Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
8cdea7c0 BS |
9 | * This program is free software; you can redistribute it and/or modify |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation; either version 2 of the License, or | |
12 | * (at your option) any later version. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | */ | |
19 | ||
20 | #include <linux/res_counter.h> | |
21 | #include <linux/memcontrol.h> | |
22 | #include <linux/cgroup.h> | |
78fb7466 | 23 | #include <linux/mm.h> |
d13d1443 | 24 | #include <linux/pagemap.h> |
d52aa412 | 25 | #include <linux/smp.h> |
8a9f3ccd | 26 | #include <linux/page-flags.h> |
66e1707b | 27 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
28 | #include <linux/bit_spinlock.h> |
29 | #include <linux/rcupdate.h> | |
8c7c6e34 | 30 | #include <linux/mutex.h> |
b6ac57d5 | 31 | #include <linux/slab.h> |
66e1707b BS |
32 | #include <linux/swap.h> |
33 | #include <linux/spinlock.h> | |
34 | #include <linux/fs.h> | |
d2ceb9b7 | 35 | #include <linux/seq_file.h> |
33327948 | 36 | #include <linux/vmalloc.h> |
b69408e8 | 37 | #include <linux/mm_inline.h> |
52d4b9ac | 38 | #include <linux/page_cgroup.h> |
08e552c6 | 39 | #include "internal.h" |
8cdea7c0 | 40 | |
8697d331 BS |
41 | #include <asm/uaccess.h> |
42 | ||
a181b0e8 | 43 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 44 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 45 | |
c077719b KH |
46 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
47 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */ | |
48 | int do_swap_account __read_mostly; | |
49 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
50 | #else | |
51 | #define do_swap_account (0) | |
52 | #endif | |
53 | ||
7f4d454d | 54 | static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */ |
c077719b | 55 | |
d52aa412 KH |
56 | /* |
57 | * Statistics for memory cgroup. | |
58 | */ | |
59 | enum mem_cgroup_stat_index { | |
60 | /* | |
61 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
62 | */ | |
63 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
64 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
55e462b0 BR |
65 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
66 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
67 | |
68 | MEM_CGROUP_STAT_NSTATS, | |
69 | }; | |
70 | ||
71 | struct mem_cgroup_stat_cpu { | |
72 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
73 | } ____cacheline_aligned_in_smp; | |
74 | ||
75 | struct mem_cgroup_stat { | |
c8dad2bb | 76 | struct mem_cgroup_stat_cpu cpustat[0]; |
d52aa412 KH |
77 | }; |
78 | ||
79 | /* | |
80 | * For accounting under irq disable, no need for increment preempt count. | |
81 | */ | |
addb9efe | 82 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
83 | enum mem_cgroup_stat_index idx, int val) |
84 | { | |
addb9efe | 85 | stat->count[idx] += val; |
d52aa412 KH |
86 | } |
87 | ||
88 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
89 | enum mem_cgroup_stat_index idx) | |
90 | { | |
91 | int cpu; | |
92 | s64 ret = 0; | |
93 | for_each_possible_cpu(cpu) | |
94 | ret += stat->cpustat[cpu].count[idx]; | |
95 | return ret; | |
96 | } | |
97 | ||
6d12e2d8 KH |
98 | /* |
99 | * per-zone information in memory controller. | |
100 | */ | |
6d12e2d8 | 101 | struct mem_cgroup_per_zone { |
072c56c1 KH |
102 | /* |
103 | * spin_lock to protect the per cgroup LRU | |
104 | */ | |
b69408e8 CL |
105 | struct list_head lists[NR_LRU_LISTS]; |
106 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
107 | |
108 | struct zone_reclaim_stat reclaim_stat; | |
6d12e2d8 KH |
109 | }; |
110 | /* Macro for accessing counter */ | |
111 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
112 | ||
113 | struct mem_cgroup_per_node { | |
114 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
115 | }; | |
116 | ||
117 | struct mem_cgroup_lru_info { | |
118 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
119 | }; | |
120 | ||
8cdea7c0 BS |
121 | /* |
122 | * The memory controller data structure. The memory controller controls both | |
123 | * page cache and RSS per cgroup. We would eventually like to provide | |
124 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
125 | * to help the administrator determine what knobs to tune. | |
126 | * | |
127 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
128 | * we hit the water mark. May be even add a low water mark, such that |
129 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
130 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
131 | */ |
132 | struct mem_cgroup { | |
133 | struct cgroup_subsys_state css; | |
134 | /* | |
135 | * the counter to account for memory usage | |
136 | */ | |
137 | struct res_counter res; | |
8c7c6e34 KH |
138 | /* |
139 | * the counter to account for mem+swap usage. | |
140 | */ | |
141 | struct res_counter memsw; | |
78fb7466 PE |
142 | /* |
143 | * Per cgroup active and inactive list, similar to the | |
144 | * per zone LRU lists. | |
78fb7466 | 145 | */ |
6d12e2d8 | 146 | struct mem_cgroup_lru_info info; |
072c56c1 | 147 | |
2733c06a KM |
148 | /* |
149 | protect against reclaim related member. | |
150 | */ | |
151 | spinlock_t reclaim_param_lock; | |
152 | ||
6c48a1d0 | 153 | int prev_priority; /* for recording reclaim priority */ |
6d61ef40 BS |
154 | |
155 | /* | |
156 | * While reclaiming in a hiearchy, we cache the last child we | |
157 | * reclaimed from. Protected by cgroup_lock() | |
158 | */ | |
159 | struct mem_cgroup *last_scanned_child; | |
18f59ea7 BS |
160 | /* |
161 | * Should the accounting and control be hierarchical, per subtree? | |
162 | */ | |
163 | bool use_hierarchy; | |
a636b327 | 164 | unsigned long last_oom_jiffies; |
8c7c6e34 | 165 | atomic_t refcnt; |
14797e23 | 166 | |
a7885eb8 KM |
167 | unsigned int swappiness; |
168 | ||
d52aa412 | 169 | /* |
c8dad2bb | 170 | * statistics. This must be placed at the end of memcg. |
d52aa412 KH |
171 | */ |
172 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
173 | }; |
174 | ||
217bc319 KH |
175 | enum charge_type { |
176 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
177 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 178 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 179 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 180 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
c05555b5 KH |
181 | NR_CHARGE_TYPE, |
182 | }; | |
183 | ||
52d4b9ac KH |
184 | /* only for here (for easy reading.) */ |
185 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
186 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 187 | #define PCGF_LOCK (1UL << PCG_LOCK) |
c05555b5 KH |
188 | static const unsigned long |
189 | pcg_default_flags[NR_CHARGE_TYPE] = { | |
08e552c6 KH |
190 | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */ |
191 | PCGF_USED | PCGF_LOCK, /* Anon */ | |
192 | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ | |
52d4b9ac | 193 | 0, /* FORCE */ |
217bc319 KH |
194 | }; |
195 | ||
8c7c6e34 KH |
196 | /* for encoding cft->private value on file */ |
197 | #define _MEM (0) | |
198 | #define _MEMSWAP (1) | |
199 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
200 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
201 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
202 | ||
203 | static void mem_cgroup_get(struct mem_cgroup *mem); | |
204 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
205 | ||
c05555b5 KH |
206 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
207 | struct page_cgroup *pc, | |
208 | bool charge) | |
d52aa412 KH |
209 | { |
210 | int val = (charge)? 1 : -1; | |
211 | struct mem_cgroup_stat *stat = &mem->stat; | |
addb9efe | 212 | struct mem_cgroup_stat_cpu *cpustat; |
08e552c6 | 213 | int cpu = get_cpu(); |
d52aa412 | 214 | |
08e552c6 | 215 | cpustat = &stat->cpustat[cpu]; |
c05555b5 | 216 | if (PageCgroupCache(pc)) |
addb9efe | 217 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 218 | else |
addb9efe | 219 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
220 | |
221 | if (charge) | |
addb9efe | 222 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
223 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
224 | else | |
addb9efe | 225 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 226 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
08e552c6 | 227 | put_cpu(); |
6d12e2d8 KH |
228 | } |
229 | ||
d5b69e38 | 230 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
231 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
232 | { | |
6d12e2d8 KH |
233 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
234 | } | |
235 | ||
d5b69e38 | 236 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
237 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
238 | { | |
239 | struct mem_cgroup *mem = pc->mem_cgroup; | |
240 | int nid = page_cgroup_nid(pc); | |
241 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 242 | |
54992762 KM |
243 | if (!mem) |
244 | return NULL; | |
245 | ||
6d12e2d8 KH |
246 | return mem_cgroup_zoneinfo(mem, nid, zid); |
247 | } | |
248 | ||
249 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
b69408e8 | 250 | enum lru_list idx) |
6d12e2d8 KH |
251 | { |
252 | int nid, zid; | |
253 | struct mem_cgroup_per_zone *mz; | |
254 | u64 total = 0; | |
255 | ||
256 | for_each_online_node(nid) | |
257 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
258 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
259 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
260 | } | |
261 | return total; | |
d52aa412 KH |
262 | } |
263 | ||
d5b69e38 | 264 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
265 | { |
266 | return container_of(cgroup_subsys_state(cont, | |
267 | mem_cgroup_subsys_id), struct mem_cgroup, | |
268 | css); | |
269 | } | |
270 | ||
cf475ad2 | 271 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 272 | { |
31a78f23 BS |
273 | /* |
274 | * mm_update_next_owner() may clear mm->owner to NULL | |
275 | * if it races with swapoff, page migration, etc. | |
276 | * So this can be called with p == NULL. | |
277 | */ | |
278 | if (unlikely(!p)) | |
279 | return NULL; | |
280 | ||
78fb7466 PE |
281 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
282 | struct mem_cgroup, css); | |
283 | } | |
284 | ||
54595fe2 KH |
285 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
286 | { | |
287 | struct mem_cgroup *mem = NULL; | |
288 | /* | |
289 | * Because we have no locks, mm->owner's may be being moved to other | |
290 | * cgroup. We use css_tryget() here even if this looks | |
291 | * pessimistic (rather than adding locks here). | |
292 | */ | |
293 | rcu_read_lock(); | |
294 | do { | |
295 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
296 | if (unlikely(!mem)) | |
297 | break; | |
298 | } while (!css_tryget(&mem->css)); | |
299 | rcu_read_unlock(); | |
300 | return mem; | |
301 | } | |
302 | ||
303 | static bool mem_cgroup_is_obsolete(struct mem_cgroup *mem) | |
304 | { | |
305 | if (!mem) | |
306 | return true; | |
307 | return css_is_removed(&mem->css); | |
308 | } | |
309 | ||
08e552c6 KH |
310 | /* |
311 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
312 | * Operations are called by routine of global LRU independently from memcg. | |
313 | * What we have to take care of here is validness of pc->mem_cgroup. | |
314 | * | |
315 | * Changes to pc->mem_cgroup happens when | |
316 | * 1. charge | |
317 | * 2. moving account | |
318 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
319 | * It is added to LRU before charge. | |
320 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
321 | * When moving account, the page is not on LRU. It's isolated. | |
322 | */ | |
4f98a2fe | 323 | |
08e552c6 KH |
324 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
325 | { | |
326 | struct page_cgroup *pc; | |
327 | struct mem_cgroup *mem; | |
328 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 329 | |
f8d66542 | 330 | if (mem_cgroup_disabled()) |
08e552c6 KH |
331 | return; |
332 | pc = lookup_page_cgroup(page); | |
333 | /* can happen while we handle swapcache. */ | |
544122e5 | 334 | if (list_empty(&pc->lru) || !pc->mem_cgroup) |
08e552c6 | 335 | return; |
544122e5 KH |
336 | /* |
337 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
338 | * removed from global LRU. | |
339 | */ | |
08e552c6 KH |
340 | mz = page_cgroup_zoneinfo(pc); |
341 | mem = pc->mem_cgroup; | |
b69408e8 | 342 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
08e552c6 KH |
343 | list_del_init(&pc->lru); |
344 | return; | |
6d12e2d8 KH |
345 | } |
346 | ||
08e552c6 | 347 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 348 | { |
08e552c6 KH |
349 | mem_cgroup_del_lru_list(page, page_lru(page)); |
350 | } | |
b69408e8 | 351 | |
08e552c6 KH |
352 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
353 | { | |
354 | struct mem_cgroup_per_zone *mz; | |
355 | struct page_cgroup *pc; | |
b69408e8 | 356 | |
f8d66542 | 357 | if (mem_cgroup_disabled()) |
08e552c6 | 358 | return; |
6d12e2d8 | 359 | |
08e552c6 KH |
360 | pc = lookup_page_cgroup(page); |
361 | smp_rmb(); | |
362 | /* unused page is not rotated. */ | |
363 | if (!PageCgroupUsed(pc)) | |
364 | return; | |
365 | mz = page_cgroup_zoneinfo(pc); | |
366 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
367 | } |
368 | ||
08e552c6 | 369 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 370 | { |
08e552c6 KH |
371 | struct page_cgroup *pc; |
372 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 373 | |
f8d66542 | 374 | if (mem_cgroup_disabled()) |
08e552c6 KH |
375 | return; |
376 | pc = lookup_page_cgroup(page); | |
377 | /* barrier to sync with "charge" */ | |
378 | smp_rmb(); | |
379 | if (!PageCgroupUsed(pc)) | |
894bc310 | 380 | return; |
b69408e8 | 381 | |
08e552c6 | 382 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 383 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
08e552c6 KH |
384 | list_add(&pc->lru, &mz->lists[lru]); |
385 | } | |
544122e5 | 386 | |
08e552c6 | 387 | /* |
544122e5 KH |
388 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
389 | * lru because the page may.be reused after it's fully uncharged (because of | |
390 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
391 | * it again. This function is only used to charge SwapCache. It's done under | |
392 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 393 | */ |
544122e5 | 394 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 395 | { |
544122e5 KH |
396 | unsigned long flags; |
397 | struct zone *zone = page_zone(page); | |
398 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
399 | ||
400 | spin_lock_irqsave(&zone->lru_lock, flags); | |
401 | /* | |
402 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
403 | * is guarded by lock_page() because the page is SwapCache. | |
404 | */ | |
405 | if (!PageCgroupUsed(pc)) | |
406 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
407 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
408 | } |
409 | ||
544122e5 KH |
410 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
411 | { | |
412 | unsigned long flags; | |
413 | struct zone *zone = page_zone(page); | |
414 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
415 | ||
416 | spin_lock_irqsave(&zone->lru_lock, flags); | |
417 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
418 | if (PageLRU(page) && list_empty(&pc->lru)) | |
419 | mem_cgroup_add_lru_list(page, page_lru(page)); | |
420 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
421 | } | |
422 | ||
423 | ||
08e552c6 KH |
424 | void mem_cgroup_move_lists(struct page *page, |
425 | enum lru_list from, enum lru_list to) | |
426 | { | |
f8d66542 | 427 | if (mem_cgroup_disabled()) |
08e552c6 KH |
428 | return; |
429 | mem_cgroup_del_lru_list(page, from); | |
430 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
431 | } |
432 | ||
4c4a2214 DR |
433 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
434 | { | |
435 | int ret; | |
436 | ||
437 | task_lock(task); | |
bd845e38 | 438 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
439 | task_unlock(task); |
440 | return ret; | |
441 | } | |
442 | ||
58ae83db KH |
443 | /* |
444 | * Calculate mapped_ratio under memory controller. This will be used in | |
445 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
446 | */ | |
447 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
448 | { | |
449 | long total, rss; | |
450 | ||
451 | /* | |
452 | * usage is recorded in bytes. But, here, we assume the number of | |
453 | * physical pages can be represented by "long" on any arch. | |
454 | */ | |
455 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
456 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
457 | return (int)((rss * 100L) / total); | |
458 | } | |
8869b8f6 | 459 | |
6c48a1d0 KH |
460 | /* |
461 | * prev_priority control...this will be used in memory reclaim path. | |
462 | */ | |
463 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
464 | { | |
2733c06a KM |
465 | int prev_priority; |
466 | ||
467 | spin_lock(&mem->reclaim_param_lock); | |
468 | prev_priority = mem->prev_priority; | |
469 | spin_unlock(&mem->reclaim_param_lock); | |
470 | ||
471 | return prev_priority; | |
6c48a1d0 KH |
472 | } |
473 | ||
474 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
475 | { | |
2733c06a | 476 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
477 | if (priority < mem->prev_priority) |
478 | mem->prev_priority = priority; | |
2733c06a | 479 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
480 | } |
481 | ||
482 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
483 | { | |
2733c06a | 484 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 | 485 | mem->prev_priority = priority; |
2733c06a | 486 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
487 | } |
488 | ||
c772be93 | 489 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
490 | { |
491 | unsigned long active; | |
492 | unsigned long inactive; | |
c772be93 KM |
493 | unsigned long gb; |
494 | unsigned long inactive_ratio; | |
14797e23 KM |
495 | |
496 | inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON); | |
497 | active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON); | |
498 | ||
c772be93 KM |
499 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
500 | if (gb) | |
501 | inactive_ratio = int_sqrt(10 * gb); | |
502 | else | |
503 | inactive_ratio = 1; | |
504 | ||
505 | if (present_pages) { | |
506 | present_pages[0] = inactive; | |
507 | present_pages[1] = active; | |
508 | } | |
509 | ||
510 | return inactive_ratio; | |
511 | } | |
512 | ||
513 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
514 | { | |
515 | unsigned long active; | |
516 | unsigned long inactive; | |
517 | unsigned long present_pages[2]; | |
518 | unsigned long inactive_ratio; | |
519 | ||
520 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
521 | ||
522 | inactive = present_pages[0]; | |
523 | active = present_pages[1]; | |
524 | ||
525 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
526 | return 1; |
527 | ||
528 | return 0; | |
529 | } | |
530 | ||
a3d8e054 KM |
531 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
532 | struct zone *zone, | |
533 | enum lru_list lru) | |
534 | { | |
535 | int nid = zone->zone_pgdat->node_id; | |
536 | int zid = zone_idx(zone); | |
537 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
538 | ||
539 | return MEM_CGROUP_ZSTAT(mz, lru); | |
540 | } | |
541 | ||
3e2f41f1 KM |
542 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
543 | struct zone *zone) | |
544 | { | |
545 | int nid = zone->zone_pgdat->node_id; | |
546 | int zid = zone_idx(zone); | |
547 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
548 | ||
549 | return &mz->reclaim_stat; | |
550 | } | |
551 | ||
552 | struct zone_reclaim_stat * | |
553 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
554 | { | |
555 | struct page_cgroup *pc; | |
556 | struct mem_cgroup_per_zone *mz; | |
557 | ||
558 | if (mem_cgroup_disabled()) | |
559 | return NULL; | |
560 | ||
561 | pc = lookup_page_cgroup(page); | |
562 | mz = page_cgroup_zoneinfo(pc); | |
563 | if (!mz) | |
564 | return NULL; | |
565 | ||
566 | return &mz->reclaim_stat; | |
567 | } | |
568 | ||
66e1707b BS |
569 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
570 | struct list_head *dst, | |
571 | unsigned long *scanned, int order, | |
572 | int mode, struct zone *z, | |
573 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 574 | int active, int file) |
66e1707b BS |
575 | { |
576 | unsigned long nr_taken = 0; | |
577 | struct page *page; | |
578 | unsigned long scan; | |
579 | LIST_HEAD(pc_list); | |
580 | struct list_head *src; | |
ff7283fa | 581 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
582 | int nid = z->zone_pgdat->node_id; |
583 | int zid = zone_idx(z); | |
584 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 585 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 586 | |
cf475ad2 | 587 | BUG_ON(!mem_cont); |
1ecaab2b | 588 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 589 | src = &mz->lists[lru]; |
66e1707b | 590 | |
ff7283fa KH |
591 | scan = 0; |
592 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 593 | if (scan >= nr_to_scan) |
ff7283fa | 594 | break; |
08e552c6 KH |
595 | |
596 | page = pc->page; | |
52d4b9ac KH |
597 | if (unlikely(!PageCgroupUsed(pc))) |
598 | continue; | |
436c6541 | 599 | if (unlikely(!PageLRU(page))) |
ff7283fa | 600 | continue; |
ff7283fa | 601 | |
436c6541 | 602 | scan++; |
4f98a2fe | 603 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
604 | list_move(&page->lru, dst); |
605 | nr_taken++; | |
606 | } | |
607 | } | |
608 | ||
66e1707b BS |
609 | *scanned = scan; |
610 | return nr_taken; | |
611 | } | |
612 | ||
6d61ef40 BS |
613 | #define mem_cgroup_from_res_counter(counter, member) \ |
614 | container_of(counter, struct mem_cgroup, member) | |
615 | ||
616 | /* | |
617 | * This routine finds the DFS walk successor. This routine should be | |
618 | * called with cgroup_mutex held | |
619 | */ | |
620 | static struct mem_cgroup * | |
621 | mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem) | |
622 | { | |
623 | struct cgroup *cgroup, *curr_cgroup, *root_cgroup; | |
624 | ||
625 | curr_cgroup = curr->css.cgroup; | |
626 | root_cgroup = root_mem->css.cgroup; | |
627 | ||
628 | if (!list_empty(&curr_cgroup->children)) { | |
629 | /* | |
630 | * Walk down to children | |
631 | */ | |
632 | mem_cgroup_put(curr); | |
633 | cgroup = list_entry(curr_cgroup->children.next, | |
634 | struct cgroup, sibling); | |
635 | curr = mem_cgroup_from_cont(cgroup); | |
636 | mem_cgroup_get(curr); | |
637 | goto done; | |
638 | } | |
639 | ||
640 | visit_parent: | |
641 | if (curr_cgroup == root_cgroup) { | |
642 | mem_cgroup_put(curr); | |
643 | curr = root_mem; | |
644 | mem_cgroup_get(curr); | |
645 | goto done; | |
646 | } | |
647 | ||
648 | /* | |
649 | * Goto next sibling | |
650 | */ | |
651 | if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) { | |
652 | mem_cgroup_put(curr); | |
653 | cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup, | |
654 | sibling); | |
655 | curr = mem_cgroup_from_cont(cgroup); | |
656 | mem_cgroup_get(curr); | |
657 | goto done; | |
658 | } | |
659 | ||
660 | /* | |
661 | * Go up to next parent and next parent's sibling if need be | |
662 | */ | |
663 | curr_cgroup = curr_cgroup->parent; | |
664 | goto visit_parent; | |
665 | ||
666 | done: | |
667 | root_mem->last_scanned_child = curr; | |
668 | return curr; | |
669 | } | |
670 | ||
671 | /* | |
672 | * Visit the first child (need not be the first child as per the ordering | |
673 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
674 | * that to reclaim free pages from. | |
675 | */ | |
676 | static struct mem_cgroup * | |
677 | mem_cgroup_get_first_node(struct mem_cgroup *root_mem) | |
678 | { | |
679 | struct cgroup *cgroup; | |
680 | struct mem_cgroup *ret; | |
54595fe2 KH |
681 | bool obsolete; |
682 | ||
683 | obsolete = mem_cgroup_is_obsolete(root_mem->last_scanned_child); | |
6d61ef40 BS |
684 | |
685 | /* | |
686 | * Scan all children under the mem_cgroup mem | |
687 | */ | |
688 | cgroup_lock(); | |
689 | if (list_empty(&root_mem->css.cgroup->children)) { | |
690 | ret = root_mem; | |
691 | goto done; | |
692 | } | |
693 | ||
694 | if (!root_mem->last_scanned_child || obsolete) { | |
695 | ||
54595fe2 | 696 | if (obsolete && root_mem->last_scanned_child) |
6d61ef40 BS |
697 | mem_cgroup_put(root_mem->last_scanned_child); |
698 | ||
699 | cgroup = list_first_entry(&root_mem->css.cgroup->children, | |
700 | struct cgroup, sibling); | |
701 | ret = mem_cgroup_from_cont(cgroup); | |
702 | mem_cgroup_get(ret); | |
703 | } else | |
704 | ret = mem_cgroup_get_next_node(root_mem->last_scanned_child, | |
705 | root_mem); | |
706 | ||
707 | done: | |
708 | root_mem->last_scanned_child = ret; | |
709 | cgroup_unlock(); | |
710 | return ret; | |
711 | } | |
712 | ||
b85a96c0 DN |
713 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
714 | { | |
715 | if (do_swap_account) { | |
716 | if (res_counter_check_under_limit(&mem->res) && | |
717 | res_counter_check_under_limit(&mem->memsw)) | |
718 | return true; | |
719 | } else | |
720 | if (res_counter_check_under_limit(&mem->res)) | |
721 | return true; | |
722 | return false; | |
723 | } | |
724 | ||
a7885eb8 KM |
725 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
726 | { | |
727 | struct cgroup *cgrp = memcg->css.cgroup; | |
728 | unsigned int swappiness; | |
729 | ||
730 | /* root ? */ | |
731 | if (cgrp->parent == NULL) | |
732 | return vm_swappiness; | |
733 | ||
734 | spin_lock(&memcg->reclaim_param_lock); | |
735 | swappiness = memcg->swappiness; | |
736 | spin_unlock(&memcg->reclaim_param_lock); | |
737 | ||
738 | return swappiness; | |
739 | } | |
740 | ||
6d61ef40 BS |
741 | /* |
742 | * Dance down the hierarchy if needed to reclaim memory. We remember the | |
743 | * last child we reclaimed from, so that we don't end up penalizing | |
744 | * one child extensively based on its position in the children list. | |
745 | * | |
746 | * root_mem is the original ancestor that we've been reclaim from. | |
747 | */ | |
748 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
749 | gfp_t gfp_mask, bool noswap) | |
750 | { | |
751 | struct mem_cgroup *next_mem; | |
752 | int ret = 0; | |
753 | ||
754 | /* | |
755 | * Reclaim unconditionally and don't check for return value. | |
756 | * We need to reclaim in the current group and down the tree. | |
757 | * One might think about checking for children before reclaiming, | |
758 | * but there might be left over accounting, even after children | |
759 | * have left. | |
760 | */ | |
a7885eb8 KM |
761 | ret = try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap, |
762 | get_swappiness(root_mem)); | |
b85a96c0 | 763 | if (mem_cgroup_check_under_limit(root_mem)) |
6d61ef40 | 764 | return 0; |
670ec2f1 DN |
765 | if (!root_mem->use_hierarchy) |
766 | return ret; | |
6d61ef40 BS |
767 | |
768 | next_mem = mem_cgroup_get_first_node(root_mem); | |
769 | ||
770 | while (next_mem != root_mem) { | |
54595fe2 | 771 | if (mem_cgroup_is_obsolete(next_mem)) { |
6d61ef40 BS |
772 | mem_cgroup_put(next_mem); |
773 | cgroup_lock(); | |
774 | next_mem = mem_cgroup_get_first_node(root_mem); | |
775 | cgroup_unlock(); | |
776 | continue; | |
777 | } | |
a7885eb8 KM |
778 | ret = try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap, |
779 | get_swappiness(next_mem)); | |
b85a96c0 | 780 | if (mem_cgroup_check_under_limit(root_mem)) |
6d61ef40 BS |
781 | return 0; |
782 | cgroup_lock(); | |
783 | next_mem = mem_cgroup_get_next_node(next_mem, root_mem); | |
784 | cgroup_unlock(); | |
785 | } | |
786 | return ret; | |
787 | } | |
788 | ||
a636b327 KH |
789 | bool mem_cgroup_oom_called(struct task_struct *task) |
790 | { | |
791 | bool ret = false; | |
792 | struct mem_cgroup *mem; | |
793 | struct mm_struct *mm; | |
794 | ||
795 | rcu_read_lock(); | |
796 | mm = task->mm; | |
797 | if (!mm) | |
798 | mm = &init_mm; | |
799 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
800 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
801 | ret = true; | |
802 | rcu_read_unlock(); | |
803 | return ret; | |
804 | } | |
f817ed48 KH |
805 | /* |
806 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
807 | * oom-killer can be invoked. | |
8a9f3ccd | 808 | */ |
f817ed48 | 809 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
8c7c6e34 KH |
810 | gfp_t gfp_mask, struct mem_cgroup **memcg, |
811 | bool oom) | |
8a9f3ccd | 812 | { |
6d61ef40 | 813 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 814 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
28dbc4b6 | 815 | struct res_counter *fail_res; |
a636b327 KH |
816 | |
817 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
818 | /* Don't account this! */ | |
819 | *memcg = NULL; | |
820 | return 0; | |
821 | } | |
822 | ||
8a9f3ccd | 823 | /* |
3be91277 HD |
824 | * We always charge the cgroup the mm_struct belongs to. |
825 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
826 | * thread group leader migrates. It's possible that mm is not |
827 | * set, if so charge the init_mm (happens for pagecache usage). | |
828 | */ | |
54595fe2 KH |
829 | mem = *memcg; |
830 | if (likely(!mem)) { | |
831 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 832 | *memcg = mem; |
e8589cc1 | 833 | } else { |
7a81b88c | 834 | css_get(&mem->css); |
e8589cc1 | 835 | } |
54595fe2 KH |
836 | if (unlikely(!mem)) |
837 | return 0; | |
838 | ||
839 | VM_BUG_ON(mem_cgroup_is_obsolete(mem)); | |
8a9f3ccd | 840 | |
8c7c6e34 KH |
841 | while (1) { |
842 | int ret; | |
843 | bool noswap = false; | |
7a81b88c | 844 | |
28dbc4b6 | 845 | ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res); |
8c7c6e34 KH |
846 | if (likely(!ret)) { |
847 | if (!do_swap_account) | |
848 | break; | |
28dbc4b6 BS |
849 | ret = res_counter_charge(&mem->memsw, PAGE_SIZE, |
850 | &fail_res); | |
8c7c6e34 KH |
851 | if (likely(!ret)) |
852 | break; | |
853 | /* mem+swap counter fails */ | |
854 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
855 | noswap = true; | |
6d61ef40 BS |
856 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
857 | memsw); | |
858 | } else | |
859 | /* mem counter fails */ | |
860 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
861 | res); | |
862 | ||
3be91277 | 863 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 864 | goto nomem; |
e1a1cd59 | 865 | |
6d61ef40 BS |
866 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask, |
867 | noswap); | |
66e1707b BS |
868 | |
869 | /* | |
8869b8f6 HD |
870 | * try_to_free_mem_cgroup_pages() might not give us a full |
871 | * picture of reclaim. Some pages are reclaimed and might be | |
872 | * moved to swap cache or just unmapped from the cgroup. | |
873 | * Check the limit again to see if the reclaim reduced the | |
874 | * current usage of the cgroup before giving up | |
8c7c6e34 | 875 | * |
8869b8f6 | 876 | */ |
b85a96c0 DN |
877 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
878 | continue; | |
3be91277 HD |
879 | |
880 | if (!nr_retries--) { | |
a636b327 | 881 | if (oom) { |
7f4d454d | 882 | mutex_lock(&memcg_tasklist); |
88700756 | 883 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); |
7f4d454d | 884 | mutex_unlock(&memcg_tasklist); |
88700756 | 885 | mem_over_limit->last_oom_jiffies = jiffies; |
a636b327 | 886 | } |
7a81b88c | 887 | goto nomem; |
66e1707b | 888 | } |
8a9f3ccd | 889 | } |
7a81b88c KH |
890 | return 0; |
891 | nomem: | |
892 | css_put(&mem->css); | |
893 | return -ENOMEM; | |
894 | } | |
8a9f3ccd | 895 | |
b5a84319 KH |
896 | static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page) |
897 | { | |
898 | struct mem_cgroup *mem; | |
899 | swp_entry_t ent; | |
900 | ||
901 | if (!PageSwapCache(page)) | |
902 | return NULL; | |
903 | ||
904 | ent.val = page_private(page); | |
905 | mem = lookup_swap_cgroup(ent); | |
906 | if (!mem) | |
907 | return NULL; | |
908 | if (!css_tryget(&mem->css)) | |
909 | return NULL; | |
910 | return mem; | |
911 | } | |
912 | ||
7a81b88c | 913 | /* |
a5e924f5 | 914 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
915 | * USED state. If already USED, uncharge and return. |
916 | */ | |
917 | ||
918 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
919 | struct page_cgroup *pc, | |
920 | enum charge_type ctype) | |
921 | { | |
7a81b88c KH |
922 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
923 | if (!mem) | |
924 | return; | |
52d4b9ac KH |
925 | |
926 | lock_page_cgroup(pc); | |
927 | if (unlikely(PageCgroupUsed(pc))) { | |
928 | unlock_page_cgroup(pc); | |
929 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
930 | if (do_swap_account) |
931 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
52d4b9ac | 932 | css_put(&mem->css); |
7a81b88c | 933 | return; |
52d4b9ac | 934 | } |
8a9f3ccd | 935 | pc->mem_cgroup = mem; |
08e552c6 | 936 | smp_wmb(); |
c05555b5 | 937 | pc->flags = pcg_default_flags[ctype]; |
3be91277 | 938 | |
08e552c6 | 939 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 940 | |
52d4b9ac | 941 | unlock_page_cgroup(pc); |
7a81b88c | 942 | } |
66e1707b | 943 | |
f817ed48 KH |
944 | /** |
945 | * mem_cgroup_move_account - move account of the page | |
946 | * @pc: page_cgroup of the page. | |
947 | * @from: mem_cgroup which the page is moved from. | |
948 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
949 | * | |
950 | * The caller must confirm following. | |
08e552c6 | 951 | * - page is not on LRU (isolate_page() is useful.) |
f817ed48 KH |
952 | * |
953 | * returns 0 at success, | |
954 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
955 | * | |
956 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
957 | * new cgroup. It should be done by a caller. | |
958 | */ | |
959 | ||
960 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
961 | struct mem_cgroup *from, struct mem_cgroup *to) | |
962 | { | |
963 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
964 | int nid, zid; | |
965 | int ret = -EBUSY; | |
966 | ||
f817ed48 | 967 | VM_BUG_ON(from == to); |
08e552c6 | 968 | VM_BUG_ON(PageLRU(pc->page)); |
f817ed48 KH |
969 | |
970 | nid = page_cgroup_nid(pc); | |
971 | zid = page_cgroup_zid(pc); | |
972 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
973 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
974 | ||
f817ed48 KH |
975 | if (!trylock_page_cgroup(pc)) |
976 | return ret; | |
977 | ||
978 | if (!PageCgroupUsed(pc)) | |
979 | goto out; | |
980 | ||
981 | if (pc->mem_cgroup != from) | |
982 | goto out; | |
983 | ||
08e552c6 KH |
984 | css_put(&from->css); |
985 | res_counter_uncharge(&from->res, PAGE_SIZE); | |
986 | mem_cgroup_charge_statistics(from, pc, false); | |
987 | if (do_swap_account) | |
988 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
989 | pc->mem_cgroup = to; | |
990 | mem_cgroup_charge_statistics(to, pc, true); | |
991 | css_get(&to->css); | |
992 | ret = 0; | |
f817ed48 KH |
993 | out: |
994 | unlock_page_cgroup(pc); | |
995 | return ret; | |
996 | } | |
997 | ||
998 | /* | |
999 | * move charges to its parent. | |
1000 | */ | |
1001 | ||
1002 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1003 | struct mem_cgroup *child, | |
1004 | gfp_t gfp_mask) | |
1005 | { | |
08e552c6 | 1006 | struct page *page = pc->page; |
f817ed48 KH |
1007 | struct cgroup *cg = child->css.cgroup; |
1008 | struct cgroup *pcg = cg->parent; | |
1009 | struct mem_cgroup *parent; | |
f817ed48 KH |
1010 | int ret; |
1011 | ||
1012 | /* Is ROOT ? */ | |
1013 | if (!pcg) | |
1014 | return -EINVAL; | |
1015 | ||
08e552c6 | 1016 | |
f817ed48 KH |
1017 | parent = mem_cgroup_from_cont(pcg); |
1018 | ||
08e552c6 | 1019 | |
f817ed48 | 1020 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); |
a636b327 | 1021 | if (ret || !parent) |
f817ed48 KH |
1022 | return ret; |
1023 | ||
08e552c6 KH |
1024 | if (!get_page_unless_zero(page)) |
1025 | return -EBUSY; | |
1026 | ||
1027 | ret = isolate_lru_page(page); | |
1028 | ||
1029 | if (ret) | |
1030 | goto cancel; | |
f817ed48 | 1031 | |
f817ed48 | 1032 | ret = mem_cgroup_move_account(pc, child, parent); |
f817ed48 | 1033 | |
08e552c6 | 1034 | /* drop extra refcnt by try_charge() (move_account increment one) */ |
f817ed48 | 1035 | css_put(&parent->css); |
08e552c6 KH |
1036 | putback_lru_page(page); |
1037 | if (!ret) { | |
1038 | put_page(page); | |
1039 | return 0; | |
8c7c6e34 | 1040 | } |
08e552c6 KH |
1041 | /* uncharge if move fails */ |
1042 | cancel: | |
1043 | res_counter_uncharge(&parent->res, PAGE_SIZE); | |
1044 | if (do_swap_account) | |
1045 | res_counter_uncharge(&parent->memsw, PAGE_SIZE); | |
1046 | put_page(page); | |
f817ed48 KH |
1047 | return ret; |
1048 | } | |
1049 | ||
7a81b88c KH |
1050 | /* |
1051 | * Charge the memory controller for page usage. | |
1052 | * Return | |
1053 | * 0 if the charge was successful | |
1054 | * < 0 if the cgroup is over its limit | |
1055 | */ | |
1056 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1057 | gfp_t gfp_mask, enum charge_type ctype, | |
1058 | struct mem_cgroup *memcg) | |
1059 | { | |
1060 | struct mem_cgroup *mem; | |
1061 | struct page_cgroup *pc; | |
1062 | int ret; | |
1063 | ||
1064 | pc = lookup_page_cgroup(page); | |
1065 | /* can happen at boot */ | |
1066 | if (unlikely(!pc)) | |
1067 | return 0; | |
1068 | prefetchw(pc); | |
1069 | ||
1070 | mem = memcg; | |
f817ed48 | 1071 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
a636b327 | 1072 | if (ret || !mem) |
7a81b88c KH |
1073 | return ret; |
1074 | ||
1075 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1076 | return 0; |
8a9f3ccd BS |
1077 | } |
1078 | ||
7a81b88c KH |
1079 | int mem_cgroup_newpage_charge(struct page *page, |
1080 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1081 | { |
f8d66542 | 1082 | if (mem_cgroup_disabled()) |
cede86ac | 1083 | return 0; |
52d4b9ac KH |
1084 | if (PageCompound(page)) |
1085 | return 0; | |
69029cd5 KH |
1086 | /* |
1087 | * If already mapped, we don't have to account. | |
1088 | * If page cache, page->mapping has address_space. | |
1089 | * But page->mapping may have out-of-use anon_vma pointer, | |
1090 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1091 | * is NULL. | |
1092 | */ | |
1093 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
1094 | return 0; | |
1095 | if (unlikely(!mm)) | |
1096 | mm = &init_mm; | |
217bc319 | 1097 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 1098 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
1099 | } |
1100 | ||
e1a1cd59 BS |
1101 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
1102 | gfp_t gfp_mask) | |
8697d331 | 1103 | { |
b5a84319 KH |
1104 | struct mem_cgroup *mem = NULL; |
1105 | int ret; | |
1106 | ||
f8d66542 | 1107 | if (mem_cgroup_disabled()) |
cede86ac | 1108 | return 0; |
52d4b9ac KH |
1109 | if (PageCompound(page)) |
1110 | return 0; | |
accf163e KH |
1111 | /* |
1112 | * Corner case handling. This is called from add_to_page_cache() | |
1113 | * in usual. But some FS (shmem) precharges this page before calling it | |
1114 | * and call add_to_page_cache() with GFP_NOWAIT. | |
1115 | * | |
1116 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
1117 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
1118 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
1119 | * And when the page is SwapCache, it should take swap information |
1120 | * into account. This is under lock_page() now. | |
accf163e KH |
1121 | */ |
1122 | if (!(gfp_mask & __GFP_WAIT)) { | |
1123 | struct page_cgroup *pc; | |
1124 | ||
52d4b9ac KH |
1125 | |
1126 | pc = lookup_page_cgroup(page); | |
1127 | if (!pc) | |
1128 | return 0; | |
1129 | lock_page_cgroup(pc); | |
1130 | if (PageCgroupUsed(pc)) { | |
1131 | unlock_page_cgroup(pc); | |
accf163e KH |
1132 | return 0; |
1133 | } | |
52d4b9ac | 1134 | unlock_page_cgroup(pc); |
accf163e KH |
1135 | } |
1136 | ||
b5a84319 KH |
1137 | if (do_swap_account && PageSwapCache(page)) { |
1138 | mem = try_get_mem_cgroup_from_swapcache(page); | |
1139 | if (mem) | |
1140 | mm = NULL; | |
1141 | else | |
1142 | mem = NULL; | |
1143 | /* SwapCache may be still linked to LRU now. */ | |
1144 | mem_cgroup_lru_del_before_commit_swapcache(page); | |
1145 | } | |
1146 | ||
1147 | if (unlikely(!mm && !mem)) | |
8697d331 | 1148 | mm = &init_mm; |
accf163e | 1149 | |
c05555b5 KH |
1150 | if (page_is_file_cache(page)) |
1151 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 1152 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 KH |
1153 | |
1154 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
1155 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
1156 | if (mem) | |
1157 | css_put(&mem->css); | |
1158 | if (PageSwapCache(page)) | |
1159 | mem_cgroup_lru_add_after_commit_swapcache(page); | |
1160 | ||
1161 | if (do_swap_account && !ret && PageSwapCache(page)) { | |
1162 | swp_entry_t ent = {.val = page_private(page)}; | |
1163 | /* avoid double counting */ | |
1164 | mem = swap_cgroup_record(ent, NULL); | |
1165 | if (mem) { | |
1166 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1167 | mem_cgroup_put(mem); | |
1168 | } | |
1169 | } | |
1170 | return ret; | |
e8589cc1 KH |
1171 | } |
1172 | ||
54595fe2 KH |
1173 | /* |
1174 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
1175 | * And when try_charge() successfully returns, one refcnt to memcg without | |
1176 | * struct page_cgroup is aquired. This refcnt will be cumsumed by | |
1177 | * "commit()" or removed by "cancel()" | |
1178 | */ | |
8c7c6e34 KH |
1179 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
1180 | struct page *page, | |
1181 | gfp_t mask, struct mem_cgroup **ptr) | |
1182 | { | |
1183 | struct mem_cgroup *mem; | |
54595fe2 | 1184 | int ret; |
8c7c6e34 | 1185 | |
f8d66542 | 1186 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
1187 | return 0; |
1188 | ||
1189 | if (!do_swap_account) | |
1190 | goto charge_cur_mm; | |
8c7c6e34 KH |
1191 | /* |
1192 | * A racing thread's fault, or swapoff, may have already updated | |
1193 | * the pte, and even removed page from swap cache: return success | |
1194 | * to go on to do_swap_page()'s pte_same() test, which should fail. | |
1195 | */ | |
1196 | if (!PageSwapCache(page)) | |
1197 | return 0; | |
b5a84319 | 1198 | mem = try_get_mem_cgroup_from_swapcache(page); |
54595fe2 KH |
1199 | if (!mem) |
1200 | goto charge_cur_mm; | |
8c7c6e34 | 1201 | *ptr = mem; |
54595fe2 KH |
1202 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true); |
1203 | /* drop extra refcnt from tryget */ | |
1204 | css_put(&mem->css); | |
1205 | return ret; | |
8c7c6e34 KH |
1206 | charge_cur_mm: |
1207 | if (unlikely(!mm)) | |
1208 | mm = &init_mm; | |
1209 | return __mem_cgroup_try_charge(mm, mask, ptr, true); | |
1210 | } | |
1211 | ||
7a81b88c KH |
1212 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
1213 | { | |
1214 | struct page_cgroup *pc; | |
1215 | ||
f8d66542 | 1216 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1217 | return; |
1218 | if (!ptr) | |
1219 | return; | |
1220 | pc = lookup_page_cgroup(page); | |
544122e5 | 1221 | mem_cgroup_lru_del_before_commit_swapcache(page); |
7a81b88c | 1222 | __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
544122e5 | 1223 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
1224 | /* |
1225 | * Now swap is on-memory. This means this page may be | |
1226 | * counted both as mem and swap....double count. | |
03f3c433 KH |
1227 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
1228 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
1229 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 1230 | */ |
03f3c433 | 1231 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 KH |
1232 | swp_entry_t ent = {.val = page_private(page)}; |
1233 | struct mem_cgroup *memcg; | |
1234 | memcg = swap_cgroup_record(ent, NULL); | |
1235 | if (memcg) { | |
8c7c6e34 KH |
1236 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
1237 | mem_cgroup_put(memcg); | |
1238 | } | |
1239 | ||
1240 | } | |
08e552c6 | 1241 | /* add this page(page_cgroup) to the LRU we want. */ |
544122e5 | 1242 | |
7a81b88c KH |
1243 | } |
1244 | ||
1245 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
1246 | { | |
f8d66542 | 1247 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1248 | return; |
1249 | if (!mem) | |
1250 | return; | |
1251 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
8c7c6e34 KH |
1252 | if (do_swap_account) |
1253 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
7a81b88c KH |
1254 | css_put(&mem->css); |
1255 | } | |
1256 | ||
1257 | ||
8a9f3ccd | 1258 | /* |
69029cd5 | 1259 | * uncharge if !page_mapped(page) |
8a9f3ccd | 1260 | */ |
8c7c6e34 | 1261 | static struct mem_cgroup * |
69029cd5 | 1262 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 1263 | { |
8289546e | 1264 | struct page_cgroup *pc; |
8c7c6e34 | 1265 | struct mem_cgroup *mem = NULL; |
072c56c1 | 1266 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 1267 | |
f8d66542 | 1268 | if (mem_cgroup_disabled()) |
8c7c6e34 | 1269 | return NULL; |
4077960e | 1270 | |
d13d1443 | 1271 | if (PageSwapCache(page)) |
8c7c6e34 | 1272 | return NULL; |
d13d1443 | 1273 | |
8697d331 | 1274 | /* |
3c541e14 | 1275 | * Check if our page_cgroup is valid |
8697d331 | 1276 | */ |
52d4b9ac KH |
1277 | pc = lookup_page_cgroup(page); |
1278 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 1279 | return NULL; |
b9c565d5 | 1280 | |
52d4b9ac | 1281 | lock_page_cgroup(pc); |
d13d1443 | 1282 | |
8c7c6e34 KH |
1283 | mem = pc->mem_cgroup; |
1284 | ||
d13d1443 KH |
1285 | if (!PageCgroupUsed(pc)) |
1286 | goto unlock_out; | |
1287 | ||
1288 | switch (ctype) { | |
1289 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1290 | if (page_mapped(page)) | |
1291 | goto unlock_out; | |
1292 | break; | |
1293 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
1294 | if (!PageAnon(page)) { /* Shared memory */ | |
1295 | if (page->mapping && !page_is_file_cache(page)) | |
1296 | goto unlock_out; | |
1297 | } else if (page_mapped(page)) /* Anon */ | |
1298 | goto unlock_out; | |
1299 | break; | |
1300 | default: | |
1301 | break; | |
52d4b9ac | 1302 | } |
d13d1443 | 1303 | |
8c7c6e34 KH |
1304 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1305 | if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)) | |
1306 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1307 | ||
08e552c6 | 1308 | mem_cgroup_charge_statistics(mem, pc, false); |
52d4b9ac | 1309 | ClearPageCgroupUsed(pc); |
544122e5 KH |
1310 | /* |
1311 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
1312 | * freed from LRU. This is safe because uncharged page is expected not | |
1313 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
1314 | * special functions. | |
1315 | */ | |
b9c565d5 | 1316 | |
69029cd5 | 1317 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 1318 | unlock_page_cgroup(pc); |
fb59e9f1 | 1319 | |
a7fe942e KH |
1320 | /* at swapout, this memcg will be accessed to record to swap */ |
1321 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
1322 | css_put(&mem->css); | |
6d12e2d8 | 1323 | |
8c7c6e34 | 1324 | return mem; |
d13d1443 KH |
1325 | |
1326 | unlock_out: | |
1327 | unlock_page_cgroup(pc); | |
8c7c6e34 | 1328 | return NULL; |
3c541e14 BS |
1329 | } |
1330 | ||
69029cd5 KH |
1331 | void mem_cgroup_uncharge_page(struct page *page) |
1332 | { | |
52d4b9ac KH |
1333 | /* early check. */ |
1334 | if (page_mapped(page)) | |
1335 | return; | |
1336 | if (page->mapping && !PageAnon(page)) | |
1337 | return; | |
69029cd5 KH |
1338 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
1339 | } | |
1340 | ||
1341 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
1342 | { | |
1343 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 1344 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
1345 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
1346 | } | |
1347 | ||
8c7c6e34 KH |
1348 | /* |
1349 | * called from __delete_from_swap_cache() and drop "page" account. | |
1350 | * memcg information is recorded to swap_cgroup of "ent" | |
1351 | */ | |
1352 | void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent) | |
1353 | { | |
1354 | struct mem_cgroup *memcg; | |
1355 | ||
1356 | memcg = __mem_cgroup_uncharge_common(page, | |
1357 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT); | |
1358 | /* record memcg information */ | |
1359 | if (do_swap_account && memcg) { | |
1360 | swap_cgroup_record(ent, memcg); | |
1361 | mem_cgroup_get(memcg); | |
1362 | } | |
a7fe942e KH |
1363 | if (memcg) |
1364 | css_put(&memcg->css); | |
8c7c6e34 KH |
1365 | } |
1366 | ||
1367 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1368 | /* | |
1369 | * called from swap_entry_free(). remove record in swap_cgroup and | |
1370 | * uncharge "memsw" account. | |
1371 | */ | |
1372 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 1373 | { |
8c7c6e34 KH |
1374 | struct mem_cgroup *memcg; |
1375 | ||
1376 | if (!do_swap_account) | |
1377 | return; | |
1378 | ||
1379 | memcg = swap_cgroup_record(ent, NULL); | |
1380 | if (memcg) { | |
1381 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
1382 | mem_cgroup_put(memcg); | |
1383 | } | |
d13d1443 | 1384 | } |
8c7c6e34 | 1385 | #endif |
d13d1443 | 1386 | |
ae41be37 | 1387 | /* |
01b1ae63 KH |
1388 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
1389 | * page belongs to. | |
ae41be37 | 1390 | */ |
01b1ae63 | 1391 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
1392 | { |
1393 | struct page_cgroup *pc; | |
e8589cc1 | 1394 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 1395 | int ret = 0; |
8869b8f6 | 1396 | |
f8d66542 | 1397 | if (mem_cgroup_disabled()) |
4077960e BS |
1398 | return 0; |
1399 | ||
52d4b9ac KH |
1400 | pc = lookup_page_cgroup(page); |
1401 | lock_page_cgroup(pc); | |
1402 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
1403 | mem = pc->mem_cgroup; |
1404 | css_get(&mem->css); | |
e8589cc1 | 1405 | } |
52d4b9ac | 1406 | unlock_page_cgroup(pc); |
01b1ae63 | 1407 | |
e8589cc1 | 1408 | if (mem) { |
3bb4edf2 | 1409 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); |
e8589cc1 KH |
1410 | css_put(&mem->css); |
1411 | } | |
01b1ae63 | 1412 | *ptr = mem; |
e8589cc1 | 1413 | return ret; |
ae41be37 | 1414 | } |
8869b8f6 | 1415 | |
69029cd5 | 1416 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
1417 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
1418 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 1419 | { |
01b1ae63 KH |
1420 | struct page *target, *unused; |
1421 | struct page_cgroup *pc; | |
1422 | enum charge_type ctype; | |
1423 | ||
1424 | if (!mem) | |
1425 | return; | |
1426 | ||
1427 | /* at migration success, oldpage->mapping is NULL. */ | |
1428 | if (oldpage->mapping) { | |
1429 | target = oldpage; | |
1430 | unused = NULL; | |
1431 | } else { | |
1432 | target = newpage; | |
1433 | unused = oldpage; | |
1434 | } | |
1435 | ||
1436 | if (PageAnon(target)) | |
1437 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
1438 | else if (page_is_file_cache(target)) | |
1439 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
1440 | else | |
1441 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
1442 | ||
1443 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 1444 | if (unused) |
01b1ae63 KH |
1445 | __mem_cgroup_uncharge_common(unused, ctype); |
1446 | ||
1447 | pc = lookup_page_cgroup(target); | |
69029cd5 | 1448 | /* |
01b1ae63 KH |
1449 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
1450 | * So, double-counting is effectively avoided. | |
1451 | */ | |
1452 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
1453 | ||
1454 | /* | |
1455 | * Both of oldpage and newpage are still under lock_page(). | |
1456 | * Then, we don't have to care about race in radix-tree. | |
1457 | * But we have to be careful that this page is unmapped or not. | |
1458 | * | |
1459 | * There is a case for !page_mapped(). At the start of | |
1460 | * migration, oldpage was mapped. But now, it's zapped. | |
1461 | * But we know *target* page is not freed/reused under us. | |
1462 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 1463 | */ |
01b1ae63 KH |
1464 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
1465 | mem_cgroup_uncharge_page(target); | |
ae41be37 | 1466 | } |
78fb7466 | 1467 | |
c9b0ed51 KH |
1468 | /* |
1469 | * A call to try to shrink memory usage under specified resource controller. | |
1470 | * This is typically used for page reclaiming for shmem for reducing side | |
1471 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
1472 | */ | |
b5a84319 KH |
1473 | int mem_cgroup_shrink_usage(struct page *page, |
1474 | struct mm_struct *mm, | |
1475 | gfp_t gfp_mask) | |
c9b0ed51 | 1476 | { |
b5a84319 | 1477 | struct mem_cgroup *mem = NULL; |
c9b0ed51 KH |
1478 | int progress = 0; |
1479 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
1480 | ||
f8d66542 | 1481 | if (mem_cgroup_disabled()) |
cede86ac | 1482 | return 0; |
b5a84319 KH |
1483 | if (page) |
1484 | mem = try_get_mem_cgroup_from_swapcache(page); | |
1485 | if (!mem && mm) | |
1486 | mem = try_get_mem_cgroup_from_mm(mm); | |
54595fe2 | 1487 | if (unlikely(!mem)) |
31a78f23 | 1488 | return 0; |
c9b0ed51 KH |
1489 | |
1490 | do { | |
42e9abb6 | 1491 | progress = mem_cgroup_hierarchical_reclaim(mem, gfp_mask, true); |
b85a96c0 | 1492 | progress += mem_cgroup_check_under_limit(mem); |
c9b0ed51 KH |
1493 | } while (!progress && --retry); |
1494 | ||
1495 | css_put(&mem->css); | |
1496 | if (!retry) | |
1497 | return -ENOMEM; | |
1498 | return 0; | |
1499 | } | |
1500 | ||
8c7c6e34 KH |
1501 | static DEFINE_MUTEX(set_limit_mutex); |
1502 | ||
d38d2a75 | 1503 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 1504 | unsigned long long val) |
628f4235 KH |
1505 | { |
1506 | ||
1507 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
1508 | int progress; | |
8c7c6e34 | 1509 | u64 memswlimit; |
628f4235 KH |
1510 | int ret = 0; |
1511 | ||
8c7c6e34 | 1512 | while (retry_count) { |
628f4235 KH |
1513 | if (signal_pending(current)) { |
1514 | ret = -EINTR; | |
1515 | break; | |
1516 | } | |
8c7c6e34 KH |
1517 | /* |
1518 | * Rather than hide all in some function, I do this in | |
1519 | * open coded manner. You see what this really does. | |
1520 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1521 | */ | |
1522 | mutex_lock(&set_limit_mutex); | |
1523 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1524 | if (memswlimit < val) { | |
1525 | ret = -EINVAL; | |
1526 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
1527 | break; |
1528 | } | |
8c7c6e34 KH |
1529 | ret = res_counter_set_limit(&memcg->res, val); |
1530 | mutex_unlock(&set_limit_mutex); | |
1531 | ||
1532 | if (!ret) | |
1533 | break; | |
1534 | ||
42e9abb6 DN |
1535 | progress = mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, |
1536 | false); | |
8c7c6e34 KH |
1537 | if (!progress) retry_count--; |
1538 | } | |
14797e23 | 1539 | |
8c7c6e34 KH |
1540 | return ret; |
1541 | } | |
1542 | ||
1543 | int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, | |
1544 | unsigned long long val) | |
1545 | { | |
1546 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
1547 | u64 memlimit, oldusage, curusage; | |
1548 | int ret; | |
1549 | ||
1550 | if (!do_swap_account) | |
1551 | return -EINVAL; | |
1552 | ||
1553 | while (retry_count) { | |
1554 | if (signal_pending(current)) { | |
1555 | ret = -EINTR; | |
1556 | break; | |
1557 | } | |
1558 | /* | |
1559 | * Rather than hide all in some function, I do this in | |
1560 | * open coded manner. You see what this really does. | |
1561 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1562 | */ | |
1563 | mutex_lock(&set_limit_mutex); | |
1564 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1565 | if (memlimit > val) { | |
1566 | ret = -EINVAL; | |
1567 | mutex_unlock(&set_limit_mutex); | |
1568 | break; | |
1569 | } | |
1570 | ret = res_counter_set_limit(&memcg->memsw, val); | |
1571 | mutex_unlock(&set_limit_mutex); | |
1572 | ||
1573 | if (!ret) | |
1574 | break; | |
1575 | ||
1576 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
42e9abb6 | 1577 | mem_cgroup_hierarchical_reclaim(memcg, GFP_KERNEL, true); |
8c7c6e34 KH |
1578 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
1579 | if (curusage >= oldusage) | |
628f4235 KH |
1580 | retry_count--; |
1581 | } | |
1582 | return ret; | |
1583 | } | |
1584 | ||
cc847582 KH |
1585 | /* |
1586 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
1587 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
1588 | */ | |
f817ed48 | 1589 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 1590 | int node, int zid, enum lru_list lru) |
cc847582 | 1591 | { |
08e552c6 KH |
1592 | struct zone *zone; |
1593 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 1594 | struct page_cgroup *pc, *busy; |
08e552c6 | 1595 | unsigned long flags, loop; |
072c56c1 | 1596 | struct list_head *list; |
f817ed48 | 1597 | int ret = 0; |
072c56c1 | 1598 | |
08e552c6 KH |
1599 | zone = &NODE_DATA(node)->node_zones[zid]; |
1600 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 1601 | list = &mz->lists[lru]; |
cc847582 | 1602 | |
f817ed48 KH |
1603 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
1604 | /* give some margin against EBUSY etc...*/ | |
1605 | loop += 256; | |
1606 | busy = NULL; | |
1607 | while (loop--) { | |
1608 | ret = 0; | |
08e552c6 | 1609 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 1610 | if (list_empty(list)) { |
08e552c6 | 1611 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 1612 | break; |
f817ed48 KH |
1613 | } |
1614 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
1615 | if (busy == pc) { | |
1616 | list_move(&pc->lru, list); | |
1617 | busy = 0; | |
08e552c6 | 1618 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
1619 | continue; |
1620 | } | |
08e552c6 | 1621 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 1622 | |
2c26fdd7 | 1623 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 1624 | if (ret == -ENOMEM) |
52d4b9ac | 1625 | break; |
f817ed48 KH |
1626 | |
1627 | if (ret == -EBUSY || ret == -EINVAL) { | |
1628 | /* found lock contention or "pc" is obsolete. */ | |
1629 | busy = pc; | |
1630 | cond_resched(); | |
1631 | } else | |
1632 | busy = NULL; | |
cc847582 | 1633 | } |
08e552c6 | 1634 | |
f817ed48 KH |
1635 | if (!ret && !list_empty(list)) |
1636 | return -EBUSY; | |
1637 | return ret; | |
cc847582 KH |
1638 | } |
1639 | ||
1640 | /* | |
1641 | * make mem_cgroup's charge to be 0 if there is no task. | |
1642 | * This enables deleting this mem_cgroup. | |
1643 | */ | |
c1e862c1 | 1644 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 1645 | { |
f817ed48 KH |
1646 | int ret; |
1647 | int node, zid, shrink; | |
1648 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 1649 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 1650 | |
cc847582 | 1651 | css_get(&mem->css); |
f817ed48 KH |
1652 | |
1653 | shrink = 0; | |
c1e862c1 KH |
1654 | /* should free all ? */ |
1655 | if (free_all) | |
1656 | goto try_to_free; | |
f817ed48 | 1657 | move_account: |
1ecaab2b | 1658 | while (mem->res.usage > 0) { |
f817ed48 | 1659 | ret = -EBUSY; |
c1e862c1 KH |
1660 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
1661 | goto out; | |
1662 | ret = -EINTR; | |
1663 | if (signal_pending(current)) | |
cc847582 | 1664 | goto out; |
52d4b9ac KH |
1665 | /* This is for making all *used* pages to be on LRU. */ |
1666 | lru_add_drain_all(); | |
f817ed48 KH |
1667 | ret = 0; |
1668 | for_each_node_state(node, N_POSSIBLE) { | |
1669 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { | |
b69408e8 | 1670 | enum lru_list l; |
f817ed48 KH |
1671 | for_each_lru(l) { |
1672 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 1673 | node, zid, l); |
f817ed48 KH |
1674 | if (ret) |
1675 | break; | |
1676 | } | |
1ecaab2b | 1677 | } |
f817ed48 KH |
1678 | if (ret) |
1679 | break; | |
1680 | } | |
1681 | /* it seems parent cgroup doesn't have enough mem */ | |
1682 | if (ret == -ENOMEM) | |
1683 | goto try_to_free; | |
52d4b9ac | 1684 | cond_resched(); |
cc847582 KH |
1685 | } |
1686 | ret = 0; | |
1687 | out: | |
1688 | css_put(&mem->css); | |
1689 | return ret; | |
f817ed48 KH |
1690 | |
1691 | try_to_free: | |
c1e862c1 KH |
1692 | /* returns EBUSY if there is a task or if we come here twice. */ |
1693 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
1694 | ret = -EBUSY; |
1695 | goto out; | |
1696 | } | |
c1e862c1 KH |
1697 | /* we call try-to-free pages for make this cgroup empty */ |
1698 | lru_add_drain_all(); | |
f817ed48 KH |
1699 | /* try to free all pages in this cgroup */ |
1700 | shrink = 1; | |
1701 | while (nr_retries && mem->res.usage > 0) { | |
1702 | int progress; | |
c1e862c1 KH |
1703 | |
1704 | if (signal_pending(current)) { | |
1705 | ret = -EINTR; | |
1706 | goto out; | |
1707 | } | |
a7885eb8 KM |
1708 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
1709 | false, get_swappiness(mem)); | |
c1e862c1 | 1710 | if (!progress) { |
f817ed48 | 1711 | nr_retries--; |
c1e862c1 KH |
1712 | /* maybe some writeback is necessary */ |
1713 | congestion_wait(WRITE, HZ/10); | |
1714 | } | |
f817ed48 KH |
1715 | |
1716 | } | |
08e552c6 | 1717 | lru_add_drain(); |
f817ed48 KH |
1718 | /* try move_account...there may be some *locked* pages. */ |
1719 | if (mem->res.usage) | |
1720 | goto move_account; | |
1721 | ret = 0; | |
1722 | goto out; | |
cc847582 KH |
1723 | } |
1724 | ||
c1e862c1 KH |
1725 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
1726 | { | |
1727 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
1728 | } | |
1729 | ||
1730 | ||
18f59ea7 BS |
1731 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
1732 | { | |
1733 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
1734 | } | |
1735 | ||
1736 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
1737 | u64 val) | |
1738 | { | |
1739 | int retval = 0; | |
1740 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1741 | struct cgroup *parent = cont->parent; | |
1742 | struct mem_cgroup *parent_mem = NULL; | |
1743 | ||
1744 | if (parent) | |
1745 | parent_mem = mem_cgroup_from_cont(parent); | |
1746 | ||
1747 | cgroup_lock(); | |
1748 | /* | |
1749 | * If parent's use_hiearchy is set, we can't make any modifications | |
1750 | * in the child subtrees. If it is unset, then the change can | |
1751 | * occur, provided the current cgroup has no children. | |
1752 | * | |
1753 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
1754 | * set if there are no children. | |
1755 | */ | |
1756 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
1757 | (val == 1 || val == 0)) { | |
1758 | if (list_empty(&cont->children)) | |
1759 | mem->use_hierarchy = val; | |
1760 | else | |
1761 | retval = -EBUSY; | |
1762 | } else | |
1763 | retval = -EINVAL; | |
1764 | cgroup_unlock(); | |
1765 | ||
1766 | return retval; | |
1767 | } | |
1768 | ||
2c3daa72 | 1769 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 1770 | { |
8c7c6e34 KH |
1771 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
1772 | u64 val = 0; | |
1773 | int type, name; | |
1774 | ||
1775 | type = MEMFILE_TYPE(cft->private); | |
1776 | name = MEMFILE_ATTR(cft->private); | |
1777 | switch (type) { | |
1778 | case _MEM: | |
1779 | val = res_counter_read_u64(&mem->res, name); | |
1780 | break; | |
1781 | case _MEMSWAP: | |
1782 | if (do_swap_account) | |
1783 | val = res_counter_read_u64(&mem->memsw, name); | |
1784 | break; | |
1785 | default: | |
1786 | BUG(); | |
1787 | break; | |
1788 | } | |
1789 | return val; | |
8cdea7c0 | 1790 | } |
628f4235 KH |
1791 | /* |
1792 | * The user of this function is... | |
1793 | * RES_LIMIT. | |
1794 | */ | |
856c13aa PM |
1795 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
1796 | const char *buffer) | |
8cdea7c0 | 1797 | { |
628f4235 | 1798 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 1799 | int type, name; |
628f4235 KH |
1800 | unsigned long long val; |
1801 | int ret; | |
1802 | ||
8c7c6e34 KH |
1803 | type = MEMFILE_TYPE(cft->private); |
1804 | name = MEMFILE_ATTR(cft->private); | |
1805 | switch (name) { | |
628f4235 KH |
1806 | case RES_LIMIT: |
1807 | /* This function does all necessary parse...reuse it */ | |
1808 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
1809 | if (ret) |
1810 | break; | |
1811 | if (type == _MEM) | |
628f4235 | 1812 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
1813 | else |
1814 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 KH |
1815 | break; |
1816 | default: | |
1817 | ret = -EINVAL; /* should be BUG() ? */ | |
1818 | break; | |
1819 | } | |
1820 | return ret; | |
8cdea7c0 BS |
1821 | } |
1822 | ||
fee7b548 KH |
1823 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
1824 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
1825 | { | |
1826 | struct cgroup *cgroup; | |
1827 | unsigned long long min_limit, min_memsw_limit, tmp; | |
1828 | ||
1829 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1830 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1831 | cgroup = memcg->css.cgroup; | |
1832 | if (!memcg->use_hierarchy) | |
1833 | goto out; | |
1834 | ||
1835 | while (cgroup->parent) { | |
1836 | cgroup = cgroup->parent; | |
1837 | memcg = mem_cgroup_from_cont(cgroup); | |
1838 | if (!memcg->use_hierarchy) | |
1839 | break; | |
1840 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1841 | min_limit = min(min_limit, tmp); | |
1842 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1843 | min_memsw_limit = min(min_memsw_limit, tmp); | |
1844 | } | |
1845 | out: | |
1846 | *mem_limit = min_limit; | |
1847 | *memsw_limit = min_memsw_limit; | |
1848 | return; | |
1849 | } | |
1850 | ||
29f2a4da | 1851 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
1852 | { |
1853 | struct mem_cgroup *mem; | |
8c7c6e34 | 1854 | int type, name; |
c84872e1 PE |
1855 | |
1856 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
1857 | type = MEMFILE_TYPE(event); |
1858 | name = MEMFILE_ATTR(event); | |
1859 | switch (name) { | |
29f2a4da | 1860 | case RES_MAX_USAGE: |
8c7c6e34 KH |
1861 | if (type == _MEM) |
1862 | res_counter_reset_max(&mem->res); | |
1863 | else | |
1864 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
1865 | break; |
1866 | case RES_FAILCNT: | |
8c7c6e34 KH |
1867 | if (type == _MEM) |
1868 | res_counter_reset_failcnt(&mem->res); | |
1869 | else | |
1870 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
1871 | break; |
1872 | } | |
85cc59db | 1873 | return 0; |
c84872e1 PE |
1874 | } |
1875 | ||
d2ceb9b7 KH |
1876 | static const struct mem_cgroup_stat_desc { |
1877 | const char *msg; | |
1878 | u64 unit; | |
1879 | } mem_cgroup_stat_desc[] = { | |
1880 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
1881 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
1882 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
1883 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
1884 | }; |
1885 | ||
c64745cf PM |
1886 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
1887 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 1888 | { |
d2ceb9b7 KH |
1889 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
1890 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
1891 | int i; | |
1892 | ||
1893 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1894 | s64 val; | |
1895 | ||
1896 | val = mem_cgroup_read_stat(stat, i); | |
1897 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 1898 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 1899 | } |
6d12e2d8 KH |
1900 | /* showing # of active pages */ |
1901 | { | |
4f98a2fe RR |
1902 | unsigned long active_anon, inactive_anon; |
1903 | unsigned long active_file, inactive_file; | |
7b854121 | 1904 | unsigned long unevictable; |
4f98a2fe RR |
1905 | |
1906 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1907 | LRU_INACTIVE_ANON); | |
1908 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1909 | LRU_ACTIVE_ANON); | |
1910 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1911 | LRU_INACTIVE_FILE); | |
1912 | active_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1913 | LRU_ACTIVE_FILE); | |
7b854121 LS |
1914 | unevictable = mem_cgroup_get_all_zonestat(mem_cont, |
1915 | LRU_UNEVICTABLE); | |
1916 | ||
4f98a2fe RR |
1917 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); |
1918 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); | |
1919 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); | |
1920 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); | |
7b854121 LS |
1921 | cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); |
1922 | ||
6d12e2d8 | 1923 | } |
fee7b548 KH |
1924 | { |
1925 | unsigned long long limit, memsw_limit; | |
1926 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
1927 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
1928 | if (do_swap_account) | |
1929 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
1930 | } | |
7f016ee8 KM |
1931 | |
1932 | #ifdef CONFIG_DEBUG_VM | |
c772be93 | 1933 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
1934 | |
1935 | { | |
1936 | int nid, zid; | |
1937 | struct mem_cgroup_per_zone *mz; | |
1938 | unsigned long recent_rotated[2] = {0, 0}; | |
1939 | unsigned long recent_scanned[2] = {0, 0}; | |
1940 | ||
1941 | for_each_online_node(nid) | |
1942 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
1943 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
1944 | ||
1945 | recent_rotated[0] += | |
1946 | mz->reclaim_stat.recent_rotated[0]; | |
1947 | recent_rotated[1] += | |
1948 | mz->reclaim_stat.recent_rotated[1]; | |
1949 | recent_scanned[0] += | |
1950 | mz->reclaim_stat.recent_scanned[0]; | |
1951 | recent_scanned[1] += | |
1952 | mz->reclaim_stat.recent_scanned[1]; | |
1953 | } | |
1954 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
1955 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
1956 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
1957 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
1958 | } | |
1959 | #endif | |
1960 | ||
d2ceb9b7 KH |
1961 | return 0; |
1962 | } | |
1963 | ||
a7885eb8 KM |
1964 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
1965 | { | |
1966 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
1967 | ||
1968 | return get_swappiness(memcg); | |
1969 | } | |
1970 | ||
1971 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
1972 | u64 val) | |
1973 | { | |
1974 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
1975 | struct mem_cgroup *parent; | |
1976 | if (val > 100) | |
1977 | return -EINVAL; | |
1978 | ||
1979 | if (cgrp->parent == NULL) | |
1980 | return -EINVAL; | |
1981 | ||
1982 | parent = mem_cgroup_from_cont(cgrp->parent); | |
1983 | /* If under hierarchy, only empty-root can set this value */ | |
1984 | if ((parent->use_hierarchy) || | |
1985 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) | |
1986 | return -EINVAL; | |
1987 | ||
1988 | spin_lock(&memcg->reclaim_param_lock); | |
1989 | memcg->swappiness = val; | |
1990 | spin_unlock(&memcg->reclaim_param_lock); | |
1991 | ||
1992 | return 0; | |
1993 | } | |
1994 | ||
c1e862c1 | 1995 | |
8cdea7c0 BS |
1996 | static struct cftype mem_cgroup_files[] = { |
1997 | { | |
0eea1030 | 1998 | .name = "usage_in_bytes", |
8c7c6e34 | 1999 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 2000 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 2001 | }, |
c84872e1 PE |
2002 | { |
2003 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 2004 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 2005 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
2006 | .read_u64 = mem_cgroup_read, |
2007 | }, | |
8cdea7c0 | 2008 | { |
0eea1030 | 2009 | .name = "limit_in_bytes", |
8c7c6e34 | 2010 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 2011 | .write_string = mem_cgroup_write, |
2c3daa72 | 2012 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
2013 | }, |
2014 | { | |
2015 | .name = "failcnt", | |
8c7c6e34 | 2016 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 2017 | .trigger = mem_cgroup_reset, |
2c3daa72 | 2018 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 2019 | }, |
d2ceb9b7 KH |
2020 | { |
2021 | .name = "stat", | |
c64745cf | 2022 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 2023 | }, |
c1e862c1 KH |
2024 | { |
2025 | .name = "force_empty", | |
2026 | .trigger = mem_cgroup_force_empty_write, | |
2027 | }, | |
18f59ea7 BS |
2028 | { |
2029 | .name = "use_hierarchy", | |
2030 | .write_u64 = mem_cgroup_hierarchy_write, | |
2031 | .read_u64 = mem_cgroup_hierarchy_read, | |
2032 | }, | |
a7885eb8 KM |
2033 | { |
2034 | .name = "swappiness", | |
2035 | .read_u64 = mem_cgroup_swappiness_read, | |
2036 | .write_u64 = mem_cgroup_swappiness_write, | |
2037 | }, | |
8cdea7c0 BS |
2038 | }; |
2039 | ||
8c7c6e34 KH |
2040 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
2041 | static struct cftype memsw_cgroup_files[] = { | |
2042 | { | |
2043 | .name = "memsw.usage_in_bytes", | |
2044 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
2045 | .read_u64 = mem_cgroup_read, | |
2046 | }, | |
2047 | { | |
2048 | .name = "memsw.max_usage_in_bytes", | |
2049 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
2050 | .trigger = mem_cgroup_reset, | |
2051 | .read_u64 = mem_cgroup_read, | |
2052 | }, | |
2053 | { | |
2054 | .name = "memsw.limit_in_bytes", | |
2055 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
2056 | .write_string = mem_cgroup_write, | |
2057 | .read_u64 = mem_cgroup_read, | |
2058 | }, | |
2059 | { | |
2060 | .name = "memsw.failcnt", | |
2061 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
2062 | .trigger = mem_cgroup_reset, | |
2063 | .read_u64 = mem_cgroup_read, | |
2064 | }, | |
2065 | }; | |
2066 | ||
2067 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
2068 | { | |
2069 | if (!do_swap_account) | |
2070 | return 0; | |
2071 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
2072 | ARRAY_SIZE(memsw_cgroup_files)); | |
2073 | }; | |
2074 | #else | |
2075 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
2076 | { | |
2077 | return 0; | |
2078 | } | |
2079 | #endif | |
2080 | ||
6d12e2d8 KH |
2081 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2082 | { | |
2083 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 2084 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 2085 | enum lru_list l; |
41e3355d | 2086 | int zone, tmp = node; |
1ecaab2b KH |
2087 | /* |
2088 | * This routine is called against possible nodes. | |
2089 | * But it's BUG to call kmalloc() against offline node. | |
2090 | * | |
2091 | * TODO: this routine can waste much memory for nodes which will | |
2092 | * never be onlined. It's better to use memory hotplug callback | |
2093 | * function. | |
2094 | */ | |
41e3355d KH |
2095 | if (!node_state(node, N_NORMAL_MEMORY)) |
2096 | tmp = -1; | |
2097 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
2098 | if (!pn) |
2099 | return 1; | |
1ecaab2b | 2100 | |
6d12e2d8 KH |
2101 | mem->info.nodeinfo[node] = pn; |
2102 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
2103 | |
2104 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
2105 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
2106 | for_each_lru(l) |
2107 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 2108 | } |
6d12e2d8 KH |
2109 | return 0; |
2110 | } | |
2111 | ||
1ecaab2b KH |
2112 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2113 | { | |
2114 | kfree(mem->info.nodeinfo[node]); | |
2115 | } | |
2116 | ||
c8dad2bb JB |
2117 | static int mem_cgroup_size(void) |
2118 | { | |
2119 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
2120 | return sizeof(struct mem_cgroup) + cpustat_size; | |
2121 | } | |
2122 | ||
33327948 KH |
2123 | static struct mem_cgroup *mem_cgroup_alloc(void) |
2124 | { | |
2125 | struct mem_cgroup *mem; | |
c8dad2bb | 2126 | int size = mem_cgroup_size(); |
33327948 | 2127 | |
c8dad2bb JB |
2128 | if (size < PAGE_SIZE) |
2129 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 2130 | else |
c8dad2bb | 2131 | mem = vmalloc(size); |
33327948 KH |
2132 | |
2133 | if (mem) | |
c8dad2bb | 2134 | memset(mem, 0, size); |
33327948 KH |
2135 | return mem; |
2136 | } | |
2137 | ||
8c7c6e34 KH |
2138 | /* |
2139 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
2140 | * (scanning all at force_empty is too costly...) | |
2141 | * | |
2142 | * Instead of clearing all references at force_empty, we remember | |
2143 | * the number of reference from swap_cgroup and free mem_cgroup when | |
2144 | * it goes down to 0. | |
2145 | * | |
8c7c6e34 KH |
2146 | * Removal of cgroup itself succeeds regardless of refs from swap. |
2147 | */ | |
2148 | ||
a7ba0eef | 2149 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 2150 | { |
08e552c6 KH |
2151 | int node; |
2152 | ||
08e552c6 KH |
2153 | for_each_node_state(node, N_POSSIBLE) |
2154 | free_mem_cgroup_per_zone_info(mem, node); | |
2155 | ||
c8dad2bb | 2156 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
2157 | kfree(mem); |
2158 | else | |
2159 | vfree(mem); | |
2160 | } | |
2161 | ||
8c7c6e34 KH |
2162 | static void mem_cgroup_get(struct mem_cgroup *mem) |
2163 | { | |
2164 | atomic_inc(&mem->refcnt); | |
2165 | } | |
2166 | ||
2167 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
2168 | { | |
a7ba0eef KH |
2169 | if (atomic_dec_and_test(&mem->refcnt)) |
2170 | __mem_cgroup_free(mem); | |
8c7c6e34 KH |
2171 | } |
2172 | ||
33327948 | 2173 | |
c077719b KH |
2174 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
2175 | static void __init enable_swap_cgroup(void) | |
2176 | { | |
f8d66542 | 2177 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
2178 | do_swap_account = 1; |
2179 | } | |
2180 | #else | |
2181 | static void __init enable_swap_cgroup(void) | |
2182 | { | |
2183 | } | |
2184 | #endif | |
2185 | ||
8cdea7c0 BS |
2186 | static struct cgroup_subsys_state * |
2187 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
2188 | { | |
28dbc4b6 | 2189 | struct mem_cgroup *mem, *parent; |
6d12e2d8 | 2190 | int node; |
8cdea7c0 | 2191 | |
c8dad2bb JB |
2192 | mem = mem_cgroup_alloc(); |
2193 | if (!mem) | |
2194 | return ERR_PTR(-ENOMEM); | |
78fb7466 | 2195 | |
6d12e2d8 KH |
2196 | for_each_node_state(node, N_POSSIBLE) |
2197 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
2198 | goto free_out; | |
c077719b | 2199 | /* root ? */ |
28dbc4b6 | 2200 | if (cont->parent == NULL) { |
c077719b | 2201 | enable_swap_cgroup(); |
28dbc4b6 | 2202 | parent = NULL; |
18f59ea7 | 2203 | } else { |
28dbc4b6 | 2204 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 BS |
2205 | mem->use_hierarchy = parent->use_hierarchy; |
2206 | } | |
28dbc4b6 | 2207 | |
18f59ea7 BS |
2208 | if (parent && parent->use_hierarchy) { |
2209 | res_counter_init(&mem->res, &parent->res); | |
2210 | res_counter_init(&mem->memsw, &parent->memsw); | |
2211 | } else { | |
2212 | res_counter_init(&mem->res, NULL); | |
2213 | res_counter_init(&mem->memsw, NULL); | |
2214 | } | |
6d61ef40 | 2215 | mem->last_scanned_child = NULL; |
2733c06a | 2216 | spin_lock_init(&mem->reclaim_param_lock); |
6d61ef40 | 2217 | |
a7885eb8 KM |
2218 | if (parent) |
2219 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 2220 | atomic_set(&mem->refcnt, 1); |
8cdea7c0 | 2221 | return &mem->css; |
6d12e2d8 | 2222 | free_out: |
a7ba0eef | 2223 | __mem_cgroup_free(mem); |
2dda81ca | 2224 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
2225 | } |
2226 | ||
df878fb0 KH |
2227 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
2228 | struct cgroup *cont) | |
2229 | { | |
2230 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
c1e862c1 | 2231 | mem_cgroup_force_empty(mem, false); |
df878fb0 KH |
2232 | } |
2233 | ||
8cdea7c0 BS |
2234 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
2235 | struct cgroup *cont) | |
2236 | { | |
a7ba0eef | 2237 | mem_cgroup_put(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
2238 | } |
2239 | ||
2240 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
2241 | struct cgroup *cont) | |
2242 | { | |
8c7c6e34 KH |
2243 | int ret; |
2244 | ||
2245 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
2246 | ARRAY_SIZE(mem_cgroup_files)); | |
2247 | ||
2248 | if (!ret) | |
2249 | ret = register_memsw_files(cont, ss); | |
2250 | return ret; | |
8cdea7c0 BS |
2251 | } |
2252 | ||
67e465a7 BS |
2253 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
2254 | struct cgroup *cont, | |
2255 | struct cgroup *old_cont, | |
2256 | struct task_struct *p) | |
2257 | { | |
7f4d454d | 2258 | mutex_lock(&memcg_tasklist); |
67e465a7 | 2259 | /* |
f9717d28 NK |
2260 | * FIXME: It's better to move charges of this process from old |
2261 | * memcg to new memcg. But it's just on TODO-List now. | |
67e465a7 | 2262 | */ |
7f4d454d | 2263 | mutex_unlock(&memcg_tasklist); |
67e465a7 BS |
2264 | } |
2265 | ||
8cdea7c0 BS |
2266 | struct cgroup_subsys mem_cgroup_subsys = { |
2267 | .name = "memory", | |
2268 | .subsys_id = mem_cgroup_subsys_id, | |
2269 | .create = mem_cgroup_create, | |
df878fb0 | 2270 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
2271 | .destroy = mem_cgroup_destroy, |
2272 | .populate = mem_cgroup_populate, | |
67e465a7 | 2273 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 2274 | .early_init = 0, |
8cdea7c0 | 2275 | }; |
c077719b KH |
2276 | |
2277 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2278 | ||
2279 | static int __init disable_swap_account(char *s) | |
2280 | { | |
2281 | really_do_swap_account = 0; | |
2282 | return 1; | |
2283 | } | |
2284 | __setup("noswapaccount", disable_swap_account); | |
2285 | #endif |