
592 | ||
593 | pc = lookup_page_cgroup(page); | |
594 | /* can happen at boot */ | |
595 | if (unlikely(!pc)) | |
596 | return 0; | |
597 | prefetchw(pc); | |
598 | ||
599 | mem = memcg; | |
600 | ret = mem_cgroup_try_charge(mm, gfp_mask, &mem); | |
601 | if (ret) | |
602 | return ret; | |
603 | ||
604 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 605 | return 0; |
8a9f3ccd BS |
606 | } |
607 | ||
7a81b88c KH |
608 | int mem_cgroup_newpage_charge(struct page *page, |
609 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 610 | { |
cede86ac LZ |
611 | if (mem_cgroup_subsys.disabled) |
612 | return 0; | |
52d4b9ac KH |
613 | if (PageCompound(page)) |
614 | return 0; | |
69029cd5 KH |
615 | /* |
616 | * If already mapped, we don't have to account. | |
617 | * If page cache, page->mapping has address_space. | |
618 | * But page->mapping may have out-of-use anon_vma pointer, | |
619 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
620 | * is NULL. | |
621 | */ | |
622 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
623 | return 0; | |
624 | if (unlikely(!mm)) | |
625 | mm = &init_mm; | |
217bc319 | 626 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 627 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
628 | } |
629 | ||
e1a1cd59 BS |
630 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
631 | gfp_t gfp_mask) | |
8697d331 | 632 | { |
cede86ac LZ |
633 | if (mem_cgroup_subsys.disabled) |
634 | return 0; | |
52d4b9ac KH |
635 | if (PageCompound(page)) |
636 | return 0; | |
accf163e KH |
637 | /* |
638 | * Corner case handling. This is called from add_to_page_cache() | |
639 | * in usual. But some FS (shmem) precharges this page before calling it | |
640 | * and call add_to_page_cache() with GFP_NOWAIT. | |
641 | * | |
642 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
643 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
644 | * charge twice. (It works but has to pay a bit larger cost.) | |
645 | */ | |
646 | if (!(gfp_mask & __GFP_WAIT)) { | |
647 | struct page_cgroup *pc; | |
648 | ||
52d4b9ac KH |
649 | |
650 | pc = lookup_page_cgroup(page); | |
651 | if (!pc) | |
652 | return 0; | |
653 | lock_page_cgroup(pc); | |
654 | if (PageCgroupUsed(pc)) { | |
655 | unlock_page_cgroup(pc); | |
accf163e KH |
656 | return 0; |
657 | } | |
52d4b9ac | 658 | unlock_page_cgroup(pc); |
accf163e KH |
659 | } |
660 | ||
69029cd5 | 661 | if (unlikely(!mm)) |
8697d331 | 662 | mm = &init_mm; |
accf163e | 663 | |
c05555b5 KH |
664 | if (page_is_file_cache(page)) |
665 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 666 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
c05555b5 KH |
667 | else |
668 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
669 | MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL); | |
e8589cc1 KH |
670 | } |
671 | ||
7a81b88c KH |
672 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
673 | { | |
674 | struct page_cgroup *pc; | |
675 | ||
676 | if (mem_cgroup_subsys.disabled) | |
677 | return; | |
678 | if (!ptr) | |
679 | return; | |
680 | pc = lookup_page_cgroup(page); | |
681 | __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
682 | } | |
683 | ||
684 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
685 | { | |
686 | if (mem_cgroup_subsys.disabled) | |
687 | return; | |
688 | if (!mem) | |
689 | return; | |
690 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
691 | css_put(&mem->css); | |
692 | } | |
693 | ||
694 | ||
8a9f3ccd | 695 | /* |
69029cd5 | 696 | * uncharge if !page_mapped(page) |
8a9f3ccd | 697 | */ |
69029cd5 KH |
698 | static void |
699 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) | |
8a9f3ccd | 700 | { |
8289546e | 701 | struct page_cgroup *pc; |
8a9f3ccd | 702 | struct mem_cgroup *mem; |
072c56c1 | 703 | struct mem_cgroup_per_zone *mz; |
66e1707b | 704 | unsigned long flags; |
8a9f3ccd | 705 | |
4077960e BS |
706 | if (mem_cgroup_subsys.disabled) |
707 | return; | |
708 | ||
8697d331 | 709 | /* |
3c541e14 | 710 | * Check if our page_cgroup is valid |
8697d331 | 711 | */ |
52d4b9ac KH |
712 | pc = lookup_page_cgroup(page); |
713 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
714 | return; | |
b9c565d5 | 715 | |
52d4b9ac KH |
716 | lock_page_cgroup(pc); |
717 | if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED && page_mapped(page)) | |
718 | || !PageCgroupUsed(pc)) { | |
719 | /* This happens at race in zap_pte_range() and do_swap_page()*/ | |
720 | unlock_page_cgroup(pc); | |
721 | return; | |
722 | } | |
723 | ClearPageCgroupUsed(pc); | |
724 | mem = pc->mem_cgroup; | |
b9c565d5 | 725 | |
69029cd5 KH |
726 | mz = page_cgroup_zoneinfo(pc); |
727 | spin_lock_irqsave(&mz->lru_lock, flags); | |
728 | __mem_cgroup_remove_list(mz, pc); | |
729 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
52d4b9ac | 730 | unlock_page_cgroup(pc); |
fb59e9f1 | 731 | |
69029cd5 KH |
732 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
733 | css_put(&mem->css); | |
6d12e2d8 | 734 | |
69029cd5 | 735 | return; |
3c541e14 BS |
736 | } |
737 | ||
69029cd5 KH |
738 | void mem_cgroup_uncharge_page(struct page *page) |
739 | { | |
52d4b9ac KH |
740 | /* early check. */ |
741 | if (page_mapped(page)) | |
742 | return; | |
743 | if (page->mapping && !PageAnon(page)) | |
744 | return; | |
69029cd5 KH |
745 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
746 | } | |
747 | ||
748 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
749 | { | |
750 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 751 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
752 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
753 | } | |
754 | ||
ae41be37 | 755 | /* |
01b1ae63 KH |
756 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
757 | * page belongs to. | |
ae41be37 | 758 | */ |
01b1ae63 | 759 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
760 | { |
761 | struct page_cgroup *pc; | |
e8589cc1 | 762 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 763 | int ret = 0; |
8869b8f6 | 764 | |
4077960e BS |
765 | if (mem_cgroup_subsys.disabled) |
766 | return 0; | |
767 | ||
52d4b9ac KH |
768 | pc = lookup_page_cgroup(page); |
769 | lock_page_cgroup(pc); | |
770 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
771 | mem = pc->mem_cgroup; |
772 | css_get(&mem->css); | |
e8589cc1 | 773 | } |
52d4b9ac | 774 | unlock_page_cgroup(pc); |
01b1ae63 | 775 | |
e8589cc1 | 776 | if (mem) { |
01b1ae63 | 777 | ret = mem_cgroup_try_charge(NULL, GFP_HIGHUSER_MOVABLE, &mem); |
e8589cc1 KH |
778 | css_put(&mem->css); |
779 | } | |
01b1ae63 | 780 | *ptr = mem; |
e8589cc1 | 781 | return ret; |
ae41be37 | 782 | } |
8869b8f6 | 783 | |
69029cd5 | 784 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
785 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
786 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 787 | { |
01b1ae63 KH |
788 | struct page *target, *unused; |
789 | struct page_cgroup *pc; | |
790 | enum charge_type ctype; | |
791 | ||
792 | if (!mem) | |
793 | return; | |
794 | ||
795 | /* at migration success, oldpage->mapping is NULL. */ | |
796 | if (oldpage->mapping) { | |
797 | target = oldpage; | |
798 | unused = NULL; | |
799 | } else { | |
800 | target = newpage; | |
801 | unused = oldpage; | |
802 | } | |
803 | ||
804 | if (PageAnon(target)) | |
805 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
806 | else if (page_is_file_cache(target)) | |
807 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
808 | else | |
809 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
810 | ||
811 | /* unused page is not on radix-tree now. */ | |
812 | if (unused && ctype != MEM_CGROUP_CHARGE_TYPE_MAPPED) | |
813 | __mem_cgroup_uncharge_common(unused, ctype); | |
814 | ||
815 | pc = lookup_page_cgroup(target); | |
69029cd5 | 816 | /* |
01b1ae63 KH |
817 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
818 | * So, double-counting is effectively avoided. | |
819 | */ | |
820 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
821 | ||
822 | /* | |
823 | * Both of oldpage and newpage are still under lock_page(). | |
824 | * Then, we don't have to care about race in radix-tree. | |
825 | * But we have to be careful that this page is unmapped or not. | |
826 | * | |
827 | * There is a case for !page_mapped(). At the start of | |
828 | * migration, oldpage was mapped. But now, it's zapped. | |
829 | * But we know *target* page is not freed/reused under us. | |
830 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 831 | */ |
01b1ae63 KH |
832 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
833 | mem_cgroup_uncharge_page(target); | |
ae41be37 | 834 | } |
78fb7466 | 835 | |
c9b0ed51 KH |
836 | /* |
837 | * A call to try to shrink memory usage under specified resource controller. | |
838 | * This is typically used for page reclaiming for shmem for reducing side | |
839 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
840 | */ | |
841 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) | |
842 | { | |
843 | struct mem_cgroup *mem; | |
844 | int progress = 0; | |
845 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
846 | ||
cede86ac LZ |
847 | if (mem_cgroup_subsys.disabled) |
848 | return 0; | |
9623e078 HD |
849 | if (!mm) |
850 | return 0; | |
cede86ac | 851 | |
c9b0ed51 KH |
852 | rcu_read_lock(); |
853 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
854 | if (unlikely(!mem)) { |
855 | rcu_read_unlock(); | |
856 | return 0; | |
857 | } | |
c9b0ed51 KH |
858 | css_get(&mem->css); |
859 | rcu_read_unlock(); | |
860 | ||
861 | do { | |
862 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask); | |
a10cebf5 | 863 | progress += res_counter_check_under_limit(&mem->res); |
c9b0ed51 KH |
864 | } while (!progress && --retry); |
865 | ||
866 | css_put(&mem->css); | |
867 | if (!retry) | |
868 | return -ENOMEM; | |
869 | return 0; | |
870 | } | |
871 | ||
d38d2a75 KM |
872 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
873 | unsigned long long val) | |
628f4235 KH |
874 | { |
875 | ||
876 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
877 | int progress; | |
878 | int ret = 0; | |
879 | ||
880 | while (res_counter_set_limit(&memcg->res, val)) { | |
881 | if (signal_pending(current)) { | |
882 | ret = -EINTR; | |
883 | break; | |
884 | } | |
885 | if (!retry_count) { | |
886 | ret = -EBUSY; | |
887 | break; | |
888 | } | |
bced0520 KH |
889 | progress = try_to_free_mem_cgroup_pages(memcg, |
890 | GFP_HIGHUSER_MOVABLE); | |
628f4235 KH |
891 | if (!progress) |
892 | retry_count--; | |
893 | } | |
894 | return ret; | |
895 | } | |
896 | ||
897 | ||
cc847582 KH |
898 | /* |
899 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
900 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
901 | */ | |
902 | #define FORCE_UNCHARGE_BATCH (128) | |
8869b8f6 | 903 | static void mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 | 904 | struct mem_cgroup_per_zone *mz, |
b69408e8 | 905 | enum lru_list lru) |
cc847582 KH |
906 | { |
907 | struct page_cgroup *pc; | |
908 | struct page *page; | |
9b3c0a07 | 909 | int count = FORCE_UNCHARGE_BATCH; |
cc847582 | 910 | unsigned long flags; |
072c56c1 KH |
911 | struct list_head *list; |
912 | ||
b69408e8 | 913 | list = &mz->lists[lru]; |
cc847582 | 914 | |
072c56c1 | 915 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 916 | while (!list_empty(list)) { |
cc847582 KH |
917 | pc = list_entry(list->prev, struct page_cgroup, lru); |
918 | page = pc->page; | |
52d4b9ac KH |
919 | if (!PageCgroupUsed(pc)) |
920 | break; | |
9b3c0a07 HT |
921 | get_page(page); |
922 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
e8589cc1 KH |
923 | /* |
924 | * Check if this page is on LRU. !LRU page can be found | |
925 | * if it's under page migration. | |
926 | */ | |
927 | if (PageLRU(page)) { | |
69029cd5 KH |
928 | __mem_cgroup_uncharge_common(page, |
929 | MEM_CGROUP_CHARGE_TYPE_FORCE); | |
e8589cc1 KH |
930 | put_page(page); |
931 | if (--count <= 0) { | |
932 | count = FORCE_UNCHARGE_BATCH; | |
933 | cond_resched(); | |
934 | } | |
52d4b9ac KH |
935 | } else { |
936 | spin_lock_irqsave(&mz->lru_lock, flags); | |
937 | break; | |
938 | } | |
9b3c0a07 | 939 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 | 940 | } |
072c56c1 | 941 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
942 | } |
943 | ||
944 | /* | |
945 | * make mem_cgroup's charge to be 0 if there is no task. | |
946 | * This enables deleting this mem_cgroup. | |
947 | */ | |
d5b69e38 | 948 | static int mem_cgroup_force_empty(struct mem_cgroup *mem) |
cc847582 KH |
949 | { |
950 | int ret = -EBUSY; | |
1ecaab2b | 951 | int node, zid; |
8869b8f6 | 952 | |
cc847582 KH |
953 | css_get(&mem->css); |
954 | /* | |
955 | * page reclaim code (kswapd etc..) will move pages between | |
8869b8f6 | 956 | * active_list <-> inactive_list while we don't take a lock. |
cc847582 KH |
957 | * So, we have to do loop here until all lists are empty. |
958 | */ | |
1ecaab2b | 959 | while (mem->res.usage > 0) { |
cc847582 KH |
960 | if (atomic_read(&mem->css.cgroup->count) > 0) |
961 | goto out; | |
52d4b9ac KH |
962 | /* This is for making all *used* pages to be on LRU. */ |
963 | lru_add_drain_all(); | |
1ecaab2b KH |
964 | for_each_node_state(node, N_POSSIBLE) |
965 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
966 | struct mem_cgroup_per_zone *mz; | |
b69408e8 | 967 | enum lru_list l; |
1ecaab2b | 968 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
b69408e8 CL |
969 | for_each_lru(l) |
970 | mem_cgroup_force_empty_list(mem, mz, l); | |
1ecaab2b | 971 | } |
52d4b9ac | 972 | cond_resched(); |
cc847582 KH |
973 | } |
974 | ret = 0; | |
975 | out: | |
976 | css_put(&mem->css); | |
977 | return ret; | |
978 | } | |
979 | ||
2c3daa72 | 980 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 981 | { |
2c3daa72 PM |
982 | return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res, |
983 | cft->private); | |
8cdea7c0 | 984 | } |
628f4235 KH |
985 | /* |
986 | * The user of this function is... | |
987 | * RES_LIMIT. | |
988 | */ | |
856c13aa PM |
989 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
990 | const char *buffer) | |
8cdea7c0 | 991 | { |
628f4235 KH |
992 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
993 | unsigned long long val; | |
994 | int ret; | |
995 | ||
996 | switch (cft->private) { | |
997 | case RES_LIMIT: | |
998 | /* This function does all necessary parse...reuse it */ | |
999 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
1000 | if (!ret) | |
1001 | ret = mem_cgroup_resize_limit(memcg, val); | |
1002 | break; | |
1003 | default: | |
1004 | ret = -EINVAL; /* should be BUG() ? */ | |
1005 | break; | |
1006 | } | |
1007 | return ret; | |
8cdea7c0 BS |
1008 | } |
1009 | ||
29f2a4da | 1010 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
1011 | { |
1012 | struct mem_cgroup *mem; | |
1013 | ||
1014 | mem = mem_cgroup_from_cont(cont); | |
29f2a4da PE |
1015 | switch (event) { |
1016 | case RES_MAX_USAGE: | |
1017 | res_counter_reset_max(&mem->res); | |
1018 | break; | |
1019 | case RES_FAILCNT: | |
1020 | res_counter_reset_failcnt(&mem->res); | |
1021 | break; | |
1022 | } | |
85cc59db | 1023 | return 0; |
c84872e1 PE |
1024 | } |
1025 | ||
85cc59db | 1026 | static int mem_force_empty_write(struct cgroup *cont, unsigned int event) |
cc847582 | 1027 | { |
85cc59db | 1028 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont)); |
cc847582 KH |
1029 | } |
1030 | ||
d2ceb9b7 KH |
1031 | static const struct mem_cgroup_stat_desc { |
1032 | const char *msg; | |
1033 | u64 unit; | |
1034 | } mem_cgroup_stat_desc[] = { | |
1035 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
1036 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
1037 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
1038 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
1039 | }; |
1040 | ||
c64745cf PM |
1041 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
1042 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 1043 | { |
d2ceb9b7 KH |
1044 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
1045 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
1046 | int i; | |
1047 | ||
1048 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1049 | s64 val; | |
1050 | ||
1051 | val = mem_cgroup_read_stat(stat, i); | |
1052 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 1053 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 1054 | } |
6d12e2d8 KH |
1055 | /* showing # of active pages */ |
1056 | { | |
4f98a2fe RR |
1057 | unsigned long active_anon, inactive_anon; |
1058 | unsigned long active_file, inactive_file; | |
7b854121 | 1059 | unsigned long unevictable; |
4f98a2fe RR |
1060 | |
1061 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1062 | LRU_INACTIVE_ANON); | |
1063 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1064 | LRU_ACTIVE_ANON); | |
1065 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1066 | LRU_INACTIVE_FILE); | |
1067 | active_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1068 | LRU_ACTIVE_FILE); | |
7b854121 LS |
1069 | unevictable = mem_cgroup_get_all_zonestat(mem_cont, |
1070 | LRU_UNEVICTABLE); | |
1071 | ||
4f98a2fe RR |
1072 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); |
1073 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); | |
1074 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); | |
1075 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); | |
7b854121 LS |
1076 | cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); |
1077 | ||
6d12e2d8 | 1078 | } |
d2ceb9b7 KH |
1079 | return 0; |
1080 | } | |
1081 | ||
8cdea7c0 BS |
1082 | static struct cftype mem_cgroup_files[] = { |
1083 | { | |
0eea1030 | 1084 | .name = "usage_in_bytes", |
8cdea7c0 | 1085 | .private = RES_USAGE, |
2c3daa72 | 1086 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1087 | }, |
c84872e1 PE |
1088 | { |
1089 | .name = "max_usage_in_bytes", | |
1090 | .private = RES_MAX_USAGE, | |
29f2a4da | 1091 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
1092 | .read_u64 = mem_cgroup_read, |
1093 | }, | |
8cdea7c0 | 1094 | { |
0eea1030 | 1095 | .name = "limit_in_bytes", |
8cdea7c0 | 1096 | .private = RES_LIMIT, |
856c13aa | 1097 | .write_string = mem_cgroup_write, |
2c3daa72 | 1098 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
1099 | }, |
1100 | { | |
1101 | .name = "failcnt", | |
1102 | .private = RES_FAILCNT, | |
29f2a4da | 1103 | .trigger = mem_cgroup_reset, |
2c3daa72 | 1104 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1105 | }, |
cc847582 KH |
1106 | { |
1107 | .name = "force_empty", | |
85cc59db | 1108 | .trigger = mem_force_empty_write, |
cc847582 | 1109 | }, |
d2ceb9b7 KH |
1110 | { |
1111 | .name = "stat", | |
c64745cf | 1112 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 1113 | }, |
8cdea7c0 BS |
1114 | }; |
1115 | ||
6d12e2d8 KH |
1116 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1117 | { | |
1118 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 1119 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1120 | enum lru_list l; |
41e3355d | 1121 | int zone, tmp = node; |
1ecaab2b KH |
1122 | /* |
1123 | * This routine is called against possible nodes. | |
1124 | * But it's BUG to call kmalloc() against offline node. | |
1125 | * | |
1126 | * TODO: this routine can waste much memory for nodes which will | |
1127 | * never be onlined. It's better to use memory hotplug callback | |
1128 | * function. | |
1129 | */ | |
41e3355d KH |
1130 | if (!node_state(node, N_NORMAL_MEMORY)) |
1131 | tmp = -1; | |
1132 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
1133 | if (!pn) |
1134 | return 1; | |
1ecaab2b | 1135 | |
6d12e2d8 KH |
1136 | mem->info.nodeinfo[node] = pn; |
1137 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1138 | |
1139 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1140 | mz = &pn->zoneinfo[zone]; | |
072c56c1 | 1141 | spin_lock_init(&mz->lru_lock); |
b69408e8 CL |
1142 | for_each_lru(l) |
1143 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 1144 | } |
6d12e2d8 KH |
1145 | return 0; |
1146 | } | |
1147 | ||
1ecaab2b KH |
1148 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1149 | { | |
1150 | kfree(mem->info.nodeinfo[node]); | |
1151 | } | |
1152 | ||
33327948 KH |
1153 | static struct mem_cgroup *mem_cgroup_alloc(void) |
1154 | { | |
1155 | struct mem_cgroup *mem; | |
1156 | ||
1157 | if (sizeof(*mem) < PAGE_SIZE) | |
1158 | mem = kmalloc(sizeof(*mem), GFP_KERNEL); | |
1159 | else | |
1160 | mem = vmalloc(sizeof(*mem)); | |
1161 | ||
1162 | if (mem) | |
1163 | memset(mem, 0, sizeof(*mem)); | |
1164 | return mem; | |
1165 | } | |
1166 | ||
1167 | static void mem_cgroup_free(struct mem_cgroup *mem) | |
1168 | { | |
1169 | if (sizeof(*mem) < PAGE_SIZE) | |
1170 | kfree(mem); | |
1171 | else | |
1172 | vfree(mem); | |
1173 | } | |
1174 | ||
1175 | ||
8cdea7c0 BS |
1176 | static struct cgroup_subsys_state * |
1177 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1178 | { | |
1179 | struct mem_cgroup *mem; | |
6d12e2d8 | 1180 | int node; |
8cdea7c0 | 1181 | |
b6ac57d5 | 1182 | if (unlikely((cont->parent) == NULL)) { |
78fb7466 | 1183 | mem = &init_mem_cgroup; |
b6ac57d5 | 1184 | } else { |
33327948 KH |
1185 | mem = mem_cgroup_alloc(); |
1186 | if (!mem) | |
1187 | return ERR_PTR(-ENOMEM); | |
b6ac57d5 | 1188 | } |
78fb7466 | 1189 | |
8cdea7c0 | 1190 | res_counter_init(&mem->res); |
1ecaab2b | 1191 | |
6d12e2d8 KH |
1192 | for_each_node_state(node, N_POSSIBLE) |
1193 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1194 | goto free_out; | |
1195 | ||
8cdea7c0 | 1196 | return &mem->css; |
6d12e2d8 KH |
1197 | free_out: |
1198 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1199 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1200 | if (cont->parent != NULL) |
33327948 | 1201 | mem_cgroup_free(mem); |
2dda81ca | 1202 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1203 | } |
1204 | ||
df878fb0 KH |
1205 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1206 | struct cgroup *cont) | |
1207 | { | |
1208 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1209 | mem_cgroup_force_empty(mem); | |
1210 | } | |
1211 | ||
8cdea7c0 BS |
1212 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1213 | struct cgroup *cont) | |
1214 | { | |
6d12e2d8 KH |
1215 | int node; |
1216 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1217 | ||
1218 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1219 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1220 | |
33327948 | 1221 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
1222 | } |
1223 | ||
1224 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1225 | struct cgroup *cont) | |
1226 | { | |
1227 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1228 | ARRAY_SIZE(mem_cgroup_files)); | |
1229 | } | |
1230 | ||
67e465a7 BS |
1231 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1232 | struct cgroup *cont, | |
1233 | struct cgroup *old_cont, | |
1234 | struct task_struct *p) | |
1235 | { | |
1236 | struct mm_struct *mm; | |
1237 | struct mem_cgroup *mem, *old_mem; | |
1238 | ||
1239 | mm = get_task_mm(p); | |
1240 | if (mm == NULL) | |
1241 | return; | |
1242 | ||
1243 | mem = mem_cgroup_from_cont(cont); | |
1244 | old_mem = mem_cgroup_from_cont(old_cont); | |
1245 | ||
67e465a7 BS |
1246 | /* |
1247 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1248 | * in effect owned by the leader | |
1249 | */ | |
52ea27eb | 1250 | if (!thread_group_leader(p)) |
67e465a7 BS |
1251 | goto out; |
1252 | ||
67e465a7 BS |
1253 | out: |
1254 | mmput(mm); | |
67e465a7 BS |
1255 | } |
1256 | ||
8cdea7c0 BS |
1257 | struct cgroup_subsys mem_cgroup_subsys = { |
1258 | .name = "memory", | |
1259 | .subsys_id = mem_cgroup_subsys_id, | |
1260 | .create = mem_cgroup_create, | |
df878fb0 | 1261 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1262 | .destroy = mem_cgroup_destroy, |
1263 | .populate = mem_cgroup_populate, | |
67e465a7 | 1264 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1265 | .early_init = 0, |
8cdea7c0 | 1266 | }; |