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 | */ | |
addb9efe | 69 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
70 | enum mem_cgroup_stat_index idx, int val) |
71 | { | |
addb9efe | 72 | stat->count[idx] += val; |
d52aa412 KH |
73 | } |
74 | ||
75 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
76 | enum mem_cgroup_stat_index idx) | |
77 | { | |
78 | int cpu; | |
79 | s64 ret = 0; | |
80 | for_each_possible_cpu(cpu) | |
81 | ret += stat->cpustat[cpu].count[idx]; | |
82 | return ret; | |
83 | } | |
84 | ||
6d12e2d8 KH |
85 | /* |
86 | * per-zone information in memory controller. | |
87 | */ | |
6d12e2d8 | 88 | struct mem_cgroup_per_zone { |
072c56c1 KH |
89 | /* |
90 | * spin_lock to protect the per cgroup LRU | |
91 | */ | |
92 | spinlock_t lru_lock; | |
b69408e8 CL |
93 | struct list_head lists[NR_LRU_LISTS]; |
94 | unsigned long count[NR_LRU_LISTS]; | |
6d12e2d8 KH |
95 | }; |
96 | /* Macro for accessing counter */ | |
97 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
98 | ||
99 | struct mem_cgroup_per_node { | |
100 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
101 | }; | |
102 | ||
103 | struct mem_cgroup_lru_info { | |
104 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
105 | }; | |
106 | ||
8cdea7c0 BS |
107 | /* |
108 | * The memory controller data structure. The memory controller controls both | |
109 | * page cache and RSS per cgroup. We would eventually like to provide | |
110 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
111 | * to help the administrator determine what knobs to tune. | |
112 | * | |
113 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
114 | * we hit the water mark. May be even add a low water mark, such that |
115 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
116 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
117 | */ |
118 | struct mem_cgroup { | |
119 | struct cgroup_subsys_state css; | |
120 | /* | |
121 | * the counter to account for memory usage | |
122 | */ | |
123 | struct res_counter res; | |
78fb7466 PE |
124 | /* |
125 | * Per cgroup active and inactive list, similar to the | |
126 | * per zone LRU lists. | |
78fb7466 | 127 | */ |
6d12e2d8 | 128 | struct mem_cgroup_lru_info info; |
072c56c1 | 129 | |
6c48a1d0 | 130 | int prev_priority; /* for recording reclaim priority */ |
d52aa412 KH |
131 | /* |
132 | * statistics. | |
133 | */ | |
134 | struct mem_cgroup_stat stat; | |
8cdea7c0 | 135 | }; |
8869b8f6 | 136 | static struct mem_cgroup init_mem_cgroup; |
8cdea7c0 | 137 | |
8a9f3ccd BS |
138 | /* |
139 | * We use the lower bit of the page->page_cgroup pointer as a bit spin | |
9442ec9d HD |
140 | * lock. We need to ensure that page->page_cgroup is at least two |
141 | * byte aligned (based on comments from Nick Piggin). But since | |
142 | * bit_spin_lock doesn't actually set that lock bit in a non-debug | |
143 | * uniprocessor kernel, we should avoid setting it here too. | |
8a9f3ccd BS |
144 | */ |
145 | #define PAGE_CGROUP_LOCK_BIT 0x0 | |
9442ec9d HD |
146 | #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) |
147 | #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT) | |
148 | #else | |
149 | #define PAGE_CGROUP_LOCK 0x0 | |
150 | #endif | |
8a9f3ccd | 151 | |
8cdea7c0 BS |
152 | /* |
153 | * A page_cgroup page is associated with every page descriptor. The | |
154 | * page_cgroup helps us identify information about the cgroup | |
155 | */ | |
156 | struct page_cgroup { | |
157 | struct list_head lru; /* per cgroup LRU list */ | |
158 | struct page *page; | |
159 | struct mem_cgroup *mem_cgroup; | |
8869b8f6 | 160 | int flags; |
8cdea7c0 | 161 | }; |
894bc310 LS |
162 | #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */ |
163 | #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */ | |
164 | #define PAGE_CGROUP_FLAG_FILE (0x4) /* page is file system backed */ | |
165 | #define PAGE_CGROUP_FLAG_UNEVICTABLE (0x8) /* page is unevictableable */ | |
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; | |
addb9efe | 192 | struct mem_cgroup_stat_cpu *cpustat; |
d52aa412 | 193 | |
8869b8f6 | 194 | VM_BUG_ON(!irqs_disabled()); |
addb9efe KH |
195 | |
196 | cpustat = &stat->cpustat[smp_processor_id()]; | |
d52aa412 | 197 | if (flags & PAGE_CGROUP_FLAG_CACHE) |
addb9efe | 198 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 199 | else |
addb9efe | 200 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
201 | |
202 | if (charge) | |
addb9efe | 203 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
204 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
205 | else | |
addb9efe | 206 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 207 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
6d12e2d8 KH |
208 | } |
209 | ||
d5b69e38 | 210 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
211 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
212 | { | |
6d12e2d8 KH |
213 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
214 | } | |
215 | ||
d5b69e38 | 216 | static struct mem_cgroup_per_zone * |
6d12e2d8 KH |
217 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
218 | { | |
219 | struct mem_cgroup *mem = pc->mem_cgroup; | |
220 | int nid = page_cgroup_nid(pc); | |
221 | int zid = page_cgroup_zid(pc); | |
d52aa412 | 222 | |
6d12e2d8 KH |
223 | return mem_cgroup_zoneinfo(mem, nid, zid); |
224 | } | |
225 | ||
226 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
b69408e8 | 227 | enum lru_list idx) |
6d12e2d8 KH |
228 | { |
229 | int nid, zid; | |
230 | struct mem_cgroup_per_zone *mz; | |
231 | u64 total = 0; | |
232 | ||
233 | for_each_online_node(nid) | |
234 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
235 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
236 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
237 | } | |
238 | return total; | |
d52aa412 KH |
239 | } |
240 | ||
d5b69e38 | 241 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
242 | { |
243 | return container_of(cgroup_subsys_state(cont, | |
244 | mem_cgroup_subsys_id), struct mem_cgroup, | |
245 | css); | |
246 | } | |
247 | ||
cf475ad2 | 248 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 249 | { |
31a78f23 BS |
250 | /* |
251 | * mm_update_next_owner() may clear mm->owner to NULL | |
252 | * if it races with swapoff, page migration, etc. | |
253 | * So this can be called with p == NULL. | |
254 | */ | |
255 | if (unlikely(!p)) | |
256 | return NULL; | |
257 | ||
78fb7466 PE |
258 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
259 | struct mem_cgroup, css); | |
260 | } | |
261 | ||
8a9f3ccd BS |
262 | static inline int page_cgroup_locked(struct page *page) |
263 | { | |
8869b8f6 | 264 | return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); |
8a9f3ccd BS |
265 | } |
266 | ||
9442ec9d | 267 | static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc) |
78fb7466 | 268 | { |
9442ec9d HD |
269 | VM_BUG_ON(!page_cgroup_locked(page)); |
270 | page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK); | |
78fb7466 PE |
271 | } |
272 | ||
273 | struct page_cgroup *page_get_page_cgroup(struct page *page) | |
274 | { | |
8869b8f6 | 275 | return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK); |
8a9f3ccd BS |
276 | } |
277 | ||
d5b69e38 | 278 | static void lock_page_cgroup(struct page *page) |
8a9f3ccd BS |
279 | { |
280 | bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
8a9f3ccd BS |
281 | } |
282 | ||
2680eed7 HD |
283 | static int try_lock_page_cgroup(struct page *page) |
284 | { | |
285 | return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
286 | } | |
287 | ||
d5b69e38 | 288 | static void unlock_page_cgroup(struct page *page) |
8a9f3ccd BS |
289 | { |
290 | bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup); | |
291 | } | |
292 | ||
3eae90c3 KH |
293 | static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, |
294 | struct page_cgroup *pc) | |
6d12e2d8 | 295 | { |
4f98a2fe RR |
296 | int lru = LRU_BASE; |
297 | ||
894bc310 LS |
298 | if (pc->flags & PAGE_CGROUP_FLAG_UNEVICTABLE) |
299 | lru = LRU_UNEVICTABLE; | |
300 | else { | |
301 | if (pc->flags & PAGE_CGROUP_FLAG_ACTIVE) | |
302 | lru += LRU_ACTIVE; | |
303 | if (pc->flags & PAGE_CGROUP_FLAG_FILE) | |
304 | lru += LRU_FILE; | |
305 | } | |
6d12e2d8 | 306 | |
b69408e8 | 307 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
6d12e2d8 KH |
308 | |
309 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false); | |
508b7be0 | 310 | list_del(&pc->lru); |
6d12e2d8 KH |
311 | } |
312 | ||
3eae90c3 KH |
313 | static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, |
314 | struct page_cgroup *pc) | |
6d12e2d8 | 315 | { |
4f98a2fe | 316 | int lru = LRU_BASE; |
b69408e8 | 317 | |
894bc310 LS |
318 | if (pc->flags & PAGE_CGROUP_FLAG_UNEVICTABLE) |
319 | lru = LRU_UNEVICTABLE; | |
320 | else { | |
321 | if (pc->flags & PAGE_CGROUP_FLAG_ACTIVE) | |
322 | lru += LRU_ACTIVE; | |
323 | if (pc->flags & PAGE_CGROUP_FLAG_FILE) | |
324 | lru += LRU_FILE; | |
325 | } | |
b69408e8 CL |
326 | |
327 | MEM_CGROUP_ZSTAT(mz, lru) += 1; | |
328 | list_add(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 | 329 | |
6d12e2d8 KH |
330 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true); |
331 | } | |
332 | ||
894bc310 | 333 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru) |
66e1707b | 334 | { |
6d12e2d8 | 335 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); |
894bc310 LS |
336 | int active = pc->flags & PAGE_CGROUP_FLAG_ACTIVE; |
337 | int file = pc->flags & PAGE_CGROUP_FLAG_FILE; | |
338 | int unevictable = pc->flags & PAGE_CGROUP_FLAG_UNEVICTABLE; | |
339 | enum lru_list from = unevictable ? LRU_UNEVICTABLE : | |
340 | (LRU_FILE * !!file + !!active); | |
6d12e2d8 | 341 | |
894bc310 LS |
342 | if (lru == from) |
343 | return; | |
b69408e8 | 344 | |
894bc310 LS |
345 | MEM_CGROUP_ZSTAT(mz, from) -= 1; |
346 | ||
347 | if (is_unevictable_lru(lru)) { | |
3564c7c4 | 348 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; |
894bc310 LS |
349 | pc->flags |= PAGE_CGROUP_FLAG_UNEVICTABLE; |
350 | } else { | |
351 | if (is_active_lru(lru)) | |
352 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; | |
353 | else | |
354 | pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE; | |
355 | pc->flags &= ~PAGE_CGROUP_FLAG_UNEVICTABLE; | |
356 | } | |
b69408e8 | 357 | |
b69408e8 CL |
358 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
359 | list_move(&pc->lru, &mz->lists[lru]); | |
66e1707b BS |
360 | } |
361 | ||
4c4a2214 DR |
362 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
363 | { | |
364 | int ret; | |
365 | ||
366 | task_lock(task); | |
bd845e38 | 367 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
4c4a2214 DR |
368 | task_unlock(task); |
369 | return ret; | |
370 | } | |
371 | ||
66e1707b BS |
372 | /* |
373 | * This routine assumes that the appropriate zone's lru lock is already held | |
374 | */ | |
894bc310 | 375 | void mem_cgroup_move_lists(struct page *page, enum lru_list lru) |
66e1707b | 376 | { |
427d5416 | 377 | struct page_cgroup *pc; |
072c56c1 KH |
378 | struct mem_cgroup_per_zone *mz; |
379 | unsigned long flags; | |
380 | ||
cede86ac LZ |
381 | if (mem_cgroup_subsys.disabled) |
382 | return; | |
383 | ||
2680eed7 HD |
384 | /* |
385 | * We cannot lock_page_cgroup while holding zone's lru_lock, | |
386 | * because other holders of lock_page_cgroup can be interrupted | |
387 | * with an attempt to rotate_reclaimable_page. But we cannot | |
388 | * safely get to page_cgroup without it, so just try_lock it: | |
389 | * mem_cgroup_isolate_pages allows for page left on wrong list. | |
390 | */ | |
391 | if (!try_lock_page_cgroup(page)) | |
66e1707b BS |
392 | return; |
393 | ||
2680eed7 HD |
394 | pc = page_get_page_cgroup(page); |
395 | if (pc) { | |
2680eed7 | 396 | mz = page_cgroup_zoneinfo(pc); |
2680eed7 | 397 | spin_lock_irqsave(&mz->lru_lock, flags); |
894bc310 | 398 | __mem_cgroup_move_lists(pc, lru); |
2680eed7 | 399 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
9b3c0a07 HT |
400 | } |
401 | unlock_page_cgroup(page); | |
66e1707b BS |
402 | } |
403 | ||
58ae83db KH |
404 | /* |
405 | * Calculate mapped_ratio under memory controller. This will be used in | |
406 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
407 | */ | |
408 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
409 | { | |
410 | long total, rss; | |
411 | ||
412 | /* | |
413 | * usage is recorded in bytes. But, here, we assume the number of | |
414 | * physical pages can be represented by "long" on any arch. | |
415 | */ | |
416 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
417 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
418 | return (int)((rss * 100L) / total); | |
419 | } | |
8869b8f6 | 420 | |
6c48a1d0 KH |
421 | /* |
422 | * prev_priority control...this will be used in memory reclaim path. | |
423 | */ | |
424 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
425 | { | |
426 | return mem->prev_priority; | |
427 | } | |
428 | ||
429 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
430 | { | |
431 | if (priority < mem->prev_priority) | |
432 | mem->prev_priority = priority; | |
433 | } | |
434 | ||
435 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
436 | { | |
437 | mem->prev_priority = priority; | |
438 | } | |
439 | ||
cc38108e KH |
440 | /* |
441 | * Calculate # of pages to be scanned in this priority/zone. | |
442 | * See also vmscan.c | |
443 | * | |
444 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
445 | * (see include/linux/mmzone.h) | |
446 | */ | |
447 | ||
b69408e8 CL |
448 | long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, |
449 | int priority, enum lru_list lru) | |
cc38108e | 450 | { |
b69408e8 | 451 | long nr_pages; |
cc38108e KH |
452 | int nid = zone->zone_pgdat->node_id; |
453 | int zid = zone_idx(zone); | |
454 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
455 | ||
b69408e8 | 456 | nr_pages = MEM_CGROUP_ZSTAT(mz, lru); |
cc38108e | 457 | |
b69408e8 | 458 | return (nr_pages >> priority); |
cc38108e KH |
459 | } |
460 | ||
66e1707b BS |
461 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
462 | struct list_head *dst, | |
463 | unsigned long *scanned, int order, | |
464 | int mode, struct zone *z, | |
465 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 466 | int active, int file) |
66e1707b BS |
467 | { |
468 | unsigned long nr_taken = 0; | |
469 | struct page *page; | |
470 | unsigned long scan; | |
471 | LIST_HEAD(pc_list); | |
472 | struct list_head *src; | |
ff7283fa | 473 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
474 | int nid = z->zone_pgdat->node_id; |
475 | int zid = zone_idx(z); | |
476 | struct mem_cgroup_per_zone *mz; | |
4f98a2fe | 477 | int lru = LRU_FILE * !!file + !!active; |
66e1707b | 478 | |
cf475ad2 | 479 | BUG_ON(!mem_cont); |
1ecaab2b | 480 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 481 | src = &mz->lists[lru]; |
66e1707b | 482 | |
072c56c1 | 483 | spin_lock(&mz->lru_lock); |
ff7283fa KH |
484 | scan = 0; |
485 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 486 | if (scan >= nr_to_scan) |
ff7283fa | 487 | break; |
66e1707b | 488 | page = pc->page; |
66e1707b | 489 | |
436c6541 | 490 | if (unlikely(!PageLRU(page))) |
ff7283fa | 491 | continue; |
ff7283fa | 492 | |
4f98a2fe RR |
493 | /* |
494 | * TODO: play better with lumpy reclaim, grabbing anything. | |
495 | */ | |
894bc310 LS |
496 | if (PageUnevictable(page) || |
497 | (PageActive(page) && !active) || | |
498 | (!PageActive(page) && active)) { | |
499 | __mem_cgroup_move_lists(pc, page_lru(page)); | |
66e1707b BS |
500 | continue; |
501 | } | |
502 | ||
436c6541 HD |
503 | scan++; |
504 | list_move(&pc->lru, &pc_list); | |
66e1707b | 505 | |
4f98a2fe | 506 | if (__isolate_lru_page(page, mode, file) == 0) { |
66e1707b BS |
507 | list_move(&page->lru, dst); |
508 | nr_taken++; | |
509 | } | |
510 | } | |
511 | ||
512 | list_splice(&pc_list, src); | |
072c56c1 | 513 | spin_unlock(&mz->lru_lock); |
66e1707b BS |
514 | |
515 | *scanned = scan; | |
516 | return nr_taken; | |
517 | } | |
518 | ||
8a9f3ccd BS |
519 | /* |
520 | * Charge the memory controller for page usage. | |
521 | * Return | |
522 | * 0 if the charge was successful | |
523 | * < 0 if the cgroup is over its limit | |
524 | */ | |
217bc319 | 525 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
e8589cc1 KH |
526 | gfp_t gfp_mask, enum charge_type ctype, |
527 | struct mem_cgroup *memcg) | |
8a9f3ccd BS |
528 | { |
529 | struct mem_cgroup *mem; | |
9175e031 | 530 | struct page_cgroup *pc; |
66e1707b BS |
531 | unsigned long flags; |
532 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
072c56c1 | 533 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 534 | |
508b7be0 | 535 | pc = kmem_cache_alloc(page_cgroup_cache, gfp_mask); |
b76734e5 | 536 | if (unlikely(pc == NULL)) |
8a9f3ccd BS |
537 | goto err; |
538 | ||
8a9f3ccd | 539 | /* |
3be91277 HD |
540 | * We always charge the cgroup the mm_struct belongs to. |
541 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
542 | * thread group leader migrates. It's possible that mm is not |
543 | * set, if so charge the init_mm (happens for pagecache usage). | |
544 | */ | |
69029cd5 | 545 | if (likely(!memcg)) { |
e8589cc1 KH |
546 | rcu_read_lock(); |
547 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
548 | if (unlikely(!mem)) { |
549 | rcu_read_unlock(); | |
550 | kmem_cache_free(page_cgroup_cache, pc); | |
551 | return 0; | |
552 | } | |
e8589cc1 KH |
553 | /* |
554 | * For every charge from the cgroup, increment reference count | |
555 | */ | |
556 | css_get(&mem->css); | |
557 | rcu_read_unlock(); | |
558 | } else { | |
559 | mem = memcg; | |
560 | css_get(&memcg->css); | |
561 | } | |
8a9f3ccd | 562 | |
addb9efe | 563 | while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) { |
3be91277 HD |
564 | if (!(gfp_mask & __GFP_WAIT)) |
565 | goto out; | |
e1a1cd59 BS |
566 | |
567 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) | |
66e1707b BS |
568 | continue; |
569 | ||
570 | /* | |
8869b8f6 HD |
571 | * try_to_free_mem_cgroup_pages() might not give us a full |
572 | * picture of reclaim. Some pages are reclaimed and might be | |
573 | * moved to swap cache or just unmapped from the cgroup. | |
574 | * Check the limit again to see if the reclaim reduced the | |
575 | * current usage of the cgroup before giving up | |
576 | */ | |
66e1707b BS |
577 | if (res_counter_check_under_limit(&mem->res)) |
578 | continue; | |
3be91277 HD |
579 | |
580 | if (!nr_retries--) { | |
581 | mem_cgroup_out_of_memory(mem, gfp_mask); | |
582 | goto out; | |
66e1707b | 583 | } |
8a9f3ccd BS |
584 | } |
585 | ||
8a9f3ccd BS |
586 | pc->mem_cgroup = mem; |
587 | pc->page = page; | |
508b7be0 KH |
588 | /* |
589 | * If a page is accounted as a page cache, insert to inactive list. | |
590 | * If anon, insert to active list. | |
591 | */ | |
4f98a2fe | 592 | if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE) { |
4a56d02e | 593 | pc->flags = PAGE_CGROUP_FLAG_CACHE; |
4f98a2fe RR |
594 | if (page_is_file_cache(page)) |
595 | pc->flags |= PAGE_CGROUP_FLAG_FILE; | |
596 | else | |
597 | pc->flags |= PAGE_CGROUP_FLAG_ACTIVE; | |
598 | } else if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) | |
508b7be0 | 599 | pc->flags = PAGE_CGROUP_FLAG_ACTIVE; |
4f98a2fe RR |
600 | else /* MEM_CGROUP_CHARGE_TYPE_SHMEM */ |
601 | pc->flags = PAGE_CGROUP_FLAG_CACHE | PAGE_CGROUP_FLAG_ACTIVE; | |
3be91277 | 602 | |
7e924aaf | 603 | lock_page_cgroup(page); |
b76734e5 | 604 | if (unlikely(page_get_page_cgroup(page))) { |
7e924aaf | 605 | unlock_page_cgroup(page); |
9175e031 KH |
606 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
607 | css_put(&mem->css); | |
b6ac57d5 | 608 | kmem_cache_free(page_cgroup_cache, pc); |
accf163e | 609 | goto done; |
9175e031 | 610 | } |
7e924aaf | 611 | page_assign_page_cgroup(page, pc); |
8a9f3ccd | 612 | |
072c56c1 KH |
613 | mz = page_cgroup_zoneinfo(pc); |
614 | spin_lock_irqsave(&mz->lru_lock, flags); | |
3eae90c3 | 615 | __mem_cgroup_add_list(mz, pc); |
072c56c1 | 616 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
66e1707b | 617 | |
fb59e9f1 | 618 | unlock_page_cgroup(page); |
8a9f3ccd | 619 | done: |
8a9f3ccd | 620 | return 0; |
3be91277 HD |
621 | out: |
622 | css_put(&mem->css); | |
b6ac57d5 | 623 | kmem_cache_free(page_cgroup_cache, pc); |
8a9f3ccd | 624 | err: |
8a9f3ccd BS |
625 | return -ENOMEM; |
626 | } | |
627 | ||
8869b8f6 | 628 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
217bc319 | 629 | { |
cede86ac LZ |
630 | if (mem_cgroup_subsys.disabled) |
631 | return 0; | |
632 | ||
69029cd5 KH |
633 | /* |
634 | * If already mapped, we don't have to account. | |
635 | * If page cache, page->mapping has address_space. | |
636 | * But page->mapping may have out-of-use anon_vma pointer, | |
637 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
638 | * is NULL. | |
639 | */ | |
640 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
641 | return 0; | |
642 | if (unlikely(!mm)) | |
643 | mm = &init_mm; | |
217bc319 | 644 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 645 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
646 | } |
647 | ||
e1a1cd59 BS |
648 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
649 | gfp_t gfp_mask) | |
8697d331 | 650 | { |
cede86ac LZ |
651 | if (mem_cgroup_subsys.disabled) |
652 | return 0; | |
653 | ||
accf163e KH |
654 | /* |
655 | * Corner case handling. This is called from add_to_page_cache() | |
656 | * in usual. But some FS (shmem) precharges this page before calling it | |
657 | * and call add_to_page_cache() with GFP_NOWAIT. | |
658 | * | |
659 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
660 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
661 | * charge twice. (It works but has to pay a bit larger cost.) | |
662 | */ | |
663 | if (!(gfp_mask & __GFP_WAIT)) { | |
664 | struct page_cgroup *pc; | |
665 | ||
666 | lock_page_cgroup(page); | |
667 | pc = page_get_page_cgroup(page); | |
668 | if (pc) { | |
669 | VM_BUG_ON(pc->page != page); | |
670 | VM_BUG_ON(!pc->mem_cgroup); | |
671 | unlock_page_cgroup(page); | |
672 | return 0; | |
673 | } | |
674 | unlock_page_cgroup(page); | |
675 | } | |
676 | ||
69029cd5 | 677 | if (unlikely(!mm)) |
8697d331 | 678 | mm = &init_mm; |
accf163e | 679 | |
8869b8f6 | 680 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 KH |
681 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
682 | } | |
683 | ||
8a9f3ccd | 684 | /* |
69029cd5 | 685 | * uncharge if !page_mapped(page) |
8a9f3ccd | 686 | */ |
69029cd5 KH |
687 | static void |
688 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) | |
8a9f3ccd | 689 | { |
8289546e | 690 | struct page_cgroup *pc; |
8a9f3ccd | 691 | struct mem_cgroup *mem; |
072c56c1 | 692 | struct mem_cgroup_per_zone *mz; |
66e1707b | 693 | unsigned long flags; |
8a9f3ccd | 694 | |
4077960e BS |
695 | if (mem_cgroup_subsys.disabled) |
696 | return; | |
697 | ||
8697d331 | 698 | /* |
3c541e14 | 699 | * Check if our page_cgroup is valid |
8697d331 | 700 | */ |
8289546e HD |
701 | lock_page_cgroup(page); |
702 | pc = page_get_page_cgroup(page); | |
b76734e5 | 703 | if (unlikely(!pc)) |
8289546e | 704 | goto unlock; |
8a9f3ccd | 705 | |
b9c565d5 | 706 | VM_BUG_ON(pc->page != page); |
b9c565d5 | 707 | |
69029cd5 KH |
708 | if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
709 | && ((pc->flags & PAGE_CGROUP_FLAG_CACHE) | |
710 | || page_mapped(page))) | |
711 | goto unlock; | |
b9c565d5 | 712 | |
69029cd5 KH |
713 | mz = page_cgroup_zoneinfo(pc); |
714 | spin_lock_irqsave(&mz->lru_lock, flags); | |
715 | __mem_cgroup_remove_list(mz, pc); | |
716 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
fb59e9f1 | 717 | |
69029cd5 KH |
718 | page_assign_page_cgroup(page, NULL); |
719 | unlock_page_cgroup(page); | |
6d48ff8b | 720 | |
69029cd5 KH |
721 | mem = pc->mem_cgroup; |
722 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
723 | css_put(&mem->css); | |
6d12e2d8 | 724 | |
69029cd5 KH |
725 | kmem_cache_free(page_cgroup_cache, pc); |
726 | return; | |
8289546e | 727 | unlock: |
3c541e14 BS |
728 | unlock_page_cgroup(page); |
729 | } | |
730 | ||
69029cd5 KH |
731 | void mem_cgroup_uncharge_page(struct page *page) |
732 | { | |
733 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
734 | } | |
735 | ||
736 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
737 | { | |
738 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 739 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
740 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
741 | } | |
742 | ||
ae41be37 | 743 | /* |
e8589cc1 | 744 | * Before starting migration, account against new page. |
ae41be37 | 745 | */ |
e8589cc1 | 746 | int mem_cgroup_prepare_migration(struct page *page, struct page *newpage) |
ae41be37 KH |
747 | { |
748 | struct page_cgroup *pc; | |
e8589cc1 KH |
749 | struct mem_cgroup *mem = NULL; |
750 | enum charge_type ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
751 | int ret = 0; | |
8869b8f6 | 752 | |
4077960e BS |
753 | if (mem_cgroup_subsys.disabled) |
754 | return 0; | |
755 | ||
ae41be37 KH |
756 | lock_page_cgroup(page); |
757 | pc = page_get_page_cgroup(page); | |
e8589cc1 KH |
758 | if (pc) { |
759 | mem = pc->mem_cgroup; | |
760 | css_get(&mem->css); | |
4f98a2fe RR |
761 | if (pc->flags & PAGE_CGROUP_FLAG_CACHE) { |
762 | if (page_is_file_cache(page)) | |
763 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
764 | else | |
765 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
766 | } | |
e8589cc1 | 767 | } |
ae41be37 | 768 | unlock_page_cgroup(page); |
e8589cc1 KH |
769 | if (mem) { |
770 | ret = mem_cgroup_charge_common(newpage, NULL, GFP_KERNEL, | |
771 | ctype, mem); | |
772 | css_put(&mem->css); | |
773 | } | |
774 | return ret; | |
ae41be37 | 775 | } |
8869b8f6 | 776 | |
69029cd5 | 777 | /* remove redundant charge if migration failed*/ |
e8589cc1 | 778 | void mem_cgroup_end_migration(struct page *newpage) |
ae41be37 | 779 | { |
69029cd5 KH |
780 | /* |
781 | * At success, page->mapping is not NULL. | |
782 | * special rollback care is necessary when | |
783 | * 1. at migration failure. (newpage->mapping is cleared in this case) | |
784 | * 2. the newpage was moved but not remapped again because the task | |
785 | * exits and the newpage is obsolete. In this case, the new page | |
786 | * may be a swapcache. So, we just call mem_cgroup_uncharge_page() | |
787 | * always for avoiding mess. The page_cgroup will be removed if | |
788 | * unnecessary. File cache pages is still on radix-tree. Don't | |
789 | * care it. | |
790 | */ | |
791 | if (!newpage->mapping) | |
792 | __mem_cgroup_uncharge_common(newpage, | |
793 | MEM_CGROUP_CHARGE_TYPE_FORCE); | |
794 | else if (PageAnon(newpage)) | |
795 | mem_cgroup_uncharge_page(newpage); | |
ae41be37 | 796 | } |
78fb7466 | 797 | |
c9b0ed51 KH |
798 | /* |
799 | * A call to try to shrink memory usage under specified resource controller. | |
800 | * This is typically used for page reclaiming for shmem for reducing side | |
801 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
802 | */ | |
803 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) | |
804 | { | |
805 | struct mem_cgroup *mem; | |
806 | int progress = 0; | |
807 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
808 | ||
cede86ac LZ |
809 | if (mem_cgroup_subsys.disabled) |
810 | return 0; | |
9623e078 HD |
811 | if (!mm) |
812 | return 0; | |
cede86ac | 813 | |
c9b0ed51 KH |
814 | rcu_read_lock(); |
815 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
31a78f23 BS |
816 | if (unlikely(!mem)) { |
817 | rcu_read_unlock(); | |
818 | return 0; | |
819 | } | |
c9b0ed51 KH |
820 | css_get(&mem->css); |
821 | rcu_read_unlock(); | |
822 | ||
823 | do { | |
824 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask); | |
a10cebf5 | 825 | progress += res_counter_check_under_limit(&mem->res); |
c9b0ed51 KH |
826 | } while (!progress && --retry); |
827 | ||
828 | css_put(&mem->css); | |
829 | if (!retry) | |
830 | return -ENOMEM; | |
831 | return 0; | |
832 | } | |
833 | ||
628f4235 KH |
834 | int mem_cgroup_resize_limit(struct mem_cgroup *memcg, unsigned long long val) |
835 | { | |
836 | ||
837 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
838 | int progress; | |
839 | int ret = 0; | |
840 | ||
841 | while (res_counter_set_limit(&memcg->res, val)) { | |
842 | if (signal_pending(current)) { | |
843 | ret = -EINTR; | |
844 | break; | |
845 | } | |
846 | if (!retry_count) { | |
847 | ret = -EBUSY; | |
848 | break; | |
849 | } | |
850 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL); | |
851 | if (!progress) | |
852 | retry_count--; | |
853 | } | |
854 | return ret; | |
855 | } | |
856 | ||
857 | ||
cc847582 KH |
858 | /* |
859 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
860 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
861 | */ | |
862 | #define FORCE_UNCHARGE_BATCH (128) | |
8869b8f6 | 863 | static void mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
072c56c1 | 864 | struct mem_cgroup_per_zone *mz, |
b69408e8 | 865 | enum lru_list lru) |
cc847582 KH |
866 | { |
867 | struct page_cgroup *pc; | |
868 | struct page *page; | |
9b3c0a07 | 869 | int count = FORCE_UNCHARGE_BATCH; |
cc847582 | 870 | unsigned long flags; |
072c56c1 KH |
871 | struct list_head *list; |
872 | ||
b69408e8 | 873 | list = &mz->lists[lru]; |
cc847582 | 874 | |
072c56c1 | 875 | spin_lock_irqsave(&mz->lru_lock, flags); |
9b3c0a07 | 876 | while (!list_empty(list)) { |
cc847582 KH |
877 | pc = list_entry(list->prev, struct page_cgroup, lru); |
878 | page = pc->page; | |
9b3c0a07 HT |
879 | get_page(page); |
880 | spin_unlock_irqrestore(&mz->lru_lock, flags); | |
e8589cc1 KH |
881 | /* |
882 | * Check if this page is on LRU. !LRU page can be found | |
883 | * if it's under page migration. | |
884 | */ | |
885 | if (PageLRU(page)) { | |
69029cd5 KH |
886 | __mem_cgroup_uncharge_common(page, |
887 | MEM_CGROUP_CHARGE_TYPE_FORCE); | |
e8589cc1 KH |
888 | put_page(page); |
889 | if (--count <= 0) { | |
890 | count = FORCE_UNCHARGE_BATCH; | |
891 | cond_resched(); | |
892 | } | |
893 | } else | |
9b3c0a07 | 894 | cond_resched(); |
9b3c0a07 | 895 | spin_lock_irqsave(&mz->lru_lock, flags); |
cc847582 | 896 | } |
072c56c1 | 897 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
cc847582 KH |
898 | } |
899 | ||
900 | /* | |
901 | * make mem_cgroup's charge to be 0 if there is no task. | |
902 | * This enables deleting this mem_cgroup. | |
903 | */ | |
d5b69e38 | 904 | static int mem_cgroup_force_empty(struct mem_cgroup *mem) |
cc847582 KH |
905 | { |
906 | int ret = -EBUSY; | |
1ecaab2b | 907 | int node, zid; |
8869b8f6 | 908 | |
cc847582 KH |
909 | css_get(&mem->css); |
910 | /* | |
911 | * page reclaim code (kswapd etc..) will move pages between | |
8869b8f6 | 912 | * active_list <-> inactive_list while we don't take a lock. |
cc847582 KH |
913 | * So, we have to do loop here until all lists are empty. |
914 | */ | |
1ecaab2b | 915 | while (mem->res.usage > 0) { |
cc847582 KH |
916 | if (atomic_read(&mem->css.cgroup->count) > 0) |
917 | goto out; | |
1ecaab2b KH |
918 | for_each_node_state(node, N_POSSIBLE) |
919 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
920 | struct mem_cgroup_per_zone *mz; | |
b69408e8 | 921 | enum lru_list l; |
1ecaab2b | 922 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
b69408e8 CL |
923 | for_each_lru(l) |
924 | mem_cgroup_force_empty_list(mem, mz, l); | |
1ecaab2b | 925 | } |
cc847582 KH |
926 | } |
927 | ret = 0; | |
928 | out: | |
929 | css_put(&mem->css); | |
930 | return ret; | |
931 | } | |
932 | ||
2c3daa72 | 933 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 934 | { |
2c3daa72 PM |
935 | return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res, |
936 | cft->private); | |
8cdea7c0 | 937 | } |
628f4235 KH |
938 | /* |
939 | * The user of this function is... | |
940 | * RES_LIMIT. | |
941 | */ | |
856c13aa PM |
942 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
943 | const char *buffer) | |
8cdea7c0 | 944 | { |
628f4235 KH |
945 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
946 | unsigned long long val; | |
947 | int ret; | |
948 | ||
949 | switch (cft->private) { | |
950 | case RES_LIMIT: | |
951 | /* This function does all necessary parse...reuse it */ | |
952 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
953 | if (!ret) | |
954 | ret = mem_cgroup_resize_limit(memcg, val); | |
955 | break; | |
956 | default: | |
957 | ret = -EINVAL; /* should be BUG() ? */ | |
958 | break; | |
959 | } | |
960 | return ret; | |
8cdea7c0 BS |
961 | } |
962 | ||
29f2a4da | 963 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
964 | { |
965 | struct mem_cgroup *mem; | |
966 | ||
967 | mem = mem_cgroup_from_cont(cont); | |
29f2a4da PE |
968 | switch (event) { |
969 | case RES_MAX_USAGE: | |
970 | res_counter_reset_max(&mem->res); | |
971 | break; | |
972 | case RES_FAILCNT: | |
973 | res_counter_reset_failcnt(&mem->res); | |
974 | break; | |
975 | } | |
85cc59db | 976 | return 0; |
c84872e1 PE |
977 | } |
978 | ||
85cc59db | 979 | static int mem_force_empty_write(struct cgroup *cont, unsigned int event) |
cc847582 | 980 | { |
85cc59db | 981 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont)); |
cc847582 KH |
982 | } |
983 | ||
d2ceb9b7 KH |
984 | static const struct mem_cgroup_stat_desc { |
985 | const char *msg; | |
986 | u64 unit; | |
987 | } mem_cgroup_stat_desc[] = { | |
988 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
989 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
55e462b0 BR |
990 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
991 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
d2ceb9b7 KH |
992 | }; |
993 | ||
c64745cf PM |
994 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
995 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 996 | { |
d2ceb9b7 KH |
997 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
998 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
999 | int i; | |
1000 | ||
1001 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1002 | s64 val; | |
1003 | ||
1004 | val = mem_cgroup_read_stat(stat, i); | |
1005 | val *= mem_cgroup_stat_desc[i].unit; | |
c64745cf | 1006 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
d2ceb9b7 | 1007 | } |
6d12e2d8 KH |
1008 | /* showing # of active pages */ |
1009 | { | |
4f98a2fe RR |
1010 | unsigned long active_anon, inactive_anon; |
1011 | unsigned long active_file, inactive_file; | |
7b854121 | 1012 | unsigned long unevictable; |
4f98a2fe RR |
1013 | |
1014 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1015 | LRU_INACTIVE_ANON); | |
1016 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1017 | LRU_ACTIVE_ANON); | |
1018 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1019 | LRU_INACTIVE_FILE); | |
1020 | active_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1021 | LRU_ACTIVE_FILE); | |
7b854121 LS |
1022 | unevictable = mem_cgroup_get_all_zonestat(mem_cont, |
1023 | LRU_UNEVICTABLE); | |
1024 | ||
4f98a2fe RR |
1025 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); |
1026 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); | |
1027 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); | |
1028 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); | |
7b854121 LS |
1029 | cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); |
1030 | ||
6d12e2d8 | 1031 | } |
d2ceb9b7 KH |
1032 | return 0; |
1033 | } | |
1034 | ||
8cdea7c0 BS |
1035 | static struct cftype mem_cgroup_files[] = { |
1036 | { | |
0eea1030 | 1037 | .name = "usage_in_bytes", |
8cdea7c0 | 1038 | .private = RES_USAGE, |
2c3daa72 | 1039 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1040 | }, |
c84872e1 PE |
1041 | { |
1042 | .name = "max_usage_in_bytes", | |
1043 | .private = RES_MAX_USAGE, | |
29f2a4da | 1044 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
1045 | .read_u64 = mem_cgroup_read, |
1046 | }, | |
8cdea7c0 | 1047 | { |
0eea1030 | 1048 | .name = "limit_in_bytes", |
8cdea7c0 | 1049 | .private = RES_LIMIT, |
856c13aa | 1050 | .write_string = mem_cgroup_write, |
2c3daa72 | 1051 | .read_u64 = mem_cgroup_read, |
8cdea7c0 BS |
1052 | }, |
1053 | { | |
1054 | .name = "failcnt", | |
1055 | .private = RES_FAILCNT, | |
29f2a4da | 1056 | .trigger = mem_cgroup_reset, |
2c3daa72 | 1057 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 1058 | }, |
cc847582 KH |
1059 | { |
1060 | .name = "force_empty", | |
85cc59db | 1061 | .trigger = mem_force_empty_write, |
cc847582 | 1062 | }, |
d2ceb9b7 KH |
1063 | { |
1064 | .name = "stat", | |
c64745cf | 1065 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 1066 | }, |
8cdea7c0 BS |
1067 | }; |
1068 | ||
6d12e2d8 KH |
1069 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1070 | { | |
1071 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 1072 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 1073 | enum lru_list l; |
41e3355d | 1074 | int zone, tmp = node; |
1ecaab2b KH |
1075 | /* |
1076 | * This routine is called against possible nodes. | |
1077 | * But it's BUG to call kmalloc() against offline node. | |
1078 | * | |
1079 | * TODO: this routine can waste much memory for nodes which will | |
1080 | * never be onlined. It's better to use memory hotplug callback | |
1081 | * function. | |
1082 | */ | |
41e3355d KH |
1083 | if (!node_state(node, N_NORMAL_MEMORY)) |
1084 | tmp = -1; | |
1085 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
1086 | if (!pn) |
1087 | return 1; | |
1ecaab2b | 1088 | |
6d12e2d8 KH |
1089 | mem->info.nodeinfo[node] = pn; |
1090 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
1091 | |
1092 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1093 | mz = &pn->zoneinfo[zone]; | |
072c56c1 | 1094 | spin_lock_init(&mz->lru_lock); |
b69408e8 CL |
1095 | for_each_lru(l) |
1096 | INIT_LIST_HEAD(&mz->lists[l]); | |
1ecaab2b | 1097 | } |
6d12e2d8 KH |
1098 | return 0; |
1099 | } | |
1100 | ||
1ecaab2b KH |
1101 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
1102 | { | |
1103 | kfree(mem->info.nodeinfo[node]); | |
1104 | } | |
1105 | ||
33327948 KH |
1106 | static struct mem_cgroup *mem_cgroup_alloc(void) |
1107 | { | |
1108 | struct mem_cgroup *mem; | |
1109 | ||
1110 | if (sizeof(*mem) < PAGE_SIZE) | |
1111 | mem = kmalloc(sizeof(*mem), GFP_KERNEL); | |
1112 | else | |
1113 | mem = vmalloc(sizeof(*mem)); | |
1114 | ||
1115 | if (mem) | |
1116 | memset(mem, 0, sizeof(*mem)); | |
1117 | return mem; | |
1118 | } | |
1119 | ||
1120 | static void mem_cgroup_free(struct mem_cgroup *mem) | |
1121 | { | |
1122 | if (sizeof(*mem) < PAGE_SIZE) | |
1123 | kfree(mem); | |
1124 | else | |
1125 | vfree(mem); | |
1126 | } | |
1127 | ||
1128 | ||
8cdea7c0 BS |
1129 | static struct cgroup_subsys_state * |
1130 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
1131 | { | |
1132 | struct mem_cgroup *mem; | |
6d12e2d8 | 1133 | int node; |
8cdea7c0 | 1134 | |
b6ac57d5 | 1135 | if (unlikely((cont->parent) == NULL)) { |
78fb7466 | 1136 | mem = &init_mem_cgroup; |
b6ac57d5 BS |
1137 | page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC); |
1138 | } else { | |
33327948 KH |
1139 | mem = mem_cgroup_alloc(); |
1140 | if (!mem) | |
1141 | return ERR_PTR(-ENOMEM); | |
b6ac57d5 | 1142 | } |
78fb7466 | 1143 | |
8cdea7c0 | 1144 | res_counter_init(&mem->res); |
1ecaab2b | 1145 | |
6d12e2d8 KH |
1146 | for_each_node_state(node, N_POSSIBLE) |
1147 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
1148 | goto free_out; | |
1149 | ||
8cdea7c0 | 1150 | return &mem->css; |
6d12e2d8 KH |
1151 | free_out: |
1152 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1153 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1154 | if (cont->parent != NULL) |
33327948 | 1155 | mem_cgroup_free(mem); |
2dda81ca | 1156 | return ERR_PTR(-ENOMEM); |
8cdea7c0 BS |
1157 | } |
1158 | ||
df878fb0 KH |
1159 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
1160 | struct cgroup *cont) | |
1161 | { | |
1162 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1163 | mem_cgroup_force_empty(mem); | |
1164 | } | |
1165 | ||
8cdea7c0 BS |
1166 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
1167 | struct cgroup *cont) | |
1168 | { | |
6d12e2d8 KH |
1169 | int node; |
1170 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1171 | ||
1172 | for_each_node_state(node, N_POSSIBLE) | |
1ecaab2b | 1173 | free_mem_cgroup_per_zone_info(mem, node); |
6d12e2d8 | 1174 | |
33327948 | 1175 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
8cdea7c0 BS |
1176 | } |
1177 | ||
1178 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
1179 | struct cgroup *cont) | |
1180 | { | |
1181 | return cgroup_add_files(cont, ss, mem_cgroup_files, | |
1182 | ARRAY_SIZE(mem_cgroup_files)); | |
1183 | } | |
1184 | ||
67e465a7 BS |
1185 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
1186 | struct cgroup *cont, | |
1187 | struct cgroup *old_cont, | |
1188 | struct task_struct *p) | |
1189 | { | |
1190 | struct mm_struct *mm; | |
1191 | struct mem_cgroup *mem, *old_mem; | |
1192 | ||
1193 | mm = get_task_mm(p); | |
1194 | if (mm == NULL) | |
1195 | return; | |
1196 | ||
1197 | mem = mem_cgroup_from_cont(cont); | |
1198 | old_mem = mem_cgroup_from_cont(old_cont); | |
1199 | ||
67e465a7 BS |
1200 | /* |
1201 | * Only thread group leaders are allowed to migrate, the mm_struct is | |
1202 | * in effect owned by the leader | |
1203 | */ | |
52ea27eb | 1204 | if (!thread_group_leader(p)) |
67e465a7 BS |
1205 | goto out; |
1206 | ||
67e465a7 BS |
1207 | out: |
1208 | mmput(mm); | |
67e465a7 BS |
1209 | } |
1210 | ||
8cdea7c0 BS |
1211 | struct cgroup_subsys mem_cgroup_subsys = { |
1212 | .name = "memory", | |
1213 | .subsys_id = mem_cgroup_subsys_id, | |
1214 | .create = mem_cgroup_create, | |
df878fb0 | 1215 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
1216 | .destroy = mem_cgroup_destroy, |
1217 | .populate = mem_cgroup_populate, | |
67e465a7 | 1218 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 1219 | .early_init = 0, |
8cdea7c0 | 1220 | }; |