]>
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 { | |
c8dad2bb | 63 | struct mem_cgroup_stat_cpu cpustat[0]; |
d52aa412 KH |
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 | 131 | /* |
c8dad2bb | 132 | * statistics. This must be placed at the end of memcg. |
d52aa412 KH |
133 | */ |
134 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
135 | }; |
136 | ||
217bc319 KH |
137 | enum charge_type { |
138 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
139 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 140 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 KH |
141 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
142 | NR_CHARGE_TYPE, | |
143 | }; | |
144 | ||
52d4b9ac KH |
145 | /* only for here (for easy reading.) */ |
146 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
147 | #define PCGF_USED (1UL << PCG_USED) | |
148 | #define PCGF_ACTIVE (1UL << PCG_ACTIVE) | |
149 | #define PCGF_LOCK (1UL << PCG_LOCK) | |
150 | #define PCGF_FILE (1UL << PCG_FILE) | |
c05555b5 KH |
151 | static const unsigned long |
152 | pcg_default_flags[NR_CHARGE_TYPE] = { | |
52d4b9ac KH |
153 | PCGF_CACHE | PCGF_FILE | PCGF_USED | PCGF_LOCK, /* File Cache */ |
154 | PCGF_ACTIVE | PCGF_USED | PCGF_LOCK, /* Anon */ | |
155 | PCGF_ACTIVE | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ | |
156 | 0, /* FORCE */ | |
217bc319 KH |
157 | }; |
158 | ||
d52aa412 KH |
159 | /* |
160 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
161 | */ | |
c05555b5 KH |
162 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
163 | struct page_cgroup *pc, | |
164 | bool charge) | |
d52aa412 KH |
165 | { |
166 | int val = (charge)? 1 : -1; | |
167 | struct mem_cgroup_stat *stat = &mem->stat; | |
addb9efe | 168 | struct mem_cgroup_stat_cpu *cpustat; |
d52aa412 | 169 | |
8869b8f6 | 170 | VM_BUG_ON(!irqs_disabled()); |
addb9efe KH |
171 | |
172 | cpustat = &stat->cpustat[smp_processor_id()]; | |
c05555b5 | 173 | if (PageCgroupCache(pc)) |
addb9efe | 174 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 175 | else |
addb9efe | 176 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
177 | |
178 | if (charge) | |
addb9efe | 179 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
180 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
181 | else | |
addb9efe | 182 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 183 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
6d12e2d8 KH |
184 | } |
185 | ||
d5b69e38 | 186 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
187 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
188 | { | |
6d12e2d8 KH |
189 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
190 | } | |
191 | ||
d5b69e38 | 192 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
193 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
194 | { | |
195 | struct mem_cgroup *mem = pc->mem_cgroup; | |
196 | int nid = page_cgroup_nid(pc); | |
197 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 198 | |
6d12e2d8 KH |
199 | return mem_cgroup_zoneinfo(mem, nid, zid); |
200 | } | |
201 | ||
202 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
b69408e8 | 203 | enum lru_list idx) |
6d12e2d8 KH |
204 | { |
205 | int nid, zid; | |
206 | struct mem_cgroup_per_zone *mz; | |
207 | u64 total = 0; | |
208 | ||
209 | for_each_online_node(nid) | |
210 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
211 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
212 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
213 | } | |
214 | return total; | |
d52aa412 KH |
215 | } |
216 | ||
d5b69e38 | 217 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
218 | { |
219 | return container_of(cgroup_subsys_state(cont, | |
220 | mem_cgroup_subsys_id), struct mem_cgroup, | |
221 | css); | |
222 | } | |
223 | ||
cf475ad2 | 224 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 225 | { |
31a78f23 BS |
226 | /* |
227 | * mm_update_next_owner() may clear mm->owner to NULL | |
228 | * if it races with swapoff, page migration, etc. | |
229 | * So this can be called with p == NULL. | |
230 | */ | |
231 | if (unlikely(!p)) | |
232 | return NULL; | |
233 | ||
78fb7466 PE |
234 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
235 | struct mem_cgroup, css); | |
236 | } | |
237 | ||
3eae90c3 KH |
238 | static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, |
239 | struct page_cgroup *pc) | |
6d12e2d8 | 240 | { |
4f98a2fe RR |
241 | int lru = LRU_BASE; |
242 | ||
c05555b5 | 243 | if (PageCgroupUnevictable(pc)) |
894bc310 LS |
244 | lru = LRU_UNEVICTABLE; |
245 | else { | |
c05555b5 | 246 | if (PageCgroupActive(pc)) |
894bc310 | 247 | lru += LRU_ACTIVE; |
c05555b5 | 248 | if (PageCgroupFile(pc)) |
894bc310 LS |
249 | lru += LRU_FILE; |
250 | } | |
6d12e2d8 | 251 | |
b69408e8 | 252 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
6d12e2d8 | 253 | |
c05555b5 | 254 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, false); |
508b7be0 | 255 | list_del(&pc->lru); |
6d12e2d8 KH |
256 | } |
257 | ||
3eae90c3 | 258 | static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, |
f817ed48 | 259 | struct page_cgroup *pc, bool hot) |
6d12e2d8 | 260 | { |
4f98a2fe | 261 | int lru = LRU_BASE; |
b69408e8 | 262 | |
c05555b5 | 263 | if (PageCgroupUnevictable(pc)) |
894bc310 LS |
264 | lru = LRU_UNEVICTABLE; |
265 | else { | |
c05555b5 | 266 | if (PageCgroupActive(pc)) |
894bc310 | 267 | lru += LRU_ACTIVE; |
c05555b5 | 268 | if (PageCgroupFile(pc)) |
894bc310 LS |
269 | lru += LRU_FILE; |
270 | } | |
b69408e8 CL |
271 | |
272 | MEM_CGROUP_ZSTAT(mz, lru) += 1; | |
f817ed48 KH |
273 | if (hot) |
274 | list_add(&pc->lru, &mz->lists[lru]); | |
275 | else | |
276 | list_add_tail(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 | 277 | |
c05555b5 | 278 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true); |
6d12e2d8 KH |
279 | } |
280 | ||
894bc310 | 281 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru) |
66e1707b | 282 | { |
6d12e2d8 | 283 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); |
c05555b5 KH |
284 | int active = PageCgroupActive(pc); |
285 | int file = PageCgroupFile(pc); | |
286 | int unevictable = PageCgroupUnevictable(pc); | |
894bc310 LS |
287 | enum lru_list from = unevictable ? LRU_UNEVICTABLE : |
288 | (LRU_FILE * !!file + !!active); | |
6d12e2d8 | 289 | |
894bc310 LS |
290 | if (lru == from) |
291 | return; | |
b69408e8 | 292 | |
894bc310 | 293 | MEM_CGROUP_ZSTAT(mz, from) -= 1; |
c05555b5 KH |
294 | /* |
295 | * However this is done under mz->lru_lock, another flags, which | |
296 | * are not related to LRU, will be modified from out-of-lock. | |
297 | * We have to use atomic set/clear flags. | |
298 | */ | |
894bc310 | 299 | if (is_unevictable_lru(lru)) { |
c05555b5 KH |
300 | ClearPageCgroupActive(pc); |
301 | SetPageCgroupUnevictable(pc); | |
894bc310 LS |
302 | } else { |
303 | if (is_active_lru(lru)) | |
c05555b5 | 304 | SetPageCgroupActive(pc); |
894bc310 | 305 | else |
c05555b5 KH |
306 | ClearPageCgroupActive(pc); |
307 | ClearPageCgroupUnevictable(pc); | |
894bc310 | 308 | } |
b69408e8 | 309 | |
b69408e8 CL |
310 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
311 | list_move(&pc->lru, &mz->lists[lru]); | |
66e1707b BS |
312 | } |
313 | ||
4c4a2214 DR |
314 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
315 | { | |
316 | int ret; | |
317 | ||
318 | task_lock(task); | |
bd845e38 | 319 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
320 | task_unlock(task); |
321 | return ret; | |
322 | } | |
323 | ||
66e1707b BS |
324 | /* |
325 | * This routine assumes that the appropriate zone's lru lock is already held | |
326 | */ | |
894bc310 | 327 | void mem_cgroup_move_lists(struct page *page, enum lru_list lru) |
66e1707b | 328 | { |
427d5416 | 329 | struct page_cgroup *pc; |
072c56c1 KH |
330 | struct mem_cgroup_per_zone *mz; |
331 | unsigned long flags; | |
332 | ||
cede86ac LZ |
333 | if (mem_cgroup_subsys.disabled) |
334 | return; | |
335 | ||
2680eed7 HD |
336 | /* |
337 | * We cannot lock_page_cgroup while holding zone's lru_lock, | |
338 | * because other holders of lock_page_cgroup can be interrupted | |
339 | * with an attempt to rotate_reclaimable_page. But we cannot | |
340 | * safely get to page_cgroup without it, so just try_lock it: | |
341 | * mem_cgroup_isolate_pages allows for page left on wrong list. | |
342 | */ | |
52d4b9ac KH |
343 | pc = lookup_page_cgroup(page); |
344 | if (!trylock_page_cgroup(pc)) | |
66e1707b | 345 | return; |
52d4b9ac | 346 | if (pc && PageCgroupUsed(pc)) { |
2680eed7 | 347 | mz = page_cgroup_zoneinfo(pc); |
2680eed7 | 348 | spin_lock_irqsave(&mz->lru_lock, flags); |
894bc310 | 349 | __mem_cgroup_move_lists(pc, lru); |
2680eed7 | 350 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9b3c0a07 | 351 | } |
52d4b9ac | 352 | unlock_page_cgroup(pc); |
66e1707b BS |
353 | } |
354 | ||
58ae83db KH |
355 | /* |
356 | * Calculate mapped_ratio under memory controller. This will be used in | |
357 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
358 | */ | |
359 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
360 | { | |
361 | long total, rss; | |
362 | ||
363 | /* | |
364 | * usage is recorded in bytes. But, here, we assume the number of | |
365 | * physical pages can be represented by "long" on any arch. | |
366 | */ | |
367 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
368 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
369 | return (int)((rss * 100L) / total); | |
370 | } | |
8869b8f6 | 371 | |
6c48a1d0 KH |
372 | /* |
373 | * prev_priority control...this will be used in memory reclaim path. | |
374 | */ | |
375 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
376 | { | |
377 | return mem->prev_priority; | |
378 | } | |
379 | ||
380 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
381 | { | |
382 | if (priority < mem->prev_priority) | |
383 | mem->prev_priority = priority; | |
384 | } | |
385 | ||
386 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
387 | { | |
388 | mem->prev_priority = priority; | |
389 | } | |
390 | ||
cc38108e KH |
391 | /* |
392 | * Calculate # of pages to be scanned in this priority/zone. | |
393 | * See also vmscan.c | |
394 | * | |
395 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
396 | * (see include/linux/mmzone.h) | |
397 | */ | |
398 | ||
b69408e8 CL |
399 | long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, |
400 | int priority, enum lru_list lru) | |
cc38108e | 401 | { |
b69408e8 | 402 | long nr_pages; |
cc38108e KH |
403 | int nid = zone->zone_pgdat->node_id; |
404 | int zid = zone_idx(zone); | |
405 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
406 | ||
b69408e8 | 407 | nr_pages = MEM_CGROUP_ZSTAT(mz, lru); |
cc38108e | 408 | |
b69408e8 | 409 | return (nr_pages >> priority); |
cc38108e KH |
410 | } |
411 | ||
66e1707b BS |
412 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
413 | struct list_head *dst, | |
414 | unsigned long *scanned, int order, | |
415 | int mode, struct zone *z, | |
416 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 417 | int active, int file) |
66e1707b BS |
418 | { |
419 | unsigned long nr_taken = 0; | |
420 | struct page *page; | |
421 | unsigned long scan; | |
422 | LIST_HEAD(pc_list); | |
423 | struct list_head *src; | |
ff7283fa | 424 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
425 | int nid = z->zone_pgdat->node_id; |
426 | int zid = zone_idx(z); | |
427 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 428 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 429 | |
cf475ad2 | 430 | BUG_ON(!mem_cont); |
1ecaab2b | 431 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 432 | src = &mz->lists[lru]; |
66e1707b | 433 | |
072c56c1 | 434 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
435 | scan = 0; |
436 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 437 | if (scan >= nr_to_scan) |
ff7283fa | 438 | break; |
52d4b9ac KH |
439 | if (unlikely(!PageCgroupUsed(pc))) |
440 | continue; | |
66e1707b | 441 | page = pc->page; |
66e1707b | 442 | |
436c6541 | 443 | if (unlikely(!PageLRU(page))) |
ff7283fa | 444 | continue; |
ff7283fa | 445 | |
4f98a2fe RR |
446 | /* |
447 | * TODO: play better with lumpy reclaim, grabbing anything. | |
448 | */ | |
894bc310 LS |
449 | if (PageUnevictable(page) || |
450 | (PageActive(page) && !active) || | |
451 | (!PageActive(page) && active)) { | |
452 | __mem_cgroup_move_lists(pc, page_lru(page)); | |
66e1707b BS |
453 | continue; |
454 | } | |
455 | ||
436c6541 HD |
456 | scan++; |
457 | list_move(&pc->lru, &pc_list); | |
66e1707b | 458 | |
4f98a2fe | 459 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
460 | list_move(&page->lru, dst); |
461 | nr_taken++; | |
462 | } | |
463 | } | |
464 | ||
465 | list_splice(&pc_list, src); | |
072c56c1 | 466 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
467 | |
468 | *scanned = scan; | |
469 | return nr_taken; | |
470 | } | |
471 | ||
f817ed48 KH |
472 | /* |
473 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
474 | * oom-killer can be invoked. | |
8a9f3ccd | 475 | */ |
f817ed48 KH |
476 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
477 | gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom) | |
8a9f3ccd BS |
478 | { |
479 | struct mem_cgroup *mem; | |
7a81b88c | 480 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
8a9f3ccd | 481 | /* |
3be91277 HD |
482 | * We always charge the cgroup the mm_struct belongs to. |
483 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
484 | * thread group leader migrates. It's possible that mm is not |
485 | * set, if so charge the init_mm (happens for pagecache usage). | |
486 | */ | |
7a81b88c | 487 | if (likely(!*memcg)) { |
e8589cc1 KH |
488 | rcu_read_lock(); |
489 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
490 | if (unlikely(!mem)) { |
491 | rcu_read_unlock(); | |
31a78f23 BS |
492 | return 0; |
493 | } | |
e8589cc1 KH |
494 | /* |
495 | * For every charge from the cgroup, increment reference count | |
496 | */ | |
497 | css_get(&mem->css); | |
7a81b88c | 498 | *memcg = mem; |
e8589cc1 KH |
499 | rcu_read_unlock(); |
500 | } else { | |
7a81b88c KH |
501 | mem = *memcg; |
502 | css_get(&mem->css); | |
e8589cc1 | 503 | } |
8a9f3ccd | 504 | |
7a81b88c | 505 | |
addb9efe | 506 | while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) { |
3be91277 | 507 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 508 | goto nomem; |
e1a1cd59 BS |
509 | |
510 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
511 | continue; |
512 | ||
513 | /* | |
8869b8f6 HD |
514 | * try_to_free_mem_cgroup_pages() might not give us a full |
515 | * picture of reclaim. Some pages are reclaimed and might be | |
516 | * moved to swap cache or just unmapped from the cgroup. | |
517 | * Check the limit again to see if the reclaim reduced the | |
518 | * current usage of the cgroup before giving up | |
519 | */ | |
66e1707b BS |
520 | if (res_counter_check_under_limit(&mem->res)) |
521 | continue; | |
3be91277 HD |
522 | |
523 | if (!nr_retries--) { | |
f817ed48 KH |
524 | if (oom) |
525 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
7a81b88c | 526 | goto nomem; |
66e1707b | 527 | } |
8a9f3ccd | 528 | } |
7a81b88c KH |
529 | return 0; |
530 | nomem: | |
531 | css_put(&mem->css); | |
532 | return -ENOMEM; | |
533 | } | |
8a9f3ccd | 534 | |
f817ed48 KH |
535 | /** |
536 | * mem_cgroup_try_charge - get charge of PAGE_SIZE. | |
537 | * @mm: an mm_struct which is charged against. (when *memcg is NULL) | |
538 | * @gfp_mask: gfp_mask for reclaim. | |
539 | * @memcg: a pointer to memory cgroup which is charged against. | |
540 | * | |
541 | * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated | |
542 | * memory cgroup from @mm is got and stored in *memcg. | |
543 | * | |
544 | * Returns 0 if success. -ENOMEM at failure. | |
545 | * This call can invoke OOM-Killer. | |
546 | */ | |
547 | ||
548 | int mem_cgroup_try_charge(struct mm_struct *mm, | |
549 | gfp_t mask, struct mem_cgroup **memcg) | |
550 | { | |
551 | return __mem_cgroup_try_charge(mm, mask, memcg, true); | |
552 | } | |
553 | ||
7a81b88c KH |
554 | /* |
555 | * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be | |
556 | * USED state. If already USED, uncharge and return. | |
557 | */ | |
558 | ||
559 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
560 | struct page_cgroup *pc, | |
561 | enum charge_type ctype) | |
562 | { | |
563 | struct mem_cgroup_per_zone *mz; | |
564 | unsigned long flags; | |
565 | ||
566 | /* try_charge() can return NULL to *memcg, taking care of it. */ | |
567 | if (!mem) | |
568 | return; | |
52d4b9ac KH |
569 | |
570 | lock_page_cgroup(pc); | |
571 | if (unlikely(PageCgroupUsed(pc))) { | |
572 | unlock_page_cgroup(pc); | |
573 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
574 | css_put(&mem->css); | |
7a81b88c | 575 | return; |
52d4b9ac | 576 | } |
8a9f3ccd | 577 | pc->mem_cgroup = mem; |
508b7be0 KH |
578 | /* |
579 | * If a page is accounted as a page cache, insert to inactive list. | |
580 | * If anon, insert to active list. | |
581 | */ | |
c05555b5 | 582 | pc->flags = pcg_default_flags[ctype]; |
3be91277 | 583 | |
072c56c1 | 584 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 585 | |
072c56c1 | 586 | spin_lock_irqsave(&mz->lru_lock, flags); |
f817ed48 | 587 | __mem_cgroup_add_list(mz, pc, true); |
072c56c1 | 588 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
52d4b9ac | 589 | unlock_page_cgroup(pc); |
7a81b88c | 590 | } |
66e1707b | 591 | |
f817ed48 KH |
592 | /** |
593 | * mem_cgroup_move_account - move account of the page | |
594 | * @pc: page_cgroup of the page. | |
595 | * @from: mem_cgroup which the page is moved from. | |
596 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
597 | * | |
598 | * The caller must confirm following. | |
599 | * 1. disable irq. | |
600 | * 2. lru_lock of old mem_cgroup(@from) should be held. | |
601 | * | |
602 | * returns 0 at success, | |
603 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
604 | * | |
605 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
606 | * new cgroup. It should be done by a caller. | |
607 | */ | |
608 | ||
609 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
610 | struct mem_cgroup *from, struct mem_cgroup *to) | |
611 | { | |
612 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
613 | int nid, zid; | |
614 | int ret = -EBUSY; | |
615 | ||
616 | VM_BUG_ON(!irqs_disabled()); | |
617 | VM_BUG_ON(from == to); | |
618 | ||
619 | nid = page_cgroup_nid(pc); | |
620 | zid = page_cgroup_zid(pc); | |
621 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
622 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
623 | ||
624 | ||
625 | if (!trylock_page_cgroup(pc)) | |
626 | return ret; | |
627 | ||
628 | if (!PageCgroupUsed(pc)) | |
629 | goto out; | |
630 | ||
631 | if (pc->mem_cgroup != from) | |
632 | goto out; | |
633 | ||
634 | if (spin_trylock(&to_mz->lru_lock)) { | |
635 | __mem_cgroup_remove_list(from_mz, pc); | |
636 | css_put(&from->css); | |
637 | res_counter_uncharge(&from->res, PAGE_SIZE); | |
638 | pc->mem_cgroup = to; | |
639 | css_get(&to->css); | |
640 | __mem_cgroup_add_list(to_mz, pc, false); | |
641 | ret = 0; | |
642 | spin_unlock(&to_mz->lru_lock); | |
643 | } | |
644 | out: | |
645 | unlock_page_cgroup(pc); | |
646 | return ret; | |
647 | } | |
648 | ||
649 | /* | |
650 | * move charges to its parent. | |
651 | */ | |
652 | ||
653 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
654 | struct mem_cgroup *child, | |
655 | gfp_t gfp_mask) | |
656 | { | |
657 | struct cgroup *cg = child->css.cgroup; | |
658 | struct cgroup *pcg = cg->parent; | |
659 | struct mem_cgroup *parent; | |
660 | struct mem_cgroup_per_zone *mz; | |
661 | unsigned long flags; | |
662 | int ret; | |
663 | ||
664 | /* Is ROOT ? */ | |
665 | if (!pcg) | |
666 | return -EINVAL; | |
667 | ||
668 | parent = mem_cgroup_from_cont(pcg); | |
669 | ||
670 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); | |
671 | if (ret) | |
672 | return ret; | |
673 | ||
674 | mz = mem_cgroup_zoneinfo(child, | |
675 | page_cgroup_nid(pc), page_cgroup_zid(pc)); | |
676 | ||
677 | spin_lock_irqsave(&mz->lru_lock, flags); | |
678 | ret = mem_cgroup_move_account(pc, child, parent); | |
679 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
680 | ||
681 | /* drop extra refcnt */ | |
682 | css_put(&parent->css); | |
683 | /* uncharge if move fails */ | |
684 | if (ret) | |
685 | res_counter_uncharge(&parent->res, PAGE_SIZE); | |
686 | ||
687 | return ret; | |
688 | } | |
689 | ||
7a81b88c KH |
690 | /* |
691 | * Charge the memory controller for page usage. | |
692 | * Return | |
693 | * 0 if the charge was successful | |
694 | * < 0 if the cgroup is over its limit | |
695 | */ | |
696 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
697 | gfp_t gfp_mask, enum charge_type ctype, | |
698 | struct mem_cgroup *memcg) | |
699 | { | |
700 | struct mem_cgroup *mem; | |
701 | struct page_cgroup *pc; | |
702 | int ret; | |
703 | ||
704 | pc = lookup_page_cgroup(page); | |
705 | /* can happen at boot */ | |
706 | if (unlikely(!pc)) | |
707 | return 0; | |
708 | prefetchw(pc); | |
709 | ||
710 | mem = memcg; | |
f817ed48 | 711 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
7a81b88c KH |
712 | if (ret) |
713 | return ret; | |
714 | ||
715 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 716 | return 0; |
8a9f3ccd BS |
717 | } |
718 | ||
7a81b88c KH |
719 | int mem_cgroup_newpage_charge(struct page *page, |
720 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 721 | { |
cede86ac LZ |
722 | if (mem_cgroup_subsys.disabled) |
723 | return 0; | |
52d4b9ac KH |
724 | if (PageCompound(page)) |
725 | return 0; | |
69029cd5 KH |
726 | /* |
727 | * If already mapped, we don't have to account. | |
728 | * If page cache, page->mapping has address_space. | |
729 | * But page->mapping may have out-of-use anon_vma pointer, | |
730 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
731 | * is NULL. | |
732 | */ | |
733 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
734 | return 0; | |
735 | if (unlikely(!mm)) | |
736 | mm = &init_mm; | |
217bc319 | 737 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 738 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
739 | } |
740 | ||
e1a1cd59 BS |
741 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
742 | gfp_t gfp_mask) | |
8697d331 | 743 | { |
cede86ac LZ |
744 | if (mem_cgroup_subsys.disabled) |
745 | return 0; | |
52d4b9ac KH |
746 | if (PageCompound(page)) |
747 | return 0; | |
accf163e KH |
748 | /* |
749 | * Corner case handling. This is called from add_to_page_cache() | |
750 | * in usual. But some FS (shmem) precharges this page before calling it | |
751 | * and call add_to_page_cache() with GFP_NOWAIT. | |
752 | * | |
753 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
754 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
755 | * charge twice. (It works but has to pay a bit larger cost.) | |
756 | */ | |
757 | if (!(gfp_mask & __GFP_WAIT)) { | |
758 | struct page_cgroup *pc; | |
759 | ||
52d4b9ac KH |
760 | |
761 | pc = lookup_page_cgroup(page); | |
762 | if (!pc) | |
763 | return 0; | |
764 | lock_page_cgroup(pc); | |
765 | if (PageCgroupUsed(pc)) { | |
766 | unlock_page_cgroup(pc); | |
accf163e KH |
767 | return 0; |
768 | } | |
52d4b9ac | 769 | unlock_page_cgroup(pc); |
accf163e KH |
770 | } |
771 | ||
69029cd5 | 772 | if (unlikely(!mm)) |
8697d331 | 773 | mm = &init_mm; |
accf163e | 774 | |
c05555b5 KH |
775 | if (page_is_file_cache(page)) |
776 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 777 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
c05555b5 KH |
778 | else |
779 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
780 | MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL); | |
e8589cc1 KH |
781 | } |
782 | ||
7a81b88c KH |
783 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
784 | { | |
785 | struct page_cgroup *pc; | |
786 | ||
787 | if (mem_cgroup_subsys.disabled) | |
788 | return; | |
789 | if (!ptr) | |
790 | return; | |
791 | pc = lookup_page_cgroup(page); | |
792 | __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
793 | } | |
794 | ||
795 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
796 | { | |
797 | if (mem_cgroup_subsys.disabled) | |
798 | return; | |
799 | if (!mem) | |
800 | return; | |
801 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
802 | css_put(&mem->css); | |
803 | } | |
804 | ||
805 | ||
8a9f3ccd | 806 | /* |
69029cd5 | 807 | * uncharge if !page_mapped(page) |
8a9f3ccd | 808 | */ |
69029cd5 KH |
809 | static void |
810 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) | |
8a9f3ccd | 811 | { |
8289546e | 812 | struct page_cgroup *pc; |
8a9f3ccd | 813 | struct mem_cgroup *mem; |
072c56c1 | 814 | struct mem_cgroup_per_zone *mz; |
66e1707b | 815 | unsigned long flags; |
8a9f3ccd | 816 | |
4077960e BS |
817 | if (mem_cgroup_subsys.disabled) |
818 | return; | |
819 | ||
8697d331 | 820 | /* |
3c541e14 | 821 | * Check if our page_cgroup is valid |
8697d331 | 822 | */ |
52d4b9ac KH |
823 | pc = lookup_page_cgroup(page); |
824 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
825 | return; | |
b9c565d5 | 826 | |
52d4b9ac KH |
827 | lock_page_cgroup(pc); |
828 | if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED && page_mapped(page)) | |
829 | || !PageCgroupUsed(pc)) { | |
830 | /* This happens at race in zap_pte_range() and do_swap_page()*/ | |
831 | unlock_page_cgroup(pc); | |
832 | return; | |
833 | } | |
834 | ClearPageCgroupUsed(pc); | |
835 | mem = pc->mem_cgroup; | |
b9c565d5 | 836 | |
69029cd5 KH |
837 | mz = page_cgroup_zoneinfo(pc); |
838 | spin_lock_irqsave(&mz->lru_lock, flags); | |
839 | __mem_cgroup_remove_list(mz, pc); | |
840 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
52d4b9ac | 841 | unlock_page_cgroup(pc); |
fb59e9f1 | 842 | |
69029cd5 KH |
843 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
844 | css_put(&mem->css); | |
6d12e2d8 | 845 | |
69029cd5 | 846 | return; |
3c541e14 BS |
847 | } |
848 | ||
69029cd5 KH |
849 | void mem_cgroup_uncharge_page(struct page *page) |
850 | { | |
52d4b9ac KH |
851 | /* early check. */ |
852 | if (page_mapped(page)) | |
853 | return; | |
854 | if (page->mapping && !PageAnon(page)) | |
855 | return; | |
69029cd5 KH |
856 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
857 | } | |
858 | ||
859 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
860 | { | |
861 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 862 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
863 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
864 | } | |
865 | ||
ae41be37 | 866 | /* |
01b1ae63 KH |
867 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
868 | * page belongs to. | |
ae41be37 | 869 | */ |
01b1ae63 | 870 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
871 | { |
872 | struct page_cgroup *pc; | |
e8589cc1 | 873 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 874 | int ret = 0; |
8869b8f6 | 875 | |
4077960e BS |
876 | if (mem_cgroup_subsys.disabled) |
877 | return 0; | |
878 | ||
52d4b9ac KH |
879 | pc = lookup_page_cgroup(page); |
880 | lock_page_cgroup(pc); | |
881 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
882 | mem = pc->mem_cgroup; |
883 | css_get(&mem->css); | |
e8589cc1 | 884 | } |
52d4b9ac | 885 | unlock_page_cgroup(pc); |
01b1ae63 | 886 | |
e8589cc1 | 887 | if (mem) { |
01b1ae63 | 888 | ret = mem_cgroup_try_charge(NULL, GFP_HIGHUSER_MOVABLE, &mem); |
e8589cc1 KH |
889 | css_put(&mem->css); |
890 | } | |
01b1ae63 | 891 | *ptr = mem; |
e8589cc1 | 892 | return ret; |
ae41be37 | 893 | } |
8869b8f6 | 894 | |
69029cd5 | 895 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
896 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
897 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 898 | { |
01b1ae63 KH |
899 | struct page *target, *unused; |
900 | struct page_cgroup *pc; | |
901 | enum charge_type ctype; | |
902 | ||
903 | if (!mem) | |
904 | return; | |
905 | ||
906 | /* at migration success, oldpage->mapping is NULL. */ | |
907 | if (oldpage->mapping) { | |
908 | target = oldpage; | |
909 | unused = NULL; | |
910 | } else { | |
911 | target = newpage; | |
912 | unused = oldpage; | |
913 | } | |
914 | ||
915 | if (PageAnon(target)) | |
916 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
917 | else if (page_is_file_cache(target)) | |
918 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
919 | else | |
920 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
921 | ||
922 | /* unused page is not on radix-tree now. */ | |
923 | if (unused && ctype != MEM_CGROUP_CHARGE_TYPE_MAPPED) | |
924 | __mem_cgroup_uncharge_common(unused, ctype); | |
925 | ||
926 | pc = lookup_page_cgroup(target); | |
69029cd5 | 927 | /* |
01b1ae63 KH |
928 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
929 | * So, double-counting is effectively avoided. | |
930 | */ | |
931 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
932 | ||
933 | /* | |
934 | * Both of oldpage and newpage are still under lock_page(). | |
935 | * Then, we don't have to care about race in radix-tree. | |
936 | * But we have to be careful that this page is unmapped or not. | |
937 | * | |
938 | * There is a case for !page_mapped(). At the start of | |
939 | * migration, oldpage was mapped. But now, it's zapped. | |
940 | * But we know *target* page is not freed/reused under us. | |
941 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 942 | */ |
01b1ae63 KH |
943 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
944 | mem_cgroup_uncharge_page(target); | |
ae41be37 | 945 | } |
78fb7466 | 946 | |
c9b0ed51 KH |
947 | /* |
948 | * A call to try to shrink memory usage under specified resource controller. | |
949 | * This is typically used for page reclaiming for shmem for reducing side | |
950 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
951 | */ | |
952 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) | |
953 | { | |
954 | struct mem_cgroup *mem; | |
955 | int progress = 0; | |
956 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
957 | ||
cede86ac LZ |
958 | if (mem_cgroup_subsys.disabled) |
959 | return 0; | |
9623e078 HD |
960 | if (!mm) |
961 | return 0; | |
cede86ac | 962 | |
c9b0ed51 KH |
963 | rcu_read_lock(); |
964 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
965 | if (unlikely(!mem)) { |
966 | rcu_read_unlock(); | |
967 | return 0; | |
968 | } | |
c9b0ed51 KH |
969 | css_get(&mem->css); |
970 | rcu_read_unlock(); | |
971 | ||
972 | do { | |
973 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask); | |
a10cebf5 | 974 | progress += res_counter_check_under_limit(&mem->res); |
c9b0ed51 KH |
975 | } while (!progress && --retry); |
976 | ||
977 | css_put(&mem->css); | |
978 | if (!retry) | |
979 | return -ENOMEM; | |
980 | return 0; | |
981 | } | |
982 | ||
d38d2a75 KM |
983 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
984 | unsigned long long val) | |
628f4235 KH |
985 | { |
986 | ||
987 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
988 | int progress; | |
989 | int ret = 0; | |
990 | ||
991 | while (res_counter_set_limit(&memcg->res, val)) { | |
992 | if (signal_pending(current)) { | |
993 | ret = -EINTR; | |
994 | break; | |
995 | } | |
996 | if (!retry_count) { | |
997 | ret = -EBUSY; | |
998 | break; | |
999 | } | |
bced0520 KH |
1000 | progress = try_to_free_mem_cgroup_pages(memcg, |
1001 | GFP_HIGHUSER_MOVABLE); | |
628f4235 KH |
1002 | if (!progress) |
1003 | retry_count--; | |
1004 | } | |
1005 | return ret; | |
1006 | } | |
1007 | ||
1008 | ||
cc847582 KH |
1009 | /* |
1010 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
1011 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
1012 | */ | |
f817ed48 | 1013 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 | 1014 | struct mem_cgroup_per_zone *mz, |
b69408e8 | 1015 | enum lru_list lru) |
cc847582 | 1016 | { |
f817ed48 | 1017 | struct page_cgroup *pc, *busy; |
cc847582 | 1018 | unsigned long flags; |
f817ed48 | 1019 | unsigned long loop; |
072c56c1 | 1020 | struct list_head *list; |
f817ed48 | 1021 | int ret = 0; |
072c56c1 | 1022 | |
b69408e8 | 1023 | list = &mz->lists[lru]; |
cc847582 | 1024 | |
f817ed48 KH |
1025 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
1026 | /* give some margin against EBUSY etc...*/ | |
1027 | loop += 256; | |
1028 | busy = NULL; | |
1029 | while (loop--) { | |
1030 | ret = 0; | |
1031 | spin_lock_irqsave(&mz->lru_lock, flags); | |
1032 | if (list_empty(list)) { | |
1033 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
52d4b9ac | 1034 | break; |
f817ed48 KH |
1035 | } |
1036 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
1037 | if (busy == pc) { | |
1038 | list_move(&pc->lru, list); | |
1039 | busy = 0; | |
1040 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
1041 | continue; | |
1042 | } | |
9b3c0a07 | 1043 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
f817ed48 KH |
1044 | |
1045 | ret = mem_cgroup_move_parent(pc, mem, GFP_HIGHUSER_MOVABLE); | |
1046 | if (ret == -ENOMEM) | |
52d4b9ac | 1047 | break; |
f817ed48 KH |
1048 | |
1049 | if (ret == -EBUSY || ret == -EINVAL) { | |
1050 | /* found lock contention or "pc" is obsolete. */ | |
1051 | busy = pc; | |
1052 | cond_resched(); | |
1053 | } else | |
1054 | busy = NULL; | |
cc847582 | 1055 | } |
f817ed48 KH |
1056 | if (!ret && !list_empty(list)) |
1057 | return -EBUSY; | |
1058 | return ret; | |
cc847582 KH |
1059 | } |
1060 | ||
1061 | /* | |
1062 | * make mem_cgroup's charge to be 0 if there is no task. | |
1063 | * This enables deleting this mem_cgroup. | |
1064 | */ | |
d5b69e38 | 1065 | static int mem_cgroup_force_empty(struct mem_cgroup *mem) |
cc847582 | 1066 | { |
f817ed48 KH |
1067 | int ret; |
1068 | int node, zid, shrink; | |
1069 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
8869b8f6 | 1070 | |
cc847582 | 1071 | css_get(&mem->css); |
f817ed48 KH |
1072 | |
1073 | shrink = 0; | |
1074 | move_account: | |
1ecaab2b | 1075 | while (mem->res.usage > 0) { |
f817ed48 | 1076 | ret = -EBUSY; |
cc847582 KH |
1077 | if (atomic_read(&mem->css.cgroup->count) > 0) |
1078 | goto out; | |
f817ed48 | 1079 | |
52d4b9ac KH |
1080 | /* This is for making all *used* pages to be on LRU. */ |
1081 | lru_add_drain_all(); | |
f817ed48 KH |
1082 | ret = 0; |
1083 | for_each_node_state(node, N_POSSIBLE) { | |
1084 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { | |
1ecaab2b | 1085 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1086 | enum lru_list l; |
1ecaab2b | 1087 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
f817ed48 KH |
1088 | for_each_lru(l) { |
1089 | ret = mem_cgroup_force_empty_list(mem, | |
1090 | mz, l); | |
1091 | if (ret) | |
1092 | break; | |
1093 | } | |
1ecaab2b | 1094 | } |
f817ed48 KH |
1095 | if (ret) |
1096 | break; | |
1097 | } | |
1098 | /* it seems parent cgroup doesn't have enough mem */ | |
1099 | if (ret == -ENOMEM) | |
1100 | goto try_to_free; | |
52d4b9ac | 1101 | cond_resched(); |
cc847582 KH |
1102 | } |
1103 | ret = 0; | |
1104 | out: | |
1105 | css_put(&mem->css); | |
1106 | return ret; | |
f817ed48 KH |
1107 | |
1108 | try_to_free: | |
1109 | /* returns EBUSY if we come here twice. */ | |
1110 | if (shrink) { | |
1111 | ret = -EBUSY; | |
1112 | goto out; | |
1113 | } | |
1114 | /* try to free all pages in this cgroup */ | |
1115 | shrink = 1; | |
1116 | while (nr_retries && mem->res.usage > 0) { | |
1117 | int progress; | |
1118 | progress = try_to_free_mem_cgroup_pages(mem, | |
1119 | GFP_HIGHUSER_MOVABLE); | |
1120 | if (!progress) | |
1121 | nr_retries--; | |
1122 | ||
1123 | } | |
1124 | /* try move_account...there may be some *locked* pages. */ | |
1125 | if (mem->res.usage) | |
1126 | goto move_account; | |
1127 | ret = 0; | |
1128 | goto out; | |
cc847582 KH |
1129 | } |
1130 | ||
2c3daa72 | 1131 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 1132 | { |
2c3daa72 PM |
1133 | return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res, |
1134 | cft->private); | |
8cdea7c0 | 1135 | } |
628f4235 KH |
1136 | /* |
1137 | * The user of this function is... | |
1138 | * RES_LIMIT. | |
1139 | */ | |
856c13aa PM |
1140 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
1141 | const char *buffer) | |
8cdea7c0 | 1142 | { |
628f4235 KH |
1143 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
1144 | unsigned long long val; | |
1145 | int ret; | |
1146 | ||
1147 | switch (cft->private) { | |
1148 | case RES_LIMIT: | |
1149 | /* This function does all necessary parse...reuse it */ | |
1150 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
1151 | if (!ret) | |
1152 | ret = mem_cgroup_resize_limit(memcg, val); | |
1153 | break; | |
1154 | default: | |
1155 | ret = -EINVAL; /* should be BUG() ? */ | |
1156 | break; | |
1157 | } | |
1158 | return ret; | |
8cdea7c0 BS |
1159 | } |
1160 | ||
29f2a4da | 1161 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
1162 | { |
1163 | struct mem_cgroup *mem; | |
1164 | ||
1165 | mem = mem_cgroup_from_cont(cont); | |
29f2a4da PE |
1166 | switch (event) { |
1167 | case RES_MAX_USAGE: | |
1168 | res_counter_reset_max(&mem->res); | |
1169 | break; | |
1170 | case RES_FAILCNT: | |
1171 | res_counter_reset_failcnt(&mem->res); | |
1172 | break; | |
1173 | } | |
85cc59db | 1174 | return 0; |
c84872e1 PE |
1175 | } |
1176 | ||
d2ceb9b7 KH |
1177 | static const struct mem_cgroup_stat_desc { |
1178 | const char *msg; | |
1179 | u64 unit; | |
1180 | } mem_cgroup_stat_desc[] = { | |
1181 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
1182 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
1183 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
1184 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
1185 | }; |
1186 | ||
c64745cf PM |
1187 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
1188 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 1189 | { |
d2ceb9b7 KH |
1190 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
1191 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
1192 | int i; | |
1193 | ||
1194 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1195 | s64 val; | |
1196 | ||
1197 | val = mem_cgroup_read_stat(stat, i); | |
1198 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 1199 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 1200 | } |
6d12e2d8 KH |
1201 | /* showing # of active pages */ |
1202 | { | |
4f98a2fe RR |
1203 | unsigned long active_anon, inactive_anon; |
1204 | unsigned long active_file, inactive_file; | |
7b854121 | 1205 | unsigned long unevictable; |
4f98a2fe RR |
1206 | |
1207 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1208 | LRU_INACTIVE_ANON); | |
1209 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1210 | LRU_ACTIVE_ANON); | |
1211 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1212 | LRU_INACTIVE_FILE); | |
1213 | active_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1214 | LRU_ACTIVE_FILE); | |
7b854121 LS |
1215 | unevictable = mem_cgroup_get_all_zonestat(mem_cont, |
1216 | LRU_UNEVICTABLE); | |
1217 | ||
4f98a2fe RR |
1218 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); |
1219 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); | |
1220 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); | |
1221 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); | |
7b854121 LS |
1222 | cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); |
1223 | ||
6d12e2d8 | 1224 | } |
d2ceb9b7 KH |
1225 | return 0; |
1226 | } | |
1227 | ||
8cdea7c0 BS |
1228 | static struct cftype mem_cgroup_files[] = { |
1229 | { | |
0eea1030 | 1230 | .name = "usage_in_bytes", |
8cdea7c0 | 1231 | .private = RES_USAGE, |
2c3daa72 | 1232 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1233 | }, |
c84872e1 PE |
1234 | { |
1235 | .name = "max_usage_in_bytes", | |
1236 | .private = RES_MAX_USAGE, | |
29f2a4da | 1237 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
1238 | .read_u64 = mem_cgroup_read, |
1239 | }, | |
8cdea7c0 | 1240 | { |
0eea1030 | 1241 | .name = "limit_in_bytes", |
8cdea7c0 | 1242 | .private = RES_LIMIT, |
856c13aa | 1243 | .write_string = mem_cgroup_write, |
2c3daa72 | 1244 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
1245 | }, |
1246 | { | |
1247 | .name = "failcnt", | |
1248 | .private = RES_FAILCNT, | |
29f2a4da | 1249 | .trigger = mem_cgroup_reset, |
2c3daa72 | 1250 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1251 | }, |
d2ceb9b7 KH |
1252 | { |
1253 | .name = "stat", | |
c64745cf | 1254 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 1255 | }, |
8cdea7c0 BS |
1256 | }; |
1257 | ||
6d12e2d8 KH |
1258 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1259 | { | |
1260 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 1261 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1262 | enum lru_list l; |
41e3355d | 1263 | int zone, tmp = node; |
1ecaab2b KH |
1264 | /* |
1265 | * This routine is called against possible nodes. | |
1266 | * But it's BUG to call kmalloc() against offline node. | |
1267 | * | |
1268 | * TODO: this routine can waste much memory for nodes which will | |
1269 | * never be onlined. It's better to use memory hotplug callback | |
1270 | * function. | |
1271 | */ | |
41e3355d KH |
1272 | if (!node_state(node, N_NORMAL_MEMORY)) |
1273 | tmp = -1; | |
1274 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
1275 | if (!pn) |
1276 | return 1; | |
1ecaab2b | 1277 | |
6d12e2d8 KH |
1278 | mem->info.nodeinfo[node] = pn; |
1279 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1280 | |
1281 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1282 | mz = &pn->zoneinfo[zone]; | |
072c56c1 | 1283 | spin_lock_init(&mz->lru_lock); |
b69408e8 CL |
1284 | for_each_lru(l) |
1285 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 1286 | } |
6d12e2d8 KH |
1287 | return 0; |
1288 | } | |
1289 | ||
1ecaab2b KH |
1290 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1291 | { | |
1292 | kfree(mem->info.nodeinfo[node]); | |
1293 | } | |
1294 | ||
c8dad2bb JB |
1295 | static int mem_cgroup_size(void) |
1296 | { | |
1297 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
1298 | return sizeof(struct mem_cgroup) + cpustat_size; | |
1299 | } | |
1300 | ||
33327948 KH |
1301 | static struct mem_cgroup *mem_cgroup_alloc(void) |
1302 | { | |
1303 | struct mem_cgroup *mem; | |
c8dad2bb | 1304 | int size = mem_cgroup_size(); |
33327948 | 1305 | |
c8dad2bb JB |
1306 | if (size < PAGE_SIZE) |
1307 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 1308 | else |
c8dad2bb | 1309 | mem = vmalloc(size); |
33327948 KH |
1310 | |
1311 | if (mem) | |
c8dad2bb | 1312 | memset(mem, 0, size); |
33327948 KH |
1313 | return mem; |
1314 | } | |
1315 | ||
1316 | static void mem_cgroup_free(struct mem_cgroup *mem) | |
1317 | { | |
c8dad2bb | 1318 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
1319 | kfree(mem); |
1320 | else | |
1321 | vfree(mem); | |
1322 | } | |
1323 | ||
1324 | ||
8cdea7c0 BS |
1325 | static struct cgroup_subsys_state * |
1326 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1327 | { | |
1328 | struct mem_cgroup *mem; | |
6d12e2d8 | 1329 | int node; |
8cdea7c0 | 1330 | |
c8dad2bb JB |
1331 | mem = mem_cgroup_alloc(); |
1332 | if (!mem) | |
1333 | return ERR_PTR(-ENOMEM); | |
78fb7466 | 1334 | |
8cdea7c0 | 1335 | res_counter_init(&mem->res); |
1ecaab2b | 1336 | |
6d12e2d8 KH |
1337 | for_each_node_state(node, N_POSSIBLE) |
1338 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1339 | goto free_out; | |
1340 | ||
8cdea7c0 | 1341 | return &mem->css; |
6d12e2d8 KH |
1342 | free_out: |
1343 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1344 | free_mem_cgroup_per_zone_info(mem, node); |
c8dad2bb | 1345 | mem_cgroup_free(mem); |
2dda81ca | 1346 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1347 | } |
1348 | ||
df878fb0 KH |
1349 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1350 | struct cgroup *cont) | |
1351 | { | |
1352 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1353 | mem_cgroup_force_empty(mem); | |
1354 | } | |
1355 | ||
8cdea7c0 BS |
1356 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1357 | struct cgroup *cont) | |
1358 | { | |
6d12e2d8 KH |
1359 | int node; |
1360 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1361 | ||
1362 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1363 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1364 | |
33327948 | 1365 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
1366 | } |
1367 | ||
1368 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1369 | struct cgroup *cont) | |
1370 | { | |
1371 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1372 | ARRAY_SIZE(mem_cgroup_files)); | |
1373 | } | |
1374 | ||
67e465a7 BS |
1375 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1376 | struct cgroup *cont, | |
1377 | struct cgroup *old_cont, | |
1378 | struct task_struct *p) | |
1379 | { | |
1380 | struct mm_struct *mm; | |
1381 | struct mem_cgroup *mem, *old_mem; | |
1382 | ||
1383 | mm = get_task_mm(p); | |
1384 | if (mm == NULL) | |
1385 | return; | |
1386 | ||
1387 | mem = mem_cgroup_from_cont(cont); | |
1388 | old_mem = mem_cgroup_from_cont(old_cont); | |
1389 | ||
67e465a7 BS |
1390 | /* |
1391 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1392 | * in effect owned by the leader | |
1393 | */ | |
52ea27eb | 1394 | if (!thread_group_leader(p)) |
67e465a7 BS |
1395 | goto out; |
1396 | ||
67e465a7 BS |
1397 | out: |
1398 | mmput(mm); | |
67e465a7 BS |
1399 | } |
1400 | ||
8cdea7c0 BS |
1401 | struct cgroup_subsys mem_cgroup_subsys = { |
1402 | .name = "memory", | |
1403 | .subsys_id = mem_cgroup_subsys_id, | |
1404 | .create = mem_cgroup_create, | |
df878fb0 | 1405 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1406 | .destroy = mem_cgroup_destroy, |
1407 | .populate = mem_cgroup_populate, | |
67e465a7 | 1408 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1409 | .early_init = 0, |
8cdea7c0 | 1410 | }; |