]>
Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
8cdea7c0 BS |
9 | * This program is free software; you can redistribute it and/or modify |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation; either version 2 of the License, or | |
12 | * (at your option) any later version. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | */ | |
19 | ||
20 | #include <linux/res_counter.h> | |
21 | #include <linux/memcontrol.h> | |
22 | #include <linux/cgroup.h> | |
78fb7466 | 23 | #include <linux/mm.h> |
d52aa412 | 24 | #include <linux/smp.h> |
8a9f3ccd | 25 | #include <linux/page-flags.h> |
66e1707b | 26 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
27 | #include <linux/bit_spinlock.h> |
28 | #include <linux/rcupdate.h> | |
b6ac57d5 | 29 | #include <linux/slab.h> |
66e1707b BS |
30 | #include <linux/swap.h> |
31 | #include <linux/spinlock.h> | |
32 | #include <linux/fs.h> | |
d2ceb9b7 | 33 | #include <linux/seq_file.h> |
33327948 | 34 | #include <linux/vmalloc.h> |
b69408e8 | 35 | #include <linux/mm_inline.h> |
8cdea7c0 | 36 | |
8697d331 BS |
37 | #include <asm/uaccess.h> |
38 | ||
a181b0e8 KH |
39 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
40 | static struct kmem_cache *page_cgroup_cache __read_mostly; | |
41 | #define MEM_CGROUP_RECLAIM_RETRIES 5 | |
8cdea7c0 | 42 | |
d52aa412 KH |
43 | /* |
44 | * Statistics for memory cgroup. | |
45 | */ | |
46 | enum mem_cgroup_stat_index { | |
47 | /* | |
48 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
49 | */ | |
50 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
51 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
55e462b0 BR |
52 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
53 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
54 | |
55 | MEM_CGROUP_STAT_NSTATS, | |
56 | }; | |
57 | ||
58 | struct mem_cgroup_stat_cpu { | |
59 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
60 | } ____cacheline_aligned_in_smp; | |
61 | ||
62 | struct mem_cgroup_stat { | |
63 | struct mem_cgroup_stat_cpu cpustat[NR_CPUS]; | |
64 | }; | |
65 | ||
66 | /* | |
67 | * For accounting under irq disable, no need for increment preempt count. | |
68 | */ | |
69 | static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat, | |
70 | enum mem_cgroup_stat_index idx, int val) | |
71 | { | |
72 | int cpu = smp_processor_id(); | |
73 | stat->cpustat[cpu].count[idx] += val; | |
74 | } | |
75 | ||
76 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
77 | enum mem_cgroup_stat_index idx) | |
78 | { | |
79 | int cpu; | |
80 | s64 ret = 0; | |
81 | for_each_possible_cpu(cpu) | |
82 | ret += stat->cpustat[cpu].count[idx]; | |
83 | return ret; | |
84 | } | |
85 | ||
6d12e2d8 KH |
86 | /* |
87 | * per-zone information in memory controller. | |
88 | */ | |
6d12e2d8 | 89 | struct mem_cgroup_per_zone { |
072c56c1 KH |
90 | /* |
91 | * spin_lock to protect the per cgroup LRU | |
92 | */ | |
93 | spinlock_t lru_lock; | |
b69408e8 CL |
94 | struct list_head lists[NR_LRU_LISTS]; |
95 | unsigned long count[NR_LRU_LISTS]; | |
6d12e2d8 KH |
96 | }; |
97 | /* Macro for accessing counter */ | |
98 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
99 | ||
100 | struct mem_cgroup_per_node { | |
101 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
102 | }; | |
103 | ||
104 | struct mem_cgroup_lru_info { | |
105 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
106 | }; | |
107 | ||
8cdea7c0 BS |
108 | /* |
109 | * The memory controller data structure. The memory controller controls both | |
110 | * page cache and RSS per cgroup. We would eventually like to provide | |
111 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
112 | * to help the administrator determine what knobs to tune. | |
113 | * | |
114 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
115 | * we hit the water mark. May be even add a low water mark, such that |
116 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
117 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
118 | */ |
119 | struct mem_cgroup { | |
120 | struct cgroup_subsys_state css; | |
121 | /* | |
122 | * the counter to account for memory usage | |
123 | */ | |
124 | struct res_counter res; | |
78fb7466 PE |
125 | /* |
126 | * Per cgroup active and inactive list, similar to the | |
127 | * per zone LRU lists. | |
78fb7466 | 128 | */ |
6d12e2d8 | 129 | struct mem_cgroup_lru_info info; |
072c56c1 | 130 | |
6c48a1d0 | 131 | int prev_priority; /* for recording reclaim priority */ |
d52aa412 KH |
132 | /* |
133 | * statistics. | |
134 | */ | |
135 | struct mem_cgroup_stat stat; | |
8cdea7c0 | 136 | }; |
8869b8f6 | 137 | static struct mem_cgroup init_mem_cgroup; |
8cdea7c0 | 138 | |
8a9f3ccd BS |
139 | /* |
140 | * We use the lower bit of the page->page_cgroup pointer as a bit spin | |
9442ec9d HD |
141 | * lock. We need to ensure that page->page_cgroup is at least two |
142 | * byte aligned (based on comments from Nick Piggin). But since | |
143 | * bit_spin_lock doesn't actually set that lock bit in a non-debug | |
144 | * uniprocessor kernel, we should avoid setting it here too. | |
8a9f3ccd BS |
145 | */ |
146 | #define PAGE_CGROUP_LOCK_BIT 0x0 | |
9442ec9d HD |
147 | #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) |
148 | #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT) | |
149 | #else | |
150 | #define PAGE_CGROUP_LOCK 0x0 | |
151 | #endif | |
8a9f3ccd | 152 | |
8cdea7c0 BS |
153 | /* |
154 | * A page_cgroup page is associated with every page descriptor. The | |
155 | * page_cgroup helps us identify information about the cgroup | |
156 | */ | |
157 | struct page_cgroup { | |
158 | struct list_head lru; /* per cgroup LRU list */ | |
159 | struct page *page; | |
160 | struct mem_cgroup *mem_cgroup; | |
8869b8f6 | 161 | int flags; |
8cdea7c0 | 162 | }; |
217bc319 | 163 | #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */ |
3564c7c4 | 164 | #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */ |
4f98a2fe | 165 | #define PAGE_CGROUP_FLAG_FILE (0x4) /* page is file system backed */ |
8cdea7c0 | 166 | |
d5b69e38 | 167 | static int page_cgroup_nid(struct page_cgroup *pc) |
c0149530 KH |
168 | { |
169 | return page_to_nid(pc->page); | |
170 | } | |
171 | ||
d5b69e38 | 172 | static enum zone_type page_cgroup_zid(struct page_cgroup *pc) |
c0149530 KH |
173 | { |
174 | return page_zonenum(pc->page); | |
175 | } | |
176 | ||
217bc319 KH |
177 | enum charge_type { |
178 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
179 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
69029cd5 | 180 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
4f98a2fe | 181 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
217bc319 KH |
182 | }; |
183 | ||
d52aa412 KH |
184 | /* |
185 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
186 | */ | |
187 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags, | |
188 | bool charge) | |
189 | { | |
190 | int val = (charge)? 1 : -1; | |
191 | struct mem_cgroup_stat *stat = &mem->stat; | |
d52aa412 | 192 | |
8869b8f6 | 193 | VM_BUG_ON(!irqs_disabled()); |
d52aa412 | 194 | if (flags & PAGE_CGROUP_FLAG_CACHE) |
8869b8f6 | 195 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 KH |
196 | else |
197 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val); | |
55e462b0 BR |
198 | |
199 | if (charge) | |
200 | __mem_cgroup_stat_add_safe(stat, | |
201 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); | |
202 | else | |
203 | __mem_cgroup_stat_add_safe(stat, | |
204 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); | |
6d12e2d8 KH |
205 | } |
206 | ||
d5b69e38 | 207 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
208 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
209 | { | |
6d12e2d8 KH |
210 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
211 | } | |
212 | ||
d5b69e38 | 213 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
214 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
215 | { | |
216 | struct mem_cgroup *mem = pc->mem_cgroup; | |
217 | int nid = page_cgroup_nid(pc); | |
218 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 219 | |
6d12e2d8 KH |
220 | return mem_cgroup_zoneinfo(mem, nid, zid); |
221 | } | |
222 | ||
223 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
b69408e8 | 224 | enum lru_list idx) |
6d12e2d8 KH |
225 | { |
226 | int nid, zid; | |
227 | struct mem_cgroup_per_zone *mz; | |
228 | u64 total = 0; | |
229 | ||
230 | for_each_online_node(nid) | |
231 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
232 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
233 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
234 | } | |
235 | return total; | |
d52aa412 KH |
236 | } |
237 | ||
d5b69e38 | 238 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
239 | { |
240 | return container_of(cgroup_subsys_state(cont, | |
241 | mem_cgroup_subsys_id), struct mem_cgroup, | |
242 | css); | |
243 | } | |
244 | ||
cf475ad2 | 245 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 246 | { |
31a78f23 BS |
247 | /* |
248 | * mm_update_next_owner() may clear mm->owner to NULL | |
249 | * if it races with swapoff, page migration, etc. | |
250 | * So this can be called with p == NULL. | |
251 | */ | |
252 | if (unlikely(!p)) | |
253 | return NULL; | |
254 | ||
78fb7466 PE |
255 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
256 | struct mem_cgroup, css); | |
257 | } | |
258 | ||
8a9f3ccd BS |
259 | static inline int page_cgroup_locked(struct page *page) |
260 | { | |
8869b8f6 | 261 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); |
8a9f3ccd BS |
262 | } |
263 | ||
9442ec9d | 264 | static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
78fb7466 | 265 | { |
9442ec9d HD |
266 | VM_BUG_ON(!page_cgroup_locked(page)); |
267 | page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK); | |
78fb7466 PE |
268 | } |
269 | ||
270 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
271 | { | |
8869b8f6 | 272 | return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK); |
8a9f3ccd BS |
273 | } |
274 | ||
d5b69e38 | 275 | static void lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
276 | { |
277 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
8a9f3ccd BS |
278 | } |
279 | ||
2680eed7 HD |
280 | static int try_lock_page_cgroup(struct page *page) |
281 | { | |
282 | return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
283 | } | |
284 | ||
d5b69e38 | 285 | static void unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
286 | { |
287 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
288 | } | |
289 | ||
3eae90c3 KH |
290 | static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, |
291 | struct page_cgroup *pc) | |
6d12e2d8 | 292 | { |
4f98a2fe RR |
293 | int lru = LRU_BASE; |
294 | ||
295 | if (pc->flags & PAGE_CGROUP_FLAG_ACTIVE) | |
296 | lru += LRU_ACTIVE; | |
297 | if (pc->flags & PAGE_CGROUP_FLAG_FILE) | |
298 | lru += LRU_FILE; | |
6d12e2d8 | 299 | |
b69408e8 | 300 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
6d12e2d8 KH |
301 | |
302 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false); | |
508b7be0 | 303 | list_del(&pc->lru); |
6d12e2d8 KH |
304 | } |
305 | ||
3eae90c3 KH |
306 | static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, |
307 | struct page_cgroup *pc) | |
6d12e2d8 | 308 | { |
4f98a2fe | 309 | int lru = LRU_BASE; |
b69408e8 CL |
310 | |
311 | if (pc->flags & PAGE_CGROUP_FLAG_ACTIVE) | |
312 | lru += LRU_ACTIVE; | |
4f98a2fe RR |
313 | if (pc->flags & PAGE_CGROUP_FLAG_FILE) |
314 | lru += LRU_FILE; | |
b69408e8 CL |
315 | |
316 | MEM_CGROUP_ZSTAT(mz, lru) += 1; | |
317 | list_add(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 | 318 | |
6d12e2d8 KH |
319 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true); |
320 | } | |
321 | ||
8697d331 | 322 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
66e1707b | 323 | { |
6d12e2d8 | 324 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); |
4f98a2fe RR |
325 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; |
326 | int file = pc->flags & PAGE_CGROUP_FLAG_FILE; | |
327 | int lru = LRU_FILE * !!file + !!from; | |
6d12e2d8 | 328 | |
b69408e8 CL |
329 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
330 | ||
331 | if (active) | |
3564c7c4 | 332 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; |
b69408e8 | 333 | else |
3564c7c4 | 334 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; |
b69408e8 | 335 | |
4f98a2fe | 336 | lru = LRU_FILE * !!file + !!active; |
b69408e8 CL |
337 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
338 | list_move(&pc->lru, &mz->lists[lru]); | |
66e1707b BS |
339 | } |
340 | ||
4c4a2214 DR |
341 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
342 | { | |
343 | int ret; | |
344 | ||
345 | task_lock(task); | |
bd845e38 | 346 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
347 | task_unlock(task); |
348 | return ret; | |
349 | } | |
350 | ||
66e1707b BS |
351 | /* |
352 | * This routine assumes that the appropriate zone's lru lock is already held | |
353 | */ | |
427d5416 | 354 | void mem_cgroup_move_lists(struct page *page, bool active) |
66e1707b | 355 | { |
427d5416 | 356 | struct page_cgroup *pc; |
072c56c1 KH |
357 | struct mem_cgroup_per_zone *mz; |
358 | unsigned long flags; | |
359 | ||
cede86ac LZ |
360 | if (mem_cgroup_subsys.disabled) |
361 | return; | |
362 | ||
2680eed7 HD |
363 | /* |
364 | * We cannot lock_page_cgroup while holding zone's lru_lock, | |
365 | * because other holders of lock_page_cgroup can be interrupted | |
366 | * with an attempt to rotate_reclaimable_page. But we cannot | |
367 | * safely get to page_cgroup without it, so just try_lock it: | |
368 | * mem_cgroup_isolate_pages allows for page left on wrong list. | |
369 | */ | |
370 | if (!try_lock_page_cgroup(page)) | |
66e1707b BS |
371 | return; |
372 | ||
2680eed7 HD |
373 | pc = page_get_page_cgroup(page); |
374 | if (pc) { | |
2680eed7 | 375 | mz = page_cgroup_zoneinfo(pc); |
2680eed7 | 376 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 377 | __mem_cgroup_move_lists(pc, active); |
2680eed7 | 378 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9b3c0a07 HT |
379 | } |
380 | unlock_page_cgroup(page); | |
66e1707b BS |
381 | } |
382 | ||
58ae83db KH |
383 | /* |
384 | * Calculate mapped_ratio under memory controller. This will be used in | |
385 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
386 | */ | |
387 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
388 | { | |
389 | long total, rss; | |
390 | ||
391 | /* | |
392 | * usage is recorded in bytes. But, here, we assume the number of | |
393 | * physical pages can be represented by "long" on any arch. | |
394 | */ | |
395 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
396 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
397 | return (int)((rss * 100L) / total); | |
398 | } | |
8869b8f6 | 399 | |
6c48a1d0 KH |
400 | /* |
401 | * prev_priority control...this will be used in memory reclaim path. | |
402 | */ | |
403 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
404 | { | |
405 | return mem->prev_priority; | |
406 | } | |
407 | ||
408 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
409 | { | |
410 | if (priority < mem->prev_priority) | |
411 | mem->prev_priority = priority; | |
412 | } | |
413 | ||
414 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
415 | { | |
416 | mem->prev_priority = priority; | |
417 | } | |
418 | ||
cc38108e KH |
419 | /* |
420 | * Calculate # of pages to be scanned in this priority/zone. | |
421 | * See also vmscan.c | |
422 | * | |
423 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
424 | * (see include/linux/mmzone.h) | |
425 | */ | |
426 | ||
b69408e8 CL |
427 | long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, |
428 | int priority, enum lru_list lru) | |
cc38108e | 429 | { |
b69408e8 | 430 | long nr_pages; |
cc38108e KH |
431 | int nid = zone->zone_pgdat->node_id; |
432 | int zid = zone_idx(zone); | |
433 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
434 | ||
b69408e8 | 435 | nr_pages = MEM_CGROUP_ZSTAT(mz, lru); |
cc38108e | 436 | |
b69408e8 | 437 | return (nr_pages >> priority); |
cc38108e KH |
438 | } |
439 | ||
66e1707b BS |
440 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
441 | struct list_head *dst, | |
442 | unsigned long *scanned, int order, | |
443 | int mode, struct zone *z, | |
444 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 445 | int active, int file) |
66e1707b BS |
446 | { |
447 | unsigned long nr_taken = 0; | |
448 | struct page *page; | |
449 | unsigned long scan; | |
450 | LIST_HEAD(pc_list); | |
451 | struct list_head *src; | |
ff7283fa | 452 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
453 | int nid = z->zone_pgdat->node_id; |
454 | int zid = zone_idx(z); | |
455 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 456 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 457 | |
cf475ad2 | 458 | BUG_ON(!mem_cont); |
1ecaab2b | 459 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 460 | src = &mz->lists[lru]; |
66e1707b | 461 | |
072c56c1 | 462 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
463 | scan = 0; |
464 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 465 | if (scan >= nr_to_scan) |
ff7283fa | 466 | break; |
66e1707b | 467 | page = pc->page; |
66e1707b | 468 | |
436c6541 | 469 | if (unlikely(!PageLRU(page))) |
ff7283fa | 470 | continue; |
ff7283fa | 471 | |
4f98a2fe RR |
472 | /* |
473 | * TODO: play better with lumpy reclaim, grabbing anything. | |
474 | */ | |
66e1707b BS |
475 | if (PageActive(page) && !active) { |
476 | __mem_cgroup_move_lists(pc, true); | |
66e1707b BS |
477 | continue; |
478 | } | |
479 | if (!PageActive(page) && active) { | |
480 | __mem_cgroup_move_lists(pc, false); | |
66e1707b BS |
481 | continue; |
482 | } | |
483 | ||
436c6541 HD |
484 | scan++; |
485 | list_move(&pc->lru, &pc_list); | |
66e1707b | 486 | |
4f98a2fe | 487 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
488 | list_move(&page->lru, dst); |
489 | nr_taken++; | |
490 | } | |
491 | } | |
492 | ||
493 | list_splice(&pc_list, src); | |
072c56c1 | 494 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
495 | |
496 | *scanned = scan; | |
497 | return nr_taken; | |
498 | } | |
499 | ||
8a9f3ccd BS |
500 | /* |
501 | * Charge the memory controller for page usage. | |
502 | * Return | |
503 | * 0 if the charge was successful | |
504 | * < 0 if the cgroup is over its limit | |
505 | */ | |
217bc319 | 506 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
e8589cc1 KH |
507 | gfp_t gfp_mask, enum charge_type ctype, |
508 | struct mem_cgroup *memcg) | |
8a9f3ccd BS |
509 | { |
510 | struct mem_cgroup *mem; | |
9175e031 | 511 | struct page_cgroup *pc; |
66e1707b BS |
512 | unsigned long flags; |
513 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
072c56c1 | 514 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 515 | |
508b7be0 | 516 | pc = kmem_cache_alloc(page_cgroup_cache, gfp_mask); |
b76734e5 | 517 | if (unlikely(pc == NULL)) |
8a9f3ccd BS |
518 | goto err; |
519 | ||
8a9f3ccd | 520 | /* |
3be91277 HD |
521 | * We always charge the cgroup the mm_struct belongs to. |
522 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
523 | * thread group leader migrates. It's possible that mm is not |
524 | * set, if so charge the init_mm (happens for pagecache usage). | |
525 | */ | |
69029cd5 | 526 | if (likely(!memcg)) { |
e8589cc1 KH |
527 | rcu_read_lock(); |
528 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
529 | if (unlikely(!mem)) { |
530 | rcu_read_unlock(); | |
531 | kmem_cache_free(page_cgroup_cache, pc); | |
532 | return 0; | |
533 | } | |
e8589cc1 KH |
534 | /* |
535 | * For every charge from the cgroup, increment reference count | |
536 | */ | |
537 | css_get(&mem->css); | |
538 | rcu_read_unlock(); | |
539 | } else { | |
540 | mem = memcg; | |
541 | css_get(&memcg->css); | |
542 | } | |
8a9f3ccd | 543 | |
0eea1030 | 544 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
3be91277 HD |
545 | if (!(gfp_mask & __GFP_WAIT)) |
546 | goto out; | |
e1a1cd59 BS |
547 | |
548 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
549 | continue; |
550 | ||
551 | /* | |
8869b8f6 HD |
552 | * try_to_free_mem_cgroup_pages() might not give us a full |
553 | * picture of reclaim. Some pages are reclaimed and might be | |
554 | * moved to swap cache or just unmapped from the cgroup. | |
555 | * Check the limit again to see if the reclaim reduced the | |
556 | * current usage of the cgroup before giving up | |
557 | */ | |
66e1707b BS |
558 | if (res_counter_check_under_limit(&mem->res)) |
559 | continue; | |
3be91277 HD |
560 | |
561 | if (!nr_retries--) { | |
562 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
563 | goto out; | |
66e1707b | 564 | } |
8a9f3ccd BS |
565 | } |
566 | ||
8a9f3ccd BS |
567 | pc->mem_cgroup = mem; |
568 | pc->page = page; | |
508b7be0 KH |
569 | /* |
570 | * If a page is accounted as a page cache, insert to inactive list. | |
571 | * If anon, insert to active list. | |
572 | */ | |
4f98a2fe | 573 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) { |
4a56d02e | 574 | pc->flags = PAGE_CGROUP_FLAG_CACHE; |
4f98a2fe RR |
575 | if (page_is_file_cache(page)) |
576 | pc->flags |= PAGE_CGROUP_FLAG_FILE; | |
577 | else | |
578 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; | |
579 | } else if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) | |
508b7be0 | 580 | pc->flags = PAGE_CGROUP_FLAG_ACTIVE; |
4f98a2fe RR |
581 | else /* MEM_CGROUP_CHARGE_TYPE_SHMEM */ |
582 | pc->flags = PAGE_CGROUP_FLAG_CACHE | PAGE_CGROUP_FLAG_ACTIVE; | |
3be91277 | 583 | |
7e924aaf | 584 | lock_page_cgroup(page); |
b76734e5 | 585 | if (unlikely(page_get_page_cgroup(page))) { |
7e924aaf | 586 | unlock_page_cgroup(page); |
9175e031 KH |
587 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
588 | css_put(&mem->css); | |
b6ac57d5 | 589 | kmem_cache_free(page_cgroup_cache, pc); |
accf163e | 590 | goto done; |
9175e031 | 591 | } |
7e924aaf | 592 | page_assign_page_cgroup(page, pc); |
8a9f3ccd | 593 | |
072c56c1 KH |
594 | mz = page_cgroup_zoneinfo(pc); |
595 | spin_lock_irqsave(&mz->lru_lock, flags); | |
3eae90c3 | 596 | __mem_cgroup_add_list(mz, pc); |
072c56c1 | 597 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b | 598 | |
fb59e9f1 | 599 | unlock_page_cgroup(page); |
8a9f3ccd | 600 | done: |
8a9f3ccd | 601 | return 0; |
3be91277 HD |
602 | out: |
603 | css_put(&mem->css); | |
b6ac57d5 | 604 | kmem_cache_free(page_cgroup_cache, pc); |
8a9f3ccd | 605 | err: |
8a9f3ccd BS |
606 | return -ENOMEM; |
607 | } | |
608 | ||
8869b8f6 | 609 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
217bc319 | 610 | { |
cede86ac LZ |
611 | if (mem_cgroup_subsys.disabled) |
612 | return 0; | |
613 | ||
69029cd5 KH |
614 | /* |
615 | * If already mapped, we don't have to account. | |
616 | * If page cache, page->mapping has address_space. | |
617 | * But page->mapping may have out-of-use anon_vma pointer, | |
618 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
619 | * is NULL. | |
620 | */ | |
621 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
622 | return 0; | |
623 | if (unlikely(!mm)) | |
624 | mm = &init_mm; | |
217bc319 | 625 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 626 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
627 | } |
628 | ||
e1a1cd59 BS |
629 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
630 | gfp_t gfp_mask) | |
8697d331 | 631 | { |
cede86ac LZ |
632 | if (mem_cgroup_subsys.disabled) |
633 | return 0; | |
634 | ||
accf163e KH |
635 | /* |
636 | * Corner case handling. This is called from add_to_page_cache() | |
637 | * in usual. But some FS (shmem) precharges this page before calling it | |
638 | * and call add_to_page_cache() with GFP_NOWAIT. | |
639 | * | |
640 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
641 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
642 | * charge twice. (It works but has to pay a bit larger cost.) | |
643 | */ | |
644 | if (!(gfp_mask & __GFP_WAIT)) { | |
645 | struct page_cgroup *pc; | |
646 | ||
647 | lock_page_cgroup(page); | |
648 | pc = page_get_page_cgroup(page); | |
649 | if (pc) { | |
650 | VM_BUG_ON(pc->page != page); | |
651 | VM_BUG_ON(!pc->mem_cgroup); | |
652 | unlock_page_cgroup(page); | |
653 | return 0; | |
654 | } | |
655 | unlock_page_cgroup(page); | |
656 | } | |
657 | ||
69029cd5 | 658 | if (unlikely(!mm)) |
8697d331 | 659 | mm = &init_mm; |
accf163e | 660 | |
8869b8f6 | 661 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 KH |
662 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
663 | } | |
664 | ||
8a9f3ccd | 665 | /* |
69029cd5 | 666 | * uncharge if !page_mapped(page) |
8a9f3ccd | 667 | */ |
69029cd5 KH |
668 | static void |
669 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) | |
8a9f3ccd | 670 | { |
8289546e | 671 | struct page_cgroup *pc; |
8a9f3ccd | 672 | struct mem_cgroup *mem; |
072c56c1 | 673 | struct mem_cgroup_per_zone *mz; |
66e1707b | 674 | unsigned long flags; |
8a9f3ccd | 675 | |
4077960e BS |
676 | if (mem_cgroup_subsys.disabled) |
677 | return; | |
678 | ||
8697d331 | 679 | /* |
3c541e14 | 680 | * Check if our page_cgroup is valid |
8697d331 | 681 | */ |
8289546e HD |
682 | lock_page_cgroup(page); |
683 | pc = page_get_page_cgroup(page); | |
b76734e5 | 684 | if (unlikely(!pc)) |
8289546e | 685 | goto unlock; |
8a9f3ccd | 686 | |
b9c565d5 | 687 | VM_BUG_ON(pc->page != page); |
b9c565d5 | 688 | |
69029cd5 KH |
689 | if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
690 | && ((pc->flags & PAGE_CGROUP_FLAG_CACHE) | |
691 | || page_mapped(page))) | |
692 | goto unlock; | |
b9c565d5 | 693 | |
69029cd5 KH |
694 | mz = page_cgroup_zoneinfo(pc); |
695 | spin_lock_irqsave(&mz->lru_lock, flags); | |
696 | __mem_cgroup_remove_list(mz, pc); | |
697 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
fb59e9f1 | 698 | |
69029cd5 KH |
699 | page_assign_page_cgroup(page, NULL); |
700 | unlock_page_cgroup(page); | |
6d48ff8b | 701 | |
69029cd5 KH |
702 | mem = pc->mem_cgroup; |
703 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
704 | css_put(&mem->css); | |
6d12e2d8 | 705 | |
69029cd5 KH |
706 | kmem_cache_free(page_cgroup_cache, pc); |
707 | return; | |
8289546e | 708 | unlock: |
3c541e14 BS |
709 | unlock_page_cgroup(page); |
710 | } | |
711 | ||
69029cd5 KH |
712 | void mem_cgroup_uncharge_page(struct page *page) |
713 | { | |
714 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
715 | } | |
716 | ||
717 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
718 | { | |
719 | VM_BUG_ON(page_mapped(page)); | |
720 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); | |
721 | } | |
722 | ||
ae41be37 | 723 | /* |
e8589cc1 | 724 | * Before starting migration, account against new page. |
ae41be37 | 725 | */ |
e8589cc1 | 726 | int mem_cgroup_prepare_migration(struct page *page, struct page *newpage) |
ae41be37 KH |
727 | { |
728 | struct page_cgroup *pc; | |
e8589cc1 KH |
729 | struct mem_cgroup *mem = NULL; |
730 | enum charge_type ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
731 | int ret = 0; | |
8869b8f6 | 732 | |
4077960e BS |
733 | if (mem_cgroup_subsys.disabled) |
734 | return 0; | |
735 | ||
ae41be37 KH |
736 | lock_page_cgroup(page); |
737 | pc = page_get_page_cgroup(page); | |
e8589cc1 KH |
738 | if (pc) { |
739 | mem = pc->mem_cgroup; | |
740 | css_get(&mem->css); | |
4f98a2fe RR |
741 | if (pc->flags & PAGE_CGROUP_FLAG_CACHE) { |
742 | if (page_is_file_cache(page)) | |
743 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
744 | else | |
745 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
746 | } | |
e8589cc1 | 747 | } |
ae41be37 | 748 | unlock_page_cgroup(page); |
e8589cc1 KH |
749 | if (mem) { |
750 | ret = mem_cgroup_charge_common(newpage, NULL, GFP_KERNEL, | |
751 | ctype, mem); | |
752 | css_put(&mem->css); | |
753 | } | |
754 | return ret; | |
ae41be37 | 755 | } |
8869b8f6 | 756 | |
69029cd5 | 757 | /* remove redundant charge if migration failed*/ |
e8589cc1 | 758 | void mem_cgroup_end_migration(struct page *newpage) |
ae41be37 | 759 | { |
69029cd5 KH |
760 | /* |
761 | * At success, page->mapping is not NULL. | |
762 | * special rollback care is necessary when | |
763 | * 1. at migration failure. (newpage->mapping is cleared in this case) | |
764 | * 2. the newpage was moved but not remapped again because the task | |
765 | * exits and the newpage is obsolete. In this case, the new page | |
766 | * may be a swapcache. So, we just call mem_cgroup_uncharge_page() | |
767 | * always for avoiding mess. The page_cgroup will be removed if | |
768 | * unnecessary. File cache pages is still on radix-tree. Don't | |
769 | * care it. | |
770 | */ | |
771 | if (!newpage->mapping) | |
772 | __mem_cgroup_uncharge_common(newpage, | |
773 | MEM_CGROUP_CHARGE_TYPE_FORCE); | |
774 | else if (PageAnon(newpage)) | |
775 | mem_cgroup_uncharge_page(newpage); | |
ae41be37 | 776 | } |
78fb7466 | 777 | |
c9b0ed51 KH |
778 | /* |
779 | * A call to try to shrink memory usage under specified resource controller. | |
780 | * This is typically used for page reclaiming for shmem for reducing side | |
781 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
782 | */ | |
783 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) | |
784 | { | |
785 | struct mem_cgroup *mem; | |
786 | int progress = 0; | |
787 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
788 | ||
cede86ac LZ |
789 | if (mem_cgroup_subsys.disabled) |
790 | return 0; | |
9623e078 HD |
791 | if (!mm) |
792 | return 0; | |
cede86ac | 793 | |
c9b0ed51 KH |
794 | rcu_read_lock(); |
795 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
796 | if (unlikely(!mem)) { |
797 | rcu_read_unlock(); | |
798 | return 0; | |
799 | } | |
c9b0ed51 KH |
800 | css_get(&mem->css); |
801 | rcu_read_unlock(); | |
802 | ||
803 | do { | |
804 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask); | |
a10cebf5 | 805 | progress += res_counter_check_under_limit(&mem->res); |
c9b0ed51 KH |
806 | } while (!progress && --retry); |
807 | ||
808 | css_put(&mem->css); | |
809 | if (!retry) | |
810 | return -ENOMEM; | |
811 | return 0; | |
812 | } | |
813 | ||
628f4235 KH |
814 | int mem_cgroup_resize_limit(struct mem_cgroup *memcg, unsigned long long val) |
815 | { | |
816 | ||
817 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
818 | int progress; | |
819 | int ret = 0; | |
820 | ||
821 | while (res_counter_set_limit(&memcg->res, val)) { | |
822 | if (signal_pending(current)) { | |
823 | ret = -EINTR; | |
824 | break; | |
825 | } | |
826 | if (!retry_count) { | |
827 | ret = -EBUSY; | |
828 | break; | |
829 | } | |
830 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL); | |
831 | if (!progress) | |
832 | retry_count--; | |
833 | } | |
834 | return ret; | |
835 | } | |
836 | ||
837 | ||
cc847582 KH |
838 | /* |
839 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
840 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
841 | */ | |
842 | #define FORCE_UNCHARGE_BATCH (128) | |
8869b8f6 | 843 | static void mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 | 844 | struct mem_cgroup_per_zone *mz, |
b69408e8 | 845 | enum lru_list lru) |
cc847582 KH |
846 | { |
847 | struct page_cgroup *pc; | |
848 | struct page *page; | |
9b3c0a07 | 849 | int count = FORCE_UNCHARGE_BATCH; |
cc847582 | 850 | unsigned long flags; |
072c56c1 KH |
851 | struct list_head *list; |
852 | ||
b69408e8 | 853 | list = &mz->lists[lru]; |
cc847582 | 854 | |
072c56c1 | 855 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 856 | while (!list_empty(list)) { |
cc847582 KH |
857 | pc = list_entry(list->prev, struct page_cgroup, lru); |
858 | page = pc->page; | |
9b3c0a07 HT |
859 | get_page(page); |
860 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
e8589cc1 KH |
861 | /* |
862 | * Check if this page is on LRU. !LRU page can be found | |
863 | * if it's under page migration. | |
864 | */ | |
865 | if (PageLRU(page)) { | |
69029cd5 KH |
866 | __mem_cgroup_uncharge_common(page, |
867 | MEM_CGROUP_CHARGE_TYPE_FORCE); | |
e8589cc1 KH |
868 | put_page(page); |
869 | if (--count <= 0) { | |
870 | count = FORCE_UNCHARGE_BATCH; | |
871 | cond_resched(); | |
872 | } | |
873 | } else | |
9b3c0a07 | 874 | cond_resched(); |
9b3c0a07 | 875 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 | 876 | } |
072c56c1 | 877 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
878 | } |
879 | ||
880 | /* | |
881 | * make mem_cgroup's charge to be 0 if there is no task. | |
882 | * This enables deleting this mem_cgroup. | |
883 | */ | |
d5b69e38 | 884 | static int mem_cgroup_force_empty(struct mem_cgroup *mem) |
cc847582 KH |
885 | { |
886 | int ret = -EBUSY; | |
1ecaab2b | 887 | int node, zid; |
8869b8f6 | 888 | |
cc847582 KH |
889 | css_get(&mem->css); |
890 | /* | |
891 | * page reclaim code (kswapd etc..) will move pages between | |
8869b8f6 | 892 | * active_list <-> inactive_list while we don't take a lock. |
cc847582 KH |
893 | * So, we have to do loop here until all lists are empty. |
894 | */ | |
1ecaab2b | 895 | while (mem->res.usage > 0) { |
cc847582 KH |
896 | if (atomic_read(&mem->css.cgroup->count) > 0) |
897 | goto out; | |
1ecaab2b KH |
898 | for_each_node_state(node, N_POSSIBLE) |
899 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
900 | struct mem_cgroup_per_zone *mz; | |
b69408e8 | 901 | enum lru_list l; |
1ecaab2b | 902 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
b69408e8 CL |
903 | for_each_lru(l) |
904 | mem_cgroup_force_empty_list(mem, mz, l); | |
1ecaab2b | 905 | } |
cc847582 KH |
906 | } |
907 | ret = 0; | |
908 | out: | |
909 | css_put(&mem->css); | |
910 | return ret; | |
911 | } | |
912 | ||
2c3daa72 | 913 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 914 | { |
2c3daa72 PM |
915 | return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res, |
916 | cft->private); | |
8cdea7c0 | 917 | } |
628f4235 KH |
918 | /* |
919 | * The user of this function is... | |
920 | * RES_LIMIT. | |
921 | */ | |
856c13aa PM |
922 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
923 | const char *buffer) | |
8cdea7c0 | 924 | { |
628f4235 KH |
925 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
926 | unsigned long long val; | |
927 | int ret; | |
928 | ||
929 | switch (cft->private) { | |
930 | case RES_LIMIT: | |
931 | /* This function does all necessary parse...reuse it */ | |
932 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
933 | if (!ret) | |
934 | ret = mem_cgroup_resize_limit(memcg, val); | |
935 | break; | |
936 | default: | |
937 | ret = -EINVAL; /* should be BUG() ? */ | |
938 | break; | |
939 | } | |
940 | return ret; | |
8cdea7c0 BS |
941 | } |
942 | ||
29f2a4da | 943 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
944 | { |
945 | struct mem_cgroup *mem; | |
946 | ||
947 | mem = mem_cgroup_from_cont(cont); | |
29f2a4da PE |
948 | switch (event) { |
949 | case RES_MAX_USAGE: | |
950 | res_counter_reset_max(&mem->res); | |
951 | break; | |
952 | case RES_FAILCNT: | |
953 | res_counter_reset_failcnt(&mem->res); | |
954 | break; | |
955 | } | |
85cc59db | 956 | return 0; |
c84872e1 PE |
957 | } |
958 | ||
85cc59db | 959 | static int mem_force_empty_write(struct cgroup *cont, unsigned int event) |
cc847582 | 960 | { |
85cc59db | 961 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont)); |
cc847582 KH |
962 | } |
963 | ||
d2ceb9b7 KH |
964 | static const struct mem_cgroup_stat_desc { |
965 | const char *msg; | |
966 | u64 unit; | |
967 | } mem_cgroup_stat_desc[] = { | |
968 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
969 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
970 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
971 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
972 | }; |
973 | ||
c64745cf PM |
974 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
975 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 976 | { |
d2ceb9b7 KH |
977 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
978 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
979 | int i; | |
980 | ||
981 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
982 | s64 val; | |
983 | ||
984 | val = mem_cgroup_read_stat(stat, i); | |
985 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 986 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 987 | } |
6d12e2d8 KH |
988 | /* showing # of active pages */ |
989 | { | |
4f98a2fe RR |
990 | unsigned long active_anon, inactive_anon; |
991 | unsigned long active_file, inactive_file; | |
992 | ||
993 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
994 | LRU_INACTIVE_ANON); | |
995 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
996 | LRU_ACTIVE_ANON); | |
997 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, | |
998 | LRU_INACTIVE_FILE); | |
999 | active_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1000 | LRU_ACTIVE_FILE); | |
1001 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); | |
1002 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); | |
1003 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); | |
1004 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); | |
6d12e2d8 | 1005 | } |
d2ceb9b7 KH |
1006 | return 0; |
1007 | } | |
1008 | ||
8cdea7c0 BS |
1009 | static struct cftype mem_cgroup_files[] = { |
1010 | { | |
0eea1030 | 1011 | .name = "usage_in_bytes", |
8cdea7c0 | 1012 | .private = RES_USAGE, |
2c3daa72 | 1013 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1014 | }, |
c84872e1 PE |
1015 | { |
1016 | .name = "max_usage_in_bytes", | |
1017 | .private = RES_MAX_USAGE, | |
29f2a4da | 1018 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
1019 | .read_u64 = mem_cgroup_read, |
1020 | }, | |
8cdea7c0 | 1021 | { |
0eea1030 | 1022 | .name = "limit_in_bytes", |
8cdea7c0 | 1023 | .private = RES_LIMIT, |
856c13aa | 1024 | .write_string = mem_cgroup_write, |
2c3daa72 | 1025 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
1026 | }, |
1027 | { | |
1028 | .name = "failcnt", | |
1029 | .private = RES_FAILCNT, | |
29f2a4da | 1030 | .trigger = mem_cgroup_reset, |
2c3daa72 | 1031 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1032 | }, |
cc847582 KH |
1033 | { |
1034 | .name = "force_empty", | |
85cc59db | 1035 | .trigger = mem_force_empty_write, |
cc847582 | 1036 | }, |
d2ceb9b7 KH |
1037 | { |
1038 | .name = "stat", | |
c64745cf | 1039 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 1040 | }, |
8cdea7c0 BS |
1041 | }; |
1042 | ||
6d12e2d8 KH |
1043 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1044 | { | |
1045 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 1046 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1047 | enum lru_list l; |
41e3355d | 1048 | int zone, tmp = node; |
1ecaab2b KH |
1049 | /* |
1050 | * This routine is called against possible nodes. | |
1051 | * But it's BUG to call kmalloc() against offline node. | |
1052 | * | |
1053 | * TODO: this routine can waste much memory for nodes which will | |
1054 | * never be onlined. It's better to use memory hotplug callback | |
1055 | * function. | |
1056 | */ | |
41e3355d KH |
1057 | if (!node_state(node, N_NORMAL_MEMORY)) |
1058 | tmp = -1; | |
1059 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
1060 | if (!pn) |
1061 | return 1; | |
1ecaab2b | 1062 | |
6d12e2d8 KH |
1063 | mem->info.nodeinfo[node] = pn; |
1064 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1065 | |
1066 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1067 | mz = &pn->zoneinfo[zone]; | |
072c56c1 | 1068 | spin_lock_init(&mz->lru_lock); |
b69408e8 CL |
1069 | for_each_lru(l) |
1070 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 1071 | } |
6d12e2d8 KH |
1072 | return 0; |
1073 | } | |
1074 | ||
1ecaab2b KH |
1075 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1076 | { | |
1077 | kfree(mem->info.nodeinfo[node]); | |
1078 | } | |
1079 | ||
33327948 KH |
1080 | static struct mem_cgroup *mem_cgroup_alloc(void) |
1081 | { | |
1082 | struct mem_cgroup *mem; | |
1083 | ||
1084 | if (sizeof(*mem) < PAGE_SIZE) | |
1085 | mem = kmalloc(sizeof(*mem), GFP_KERNEL); | |
1086 | else | |
1087 | mem = vmalloc(sizeof(*mem)); | |
1088 | ||
1089 | if (mem) | |
1090 | memset(mem, 0, sizeof(*mem)); | |
1091 | return mem; | |
1092 | } | |
1093 | ||
1094 | static void mem_cgroup_free(struct mem_cgroup *mem) | |
1095 | { | |
1096 | if (sizeof(*mem) < PAGE_SIZE) | |
1097 | kfree(mem); | |
1098 | else | |
1099 | vfree(mem); | |
1100 | } | |
1101 | ||
1102 | ||
8cdea7c0 BS |
1103 | static struct cgroup_subsys_state * |
1104 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1105 | { | |
1106 | struct mem_cgroup *mem; | |
6d12e2d8 | 1107 | int node; |
8cdea7c0 | 1108 | |
b6ac57d5 | 1109 | if (unlikely((cont->parent) == NULL)) { |
78fb7466 | 1110 | mem = &init_mem_cgroup; |
b6ac57d5 BS |
1111 | page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC); |
1112 | } else { | |
33327948 KH |
1113 | mem = mem_cgroup_alloc(); |
1114 | if (!mem) | |
1115 | return ERR_PTR(-ENOMEM); | |
b6ac57d5 | 1116 | } |
78fb7466 | 1117 | |
8cdea7c0 | 1118 | res_counter_init(&mem->res); |
1ecaab2b | 1119 | |
6d12e2d8 KH |
1120 | for_each_node_state(node, N_POSSIBLE) |
1121 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1122 | goto free_out; | |
1123 | ||
8cdea7c0 | 1124 | return &mem->css; |
6d12e2d8 KH |
1125 | free_out: |
1126 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1127 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1128 | if (cont->parent != NULL) |
33327948 | 1129 | mem_cgroup_free(mem); |
2dda81ca | 1130 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1131 | } |
1132 | ||
df878fb0 KH |
1133 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1134 | struct cgroup *cont) | |
1135 | { | |
1136 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1137 | mem_cgroup_force_empty(mem); | |
1138 | } | |
1139 | ||
8cdea7c0 BS |
1140 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1141 | struct cgroup *cont) | |
1142 | { | |
6d12e2d8 KH |
1143 | int node; |
1144 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1145 | ||
1146 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1147 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1148 | |
33327948 | 1149 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
1150 | } |
1151 | ||
1152 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1153 | struct cgroup *cont) | |
1154 | { | |
1155 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1156 | ARRAY_SIZE(mem_cgroup_files)); | |
1157 | } | |
1158 | ||
67e465a7 BS |
1159 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1160 | struct cgroup *cont, | |
1161 | struct cgroup *old_cont, | |
1162 | struct task_struct *p) | |
1163 | { | |
1164 | struct mm_struct *mm; | |
1165 | struct mem_cgroup *mem, *old_mem; | |
1166 | ||
1167 | mm = get_task_mm(p); | |
1168 | if (mm == NULL) | |
1169 | return; | |
1170 | ||
1171 | mem = mem_cgroup_from_cont(cont); | |
1172 | old_mem = mem_cgroup_from_cont(old_cont); | |
1173 | ||
67e465a7 BS |
1174 | /* |
1175 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1176 | * in effect owned by the leader | |
1177 | */ | |
52ea27eb | 1178 | if (!thread_group_leader(p)) |
67e465a7 BS |
1179 | goto out; |
1180 | ||
67e465a7 BS |
1181 | out: |
1182 | mmput(mm); | |
67e465a7 BS |
1183 | } |
1184 | ||
8cdea7c0 BS |
1185 | struct cgroup_subsys mem_cgroup_subsys = { |
1186 | .name = "memory", | |
1187 | .subsys_id = mem_cgroup_subsys_id, | |
1188 | .create = mem_cgroup_create, | |
df878fb0 | 1189 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1190 | .destroy = mem_cgroup_destroy, |
1191 | .populate = mem_cgroup_populate, | |
67e465a7 | 1192 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1193 | .early_init = 0, |
8cdea7c0 | 1194 | }; |