Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
8cdea7c0 BS |
9 | * This program is free software; you can redistribute it and/or modify |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation; either version 2 of the License, or | |
12 | * (at your option) any later version. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | */ | |
19 | ||
20 | #include <linux/res_counter.h> | |
21 | #include <linux/memcontrol.h> | |
22 | #include <linux/cgroup.h> | |
78fb7466 | 23 | #include <linux/mm.h> |
d13d1443 | 24 | #include <linux/pagemap.h> |
d52aa412 | 25 | #include <linux/smp.h> |
8a9f3ccd | 26 | #include <linux/page-flags.h> |
66e1707b | 27 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
28 | #include <linux/bit_spinlock.h> |
29 | #include <linux/rcupdate.h> | |
8c7c6e34 | 30 | #include <linux/mutex.h> |
b6ac57d5 | 31 | #include <linux/slab.h> |
66e1707b BS |
32 | #include <linux/swap.h> |
33 | #include <linux/spinlock.h> | |
34 | #include <linux/fs.h> | |
d2ceb9b7 | 35 | #include <linux/seq_file.h> |
33327948 | 36 | #include <linux/vmalloc.h> |
b69408e8 | 37 | #include <linux/mm_inline.h> |
52d4b9ac | 38 | #include <linux/page_cgroup.h> |
8cdea7c0 | 39 | |
8697d331 BS |
40 | #include <asm/uaccess.h> |
41 | ||
a181b0e8 | 42 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 43 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 44 | |
c077719b KH |
45 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
46 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */ | |
47 | int do_swap_account __read_mostly; | |
48 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
49 | #else | |
50 | #define do_swap_account (0) | |
51 | #endif | |
52 | ||
53 | ||
d52aa412 KH |
54 | /* |
55 | * Statistics for memory cgroup. | |
56 | */ | |
57 | enum mem_cgroup_stat_index { | |
58 | /* | |
59 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
60 | */ | |
61 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
62 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
55e462b0 BR |
63 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
64 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
65 | |
66 | MEM_CGROUP_STAT_NSTATS, | |
67 | }; | |
68 | ||
69 | struct mem_cgroup_stat_cpu { | |
70 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
71 | } ____cacheline_aligned_in_smp; | |
72 | ||
73 | struct mem_cgroup_stat { | |
c8dad2bb | 74 | struct mem_cgroup_stat_cpu cpustat[0]; |
d52aa412 KH |
75 | }; |
76 | ||
77 | /* | |
78 | * For accounting under irq disable, no need for increment preempt count. | |
79 | */ | |
addb9efe | 80 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
81 | enum mem_cgroup_stat_index idx, int val) |
82 | { | |
addb9efe | 83 | stat->count[idx] += val; |
d52aa412 KH |
84 | } |
85 | ||
86 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
87 | enum mem_cgroup_stat_index idx) | |
88 | { | |
89 | int cpu; | |
90 | s64 ret = 0; | |
91 | for_each_possible_cpu(cpu) | |
92 | ret += stat->cpustat[cpu].count[idx]; | |
93 | return ret; | |
94 | } | |
95 | ||
6d12e2d8 KH |
96 | /* |
97 | * per-zone information in memory controller. | |
98 | */ | |
6d12e2d8 | 99 | struct mem_cgroup_per_zone { |
072c56c1 KH |
100 | /* |
101 | * spin_lock to protect the per cgroup LRU | |
102 | */ | |
103 | spinlock_t lru_lock; | |
b69408e8 CL |
104 | struct list_head lists[NR_LRU_LISTS]; |
105 | unsigned long count[NR_LRU_LISTS]; | |
6d12e2d8 KH |
106 | }; |
107 | /* Macro for accessing counter */ | |
108 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
109 | ||
110 | struct mem_cgroup_per_node { | |
111 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
112 | }; | |
113 | ||
114 | struct mem_cgroup_lru_info { | |
115 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
116 | }; | |
117 | ||
8cdea7c0 BS |
118 | /* |
119 | * The memory controller data structure. The memory controller controls both | |
120 | * page cache and RSS per cgroup. We would eventually like to provide | |
121 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
122 | * to help the administrator determine what knobs to tune. | |
123 | * | |
124 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
125 | * we hit the water mark. May be even add a low water mark, such that |
126 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
127 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
128 | */ |
129 | struct mem_cgroup { | |
130 | struct cgroup_subsys_state css; | |
131 | /* | |
132 | * the counter to account for memory usage | |
133 | */ | |
134 | struct res_counter res; | |
8c7c6e34 KH |
135 | /* |
136 | * the counter to account for mem+swap usage. | |
137 | */ | |
138 | struct res_counter memsw; | |
78fb7466 PE |
139 | /* |
140 | * Per cgroup active and inactive list, similar to the | |
141 | * per zone LRU lists. | |
78fb7466 | 142 | */ |
6d12e2d8 | 143 | struct mem_cgroup_lru_info info; |
072c56c1 | 144 | |
6c48a1d0 | 145 | int prev_priority; /* for recording reclaim priority */ |
8c7c6e34 KH |
146 | int obsolete; |
147 | atomic_t refcnt; | |
d52aa412 | 148 | /* |
c8dad2bb | 149 | * statistics. This must be placed at the end of memcg. |
d52aa412 KH |
150 | */ |
151 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
152 | }; |
153 | ||
217bc319 KH |
154 | enum charge_type { |
155 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
156 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 157 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 158 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 159 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
c05555b5 KH |
160 | NR_CHARGE_TYPE, |
161 | }; | |
162 | ||
52d4b9ac KH |
163 | /* only for here (for easy reading.) */ |
164 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
165 | #define PCGF_USED (1UL << PCG_USED) | |
166 | #define PCGF_ACTIVE (1UL << PCG_ACTIVE) | |
167 | #define PCGF_LOCK (1UL << PCG_LOCK) | |
168 | #define PCGF_FILE (1UL << PCG_FILE) | |
c05555b5 KH |
169 | static const unsigned long |
170 | pcg_default_flags[NR_CHARGE_TYPE] = { | |
52d4b9ac KH |
171 | PCGF_CACHE | PCGF_FILE | PCGF_USED | PCGF_LOCK, /* File Cache */ |
172 | PCGF_ACTIVE | PCGF_USED | PCGF_LOCK, /* Anon */ | |
173 | PCGF_ACTIVE | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ | |
174 | 0, /* FORCE */ | |
217bc319 KH |
175 | }; |
176 | ||
8c7c6e34 KH |
177 | |
178 | /* for encoding cft->private value on file */ | |
179 | #define _MEM (0) | |
180 | #define _MEMSWAP (1) | |
181 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
182 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
183 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
184 | ||
185 | static void mem_cgroup_get(struct mem_cgroup *mem); | |
186 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
187 | ||
d52aa412 KH |
188 | /* |
189 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
190 | */ | |
c05555b5 KH |
191 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
192 | struct page_cgroup *pc, | |
193 | bool charge) | |
d52aa412 KH |
194 | { |
195 | int val = (charge)? 1 : -1; | |
196 | struct mem_cgroup_stat *stat = &mem->stat; | |
addb9efe | 197 | struct mem_cgroup_stat_cpu *cpustat; |
d52aa412 | 198 | |
8869b8f6 | 199 | VM_BUG_ON(!irqs_disabled()); |
addb9efe KH |
200 | |
201 | cpustat = &stat->cpustat[smp_processor_id()]; | |
c05555b5 | 202 | if (PageCgroupCache(pc)) |
addb9efe | 203 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 204 | else |
addb9efe | 205 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
206 | |
207 | if (charge) | |
addb9efe | 208 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
209 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
210 | else | |
addb9efe | 211 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 212 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
6d12e2d8 KH |
213 | } |
214 | ||
d5b69e38 | 215 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
216 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
217 | { | |
6d12e2d8 KH |
218 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
219 | } | |
220 | ||
d5b69e38 | 221 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
222 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
223 | { | |
224 | struct mem_cgroup *mem = pc->mem_cgroup; | |
225 | int nid = page_cgroup_nid(pc); | |
226 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 227 | |
6d12e2d8 KH |
228 | return mem_cgroup_zoneinfo(mem, nid, zid); |
229 | } | |
230 | ||
231 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
b69408e8 | 232 | enum lru_list idx) |
6d12e2d8 KH |
233 | { |
234 | int nid, zid; | |
235 | struct mem_cgroup_per_zone *mz; | |
236 | u64 total = 0; | |
237 | ||
238 | for_each_online_node(nid) | |
239 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
240 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
241 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
242 | } | |
243 | return total; | |
d52aa412 KH |
244 | } |
245 | ||
d5b69e38 | 246 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
247 | { |
248 | return container_of(cgroup_subsys_state(cont, | |
249 | mem_cgroup_subsys_id), struct mem_cgroup, | |
250 | css); | |
251 | } | |
252 | ||
cf475ad2 | 253 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 254 | { |
31a78f23 BS |
255 | /* |
256 | * mm_update_next_owner() may clear mm->owner to NULL | |
257 | * if it races with swapoff, page migration, etc. | |
258 | * So this can be called with p == NULL. | |
259 | */ | |
260 | if (unlikely(!p)) | |
261 | return NULL; | |
262 | ||
78fb7466 PE |
263 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
264 | struct mem_cgroup, css); | |
265 | } | |
266 | ||
3eae90c3 KH |
267 | static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, |
268 | struct page_cgroup *pc) | |
6d12e2d8 | 269 | { |
4f98a2fe RR |
270 | int lru = LRU_BASE; |
271 | ||
c05555b5 | 272 | if (PageCgroupUnevictable(pc)) |
894bc310 LS |
273 | lru = LRU_UNEVICTABLE; |
274 | else { | |
c05555b5 | 275 | if (PageCgroupActive(pc)) |
894bc310 | 276 | lru += LRU_ACTIVE; |
c05555b5 | 277 | if (PageCgroupFile(pc)) |
894bc310 LS |
278 | lru += LRU_FILE; |
279 | } | |
6d12e2d8 | 280 | |
b69408e8 | 281 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
6d12e2d8 | 282 | |
c05555b5 | 283 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, false); |
508b7be0 | 284 | list_del(&pc->lru); |
6d12e2d8 KH |
285 | } |
286 | ||
3eae90c3 | 287 | static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, |
f817ed48 | 288 | struct page_cgroup *pc, bool hot) |
6d12e2d8 | 289 | { |
4f98a2fe | 290 | int lru = LRU_BASE; |
b69408e8 | 291 | |
c05555b5 | 292 | if (PageCgroupUnevictable(pc)) |
894bc310 LS |
293 | lru = LRU_UNEVICTABLE; |
294 | else { | |
c05555b5 | 295 | if (PageCgroupActive(pc)) |
894bc310 | 296 | lru += LRU_ACTIVE; |
c05555b5 | 297 | if (PageCgroupFile(pc)) |
894bc310 LS |
298 | lru += LRU_FILE; |
299 | } | |
b69408e8 CL |
300 | |
301 | MEM_CGROUP_ZSTAT(mz, lru) += 1; | |
f817ed48 KH |
302 | if (hot) |
303 | list_add(&pc->lru, &mz->lists[lru]); | |
304 | else | |
305 | list_add_tail(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 | 306 | |
c05555b5 | 307 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true); |
6d12e2d8 KH |
308 | } |
309 | ||
894bc310 | 310 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru) |
66e1707b | 311 | { |
6d12e2d8 | 312 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); |
c05555b5 KH |
313 | int active = PageCgroupActive(pc); |
314 | int file = PageCgroupFile(pc); | |
315 | int unevictable = PageCgroupUnevictable(pc); | |
894bc310 LS |
316 | enum lru_list from = unevictable ? LRU_UNEVICTABLE : |
317 | (LRU_FILE * !!file + !!active); | |
6d12e2d8 | 318 | |
894bc310 LS |
319 | if (lru == from) |
320 | return; | |
b69408e8 | 321 | |
894bc310 | 322 | MEM_CGROUP_ZSTAT(mz, from) -= 1; |
c05555b5 KH |
323 | /* |
324 | * However this is done under mz->lru_lock, another flags, which | |
325 | * are not related to LRU, will be modified from out-of-lock. | |
326 | * We have to use atomic set/clear flags. | |
327 | */ | |
894bc310 | 328 | if (is_unevictable_lru(lru)) { |
c05555b5 KH |
329 | ClearPageCgroupActive(pc); |
330 | SetPageCgroupUnevictable(pc); | |
894bc310 LS |
331 | } else { |
332 | if (is_active_lru(lru)) | |
c05555b5 | 333 | SetPageCgroupActive(pc); |
894bc310 | 334 | else |
c05555b5 KH |
335 | ClearPageCgroupActive(pc); |
336 | ClearPageCgroupUnevictable(pc); | |
894bc310 | 337 | } |
b69408e8 | 338 | |
b69408e8 CL |
339 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
340 | list_move(&pc->lru, &mz->lists[lru]); | |
66e1707b BS |
341 | } |
342 | ||
4c4a2214 DR |
343 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
344 | { | |
345 | int ret; | |
346 | ||
347 | task_lock(task); | |
bd845e38 | 348 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
349 | task_unlock(task); |
350 | return ret; | |
351 | } | |
352 | ||
66e1707b BS |
353 | /* |
354 | * This routine assumes that the appropriate zone's lru lock is already held | |
355 | */ | |
894bc310 | 356 | void mem_cgroup_move_lists(struct page *page, enum lru_list lru) |
66e1707b | 357 | { |
427d5416 | 358 | struct page_cgroup *pc; |
072c56c1 KH |
359 | struct mem_cgroup_per_zone *mz; |
360 | unsigned long flags; | |
361 | ||
cede86ac LZ |
362 | if (mem_cgroup_subsys.disabled) |
363 | return; | |
364 | ||
2680eed7 HD |
365 | /* |
366 | * We cannot lock_page_cgroup while holding zone's lru_lock, | |
367 | * because other holders of lock_page_cgroup can be interrupted | |
368 | * with an attempt to rotate_reclaimable_page. But we cannot | |
369 | * safely get to page_cgroup without it, so just try_lock it: | |
370 | * mem_cgroup_isolate_pages allows for page left on wrong list. | |
371 | */ | |
52d4b9ac KH |
372 | pc = lookup_page_cgroup(page); |
373 | if (!trylock_page_cgroup(pc)) | |
66e1707b | 374 | return; |
52d4b9ac | 375 | if (pc && PageCgroupUsed(pc)) { |
2680eed7 | 376 | mz = page_cgroup_zoneinfo(pc); |
2680eed7 | 377 | spin_lock_irqsave(&mz->lru_lock, flags); |
894bc310 | 378 | __mem_cgroup_move_lists(pc, lru); |
2680eed7 | 379 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9b3c0a07 | 380 | } |
52d4b9ac | 381 | unlock_page_cgroup(pc); |
66e1707b BS |
382 | } |
383 | ||
58ae83db KH |
384 | /* |
385 | * Calculate mapped_ratio under memory controller. This will be used in | |
386 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
387 | */ | |
388 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
389 | { | |
390 | long total, rss; | |
391 | ||
392 | /* | |
393 | * usage is recorded in bytes. But, here, we assume the number of | |
394 | * physical pages can be represented by "long" on any arch. | |
395 | */ | |
396 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
397 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
398 | return (int)((rss * 100L) / total); | |
399 | } | |
8869b8f6 | 400 | |
6c48a1d0 KH |
401 | /* |
402 | * prev_priority control...this will be used in memory reclaim path. | |
403 | */ | |
404 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
405 | { | |
406 | return mem->prev_priority; | |
407 | } | |
408 | ||
409 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
410 | { | |
411 | if (priority < mem->prev_priority) | |
412 | mem->prev_priority = priority; | |
413 | } | |
414 | ||
415 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
416 | { | |
417 | mem->prev_priority = priority; | |
418 | } | |
419 | ||
cc38108e KH |
420 | /* |
421 | * Calculate # of pages to be scanned in this priority/zone. | |
422 | * See also vmscan.c | |
423 | * | |
424 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
425 | * (see include/linux/mmzone.h) | |
426 | */ | |
427 | ||
b69408e8 CL |
428 | long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, |
429 | int priority, enum lru_list lru) | |
cc38108e | 430 | { |
b69408e8 | 431 | long nr_pages; |
cc38108e KH |
432 | int nid = zone->zone_pgdat->node_id; |
433 | int zid = zone_idx(zone); | |
434 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
435 | ||
b69408e8 | 436 | nr_pages = MEM_CGROUP_ZSTAT(mz, lru); |
cc38108e | 437 | |
b69408e8 | 438 | return (nr_pages >> priority); |
cc38108e KH |
439 | } |
440 | ||
66e1707b BS |
441 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
442 | struct list_head *dst, | |
443 | unsigned long *scanned, int order, | |
444 | int mode, struct zone *z, | |
445 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 446 | int active, int file) |
66e1707b BS |
447 | { |
448 | unsigned long nr_taken = 0; | |
449 | struct page *page; | |
450 | unsigned long scan; | |
451 | LIST_HEAD(pc_list); | |
452 | struct list_head *src; | |
ff7283fa | 453 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
454 | int nid = z->zone_pgdat->node_id; |
455 | int zid = zone_idx(z); | |
456 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 457 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 458 | |
cf475ad2 | 459 | BUG_ON(!mem_cont); |
1ecaab2b | 460 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 461 | src = &mz->lists[lru]; |
66e1707b | 462 | |
072c56c1 | 463 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
464 | scan = 0; |
465 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 466 | if (scan >= nr_to_scan) |
ff7283fa | 467 | break; |
52d4b9ac KH |
468 | if (unlikely(!PageCgroupUsed(pc))) |
469 | continue; | |
66e1707b | 470 | page = pc->page; |
66e1707b | 471 | |
436c6541 | 472 | if (unlikely(!PageLRU(page))) |
ff7283fa | 473 | continue; |
ff7283fa | 474 | |
4f98a2fe RR |
475 | /* |
476 | * TODO: play better with lumpy reclaim, grabbing anything. | |
477 | */ | |
894bc310 LS |
478 | if (PageUnevictable(page) || |
479 | (PageActive(page) && !active) || | |
480 | (!PageActive(page) && active)) { | |
481 | __mem_cgroup_move_lists(pc, page_lru(page)); | |
66e1707b BS |
482 | continue; |
483 | } | |
484 | ||
436c6541 HD |
485 | scan++; |
486 | list_move(&pc->lru, &pc_list); | |
66e1707b | 487 | |
4f98a2fe | 488 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
489 | list_move(&page->lru, dst); |
490 | nr_taken++; | |
491 | } | |
492 | } | |
493 | ||
494 | list_splice(&pc_list, src); | |
072c56c1 | 495 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
496 | |
497 | *scanned = scan; | |
498 | return nr_taken; | |
499 | } | |
500 | ||
f817ed48 KH |
501 | /* |
502 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
503 | * oom-killer can be invoked. | |
8a9f3ccd | 504 | */ |
f817ed48 | 505 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
8c7c6e34 KH |
506 | gfp_t gfp_mask, struct mem_cgroup **memcg, |
507 | bool oom) | |
8a9f3ccd BS |
508 | { |
509 | struct mem_cgroup *mem; | |
7a81b88c | 510 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
8a9f3ccd | 511 | /* |
3be91277 HD |
512 | * We always charge the cgroup the mm_struct belongs to. |
513 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
514 | * thread group leader migrates. It's possible that mm is not |
515 | * set, if so charge the init_mm (happens for pagecache usage). | |
516 | */ | |
7a81b88c | 517 | if (likely(!*memcg)) { |
e8589cc1 KH |
518 | rcu_read_lock(); |
519 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
520 | if (unlikely(!mem)) { |
521 | rcu_read_unlock(); | |
31a78f23 BS |
522 | return 0; |
523 | } | |
e8589cc1 KH |
524 | /* |
525 | * For every charge from the cgroup, increment reference count | |
526 | */ | |
527 | css_get(&mem->css); | |
7a81b88c | 528 | *memcg = mem; |
e8589cc1 KH |
529 | rcu_read_unlock(); |
530 | } else { | |
7a81b88c KH |
531 | mem = *memcg; |
532 | css_get(&mem->css); | |
e8589cc1 | 533 | } |
8a9f3ccd | 534 | |
8c7c6e34 KH |
535 | while (1) { |
536 | int ret; | |
537 | bool noswap = false; | |
7a81b88c | 538 | |
8c7c6e34 KH |
539 | ret = res_counter_charge(&mem->res, PAGE_SIZE); |
540 | if (likely(!ret)) { | |
541 | if (!do_swap_account) | |
542 | break; | |
543 | ret = res_counter_charge(&mem->memsw, PAGE_SIZE); | |
544 | if (likely(!ret)) | |
545 | break; | |
546 | /* mem+swap counter fails */ | |
547 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
548 | noswap = true; | |
549 | } | |
3be91277 | 550 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 551 | goto nomem; |
e1a1cd59 | 552 | |
8c7c6e34 | 553 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask, noswap)) |
66e1707b BS |
554 | continue; |
555 | ||
556 | /* | |
8869b8f6 HD |
557 | * try_to_free_mem_cgroup_pages() might not give us a full |
558 | * picture of reclaim. Some pages are reclaimed and might be | |
559 | * moved to swap cache or just unmapped from the cgroup. | |
560 | * Check the limit again to see if the reclaim reduced the | |
561 | * current usage of the cgroup before giving up | |
8c7c6e34 | 562 | * |
8869b8f6 | 563 | */ |
8c7c6e34 KH |
564 | if (!do_swap_account && |
565 | res_counter_check_under_limit(&mem->res)) | |
566 | continue; | |
567 | if (do_swap_account && | |
568 | res_counter_check_under_limit(&mem->memsw)) | |
66e1707b | 569 | continue; |
3be91277 HD |
570 | |
571 | if (!nr_retries--) { | |
f817ed48 KH |
572 | if (oom) |
573 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
7a81b88c | 574 | goto nomem; |
66e1707b | 575 | } |
8a9f3ccd | 576 | } |
7a81b88c KH |
577 | return 0; |
578 | nomem: | |
579 | css_put(&mem->css); | |
580 | return -ENOMEM; | |
581 | } | |
8a9f3ccd | 582 | |
f817ed48 KH |
583 | /** |
584 | * mem_cgroup_try_charge - get charge of PAGE_SIZE. | |
585 | * @mm: an mm_struct which is charged against. (when *memcg is NULL) | |
586 | * @gfp_mask: gfp_mask for reclaim. | |
587 | * @memcg: a pointer to memory cgroup which is charged against. | |
588 | * | |
589 | * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated | |
590 | * memory cgroup from @mm is got and stored in *memcg. | |
591 | * | |
592 | * Returns 0 if success. -ENOMEM at failure. | |
593 | * This call can invoke OOM-Killer. | |
594 | */ | |
595 | ||
596 | int mem_cgroup_try_charge(struct mm_struct *mm, | |
597 | gfp_t mask, struct mem_cgroup **memcg) | |
598 | { | |
599 | return __mem_cgroup_try_charge(mm, mask, memcg, true); | |
600 | } | |
601 | ||
7a81b88c KH |
602 | /* |
603 | * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be | |
604 | * USED state. If already USED, uncharge and return. | |
605 | */ | |
606 | ||
607 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
608 | struct page_cgroup *pc, | |
609 | enum charge_type ctype) | |
610 | { | |
611 | struct mem_cgroup_per_zone *mz; | |
612 | unsigned long flags; | |
613 | ||
614 | /* try_charge() can return NULL to *memcg, taking care of it. */ | |
615 | if (!mem) | |
616 | return; | |
52d4b9ac KH |
617 | |
618 | lock_page_cgroup(pc); | |
619 | if (unlikely(PageCgroupUsed(pc))) { | |
620 | unlock_page_cgroup(pc); | |
621 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
622 | if (do_swap_account) |
623 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
52d4b9ac | 624 | css_put(&mem->css); |
7a81b88c | 625 | return; |
52d4b9ac | 626 | } |
8a9f3ccd | 627 | pc->mem_cgroup = mem; |
508b7be0 KH |
628 | /* |
629 | * If a page is accounted as a page cache, insert to inactive list. | |
630 | * If anon, insert to active list. | |
631 | */ | |
c05555b5 | 632 | pc->flags = pcg_default_flags[ctype]; |
3be91277 | 633 | |
072c56c1 | 634 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 635 | |
072c56c1 | 636 | spin_lock_irqsave(&mz->lru_lock, flags); |
f817ed48 | 637 | __mem_cgroup_add_list(mz, pc, true); |
072c56c1 | 638 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
52d4b9ac | 639 | unlock_page_cgroup(pc); |
7a81b88c | 640 | } |
66e1707b | 641 | |
f817ed48 KH |
642 | /** |
643 | * mem_cgroup_move_account - move account of the page | |
644 | * @pc: page_cgroup of the page. | |
645 | * @from: mem_cgroup which the page is moved from. | |
646 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
647 | * | |
648 | * The caller must confirm following. | |
649 | * 1. disable irq. | |
650 | * 2. lru_lock of old mem_cgroup(@from) should be held. | |
651 | * | |
652 | * returns 0 at success, | |
653 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
654 | * | |
655 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
656 | * new cgroup. It should be done by a caller. | |
657 | */ | |
658 | ||
659 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
660 | struct mem_cgroup *from, struct mem_cgroup *to) | |
661 | { | |
662 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
663 | int nid, zid; | |
664 | int ret = -EBUSY; | |
665 | ||
666 | VM_BUG_ON(!irqs_disabled()); | |
667 | VM_BUG_ON(from == to); | |
668 | ||
669 | nid = page_cgroup_nid(pc); | |
670 | zid = page_cgroup_zid(pc); | |
671 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
672 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
673 | ||
674 | ||
675 | if (!trylock_page_cgroup(pc)) | |
676 | return ret; | |
677 | ||
678 | if (!PageCgroupUsed(pc)) | |
679 | goto out; | |
680 | ||
681 | if (pc->mem_cgroup != from) | |
682 | goto out; | |
683 | ||
684 | if (spin_trylock(&to_mz->lru_lock)) { | |
685 | __mem_cgroup_remove_list(from_mz, pc); | |
686 | css_put(&from->css); | |
687 | res_counter_uncharge(&from->res, PAGE_SIZE); | |
8c7c6e34 KH |
688 | if (do_swap_account) |
689 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
f817ed48 KH |
690 | pc->mem_cgroup = to; |
691 | css_get(&to->css); | |
692 | __mem_cgroup_add_list(to_mz, pc, false); | |
693 | ret = 0; | |
694 | spin_unlock(&to_mz->lru_lock); | |
695 | } | |
696 | out: | |
697 | unlock_page_cgroup(pc); | |
698 | return ret; | |
699 | } | |
700 | ||
701 | /* | |
702 | * move charges to its parent. | |
703 | */ | |
704 | ||
705 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
706 | struct mem_cgroup *child, | |
707 | gfp_t gfp_mask) | |
708 | { | |
709 | struct cgroup *cg = child->css.cgroup; | |
710 | struct cgroup *pcg = cg->parent; | |
711 | struct mem_cgroup *parent; | |
712 | struct mem_cgroup_per_zone *mz; | |
713 | unsigned long flags; | |
714 | int ret; | |
715 | ||
716 | /* Is ROOT ? */ | |
717 | if (!pcg) | |
718 | return -EINVAL; | |
719 | ||
720 | parent = mem_cgroup_from_cont(pcg); | |
721 | ||
722 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); | |
723 | if (ret) | |
724 | return ret; | |
725 | ||
726 | mz = mem_cgroup_zoneinfo(child, | |
727 | page_cgroup_nid(pc), page_cgroup_zid(pc)); | |
728 | ||
729 | spin_lock_irqsave(&mz->lru_lock, flags); | |
730 | ret = mem_cgroup_move_account(pc, child, parent); | |
731 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
732 | ||
733 | /* drop extra refcnt */ | |
734 | css_put(&parent->css); | |
735 | /* uncharge if move fails */ | |
8c7c6e34 | 736 | if (ret) { |
f817ed48 | 737 | res_counter_uncharge(&parent->res, PAGE_SIZE); |
8c7c6e34 KH |
738 | if (do_swap_account) |
739 | res_counter_uncharge(&parent->memsw, PAGE_SIZE); | |
740 | } | |
f817ed48 KH |
741 | |
742 | return ret; | |
743 | } | |
744 | ||
7a81b88c KH |
745 | /* |
746 | * Charge the memory controller for page usage. | |
747 | * Return | |
748 | * 0 if the charge was successful | |
749 | * < 0 if the cgroup is over its limit | |
750 | */ | |
751 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
752 | gfp_t gfp_mask, enum charge_type ctype, | |
753 | struct mem_cgroup *memcg) | |
754 | { | |
755 | struct mem_cgroup *mem; | |
756 | struct page_cgroup *pc; | |
757 | int ret; | |
758 | ||
759 | pc = lookup_page_cgroup(page); | |
760 | /* can happen at boot */ | |
761 | if (unlikely(!pc)) | |
762 | return 0; | |
763 | prefetchw(pc); | |
764 | ||
765 | mem = memcg; | |
f817ed48 | 766 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
7a81b88c KH |
767 | if (ret) |
768 | return ret; | |
769 | ||
770 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 771 | return 0; |
8a9f3ccd BS |
772 | } |
773 | ||
7a81b88c KH |
774 | int mem_cgroup_newpage_charge(struct page *page, |
775 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 776 | { |
cede86ac LZ |
777 | if (mem_cgroup_subsys.disabled) |
778 | return 0; | |
52d4b9ac KH |
779 | if (PageCompound(page)) |
780 | return 0; | |
69029cd5 KH |
781 | /* |
782 | * If already mapped, we don't have to account. | |
783 | * If page cache, page->mapping has address_space. | |
784 | * But page->mapping may have out-of-use anon_vma pointer, | |
785 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
786 | * is NULL. | |
787 | */ | |
788 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
789 | return 0; | |
790 | if (unlikely(!mm)) | |
791 | mm = &init_mm; | |
217bc319 | 792 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 793 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
794 | } |
795 | ||
e1a1cd59 BS |
796 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
797 | gfp_t gfp_mask) | |
8697d331 | 798 | { |
cede86ac LZ |
799 | if (mem_cgroup_subsys.disabled) |
800 | return 0; | |
52d4b9ac KH |
801 | if (PageCompound(page)) |
802 | return 0; | |
accf163e KH |
803 | /* |
804 | * Corner case handling. This is called from add_to_page_cache() | |
805 | * in usual. But some FS (shmem) precharges this page before calling it | |
806 | * and call add_to_page_cache() with GFP_NOWAIT. | |
807 | * | |
808 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
809 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
810 | * charge twice. (It works but has to pay a bit larger cost.) | |
811 | */ | |
812 | if (!(gfp_mask & __GFP_WAIT)) { | |
813 | struct page_cgroup *pc; | |
814 | ||
52d4b9ac KH |
815 | |
816 | pc = lookup_page_cgroup(page); | |
817 | if (!pc) | |
818 | return 0; | |
819 | lock_page_cgroup(pc); | |
820 | if (PageCgroupUsed(pc)) { | |
821 | unlock_page_cgroup(pc); | |
accf163e KH |
822 | return 0; |
823 | } | |
52d4b9ac | 824 | unlock_page_cgroup(pc); |
accf163e KH |
825 | } |
826 | ||
69029cd5 | 827 | if (unlikely(!mm)) |
8697d331 | 828 | mm = &init_mm; |
accf163e | 829 | |
c05555b5 KH |
830 | if (page_is_file_cache(page)) |
831 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 832 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
c05555b5 KH |
833 | else |
834 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
835 | MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL); | |
e8589cc1 KH |
836 | } |
837 | ||
8c7c6e34 KH |
838 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
839 | struct page *page, | |
840 | gfp_t mask, struct mem_cgroup **ptr) | |
841 | { | |
842 | struct mem_cgroup *mem; | |
843 | swp_entry_t ent; | |
844 | ||
845 | if (mem_cgroup_subsys.disabled) | |
846 | return 0; | |
847 | ||
848 | if (!do_swap_account) | |
849 | goto charge_cur_mm; | |
850 | ||
851 | /* | |
852 | * A racing thread's fault, or swapoff, may have already updated | |
853 | * the pte, and even removed page from swap cache: return success | |
854 | * to go on to do_swap_page()'s pte_same() test, which should fail. | |
855 | */ | |
856 | if (!PageSwapCache(page)) | |
857 | return 0; | |
858 | ||
859 | ent.val = page_private(page); | |
860 | ||
861 | mem = lookup_swap_cgroup(ent); | |
862 | if (!mem || mem->obsolete) | |
863 | goto charge_cur_mm; | |
864 | *ptr = mem; | |
865 | return __mem_cgroup_try_charge(NULL, mask, ptr, true); | |
866 | charge_cur_mm: | |
867 | if (unlikely(!mm)) | |
868 | mm = &init_mm; | |
869 | return __mem_cgroup_try_charge(mm, mask, ptr, true); | |
870 | } | |
871 | ||
d13d1443 | 872 | #ifdef CONFIG_SWAP |
8c7c6e34 | 873 | |
d13d1443 KH |
874 | int mem_cgroup_cache_charge_swapin(struct page *page, |
875 | struct mm_struct *mm, gfp_t mask, bool locked) | |
876 | { | |
877 | int ret = 0; | |
878 | ||
879 | if (mem_cgroup_subsys.disabled) | |
880 | return 0; | |
881 | if (unlikely(!mm)) | |
882 | mm = &init_mm; | |
883 | if (!locked) | |
884 | lock_page(page); | |
885 | /* | |
886 | * If not locked, the page can be dropped from SwapCache until | |
887 | * we reach here. | |
888 | */ | |
889 | if (PageSwapCache(page)) { | |
8c7c6e34 KH |
890 | struct mem_cgroup *mem = NULL; |
891 | swp_entry_t ent; | |
892 | ||
893 | ent.val = page_private(page); | |
894 | if (do_swap_account) { | |
895 | mem = lookup_swap_cgroup(ent); | |
896 | if (mem && mem->obsolete) | |
897 | mem = NULL; | |
898 | if (mem) | |
899 | mm = NULL; | |
900 | } | |
d13d1443 | 901 | ret = mem_cgroup_charge_common(page, mm, mask, |
8c7c6e34 KH |
902 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); |
903 | ||
904 | if (!ret && do_swap_account) { | |
905 | /* avoid double counting */ | |
906 | mem = swap_cgroup_record(ent, NULL); | |
907 | if (mem) { | |
908 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
909 | mem_cgroup_put(mem); | |
910 | } | |
911 | } | |
d13d1443 KH |
912 | } |
913 | if (!locked) | |
914 | unlock_page(page); | |
915 | ||
916 | return ret; | |
917 | } | |
918 | #endif | |
919 | ||
7a81b88c KH |
920 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
921 | { | |
922 | struct page_cgroup *pc; | |
923 | ||
924 | if (mem_cgroup_subsys.disabled) | |
925 | return; | |
926 | if (!ptr) | |
927 | return; | |
928 | pc = lookup_page_cgroup(page); | |
929 | __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
8c7c6e34 KH |
930 | /* |
931 | * Now swap is on-memory. This means this page may be | |
932 | * counted both as mem and swap....double count. | |
933 | * Fix it by uncharging from memsw. This SwapCache is stable | |
934 | * because we're still under lock_page(). | |
935 | */ | |
936 | if (do_swap_account) { | |
937 | swp_entry_t ent = {.val = page_private(page)}; | |
938 | struct mem_cgroup *memcg; | |
939 | memcg = swap_cgroup_record(ent, NULL); | |
940 | if (memcg) { | |
941 | /* If memcg is obsolete, memcg can be != ptr */ | |
942 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
943 | mem_cgroup_put(memcg); | |
944 | } | |
945 | ||
946 | } | |
7a81b88c KH |
947 | } |
948 | ||
949 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
950 | { | |
951 | if (mem_cgroup_subsys.disabled) | |
952 | return; | |
953 | if (!mem) | |
954 | return; | |
955 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
956 | if (do_swap_account) |
957 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
7a81b88c KH |
958 | css_put(&mem->css); |
959 | } | |
960 | ||
961 | ||
8a9f3ccd | 962 | /* |
69029cd5 | 963 | * uncharge if !page_mapped(page) |
8a9f3ccd | 964 | */ |
8c7c6e34 | 965 | static struct mem_cgroup * |
69029cd5 | 966 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 967 | { |
8289546e | 968 | struct page_cgroup *pc; |
8c7c6e34 | 969 | struct mem_cgroup *mem = NULL; |
072c56c1 | 970 | struct mem_cgroup_per_zone *mz; |
66e1707b | 971 | unsigned long flags; |
8a9f3ccd | 972 | |
4077960e | 973 | if (mem_cgroup_subsys.disabled) |
8c7c6e34 | 974 | return NULL; |
4077960e | 975 | |
d13d1443 | 976 | if (PageSwapCache(page)) |
8c7c6e34 | 977 | return NULL; |
d13d1443 | 978 | |
8697d331 | 979 | /* |
3c541e14 | 980 | * Check if our page_cgroup is valid |
8697d331 | 981 | */ |
52d4b9ac KH |
982 | pc = lookup_page_cgroup(page); |
983 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 984 | return NULL; |
b9c565d5 | 985 | |
52d4b9ac | 986 | lock_page_cgroup(pc); |
d13d1443 | 987 | |
8c7c6e34 KH |
988 | mem = pc->mem_cgroup; |
989 | ||
d13d1443 KH |
990 | if (!PageCgroupUsed(pc)) |
991 | goto unlock_out; | |
992 | ||
993 | switch (ctype) { | |
994 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
995 | if (page_mapped(page)) | |
996 | goto unlock_out; | |
997 | break; | |
998 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
999 | if (!PageAnon(page)) { /* Shared memory */ | |
1000 | if (page->mapping && !page_is_file_cache(page)) | |
1001 | goto unlock_out; | |
1002 | } else if (page_mapped(page)) /* Anon */ | |
1003 | goto unlock_out; | |
1004 | break; | |
1005 | default: | |
1006 | break; | |
52d4b9ac | 1007 | } |
d13d1443 | 1008 | |
8c7c6e34 KH |
1009 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1010 | if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)) | |
1011 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1012 | ||
52d4b9ac | 1013 | ClearPageCgroupUsed(pc); |
b9c565d5 | 1014 | |
69029cd5 KH |
1015 | mz = page_cgroup_zoneinfo(pc); |
1016 | spin_lock_irqsave(&mz->lru_lock, flags); | |
1017 | __mem_cgroup_remove_list(mz, pc); | |
1018 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
52d4b9ac | 1019 | unlock_page_cgroup(pc); |
fb59e9f1 | 1020 | |
69029cd5 | 1021 | css_put(&mem->css); |
6d12e2d8 | 1022 | |
8c7c6e34 | 1023 | return mem; |
d13d1443 KH |
1024 | |
1025 | unlock_out: | |
1026 | unlock_page_cgroup(pc); | |
8c7c6e34 | 1027 | return NULL; |
3c541e14 BS |
1028 | } |
1029 | ||
69029cd5 KH |
1030 | void mem_cgroup_uncharge_page(struct page *page) |
1031 | { | |
52d4b9ac KH |
1032 | /* early check. */ |
1033 | if (page_mapped(page)) | |
1034 | return; | |
1035 | if (page->mapping && !PageAnon(page)) | |
1036 | return; | |
69029cd5 KH |
1037 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
1038 | } | |
1039 | ||
1040 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
1041 | { | |
1042 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 1043 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
1044 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
1045 | } | |
1046 | ||
8c7c6e34 KH |
1047 | /* |
1048 | * called from __delete_from_swap_cache() and drop "page" account. | |
1049 | * memcg information is recorded to swap_cgroup of "ent" | |
1050 | */ | |
1051 | void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent) | |
1052 | { | |
1053 | struct mem_cgroup *memcg; | |
1054 | ||
1055 | memcg = __mem_cgroup_uncharge_common(page, | |
1056 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT); | |
1057 | /* record memcg information */ | |
1058 | if (do_swap_account && memcg) { | |
1059 | swap_cgroup_record(ent, memcg); | |
1060 | mem_cgroup_get(memcg); | |
1061 | } | |
1062 | } | |
1063 | ||
1064 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1065 | /* | |
1066 | * called from swap_entry_free(). remove record in swap_cgroup and | |
1067 | * uncharge "memsw" account. | |
1068 | */ | |
1069 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 1070 | { |
8c7c6e34 KH |
1071 | struct mem_cgroup *memcg; |
1072 | ||
1073 | if (!do_swap_account) | |
1074 | return; | |
1075 | ||
1076 | memcg = swap_cgroup_record(ent, NULL); | |
1077 | if (memcg) { | |
1078 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
1079 | mem_cgroup_put(memcg); | |
1080 | } | |
d13d1443 | 1081 | } |
8c7c6e34 | 1082 | #endif |
d13d1443 | 1083 | |
ae41be37 | 1084 | /* |
01b1ae63 KH |
1085 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
1086 | * page belongs to. | |
ae41be37 | 1087 | */ |
01b1ae63 | 1088 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
1089 | { |
1090 | struct page_cgroup *pc; | |
e8589cc1 | 1091 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 1092 | int ret = 0; |
8869b8f6 | 1093 | |
4077960e BS |
1094 | if (mem_cgroup_subsys.disabled) |
1095 | return 0; | |
1096 | ||
52d4b9ac KH |
1097 | pc = lookup_page_cgroup(page); |
1098 | lock_page_cgroup(pc); | |
1099 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
1100 | mem = pc->mem_cgroup; |
1101 | css_get(&mem->css); | |
e8589cc1 | 1102 | } |
52d4b9ac | 1103 | unlock_page_cgroup(pc); |
01b1ae63 | 1104 | |
e8589cc1 | 1105 | if (mem) { |
01b1ae63 | 1106 | ret = mem_cgroup_try_charge(NULL, GFP_HIGHUSER_MOVABLE, &mem); |
e8589cc1 KH |
1107 | css_put(&mem->css); |
1108 | } | |
01b1ae63 | 1109 | *ptr = mem; |
e8589cc1 | 1110 | return ret; |
ae41be37 | 1111 | } |
8869b8f6 | 1112 | |
69029cd5 | 1113 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
1114 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
1115 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 1116 | { |
01b1ae63 KH |
1117 | struct page *target, *unused; |
1118 | struct page_cgroup *pc; | |
1119 | enum charge_type ctype; | |
1120 | ||
1121 | if (!mem) | |
1122 | return; | |
1123 | ||
1124 | /* at migration success, oldpage->mapping is NULL. */ | |
1125 | if (oldpage->mapping) { | |
1126 | target = oldpage; | |
1127 | unused = NULL; | |
1128 | } else { | |
1129 | target = newpage; | |
1130 | unused = oldpage; | |
1131 | } | |
1132 | ||
1133 | if (PageAnon(target)) | |
1134 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
1135 | else if (page_is_file_cache(target)) | |
1136 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
1137 | else | |
1138 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
1139 | ||
1140 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 1141 | if (unused) |
01b1ae63 KH |
1142 | __mem_cgroup_uncharge_common(unused, ctype); |
1143 | ||
1144 | pc = lookup_page_cgroup(target); | |
69029cd5 | 1145 | /* |
01b1ae63 KH |
1146 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
1147 | * So, double-counting is effectively avoided. | |
1148 | */ | |
1149 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
1150 | ||
1151 | /* | |
1152 | * Both of oldpage and newpage are still under lock_page(). | |
1153 | * Then, we don't have to care about race in radix-tree. | |
1154 | * But we have to be careful that this page is unmapped or not. | |
1155 | * | |
1156 | * There is a case for !page_mapped(). At the start of | |
1157 | * migration, oldpage was mapped. But now, it's zapped. | |
1158 | * But we know *target* page is not freed/reused under us. | |
1159 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 1160 | */ |
01b1ae63 KH |
1161 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
1162 | mem_cgroup_uncharge_page(target); | |
ae41be37 | 1163 | } |
78fb7466 | 1164 | |
c9b0ed51 KH |
1165 | /* |
1166 | * A call to try to shrink memory usage under specified resource controller. | |
1167 | * This is typically used for page reclaiming for shmem for reducing side | |
1168 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
1169 | */ | |
1170 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) | |
1171 | { | |
1172 | struct mem_cgroup *mem; | |
1173 | int progress = 0; | |
1174 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
1175 | ||
cede86ac LZ |
1176 | if (mem_cgroup_subsys.disabled) |
1177 | return 0; | |
9623e078 HD |
1178 | if (!mm) |
1179 | return 0; | |
cede86ac | 1180 | |
c9b0ed51 KH |
1181 | rcu_read_lock(); |
1182 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
1183 | if (unlikely(!mem)) { |
1184 | rcu_read_unlock(); | |
1185 | return 0; | |
1186 | } | |
c9b0ed51 KH |
1187 | css_get(&mem->css); |
1188 | rcu_read_unlock(); | |
1189 | ||
1190 | do { | |
8c7c6e34 | 1191 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask, true); |
a10cebf5 | 1192 | progress += res_counter_check_under_limit(&mem->res); |
c9b0ed51 KH |
1193 | } while (!progress && --retry); |
1194 | ||
1195 | css_put(&mem->css); | |
1196 | if (!retry) | |
1197 | return -ENOMEM; | |
1198 | return 0; | |
1199 | } | |
1200 | ||
8c7c6e34 KH |
1201 | static DEFINE_MUTEX(set_limit_mutex); |
1202 | ||
d38d2a75 | 1203 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 1204 | unsigned long long val) |
628f4235 KH |
1205 | { |
1206 | ||
1207 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
1208 | int progress; | |
8c7c6e34 | 1209 | u64 memswlimit; |
628f4235 KH |
1210 | int ret = 0; |
1211 | ||
8c7c6e34 | 1212 | while (retry_count) { |
628f4235 KH |
1213 | if (signal_pending(current)) { |
1214 | ret = -EINTR; | |
1215 | break; | |
1216 | } | |
8c7c6e34 KH |
1217 | /* |
1218 | * Rather than hide all in some function, I do this in | |
1219 | * open coded manner. You see what this really does. | |
1220 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1221 | */ | |
1222 | mutex_lock(&set_limit_mutex); | |
1223 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1224 | if (memswlimit < val) { | |
1225 | ret = -EINVAL; | |
1226 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
1227 | break; |
1228 | } | |
8c7c6e34 KH |
1229 | ret = res_counter_set_limit(&memcg->res, val); |
1230 | mutex_unlock(&set_limit_mutex); | |
1231 | ||
1232 | if (!ret) | |
1233 | break; | |
1234 | ||
bced0520 | 1235 | progress = try_to_free_mem_cgroup_pages(memcg, |
8c7c6e34 KH |
1236 | GFP_HIGHUSER_MOVABLE, false); |
1237 | if (!progress) retry_count--; | |
1238 | } | |
1239 | return ret; | |
1240 | } | |
1241 | ||
1242 | int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, | |
1243 | unsigned long long val) | |
1244 | { | |
1245 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
1246 | u64 memlimit, oldusage, curusage; | |
1247 | int ret; | |
1248 | ||
1249 | if (!do_swap_account) | |
1250 | return -EINVAL; | |
1251 | ||
1252 | while (retry_count) { | |
1253 | if (signal_pending(current)) { | |
1254 | ret = -EINTR; | |
1255 | break; | |
1256 | } | |
1257 | /* | |
1258 | * Rather than hide all in some function, I do this in | |
1259 | * open coded manner. You see what this really does. | |
1260 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1261 | */ | |
1262 | mutex_lock(&set_limit_mutex); | |
1263 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1264 | if (memlimit > val) { | |
1265 | ret = -EINVAL; | |
1266 | mutex_unlock(&set_limit_mutex); | |
1267 | break; | |
1268 | } | |
1269 | ret = res_counter_set_limit(&memcg->memsw, val); | |
1270 | mutex_unlock(&set_limit_mutex); | |
1271 | ||
1272 | if (!ret) | |
1273 | break; | |
1274 | ||
1275 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
1276 | try_to_free_mem_cgroup_pages(memcg, GFP_HIGHUSER_MOVABLE, true); | |
1277 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
1278 | if (curusage >= oldusage) | |
628f4235 KH |
1279 | retry_count--; |
1280 | } | |
1281 | return ret; | |
1282 | } | |
1283 | ||
1284 | ||
cc847582 KH |
1285 | /* |
1286 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
1287 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
1288 | */ | |
f817ed48 | 1289 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 | 1290 | struct mem_cgroup_per_zone *mz, |
b69408e8 | 1291 | enum lru_list lru) |
cc847582 | 1292 | { |
f817ed48 | 1293 | struct page_cgroup *pc, *busy; |
cc847582 | 1294 | unsigned long flags; |
f817ed48 | 1295 | unsigned long loop; |
072c56c1 | 1296 | struct list_head *list; |
f817ed48 | 1297 | int ret = 0; |
072c56c1 | 1298 | |
b69408e8 | 1299 | list = &mz->lists[lru]; |
cc847582 | 1300 | |
f817ed48 KH |
1301 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
1302 | /* give some margin against EBUSY etc...*/ | |
1303 | loop += 256; | |
1304 | busy = NULL; | |
1305 | while (loop--) { | |
1306 | ret = 0; | |
1307 | spin_lock_irqsave(&mz->lru_lock, flags); | |
1308 | if (list_empty(list)) { | |
1309 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
52d4b9ac | 1310 | break; |
f817ed48 KH |
1311 | } |
1312 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
1313 | if (busy == pc) { | |
1314 | list_move(&pc->lru, list); | |
1315 | busy = 0; | |
1316 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
1317 | continue; | |
1318 | } | |
9b3c0a07 | 1319 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
f817ed48 KH |
1320 | |
1321 | ret = mem_cgroup_move_parent(pc, mem, GFP_HIGHUSER_MOVABLE); | |
1322 | if (ret == -ENOMEM) | |
52d4b9ac | 1323 | break; |
f817ed48 KH |
1324 | |
1325 | if (ret == -EBUSY || ret == -EINVAL) { | |
1326 | /* found lock contention or "pc" is obsolete. */ | |
1327 | busy = pc; | |
1328 | cond_resched(); | |
1329 | } else | |
1330 | busy = NULL; | |
cc847582 | 1331 | } |
f817ed48 KH |
1332 | if (!ret && !list_empty(list)) |
1333 | return -EBUSY; | |
1334 | return ret; | |
cc847582 KH |
1335 | } |
1336 | ||
1337 | /* | |
1338 | * make mem_cgroup's charge to be 0 if there is no task. | |
1339 | * This enables deleting this mem_cgroup. | |
1340 | */ | |
c1e862c1 | 1341 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 1342 | { |
f817ed48 KH |
1343 | int ret; |
1344 | int node, zid, shrink; | |
1345 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 1346 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 1347 | |
cc847582 | 1348 | css_get(&mem->css); |
f817ed48 KH |
1349 | |
1350 | shrink = 0; | |
c1e862c1 KH |
1351 | /* should free all ? */ |
1352 | if (free_all) | |
1353 | goto try_to_free; | |
f817ed48 | 1354 | move_account: |
1ecaab2b | 1355 | while (mem->res.usage > 0) { |
f817ed48 | 1356 | ret = -EBUSY; |
c1e862c1 KH |
1357 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
1358 | goto out; | |
1359 | ret = -EINTR; | |
1360 | if (signal_pending(current)) | |
cc847582 | 1361 | goto out; |
52d4b9ac KH |
1362 | /* This is for making all *used* pages to be on LRU. */ |
1363 | lru_add_drain_all(); | |
f817ed48 KH |
1364 | ret = 0; |
1365 | for_each_node_state(node, N_POSSIBLE) { | |
1366 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { | |
1ecaab2b | 1367 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1368 | enum lru_list l; |
1ecaab2b | 1369 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
f817ed48 KH |
1370 | for_each_lru(l) { |
1371 | ret = mem_cgroup_force_empty_list(mem, | |
1372 | mz, l); | |
1373 | if (ret) | |
1374 | break; | |
1375 | } | |
1ecaab2b | 1376 | } |
f817ed48 KH |
1377 | if (ret) |
1378 | break; | |
1379 | } | |
1380 | /* it seems parent cgroup doesn't have enough mem */ | |
1381 | if (ret == -ENOMEM) | |
1382 | goto try_to_free; | |
52d4b9ac | 1383 | cond_resched(); |
cc847582 KH |
1384 | } |
1385 | ret = 0; | |
1386 | out: | |
1387 | css_put(&mem->css); | |
1388 | return ret; | |
f817ed48 KH |
1389 | |
1390 | try_to_free: | |
c1e862c1 KH |
1391 | /* returns EBUSY if there is a task or if we come here twice. */ |
1392 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
1393 | ret = -EBUSY; |
1394 | goto out; | |
1395 | } | |
c1e862c1 KH |
1396 | /* we call try-to-free pages for make this cgroup empty */ |
1397 | lru_add_drain_all(); | |
f817ed48 KH |
1398 | /* try to free all pages in this cgroup */ |
1399 | shrink = 1; | |
1400 | while (nr_retries && mem->res.usage > 0) { | |
1401 | int progress; | |
c1e862c1 KH |
1402 | |
1403 | if (signal_pending(current)) { | |
1404 | ret = -EINTR; | |
1405 | goto out; | |
1406 | } | |
f817ed48 | 1407 | progress = try_to_free_mem_cgroup_pages(mem, |
8c7c6e34 | 1408 | GFP_HIGHUSER_MOVABLE, false); |
c1e862c1 | 1409 | if (!progress) { |
f817ed48 | 1410 | nr_retries--; |
c1e862c1 KH |
1411 | /* maybe some writeback is necessary */ |
1412 | congestion_wait(WRITE, HZ/10); | |
1413 | } | |
f817ed48 KH |
1414 | |
1415 | } | |
1416 | /* try move_account...there may be some *locked* pages. */ | |
1417 | if (mem->res.usage) | |
1418 | goto move_account; | |
1419 | ret = 0; | |
1420 | goto out; | |
cc847582 KH |
1421 | } |
1422 | ||
c1e862c1 KH |
1423 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
1424 | { | |
1425 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
1426 | } | |
1427 | ||
1428 | ||
2c3daa72 | 1429 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 1430 | { |
8c7c6e34 KH |
1431 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
1432 | u64 val = 0; | |
1433 | int type, name; | |
1434 | ||
1435 | type = MEMFILE_TYPE(cft->private); | |
1436 | name = MEMFILE_ATTR(cft->private); | |
1437 | switch (type) { | |
1438 | case _MEM: | |
1439 | val = res_counter_read_u64(&mem->res, name); | |
1440 | break; | |
1441 | case _MEMSWAP: | |
1442 | if (do_swap_account) | |
1443 | val = res_counter_read_u64(&mem->memsw, name); | |
1444 | break; | |
1445 | default: | |
1446 | BUG(); | |
1447 | break; | |
1448 | } | |
1449 | return val; | |
8cdea7c0 | 1450 | } |
628f4235 KH |
1451 | /* |
1452 | * The user of this function is... | |
1453 | * RES_LIMIT. | |
1454 | */ | |
856c13aa PM |
1455 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
1456 | const char *buffer) | |
8cdea7c0 | 1457 | { |
628f4235 | 1458 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 1459 | int type, name; |
628f4235 KH |
1460 | unsigned long long val; |
1461 | int ret; | |
1462 | ||
8c7c6e34 KH |
1463 | type = MEMFILE_TYPE(cft->private); |
1464 | name = MEMFILE_ATTR(cft->private); | |
1465 | switch (name) { | |
628f4235 KH |
1466 | case RES_LIMIT: |
1467 | /* This function does all necessary parse...reuse it */ | |
1468 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
1469 | if (ret) |
1470 | break; | |
1471 | if (type == _MEM) | |
628f4235 | 1472 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
1473 | else |
1474 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 KH |
1475 | break; |
1476 | default: | |
1477 | ret = -EINVAL; /* should be BUG() ? */ | |
1478 | break; | |
1479 | } | |
1480 | return ret; | |
8cdea7c0 BS |
1481 | } |
1482 | ||
29f2a4da | 1483 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
1484 | { |
1485 | struct mem_cgroup *mem; | |
8c7c6e34 | 1486 | int type, name; |
c84872e1 PE |
1487 | |
1488 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
1489 | type = MEMFILE_TYPE(event); |
1490 | name = MEMFILE_ATTR(event); | |
1491 | switch (name) { | |
29f2a4da | 1492 | case RES_MAX_USAGE: |
8c7c6e34 KH |
1493 | if (type == _MEM) |
1494 | res_counter_reset_max(&mem->res); | |
1495 | else | |
1496 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
1497 | break; |
1498 | case RES_FAILCNT: | |
8c7c6e34 KH |
1499 | if (type == _MEM) |
1500 | res_counter_reset_failcnt(&mem->res); | |
1501 | else | |
1502 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
1503 | break; |
1504 | } | |
85cc59db | 1505 | return 0; |
c84872e1 PE |
1506 | } |
1507 | ||
d2ceb9b7 KH |
1508 | static const struct mem_cgroup_stat_desc { |
1509 | const char *msg; | |
1510 | u64 unit; | |
1511 | } mem_cgroup_stat_desc[] = { | |
1512 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
1513 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
1514 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
1515 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
1516 | }; |
1517 | ||
c64745cf PM |
1518 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
1519 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 1520 | { |
d2ceb9b7 KH |
1521 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
1522 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
1523 | int i; | |
1524 | ||
1525 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1526 | s64 val; | |
1527 | ||
1528 | val = mem_cgroup_read_stat(stat, i); | |
1529 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 1530 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 1531 | } |
6d12e2d8 KH |
1532 | /* showing # of active pages */ |
1533 | { | |
4f98a2fe RR |
1534 | unsigned long active_anon, inactive_anon; |
1535 | unsigned long active_file, inactive_file; | |
7b854121 | 1536 | unsigned long unevictable; |
4f98a2fe RR |
1537 | |
1538 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1539 | LRU_INACTIVE_ANON); | |
1540 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1541 | LRU_ACTIVE_ANON); | |
1542 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1543 | LRU_INACTIVE_FILE); | |
1544 | active_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1545 | LRU_ACTIVE_FILE); | |
7b854121 LS |
1546 | unevictable = mem_cgroup_get_all_zonestat(mem_cont, |
1547 | LRU_UNEVICTABLE); | |
1548 | ||
4f98a2fe RR |
1549 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); |
1550 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); | |
1551 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); | |
1552 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); | |
7b854121 LS |
1553 | cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); |
1554 | ||
6d12e2d8 | 1555 | } |
d2ceb9b7 KH |
1556 | return 0; |
1557 | } | |
1558 | ||
c1e862c1 | 1559 | |
8cdea7c0 BS |
1560 | static struct cftype mem_cgroup_files[] = { |
1561 | { | |
0eea1030 | 1562 | .name = "usage_in_bytes", |
8c7c6e34 | 1563 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 1564 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1565 | }, |
c84872e1 PE |
1566 | { |
1567 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 1568 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 1569 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
1570 | .read_u64 = mem_cgroup_read, |
1571 | }, | |
8cdea7c0 | 1572 | { |
0eea1030 | 1573 | .name = "limit_in_bytes", |
8c7c6e34 | 1574 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 1575 | .write_string = mem_cgroup_write, |
2c3daa72 | 1576 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
1577 | }, |
1578 | { | |
1579 | .name = "failcnt", | |
8c7c6e34 | 1580 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 1581 | .trigger = mem_cgroup_reset, |
2c3daa72 | 1582 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1583 | }, |
d2ceb9b7 KH |
1584 | { |
1585 | .name = "stat", | |
c64745cf | 1586 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 1587 | }, |
c1e862c1 KH |
1588 | { |
1589 | .name = "force_empty", | |
1590 | .trigger = mem_cgroup_force_empty_write, | |
1591 | }, | |
8cdea7c0 BS |
1592 | }; |
1593 | ||
8c7c6e34 KH |
1594 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
1595 | static struct cftype memsw_cgroup_files[] = { | |
1596 | { | |
1597 | .name = "memsw.usage_in_bytes", | |
1598 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
1599 | .read_u64 = mem_cgroup_read, | |
1600 | }, | |
1601 | { | |
1602 | .name = "memsw.max_usage_in_bytes", | |
1603 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
1604 | .trigger = mem_cgroup_reset, | |
1605 | .read_u64 = mem_cgroup_read, | |
1606 | }, | |
1607 | { | |
1608 | .name = "memsw.limit_in_bytes", | |
1609 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
1610 | .write_string = mem_cgroup_write, | |
1611 | .read_u64 = mem_cgroup_read, | |
1612 | }, | |
1613 | { | |
1614 | .name = "memsw.failcnt", | |
1615 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
1616 | .trigger = mem_cgroup_reset, | |
1617 | .read_u64 = mem_cgroup_read, | |
1618 | }, | |
1619 | }; | |
1620 | ||
1621 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
1622 | { | |
1623 | if (!do_swap_account) | |
1624 | return 0; | |
1625 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
1626 | ARRAY_SIZE(memsw_cgroup_files)); | |
1627 | }; | |
1628 | #else | |
1629 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
1630 | { | |
1631 | return 0; | |
1632 | } | |
1633 | #endif | |
1634 | ||
6d12e2d8 KH |
1635 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1636 | { | |
1637 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 1638 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1639 | enum lru_list l; |
41e3355d | 1640 | int zone, tmp = node; |
1ecaab2b KH |
1641 | /* |
1642 | * This routine is called against possible nodes. | |
1643 | * But it's BUG to call kmalloc() against offline node. | |
1644 | * | |
1645 | * TODO: this routine can waste much memory for nodes which will | |
1646 | * never be onlined. It's better to use memory hotplug callback | |
1647 | * function. | |
1648 | */ | |
41e3355d KH |
1649 | if (!node_state(node, N_NORMAL_MEMORY)) |
1650 | tmp = -1; | |
1651 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
1652 | if (!pn) |
1653 | return 1; | |
1ecaab2b | 1654 | |
6d12e2d8 KH |
1655 | mem->info.nodeinfo[node] = pn; |
1656 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1657 | |
1658 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1659 | mz = &pn->zoneinfo[zone]; | |
072c56c1 | 1660 | spin_lock_init(&mz->lru_lock); |
b69408e8 CL |
1661 | for_each_lru(l) |
1662 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 1663 | } |
6d12e2d8 KH |
1664 | return 0; |
1665 | } | |
1666 | ||
1ecaab2b KH |
1667 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1668 | { | |
1669 | kfree(mem->info.nodeinfo[node]); | |
1670 | } | |
1671 | ||
c8dad2bb JB |
1672 | static int mem_cgroup_size(void) |
1673 | { | |
1674 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
1675 | return sizeof(struct mem_cgroup) + cpustat_size; | |
1676 | } | |
1677 | ||
33327948 KH |
1678 | static struct mem_cgroup *mem_cgroup_alloc(void) |
1679 | { | |
1680 | struct mem_cgroup *mem; | |
c8dad2bb | 1681 | int size = mem_cgroup_size(); |
33327948 | 1682 | |
c8dad2bb JB |
1683 | if (size < PAGE_SIZE) |
1684 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 1685 | else |
c8dad2bb | 1686 | mem = vmalloc(size); |
33327948 KH |
1687 | |
1688 | if (mem) | |
c8dad2bb | 1689 | memset(mem, 0, size); |
33327948 KH |
1690 | return mem; |
1691 | } | |
1692 | ||
8c7c6e34 KH |
1693 | /* |
1694 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
1695 | * (scanning all at force_empty is too costly...) | |
1696 | * | |
1697 | * Instead of clearing all references at force_empty, we remember | |
1698 | * the number of reference from swap_cgroup and free mem_cgroup when | |
1699 | * it goes down to 0. | |
1700 | * | |
1701 | * When mem_cgroup is destroyed, mem->obsolete will be set to 0 and | |
1702 | * entry which points to this memcg will be ignore at swapin. | |
1703 | * | |
1704 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
1705 | */ | |
1706 | ||
33327948 KH |
1707 | static void mem_cgroup_free(struct mem_cgroup *mem) |
1708 | { | |
8c7c6e34 KH |
1709 | if (atomic_read(&mem->refcnt) > 0) |
1710 | return; | |
c8dad2bb | 1711 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
1712 | kfree(mem); |
1713 | else | |
1714 | vfree(mem); | |
1715 | } | |
1716 | ||
8c7c6e34 KH |
1717 | static void mem_cgroup_get(struct mem_cgroup *mem) |
1718 | { | |
1719 | atomic_inc(&mem->refcnt); | |
1720 | } | |
1721 | ||
1722 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
1723 | { | |
1724 | if (atomic_dec_and_test(&mem->refcnt)) { | |
1725 | if (!mem->obsolete) | |
1726 | return; | |
1727 | mem_cgroup_free(mem); | |
1728 | } | |
1729 | } | |
1730 | ||
33327948 | 1731 | |
c077719b KH |
1732 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
1733 | static void __init enable_swap_cgroup(void) | |
1734 | { | |
1735 | if (!mem_cgroup_subsys.disabled && really_do_swap_account) | |
1736 | do_swap_account = 1; | |
1737 | } | |
1738 | #else | |
1739 | static void __init enable_swap_cgroup(void) | |
1740 | { | |
1741 | } | |
1742 | #endif | |
1743 | ||
8cdea7c0 BS |
1744 | static struct cgroup_subsys_state * |
1745 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1746 | { | |
1747 | struct mem_cgroup *mem; | |
6d12e2d8 | 1748 | int node; |
8cdea7c0 | 1749 | |
c8dad2bb JB |
1750 | mem = mem_cgroup_alloc(); |
1751 | if (!mem) | |
1752 | return ERR_PTR(-ENOMEM); | |
78fb7466 | 1753 | |
8cdea7c0 | 1754 | res_counter_init(&mem->res); |
8c7c6e34 | 1755 | res_counter_init(&mem->memsw); |
1ecaab2b | 1756 | |
6d12e2d8 KH |
1757 | for_each_node_state(node, N_POSSIBLE) |
1758 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1759 | goto free_out; | |
c077719b KH |
1760 | /* root ? */ |
1761 | if (cont->parent == NULL) | |
1762 | enable_swap_cgroup(); | |
6d12e2d8 | 1763 | |
8cdea7c0 | 1764 | return &mem->css; |
6d12e2d8 KH |
1765 | free_out: |
1766 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1767 | free_mem_cgroup_per_zone_info(mem, node); |
c8dad2bb | 1768 | mem_cgroup_free(mem); |
2dda81ca | 1769 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1770 | } |
1771 | ||
df878fb0 KH |
1772 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1773 | struct cgroup *cont) | |
1774 | { | |
1775 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 | 1776 | mem->obsolete = 1; |
c1e862c1 | 1777 | mem_cgroup_force_empty(mem, false); |
df878fb0 KH |
1778 | } |
1779 | ||
8cdea7c0 BS |
1780 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1781 | struct cgroup *cont) | |
1782 | { | |
6d12e2d8 KH |
1783 | int node; |
1784 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1785 | ||
1786 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1787 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1788 | |
33327948 | 1789 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
1790 | } |
1791 | ||
1792 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1793 | struct cgroup *cont) | |
1794 | { | |
8c7c6e34 KH |
1795 | int ret; |
1796 | ||
1797 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
1798 | ARRAY_SIZE(mem_cgroup_files)); | |
1799 | ||
1800 | if (!ret) | |
1801 | ret = register_memsw_files(cont, ss); | |
1802 | return ret; | |
8cdea7c0 BS |
1803 | } |
1804 | ||
67e465a7 BS |
1805 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1806 | struct cgroup *cont, | |
1807 | struct cgroup *old_cont, | |
1808 | struct task_struct *p) | |
1809 | { | |
1810 | struct mm_struct *mm; | |
1811 | struct mem_cgroup *mem, *old_mem; | |
1812 | ||
1813 | mm = get_task_mm(p); | |
1814 | if (mm == NULL) | |
1815 | return; | |
1816 | ||
1817 | mem = mem_cgroup_from_cont(cont); | |
1818 | old_mem = mem_cgroup_from_cont(old_cont); | |
1819 | ||
67e465a7 BS |
1820 | /* |
1821 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1822 | * in effect owned by the leader | |
1823 | */ | |
52ea27eb | 1824 | if (!thread_group_leader(p)) |
67e465a7 BS |
1825 | goto out; |
1826 | ||
67e465a7 BS |
1827 | out: |
1828 | mmput(mm); | |
67e465a7 BS |
1829 | } |
1830 | ||
8cdea7c0 BS |
1831 | struct cgroup_subsys mem_cgroup_subsys = { |
1832 | .name = "memory", | |
1833 | .subsys_id = mem_cgroup_subsys_id, | |
1834 | .create = mem_cgroup_create, | |
df878fb0 | 1835 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1836 | .destroy = mem_cgroup_destroy, |
1837 | .populate = mem_cgroup_populate, | |
67e465a7 | 1838 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1839 | .early_init = 0, |
8cdea7c0 | 1840 | }; |
c077719b KH |
1841 | |
1842 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1843 | ||
1844 | static int __init disable_swap_account(char *s) | |
1845 | { | |
1846 | really_do_swap_account = 0; | |
1847 | return 1; | |
1848 | } | |
1849 | __setup("noswapaccount", disable_swap_account); | |
1850 | #endif |