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> |
8cdea7c0 | 35 | |
8697d331 BS |
36 | #include <asm/uaccess.h> |
37 | ||
a181b0e8 KH |
38 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
39 | static struct kmem_cache *page_cgroup_cache __read_mostly; | |
40 | #define MEM_CGROUP_RECLAIM_RETRIES 5 | |
8cdea7c0 | 41 | |
d52aa412 KH |
42 | /* |
43 | * Statistics for memory cgroup. | |
44 | */ | |
45 | enum mem_cgroup_stat_index { | |
46 | /* | |
47 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
48 | */ | |
49 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
50 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
55e462b0 BR |
51 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
52 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
d52aa412 KH |
53 | |
54 | MEM_CGROUP_STAT_NSTATS, | |
55 | }; | |
56 | ||
57 | struct mem_cgroup_stat_cpu { | |
58 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
59 | } ____cacheline_aligned_in_smp; | |
60 | ||
61 | struct mem_cgroup_stat { | |
62 | struct mem_cgroup_stat_cpu cpustat[NR_CPUS]; | |
63 | }; | |
64 | ||
65 | /* | |
66 | * For accounting under irq disable, no need for increment preempt count. | |
67 | */ | |
68 | static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat, | |
69 | enum mem_cgroup_stat_index idx, int val) | |
70 | { | |
71 | int cpu = smp_processor_id(); | |
72 | stat->cpustat[cpu].count[idx] += val; | |
73 | } | |
74 | ||
75 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
76 | enum mem_cgroup_stat_index idx) | |
77 | { | |
78 | int cpu; | |
79 | s64 ret = 0; | |
80 | for_each_possible_cpu(cpu) | |
81 | ret += stat->cpustat[cpu].count[idx]; | |
82 | return ret; | |
83 | } | |
84 | ||
6d12e2d8 KH |
85 | /* |
86 | * per-zone information in memory controller. | |
87 | */ | |
88 | ||
89 | enum mem_cgroup_zstat_index { | |
90 | MEM_CGROUP_ZSTAT_ACTIVE, | |
91 | MEM_CGROUP_ZSTAT_INACTIVE, | |
92 | ||
93 | NR_MEM_CGROUP_ZSTAT, | |
94 | }; | |
95 | ||
96 | struct mem_cgroup_per_zone { | |
072c56c1 KH |
97 | /* |
98 | * spin_lock to protect the per cgroup LRU | |
99 | */ | |
100 | spinlock_t lru_lock; | |
1ecaab2b KH |
101 | struct list_head active_list; |
102 | struct list_head inactive_list; | |
6d12e2d8 KH |
103 | unsigned long count[NR_MEM_CGROUP_ZSTAT]; |
104 | }; | |
105 | /* Macro for accessing counter */ | |
106 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
107 | ||
108 | struct mem_cgroup_per_node { | |
109 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
110 | }; | |
111 | ||
112 | struct mem_cgroup_lru_info { | |
113 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
114 | }; | |
115 | ||
8cdea7c0 BS |
116 | /* |
117 | * The memory controller data structure. The memory controller controls both | |
118 | * page cache and RSS per cgroup. We would eventually like to provide | |
119 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
120 | * to help the administrator determine what knobs to tune. | |
121 | * | |
122 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
123 | * we hit the water mark. May be even add a low water mark, such that |
124 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
125 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
126 | */ |
127 | struct mem_cgroup { | |
128 | struct cgroup_subsys_state css; | |
129 | /* | |
130 | * the counter to account for memory usage | |
131 | */ | |
132 | struct res_counter res; | |
78fb7466 PE |
133 | /* |
134 | * Per cgroup active and inactive list, similar to the | |
135 | * per zone LRU lists. | |
78fb7466 | 136 | */ |
6d12e2d8 | 137 | struct mem_cgroup_lru_info info; |
072c56c1 | 138 | |
6c48a1d0 | 139 | int prev_priority; /* for recording reclaim priority */ |
d52aa412 KH |
140 | /* |
141 | * statistics. | |
142 | */ | |
143 | struct mem_cgroup_stat stat; | |
8cdea7c0 | 144 | }; |
8869b8f6 | 145 | static struct mem_cgroup init_mem_cgroup; |
8cdea7c0 | 146 | |
8a9f3ccd BS |
147 | /* |
148 | * We use the lower bit of the page->page_cgroup pointer as a bit spin | |
9442ec9d HD |
149 | * lock. We need to ensure that page->page_cgroup is at least two |
150 | * byte aligned (based on comments from Nick Piggin). But since | |
151 | * bit_spin_lock doesn't actually set that lock bit in a non-debug | |
152 | * uniprocessor kernel, we should avoid setting it here too. | |
8a9f3ccd BS |
153 | */ |
154 | #define PAGE_CGROUP_LOCK_BIT 0x0 | |
9442ec9d HD |
155 | #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) |
156 | #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT) | |
157 | #else | |
158 | #define PAGE_CGROUP_LOCK 0x0 | |
159 | #endif | |
8a9f3ccd | 160 | |
8cdea7c0 BS |
161 | /* |
162 | * A page_cgroup page is associated with every page descriptor. The | |
163 | * page_cgroup helps us identify information about the cgroup | |
164 | */ | |
165 | struct page_cgroup { | |
166 | struct list_head lru; /* per cgroup LRU list */ | |
167 | struct page *page; | |
168 | struct mem_cgroup *mem_cgroup; | |
8869b8f6 | 169 | int flags; |
8cdea7c0 | 170 | }; |
217bc319 | 171 | #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */ |
3564c7c4 | 172 | #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */ |
8cdea7c0 | 173 | |
d5b69e38 | 174 | static int page_cgroup_nid(struct page_cgroup *pc) |
c0149530 KH |
175 | { |
176 | return page_to_nid(pc->page); | |
177 | } | |
178 | ||
d5b69e38 | 179 | static enum zone_type page_cgroup_zid(struct page_cgroup *pc) |
c0149530 KH |
180 | { |
181 | return page_zonenum(pc->page); | |
182 | } | |
183 | ||
217bc319 KH |
184 | enum charge_type { |
185 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
186 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
69029cd5 | 187 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
217bc319 KH |
188 | }; |
189 | ||
d52aa412 KH |
190 | /* |
191 | * Always modified under lru lock. Then, not necessary to preempt_disable() | |
192 | */ | |
193 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags, | |
194 | bool charge) | |
195 | { | |
196 | int val = (charge)? 1 : -1; | |
197 | struct mem_cgroup_stat *stat = &mem->stat; | |
d52aa412 | 198 | |
8869b8f6 | 199 | VM_BUG_ON(!irqs_disabled()); |
d52aa412 | 200 | if (flags & PAGE_CGROUP_FLAG_CACHE) |
8869b8f6 | 201 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 KH |
202 | else |
203 | __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val); | |
55e462b0 BR |
204 | |
205 | if (charge) | |
206 | __mem_cgroup_stat_add_safe(stat, | |
207 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); | |
208 | else | |
209 | __mem_cgroup_stat_add_safe(stat, | |
210 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); | |
6d12e2d8 KH |
211 | } |
212 | ||
d5b69e38 | 213 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
214 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
215 | { | |
6d12e2d8 KH |
216 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
217 | } | |
218 | ||
d5b69e38 | 219 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
220 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
221 | { | |
222 | struct mem_cgroup *mem = pc->mem_cgroup; | |
223 | int nid = page_cgroup_nid(pc); | |
224 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 225 | |
6d12e2d8 KH |
226 | return mem_cgroup_zoneinfo(mem, nid, zid); |
227 | } | |
228 | ||
229 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
230 | enum mem_cgroup_zstat_index idx) | |
231 | { | |
232 | int nid, zid; | |
233 | struct mem_cgroup_per_zone *mz; | |
234 | u64 total = 0; | |
235 | ||
236 | for_each_online_node(nid) | |
237 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
238 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
239 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
240 | } | |
241 | return total; | |
d52aa412 KH |
242 | } |
243 | ||
d5b69e38 | 244 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
245 | { |
246 | return container_of(cgroup_subsys_state(cont, | |
247 | mem_cgroup_subsys_id), struct mem_cgroup, | |
248 | css); | |
249 | } | |
250 | ||
cf475ad2 | 251 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 252 | { |
31a78f23 BS |
253 | /* |
254 | * mm_update_next_owner() may clear mm->owner to NULL | |
255 | * if it races with swapoff, page migration, etc. | |
256 | * So this can be called with p == NULL. | |
257 | */ | |
258 | if (unlikely(!p)) | |
259 | return NULL; | |
260 | ||
78fb7466 PE |
261 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
262 | struct mem_cgroup, css); | |
263 | } | |
264 | ||
8a9f3ccd BS |
265 | static inline int page_cgroup_locked(struct page *page) |
266 | { | |
8869b8f6 | 267 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); |
8a9f3ccd BS |
268 | } |
269 | ||
9442ec9d | 270 | static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
78fb7466 | 271 | { |
9442ec9d HD |
272 | VM_BUG_ON(!page_cgroup_locked(page)); |
273 | page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK); | |
78fb7466 PE |
274 | } |
275 | ||
276 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
277 | { | |
8869b8f6 | 278 | return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK); |
8a9f3ccd BS |
279 | } |
280 | ||
d5b69e38 | 281 | static void lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
282 | { |
283 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
8a9f3ccd BS |
284 | } |
285 | ||
2680eed7 HD |
286 | static int try_lock_page_cgroup(struct page *page) |
287 | { | |
288 | return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
289 | } | |
290 | ||
d5b69e38 | 291 | static void unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
292 | { |
293 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
294 | } | |
295 | ||
3eae90c3 KH |
296 | static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, |
297 | struct page_cgroup *pc) | |
6d12e2d8 KH |
298 | { |
299 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
6d12e2d8 KH |
300 | |
301 | if (from) | |
302 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
303 | else | |
304 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
305 | ||
306 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false); | |
508b7be0 | 307 | list_del(&pc->lru); |
6d12e2d8 KH |
308 | } |
309 | ||
3eae90c3 KH |
310 | static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, |
311 | struct page_cgroup *pc) | |
6d12e2d8 KH |
312 | { |
313 | int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; | |
6d12e2d8 KH |
314 | |
315 | if (!to) { | |
316 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; | |
1ecaab2b | 317 | list_add(&pc->lru, &mz->inactive_list); |
6d12e2d8 KH |
318 | } else { |
319 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; | |
1ecaab2b | 320 | list_add(&pc->lru, &mz->active_list); |
6d12e2d8 KH |
321 | } |
322 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true); | |
323 | } | |
324 | ||
8697d331 | 325 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active) |
66e1707b | 326 | { |
6d12e2d8 KH |
327 | int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; |
328 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); | |
329 | ||
330 | if (from) | |
331 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1; | |
332 | else | |
333 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1; | |
334 | ||
3564c7c4 | 335 | if (active) { |
6d12e2d8 | 336 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1; |
3564c7c4 | 337 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 338 | list_move(&pc->lru, &mz->active_list); |
3564c7c4 | 339 | } else { |
6d12e2d8 | 340 | MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1; |
3564c7c4 | 341 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; |
1ecaab2b | 342 | list_move(&pc->lru, &mz->inactive_list); |
3564c7c4 | 343 | } |
66e1707b BS |
344 | } |
345 | ||
4c4a2214 DR |
346 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
347 | { | |
348 | int ret; | |
349 | ||
350 | task_lock(task); | |
bd845e38 | 351 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
352 | task_unlock(task); |
353 | return ret; | |
354 | } | |
355 | ||
66e1707b BS |
356 | /* |
357 | * This routine assumes that the appropriate zone's lru lock is already held | |
358 | */ | |
427d5416 | 359 | void mem_cgroup_move_lists(struct page *page, bool active) |
66e1707b | 360 | { |
427d5416 | 361 | struct page_cgroup *pc; |
072c56c1 KH |
362 | struct mem_cgroup_per_zone *mz; |
363 | unsigned long flags; | |
364 | ||
cede86ac LZ |
365 | if (mem_cgroup_subsys.disabled) |
366 | return; | |
367 | ||
2680eed7 HD |
368 | /* |
369 | * We cannot lock_page_cgroup while holding zone's lru_lock, | |
370 | * because other holders of lock_page_cgroup can be interrupted | |
371 | * with an attempt to rotate_reclaimable_page. But we cannot | |
372 | * safely get to page_cgroup without it, so just try_lock it: | |
373 | * mem_cgroup_isolate_pages allows for page left on wrong list. | |
374 | */ | |
375 | if (!try_lock_page_cgroup(page)) | |
66e1707b BS |
376 | return; |
377 | ||
2680eed7 HD |
378 | pc = page_get_page_cgroup(page); |
379 | if (pc) { | |
2680eed7 | 380 | mz = page_cgroup_zoneinfo(pc); |
2680eed7 | 381 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 382 | __mem_cgroup_move_lists(pc, active); |
2680eed7 | 383 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9b3c0a07 HT |
384 | } |
385 | unlock_page_cgroup(page); | |
66e1707b BS |
386 | } |
387 | ||
58ae83db KH |
388 | /* |
389 | * Calculate mapped_ratio under memory controller. This will be used in | |
390 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
391 | */ | |
392 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
393 | { | |
394 | long total, rss; | |
395 | ||
396 | /* | |
397 | * usage is recorded in bytes. But, here, we assume the number of | |
398 | * physical pages can be represented by "long" on any arch. | |
399 | */ | |
400 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
401 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
402 | return (int)((rss * 100L) / total); | |
403 | } | |
8869b8f6 | 404 | |
5932f367 KH |
405 | /* |
406 | * This function is called from vmscan.c. In page reclaiming loop. balance | |
407 | * between active and inactive list is calculated. For memory controller | |
408 | * page reclaiming, we should use using mem_cgroup's imbalance rather than | |
409 | * zone's global lru imbalance. | |
410 | */ | |
411 | long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem) | |
412 | { | |
413 | unsigned long active, inactive; | |
414 | /* active and inactive are the number of pages. 'long' is ok.*/ | |
415 | active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE); | |
416 | inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE); | |
417 | return (long) (active / (inactive + 1)); | |
418 | } | |
58ae83db | 419 | |
6c48a1d0 KH |
420 | /* |
421 | * prev_priority control...this will be used in memory reclaim path. | |
422 | */ | |
423 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
424 | { | |
425 | return mem->prev_priority; | |
426 | } | |
427 | ||
428 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
429 | { | |
430 | if (priority < mem->prev_priority) | |
431 | mem->prev_priority = priority; | |
432 | } | |
433 | ||
434 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
435 | { | |
436 | mem->prev_priority = priority; | |
437 | } | |
438 | ||
cc38108e KH |
439 | /* |
440 | * Calculate # of pages to be scanned in this priority/zone. | |
441 | * See also vmscan.c | |
442 | * | |
443 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
444 | * (see include/linux/mmzone.h) | |
445 | */ | |
446 | ||
447 | long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem, | |
448 | struct zone *zone, int priority) | |
449 | { | |
450 | long nr_active; | |
451 | int nid = zone->zone_pgdat->node_id; | |
452 | int zid = zone_idx(zone); | |
453 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
454 | ||
455 | nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE); | |
456 | return (nr_active >> priority); | |
457 | } | |
458 | ||
459 | long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem, | |
460 | struct zone *zone, int priority) | |
461 | { | |
462 | long nr_inactive; | |
463 | int nid = zone->zone_pgdat->node_id; | |
464 | int zid = zone_idx(zone); | |
465 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
466 | ||
467 | nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE); | |
cc38108e KH |
468 | return (nr_inactive >> priority); |
469 | } | |
470 | ||
66e1707b BS |
471 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
472 | struct list_head *dst, | |
473 | unsigned long *scanned, int order, | |
474 | int mode, struct zone *z, | |
475 | struct mem_cgroup *mem_cont, | |
476 | int active) | |
477 | { | |
478 | unsigned long nr_taken = 0; | |
479 | struct page *page; | |
480 | unsigned long scan; | |
481 | LIST_HEAD(pc_list); | |
482 | struct list_head *src; | |
ff7283fa | 483 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
484 | int nid = z->zone_pgdat->node_id; |
485 | int zid = zone_idx(z); | |
486 | struct mem_cgroup_per_zone *mz; | |
66e1707b | 487 | |
cf475ad2 | 488 | BUG_ON(!mem_cont); |
1ecaab2b | 489 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
66e1707b | 490 | if (active) |
1ecaab2b | 491 | src = &mz->active_list; |
66e1707b | 492 | else |
1ecaab2b KH |
493 | src = &mz->inactive_list; |
494 | ||
66e1707b | 495 | |
072c56c1 | 496 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
497 | scan = 0; |
498 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 499 | if (scan >= nr_to_scan) |
ff7283fa | 500 | break; |
66e1707b | 501 | page = pc->page; |
66e1707b | 502 | |
436c6541 | 503 | if (unlikely(!PageLRU(page))) |
ff7283fa | 504 | continue; |
ff7283fa | 505 | |
66e1707b BS |
506 | if (PageActive(page) && !active) { |
507 | __mem_cgroup_move_lists(pc, true); | |
66e1707b BS |
508 | continue; |
509 | } | |
510 | if (!PageActive(page) && active) { | |
511 | __mem_cgroup_move_lists(pc, false); | |
66e1707b BS |
512 | continue; |
513 | } | |
514 | ||
436c6541 HD |
515 | scan++; |
516 | list_move(&pc->lru, &pc_list); | |
66e1707b BS |
517 | |
518 | if (__isolate_lru_page(page, mode) == 0) { | |
519 | list_move(&page->lru, dst); | |
520 | nr_taken++; | |
521 | } | |
522 | } | |
523 | ||
524 | list_splice(&pc_list, src); | |
072c56c1 | 525 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
526 | |
527 | *scanned = scan; | |
528 | return nr_taken; | |
529 | } | |
530 | ||
8a9f3ccd BS |
531 | /* |
532 | * Charge the memory controller for page usage. | |
533 | * Return | |
534 | * 0 if the charge was successful | |
535 | * < 0 if the cgroup is over its limit | |
536 | */ | |
217bc319 | 537 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
e8589cc1 KH |
538 | gfp_t gfp_mask, enum charge_type ctype, |
539 | struct mem_cgroup *memcg) | |
8a9f3ccd BS |
540 | { |
541 | struct mem_cgroup *mem; | |
9175e031 | 542 | struct page_cgroup *pc; |
66e1707b BS |
543 | unsigned long flags; |
544 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
072c56c1 | 545 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 546 | |
508b7be0 | 547 | pc = kmem_cache_alloc(page_cgroup_cache, gfp_mask); |
b76734e5 | 548 | if (unlikely(pc == NULL)) |
8a9f3ccd BS |
549 | goto err; |
550 | ||
8a9f3ccd | 551 | /* |
3be91277 HD |
552 | * We always charge the cgroup the mm_struct belongs to. |
553 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
554 | * thread group leader migrates. It's possible that mm is not |
555 | * set, if so charge the init_mm (happens for pagecache usage). | |
556 | */ | |
69029cd5 | 557 | if (likely(!memcg)) { |
e8589cc1 KH |
558 | rcu_read_lock(); |
559 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
560 | if (unlikely(!mem)) { |
561 | rcu_read_unlock(); | |
562 | kmem_cache_free(page_cgroup_cache, pc); | |
563 | return 0; | |
564 | } | |
e8589cc1 KH |
565 | /* |
566 | * For every charge from the cgroup, increment reference count | |
567 | */ | |
568 | css_get(&mem->css); | |
569 | rcu_read_unlock(); | |
570 | } else { | |
571 | mem = memcg; | |
572 | css_get(&memcg->css); | |
573 | } | |
8a9f3ccd | 574 | |
0eea1030 | 575 | while (res_counter_charge(&mem->res, PAGE_SIZE)) { |
3be91277 HD |
576 | if (!(gfp_mask & __GFP_WAIT)) |
577 | goto out; | |
e1a1cd59 BS |
578 | |
579 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
580 | continue; |
581 | ||
582 | /* | |
8869b8f6 HD |
583 | * try_to_free_mem_cgroup_pages() might not give us a full |
584 | * picture of reclaim. Some pages are reclaimed and might be | |
585 | * moved to swap cache or just unmapped from the cgroup. | |
586 | * Check the limit again to see if the reclaim reduced the | |
587 | * current usage of the cgroup before giving up | |
588 | */ | |
66e1707b BS |
589 | if (res_counter_check_under_limit(&mem->res)) |
590 | continue; | |
3be91277 HD |
591 | |
592 | if (!nr_retries--) { | |
593 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
594 | goto out; | |
66e1707b | 595 | } |
8a9f3ccd BS |
596 | } |
597 | ||
8a9f3ccd BS |
598 | pc->mem_cgroup = mem; |
599 | pc->page = page; | |
508b7be0 KH |
600 | /* |
601 | * If a page is accounted as a page cache, insert to inactive list. | |
602 | * If anon, insert to active list. | |
603 | */ | |
217bc319 | 604 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) |
4a56d02e | 605 | pc->flags = PAGE_CGROUP_FLAG_CACHE; |
508b7be0 KH |
606 | else |
607 | pc->flags = PAGE_CGROUP_FLAG_ACTIVE; | |
3be91277 | 608 | |
7e924aaf | 609 | lock_page_cgroup(page); |
b76734e5 | 610 | if (unlikely(page_get_page_cgroup(page))) { |
7e924aaf | 611 | unlock_page_cgroup(page); |
9175e031 KH |
612 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
613 | css_put(&mem->css); | |
b6ac57d5 | 614 | kmem_cache_free(page_cgroup_cache, pc); |
accf163e | 615 | goto done; |
9175e031 | 616 | } |
7e924aaf | 617 | page_assign_page_cgroup(page, pc); |
8a9f3ccd | 618 | |
072c56c1 KH |
619 | mz = page_cgroup_zoneinfo(pc); |
620 | spin_lock_irqsave(&mz->lru_lock, flags); | |
3eae90c3 | 621 | __mem_cgroup_add_list(mz, pc); |
072c56c1 | 622 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b | 623 | |
fb59e9f1 | 624 | unlock_page_cgroup(page); |
8a9f3ccd | 625 | done: |
8a9f3ccd | 626 | return 0; |
3be91277 HD |
627 | out: |
628 | css_put(&mem->css); | |
b6ac57d5 | 629 | kmem_cache_free(page_cgroup_cache, pc); |
8a9f3ccd | 630 | err: |
8a9f3ccd BS |
631 | return -ENOMEM; |
632 | } | |
633 | ||
8869b8f6 | 634 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
217bc319 | 635 | { |
cede86ac LZ |
636 | if (mem_cgroup_subsys.disabled) |
637 | return 0; | |
638 | ||
69029cd5 KH |
639 | /* |
640 | * If already mapped, we don't have to account. | |
641 | * If page cache, page->mapping has address_space. | |
642 | * But page->mapping may have out-of-use anon_vma pointer, | |
643 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
644 | * is NULL. | |
645 | */ | |
646 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
647 | return 0; | |
648 | if (unlikely(!mm)) | |
649 | mm = &init_mm; | |
217bc319 | 650 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 651 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
652 | } |
653 | ||
e1a1cd59 BS |
654 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
655 | gfp_t gfp_mask) | |
8697d331 | 656 | { |
cede86ac LZ |
657 | if (mem_cgroup_subsys.disabled) |
658 | return 0; | |
659 | ||
accf163e KH |
660 | /* |
661 | * Corner case handling. This is called from add_to_page_cache() | |
662 | * in usual. But some FS (shmem) precharges this page before calling it | |
663 | * and call add_to_page_cache() with GFP_NOWAIT. | |
664 | * | |
665 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
666 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
667 | * charge twice. (It works but has to pay a bit larger cost.) | |
668 | */ | |
669 | if (!(gfp_mask & __GFP_WAIT)) { | |
670 | struct page_cgroup *pc; | |
671 | ||
672 | lock_page_cgroup(page); | |
673 | pc = page_get_page_cgroup(page); | |
674 | if (pc) { | |
675 | VM_BUG_ON(pc->page != page); | |
676 | VM_BUG_ON(!pc->mem_cgroup); | |
677 | unlock_page_cgroup(page); | |
678 | return 0; | |
679 | } | |
680 | unlock_page_cgroup(page); | |
681 | } | |
682 | ||
69029cd5 | 683 | if (unlikely(!mm)) |
8697d331 | 684 | mm = &init_mm; |
accf163e | 685 | |
8869b8f6 | 686 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 KH |
687 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
688 | } | |
689 | ||
8a9f3ccd | 690 | /* |
69029cd5 | 691 | * uncharge if !page_mapped(page) |
8a9f3ccd | 692 | */ |
69029cd5 KH |
693 | static void |
694 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) | |
8a9f3ccd | 695 | { |
8289546e | 696 | struct page_cgroup *pc; |
8a9f3ccd | 697 | struct mem_cgroup *mem; |
072c56c1 | 698 | struct mem_cgroup_per_zone *mz; |
66e1707b | 699 | unsigned long flags; |
8a9f3ccd | 700 | |
4077960e BS |
701 | if (mem_cgroup_subsys.disabled) |
702 | return; | |
703 | ||
8697d331 | 704 | /* |
3c541e14 | 705 | * Check if our page_cgroup is valid |
8697d331 | 706 | */ |
8289546e HD |
707 | lock_page_cgroup(page); |
708 | pc = page_get_page_cgroup(page); | |
b76734e5 | 709 | if (unlikely(!pc)) |
8289546e | 710 | goto unlock; |
8a9f3ccd | 711 | |
b9c565d5 | 712 | VM_BUG_ON(pc->page != page); |
b9c565d5 | 713 | |
69029cd5 KH |
714 | if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
715 | && ((pc->flags & PAGE_CGROUP_FLAG_CACHE) | |
716 | || page_mapped(page))) | |
717 | goto unlock; | |
b9c565d5 | 718 | |
69029cd5 KH |
719 | mz = page_cgroup_zoneinfo(pc); |
720 | spin_lock_irqsave(&mz->lru_lock, flags); | |
721 | __mem_cgroup_remove_list(mz, pc); | |
722 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
fb59e9f1 | 723 | |
69029cd5 KH |
724 | page_assign_page_cgroup(page, NULL); |
725 | unlock_page_cgroup(page); | |
6d48ff8b | 726 | |
69029cd5 KH |
727 | mem = pc->mem_cgroup; |
728 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
729 | css_put(&mem->css); | |
6d12e2d8 | 730 | |
69029cd5 KH |
731 | kmem_cache_free(page_cgroup_cache, pc); |
732 | return; | |
8289546e | 733 | unlock: |
3c541e14 BS |
734 | unlock_page_cgroup(page); |
735 | } | |
736 | ||
69029cd5 KH |
737 | void mem_cgroup_uncharge_page(struct page *page) |
738 | { | |
739 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
740 | } | |
741 | ||
742 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
743 | { | |
744 | VM_BUG_ON(page_mapped(page)); | |
745 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); | |
746 | } | |
747 | ||
ae41be37 | 748 | /* |
e8589cc1 | 749 | * Before starting migration, account against new page. |
ae41be37 | 750 | */ |
e8589cc1 | 751 | int mem_cgroup_prepare_migration(struct page *page, struct page *newpage) |
ae41be37 KH |
752 | { |
753 | struct page_cgroup *pc; | |
e8589cc1 KH |
754 | struct mem_cgroup *mem = NULL; |
755 | enum charge_type ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
756 | int ret = 0; | |
8869b8f6 | 757 | |
4077960e BS |
758 | if (mem_cgroup_subsys.disabled) |
759 | return 0; | |
760 | ||
ae41be37 KH |
761 | lock_page_cgroup(page); |
762 | pc = page_get_page_cgroup(page); | |
e8589cc1 KH |
763 | if (pc) { |
764 | mem = pc->mem_cgroup; | |
765 | css_get(&mem->css); | |
766 | if (pc->flags & PAGE_CGROUP_FLAG_CACHE) | |
767 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
768 | } | |
ae41be37 | 769 | unlock_page_cgroup(page); |
e8589cc1 KH |
770 | if (mem) { |
771 | ret = mem_cgroup_charge_common(newpage, NULL, GFP_KERNEL, | |
772 | ctype, mem); | |
773 | css_put(&mem->css); | |
774 | } | |
775 | return ret; | |
ae41be37 | 776 | } |
8869b8f6 | 777 | |
69029cd5 | 778 | /* remove redundant charge if migration failed*/ |
e8589cc1 | 779 | void mem_cgroup_end_migration(struct page *newpage) |
ae41be37 | 780 | { |
69029cd5 KH |
781 | /* |
782 | * At success, page->mapping is not NULL. | |
783 | * special rollback care is necessary when | |
784 | * 1. at migration failure. (newpage->mapping is cleared in this case) | |
785 | * 2. the newpage was moved but not remapped again because the task | |
786 | * exits and the newpage is obsolete. In this case, the new page | |
787 | * may be a swapcache. So, we just call mem_cgroup_uncharge_page() | |
788 | * always for avoiding mess. The page_cgroup will be removed if | |
789 | * unnecessary. File cache pages is still on radix-tree. Don't | |
790 | * care it. | |
791 | */ | |
792 | if (!newpage->mapping) | |
793 | __mem_cgroup_uncharge_common(newpage, | |
794 | MEM_CGROUP_CHARGE_TYPE_FORCE); | |
795 | else if (PageAnon(newpage)) | |
796 | mem_cgroup_uncharge_page(newpage); | |
ae41be37 | 797 | } |
78fb7466 | 798 | |
c9b0ed51 KH |
799 | /* |
800 | * A call to try to shrink memory usage under specified resource controller. | |
801 | * This is typically used for page reclaiming for shmem for reducing side | |
802 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
803 | */ | |
804 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) | |
805 | { | |
806 | struct mem_cgroup *mem; | |
807 | int progress = 0; | |
808 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
809 | ||
cede86ac LZ |
810 | if (mem_cgroup_subsys.disabled) |
811 | return 0; | |
9623e078 HD |
812 | if (!mm) |
813 | return 0; | |
cede86ac | 814 | |
c9b0ed51 KH |
815 | rcu_read_lock(); |
816 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
817 | if (unlikely(!mem)) { |
818 | rcu_read_unlock(); | |
819 | return 0; | |
820 | } | |
c9b0ed51 KH |
821 | css_get(&mem->css); |
822 | rcu_read_unlock(); | |
823 | ||
824 | do { | |
825 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask); | |
a10cebf5 | 826 | progress += res_counter_check_under_limit(&mem->res); |
c9b0ed51 KH |
827 | } while (!progress && --retry); |
828 | ||
829 | css_put(&mem->css); | |
830 | if (!retry) | |
831 | return -ENOMEM; | |
832 | return 0; | |
833 | } | |
834 | ||
628f4235 KH |
835 | int mem_cgroup_resize_limit(struct mem_cgroup *memcg, unsigned long long val) |
836 | { | |
837 | ||
838 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
839 | int progress; | |
840 | int ret = 0; | |
841 | ||
842 | while (res_counter_set_limit(&memcg->res, val)) { | |
843 | if (signal_pending(current)) { | |
844 | ret = -EINTR; | |
845 | break; | |
846 | } | |
847 | if (!retry_count) { | |
848 | ret = -EBUSY; | |
849 | break; | |
850 | } | |
851 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL); | |
852 | if (!progress) | |
853 | retry_count--; | |
854 | } | |
855 | return ret; | |
856 | } | |
857 | ||
858 | ||
cc847582 KH |
859 | /* |
860 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
861 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
862 | */ | |
863 | #define FORCE_UNCHARGE_BATCH (128) | |
8869b8f6 | 864 | static void mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 KH |
865 | struct mem_cgroup_per_zone *mz, |
866 | int active) | |
cc847582 KH |
867 | { |
868 | struct page_cgroup *pc; | |
869 | struct page *page; | |
9b3c0a07 | 870 | int count = FORCE_UNCHARGE_BATCH; |
cc847582 | 871 | unsigned long flags; |
072c56c1 KH |
872 | struct list_head *list; |
873 | ||
874 | if (active) | |
875 | list = &mz->active_list; | |
876 | else | |
877 | list = &mz->inactive_list; | |
cc847582 | 878 | |
072c56c1 | 879 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 880 | while (!list_empty(list)) { |
cc847582 KH |
881 | pc = list_entry(list->prev, struct page_cgroup, lru); |
882 | page = pc->page; | |
9b3c0a07 HT |
883 | get_page(page); |
884 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
e8589cc1 KH |
885 | /* |
886 | * Check if this page is on LRU. !LRU page can be found | |
887 | * if it's under page migration. | |
888 | */ | |
889 | if (PageLRU(page)) { | |
69029cd5 KH |
890 | __mem_cgroup_uncharge_common(page, |
891 | MEM_CGROUP_CHARGE_TYPE_FORCE); | |
e8589cc1 KH |
892 | put_page(page); |
893 | if (--count <= 0) { | |
894 | count = FORCE_UNCHARGE_BATCH; | |
895 | cond_resched(); | |
896 | } | |
897 | } else | |
9b3c0a07 | 898 | cond_resched(); |
9b3c0a07 | 899 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 | 900 | } |
072c56c1 | 901 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
902 | } |
903 | ||
904 | /* | |
905 | * make mem_cgroup's charge to be 0 if there is no task. | |
906 | * This enables deleting this mem_cgroup. | |
907 | */ | |
d5b69e38 | 908 | static int mem_cgroup_force_empty(struct mem_cgroup *mem) |
cc847582 KH |
909 | { |
910 | int ret = -EBUSY; | |
1ecaab2b | 911 | int node, zid; |
8869b8f6 | 912 | |
cc847582 KH |
913 | css_get(&mem->css); |
914 | /* | |
915 | * page reclaim code (kswapd etc..) will move pages between | |
8869b8f6 | 916 | * active_list <-> inactive_list while we don't take a lock. |
cc847582 KH |
917 | * So, we have to do loop here until all lists are empty. |
918 | */ | |
1ecaab2b | 919 | while (mem->res.usage > 0) { |
cc847582 KH |
920 | if (atomic_read(&mem->css.cgroup->count) > 0) |
921 | goto out; | |
1ecaab2b KH |
922 | for_each_node_state(node, N_POSSIBLE) |
923 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
924 | struct mem_cgroup_per_zone *mz; | |
925 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
926 | /* drop all page_cgroup in active_list */ | |
072c56c1 | 927 | mem_cgroup_force_empty_list(mem, mz, 1); |
1ecaab2b | 928 | /* drop all page_cgroup in inactive_list */ |
072c56c1 | 929 | mem_cgroup_force_empty_list(mem, mz, 0); |
1ecaab2b | 930 | } |
cc847582 KH |
931 | } |
932 | ret = 0; | |
933 | out: | |
934 | css_put(&mem->css); | |
935 | return ret; | |
936 | } | |
937 | ||
2c3daa72 | 938 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 939 | { |
2c3daa72 PM |
940 | return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res, |
941 | cft->private); | |
8cdea7c0 | 942 | } |
628f4235 KH |
943 | /* |
944 | * The user of this function is... | |
945 | * RES_LIMIT. | |
946 | */ | |
856c13aa PM |
947 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
948 | const char *buffer) | |
8cdea7c0 | 949 | { |
628f4235 KH |
950 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
951 | unsigned long long val; | |
952 | int ret; | |
953 | ||
954 | switch (cft->private) { | |
955 | case RES_LIMIT: | |
956 | /* This function does all necessary parse...reuse it */ | |
957 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
958 | if (!ret) | |
959 | ret = mem_cgroup_resize_limit(memcg, val); | |
960 | break; | |
961 | default: | |
962 | ret = -EINVAL; /* should be BUG() ? */ | |
963 | break; | |
964 | } | |
965 | return ret; | |
8cdea7c0 BS |
966 | } |
967 | ||
29f2a4da | 968 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
969 | { |
970 | struct mem_cgroup *mem; | |
971 | ||
972 | mem = mem_cgroup_from_cont(cont); | |
29f2a4da PE |
973 | switch (event) { |
974 | case RES_MAX_USAGE: | |
975 | res_counter_reset_max(&mem->res); | |
976 | break; | |
977 | case RES_FAILCNT: | |
978 | res_counter_reset_failcnt(&mem->res); | |
979 | break; | |
980 | } | |
85cc59db | 981 | return 0; |
c84872e1 PE |
982 | } |
983 | ||
85cc59db | 984 | static int mem_force_empty_write(struct cgroup *cont, unsigned int event) |
cc847582 | 985 | { |
85cc59db | 986 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont)); |
cc847582 KH |
987 | } |
988 | ||
d2ceb9b7 KH |
989 | static const struct mem_cgroup_stat_desc { |
990 | const char *msg; | |
991 | u64 unit; | |
992 | } mem_cgroup_stat_desc[] = { | |
993 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
994 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
995 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
996 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
997 | }; |
998 | ||
c64745cf PM |
999 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
1000 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 1001 | { |
d2ceb9b7 KH |
1002 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
1003 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
1004 | int i; | |
1005 | ||
1006 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1007 | s64 val; | |
1008 | ||
1009 | val = mem_cgroup_read_stat(stat, i); | |
1010 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 1011 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 1012 | } |
6d12e2d8 KH |
1013 | /* showing # of active pages */ |
1014 | { | |
1015 | unsigned long active, inactive; | |
1016 | ||
1017 | inactive = mem_cgroup_get_all_zonestat(mem_cont, | |
1018 | MEM_CGROUP_ZSTAT_INACTIVE); | |
1019 | active = mem_cgroup_get_all_zonestat(mem_cont, | |
1020 | MEM_CGROUP_ZSTAT_ACTIVE); | |
c64745cf PM |
1021 | cb->fill(cb, "active", (active) * PAGE_SIZE); |
1022 | cb->fill(cb, "inactive", (inactive) * PAGE_SIZE); | |
6d12e2d8 | 1023 | } |
d2ceb9b7 KH |
1024 | return 0; |
1025 | } | |
1026 | ||
8cdea7c0 BS |
1027 | static struct cftype mem_cgroup_files[] = { |
1028 | { | |
0eea1030 | 1029 | .name = "usage_in_bytes", |
8cdea7c0 | 1030 | .private = RES_USAGE, |
2c3daa72 | 1031 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1032 | }, |
c84872e1 PE |
1033 | { |
1034 | .name = "max_usage_in_bytes", | |
1035 | .private = RES_MAX_USAGE, | |
29f2a4da | 1036 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
1037 | .read_u64 = mem_cgroup_read, |
1038 | }, | |
8cdea7c0 | 1039 | { |
0eea1030 | 1040 | .name = "limit_in_bytes", |
8cdea7c0 | 1041 | .private = RES_LIMIT, |
856c13aa | 1042 | .write_string = mem_cgroup_write, |
2c3daa72 | 1043 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
1044 | }, |
1045 | { | |
1046 | .name = "failcnt", | |
1047 | .private = RES_FAILCNT, | |
29f2a4da | 1048 | .trigger = mem_cgroup_reset, |
2c3daa72 | 1049 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1050 | }, |
cc847582 KH |
1051 | { |
1052 | .name = "force_empty", | |
85cc59db | 1053 | .trigger = mem_force_empty_write, |
cc847582 | 1054 | }, |
d2ceb9b7 KH |
1055 | { |
1056 | .name = "stat", | |
c64745cf | 1057 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 1058 | }, |
8cdea7c0 BS |
1059 | }; |
1060 | ||
6d12e2d8 KH |
1061 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1062 | { | |
1063 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 1064 | struct mem_cgroup_per_zone *mz; |
41e3355d | 1065 | int zone, tmp = node; |
1ecaab2b KH |
1066 | /* |
1067 | * This routine is called against possible nodes. | |
1068 | * But it's BUG to call kmalloc() against offline node. | |
1069 | * | |
1070 | * TODO: this routine can waste much memory for nodes which will | |
1071 | * never be onlined. It's better to use memory hotplug callback | |
1072 | * function. | |
1073 | */ | |
41e3355d KH |
1074 | if (!node_state(node, N_NORMAL_MEMORY)) |
1075 | tmp = -1; | |
1076 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
1077 | if (!pn) |
1078 | return 1; | |
1ecaab2b | 1079 | |
6d12e2d8 KH |
1080 | mem->info.nodeinfo[node] = pn; |
1081 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1082 | |
1083 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1084 | mz = &pn->zoneinfo[zone]; | |
1085 | INIT_LIST_HEAD(&mz->active_list); | |
1086 | INIT_LIST_HEAD(&mz->inactive_list); | |
072c56c1 | 1087 | spin_lock_init(&mz->lru_lock); |
1ecaab2b | 1088 | } |
6d12e2d8 KH |
1089 | return 0; |
1090 | } | |
1091 | ||
1ecaab2b KH |
1092 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1093 | { | |
1094 | kfree(mem->info.nodeinfo[node]); | |
1095 | } | |
1096 | ||
33327948 KH |
1097 | static struct mem_cgroup *mem_cgroup_alloc(void) |
1098 | { | |
1099 | struct mem_cgroup *mem; | |
1100 | ||
1101 | if (sizeof(*mem) < PAGE_SIZE) | |
1102 | mem = kmalloc(sizeof(*mem), GFP_KERNEL); | |
1103 | else | |
1104 | mem = vmalloc(sizeof(*mem)); | |
1105 | ||
1106 | if (mem) | |
1107 | memset(mem, 0, sizeof(*mem)); | |
1108 | return mem; | |
1109 | } | |
1110 | ||
1111 | static void mem_cgroup_free(struct mem_cgroup *mem) | |
1112 | { | |
1113 | if (sizeof(*mem) < PAGE_SIZE) | |
1114 | kfree(mem); | |
1115 | else | |
1116 | vfree(mem); | |
1117 | } | |
1118 | ||
1119 | ||
8cdea7c0 BS |
1120 | static struct cgroup_subsys_state * |
1121 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1122 | { | |
1123 | struct mem_cgroup *mem; | |
6d12e2d8 | 1124 | int node; |
8cdea7c0 | 1125 | |
b6ac57d5 | 1126 | if (unlikely((cont->parent) == NULL)) { |
78fb7466 | 1127 | mem = &init_mem_cgroup; |
b6ac57d5 BS |
1128 | page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC); |
1129 | } else { | |
33327948 KH |
1130 | mem = mem_cgroup_alloc(); |
1131 | if (!mem) | |
1132 | return ERR_PTR(-ENOMEM); | |
b6ac57d5 | 1133 | } |
78fb7466 | 1134 | |
8cdea7c0 | 1135 | res_counter_init(&mem->res); |
1ecaab2b | 1136 | |
6d12e2d8 KH |
1137 | for_each_node_state(node, N_POSSIBLE) |
1138 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1139 | goto free_out; | |
1140 | ||
8cdea7c0 | 1141 | return &mem->css; |
6d12e2d8 KH |
1142 | free_out: |
1143 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1144 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1145 | if (cont->parent != NULL) |
33327948 | 1146 | mem_cgroup_free(mem); |
2dda81ca | 1147 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1148 | } |
1149 | ||
df878fb0 KH |
1150 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1151 | struct cgroup *cont) | |
1152 | { | |
1153 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1154 | mem_cgroup_force_empty(mem); | |
1155 | } | |
1156 | ||
8cdea7c0 BS |
1157 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1158 | struct cgroup *cont) | |
1159 | { | |
6d12e2d8 KH |
1160 | int node; |
1161 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1162 | ||
1163 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1164 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1165 | |
33327948 | 1166 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
1167 | } |
1168 | ||
1169 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1170 | struct cgroup *cont) | |
1171 | { | |
1172 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1173 | ARRAY_SIZE(mem_cgroup_files)); | |
1174 | } | |
1175 | ||
67e465a7 BS |
1176 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1177 | struct cgroup *cont, | |
1178 | struct cgroup *old_cont, | |
1179 | struct task_struct *p) | |
1180 | { | |
1181 | struct mm_struct *mm; | |
1182 | struct mem_cgroup *mem, *old_mem; | |
1183 | ||
1184 | mm = get_task_mm(p); | |
1185 | if (mm == NULL) | |
1186 | return; | |
1187 | ||
1188 | mem = mem_cgroup_from_cont(cont); | |
1189 | old_mem = mem_cgroup_from_cont(old_cont); | |
1190 | ||
67e465a7 BS |
1191 | /* |
1192 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1193 | * in effect owned by the leader | |
1194 | */ | |
52ea27eb | 1195 | if (!thread_group_leader(p)) |
67e465a7 BS |
1196 | goto out; |
1197 | ||
67e465a7 BS |
1198 | out: |
1199 | mmput(mm); | |
67e465a7 BS |
1200 | } |
1201 | ||
8cdea7c0 BS |
1202 | struct cgroup_subsys mem_cgroup_subsys = { |
1203 | .name = "memory", | |
1204 | .subsys_id = mem_cgroup_subsys_id, | |
1205 | .create = mem_cgroup_create, | |
df878fb0 | 1206 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1207 | .destroy = mem_cgroup_destroy, |
1208 | .populate = mem_cgroup_populate, | |
67e465a7 | 1209 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1210 | .early_init = 0, |
8cdea7c0 | 1211 | }; |