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 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
8cdea7c0 BS |
17 | * This program is free software; you can redistribute it and/or modify |
18 | * it under the terms of the GNU General Public License as published by | |
19 | * the Free Software Foundation; either version 2 of the License, or | |
20 | * (at your option) any later version. | |
21 | * | |
22 | * This program is distributed in the hope that it will be useful, | |
23 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
24 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
25 | * GNU General Public License for more details. | |
26 | */ | |
27 | ||
28 | #include <linux/res_counter.h> | |
29 | #include <linux/memcontrol.h> | |
30 | #include <linux/cgroup.h> | |
78fb7466 | 31 | #include <linux/mm.h> |
4ffef5fe | 32 | #include <linux/hugetlb.h> |
d13d1443 | 33 | #include <linux/pagemap.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
f64c3f54 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 KS |
47 | #include <linux/eventfd.h> |
48 | #include <linux/sort.h> | |
66e1707b | 49 | #include <linux/fs.h> |
d2ceb9b7 | 50 | #include <linux/seq_file.h> |
33327948 | 51 | #include <linux/vmalloc.h> |
b69408e8 | 52 | #include <linux/mm_inline.h> |
52d4b9ac | 53 | #include <linux/page_cgroup.h> |
cdec2e42 | 54 | #include <linux/cpu.h> |
158e0a2d | 55 | #include <linux/oom.h> |
08e552c6 | 56 | #include "internal.h" |
d1a4c0b3 | 57 | #include <net/sock.h> |
4bd2c1ee | 58 | #include <net/ip.h> |
d1a4c0b3 | 59 | #include <net/tcp_memcontrol.h> |
8cdea7c0 | 60 | |
8697d331 BS |
61 | #include <asm/uaccess.h> |
62 | ||
cc8e970c KM |
63 | #include <trace/events/vmscan.h> |
64 | ||
a181b0e8 | 65 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
68ae564b DR |
66 | EXPORT_SYMBOL(mem_cgroup_subsys); |
67 | ||
a181b0e8 | 68 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 69 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 70 | |
c255a458 | 71 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 72 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 73 | int do_swap_account __read_mostly; |
a42c390c MH |
74 | |
75 | /* for remember boot option*/ | |
c255a458 | 76 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
77 | static int really_do_swap_account __initdata = 1; |
78 | #else | |
79 | static int really_do_swap_account __initdata = 0; | |
80 | #endif | |
81 | ||
c077719b | 82 | #else |
a0db00fc | 83 | #define do_swap_account 0 |
c077719b KH |
84 | #endif |
85 | ||
86 | ||
d52aa412 KH |
87 | /* |
88 | * Statistics for memory cgroup. | |
89 | */ | |
90 | enum mem_cgroup_stat_index { | |
91 | /* | |
92 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
93 | */ | |
94 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 95 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 96 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
bff6bb83 | 97 | MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */ |
d52aa412 KH |
98 | MEM_CGROUP_STAT_NSTATS, |
99 | }; | |
100 | ||
af7c4b0e JW |
101 | static const char * const mem_cgroup_stat_names[] = { |
102 | "cache", | |
103 | "rss", | |
104 | "mapped_file", | |
105 | "swap", | |
106 | }; | |
107 | ||
e9f8974f JW |
108 | enum mem_cgroup_events_index { |
109 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
110 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
111 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
112 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
113 | MEM_CGROUP_EVENTS_NSTATS, |
114 | }; | |
af7c4b0e JW |
115 | |
116 | static const char * const mem_cgroup_events_names[] = { | |
117 | "pgpgin", | |
118 | "pgpgout", | |
119 | "pgfault", | |
120 | "pgmajfault", | |
121 | }; | |
122 | ||
58cf188e SZ |
123 | static const char * const mem_cgroup_lru_names[] = { |
124 | "inactive_anon", | |
125 | "active_anon", | |
126 | "inactive_file", | |
127 | "active_file", | |
128 | "unevictable", | |
129 | }; | |
130 | ||
7a159cc9 JW |
131 | /* |
132 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
133 | * it will be incremated by the number of pages. This counter is used for | |
134 | * for trigger some periodic events. This is straightforward and better | |
135 | * than using jiffies etc. to handle periodic memcg event. | |
136 | */ | |
137 | enum mem_cgroup_events_target { | |
138 | MEM_CGROUP_TARGET_THRESH, | |
139 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 140 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
141 | MEM_CGROUP_NTARGETS, |
142 | }; | |
a0db00fc KS |
143 | #define THRESHOLDS_EVENTS_TARGET 128 |
144 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
145 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 146 | |
d52aa412 | 147 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 148 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 149 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 150 | unsigned long nr_page_events; |
7a159cc9 | 151 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
152 | }; |
153 | ||
527a5ec9 JW |
154 | struct mem_cgroup_reclaim_iter { |
155 | /* css_id of the last scanned hierarchy member */ | |
156 | int position; | |
157 | /* scan generation, increased every round-trip */ | |
158 | unsigned int generation; | |
159 | }; | |
160 | ||
6d12e2d8 KH |
161 | /* |
162 | * per-zone information in memory controller. | |
163 | */ | |
6d12e2d8 | 164 | struct mem_cgroup_per_zone { |
6290df54 | 165 | struct lruvec lruvec; |
1eb49272 | 166 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 167 | |
527a5ec9 JW |
168 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
169 | ||
f64c3f54 BS |
170 | struct rb_node tree_node; /* RB tree node */ |
171 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
172 | /* the soft limit is exceeded*/ | |
173 | bool on_tree; | |
d79154bb | 174 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 175 | /* use container_of */ |
6d12e2d8 | 176 | }; |
6d12e2d8 KH |
177 | |
178 | struct mem_cgroup_per_node { | |
179 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
180 | }; | |
181 | ||
182 | struct mem_cgroup_lru_info { | |
45cf7ebd | 183 | struct mem_cgroup_per_node *nodeinfo[0]; |
6d12e2d8 KH |
184 | }; |
185 | ||
f64c3f54 BS |
186 | /* |
187 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
188 | * their hierarchy representation | |
189 | */ | |
190 | ||
191 | struct mem_cgroup_tree_per_zone { | |
192 | struct rb_root rb_root; | |
193 | spinlock_t lock; | |
194 | }; | |
195 | ||
196 | struct mem_cgroup_tree_per_node { | |
197 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
198 | }; | |
199 | ||
200 | struct mem_cgroup_tree { | |
201 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
202 | }; | |
203 | ||
204 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
205 | ||
2e72b634 KS |
206 | struct mem_cgroup_threshold { |
207 | struct eventfd_ctx *eventfd; | |
208 | u64 threshold; | |
209 | }; | |
210 | ||
9490ff27 | 211 | /* For threshold */ |
2e72b634 | 212 | struct mem_cgroup_threshold_ary { |
748dad36 | 213 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 214 | int current_threshold; |
2e72b634 KS |
215 | /* Size of entries[] */ |
216 | unsigned int size; | |
217 | /* Array of thresholds */ | |
218 | struct mem_cgroup_threshold entries[0]; | |
219 | }; | |
2c488db2 KS |
220 | |
221 | struct mem_cgroup_thresholds { | |
222 | /* Primary thresholds array */ | |
223 | struct mem_cgroup_threshold_ary *primary; | |
224 | /* | |
225 | * Spare threshold array. | |
226 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
227 | * It must be able to store at least primary->size - 1 entries. | |
228 | */ | |
229 | struct mem_cgroup_threshold_ary *spare; | |
230 | }; | |
231 | ||
9490ff27 KH |
232 | /* for OOM */ |
233 | struct mem_cgroup_eventfd_list { | |
234 | struct list_head list; | |
235 | struct eventfd_ctx *eventfd; | |
236 | }; | |
2e72b634 | 237 | |
c0ff4b85 R |
238 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
239 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 240 | |
8cdea7c0 BS |
241 | /* |
242 | * The memory controller data structure. The memory controller controls both | |
243 | * page cache and RSS per cgroup. We would eventually like to provide | |
244 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
245 | * to help the administrator determine what knobs to tune. | |
246 | * | |
247 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
248 | * we hit the water mark. May be even add a low water mark, such that |
249 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
250 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
251 | */ |
252 | struct mem_cgroup { | |
253 | struct cgroup_subsys_state css; | |
254 | /* | |
255 | * the counter to account for memory usage | |
256 | */ | |
257 | struct res_counter res; | |
59927fb9 HD |
258 | |
259 | union { | |
260 | /* | |
261 | * the counter to account for mem+swap usage. | |
262 | */ | |
263 | struct res_counter memsw; | |
264 | ||
265 | /* | |
266 | * rcu_freeing is used only when freeing struct mem_cgroup, | |
267 | * so put it into a union to avoid wasting more memory. | |
268 | * It must be disjoint from the css field. It could be | |
269 | * in a union with the res field, but res plays a much | |
270 | * larger part in mem_cgroup life than memsw, and might | |
271 | * be of interest, even at time of free, when debugging. | |
272 | * So share rcu_head with the less interesting memsw. | |
273 | */ | |
274 | struct rcu_head rcu_freeing; | |
275 | /* | |
3afe36b1 GC |
276 | * We also need some space for a worker in deferred freeing. |
277 | * By the time we call it, rcu_freeing is no longer in use. | |
59927fb9 HD |
278 | */ |
279 | struct work_struct work_freeing; | |
280 | }; | |
281 | ||
510fc4e1 GC |
282 | /* |
283 | * the counter to account for kernel memory usage. | |
284 | */ | |
285 | struct res_counter kmem; | |
18f59ea7 BS |
286 | /* |
287 | * Should the accounting and control be hierarchical, per subtree? | |
288 | */ | |
289 | bool use_hierarchy; | |
510fc4e1 | 290 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
291 | |
292 | bool oom_lock; | |
293 | atomic_t under_oom; | |
294 | ||
8c7c6e34 | 295 | atomic_t refcnt; |
14797e23 | 296 | |
1f4c025b | 297 | int swappiness; |
3c11ecf4 KH |
298 | /* OOM-Killer disable */ |
299 | int oom_kill_disable; | |
a7885eb8 | 300 | |
22a668d7 KH |
301 | /* set when res.limit == memsw.limit */ |
302 | bool memsw_is_minimum; | |
303 | ||
2e72b634 KS |
304 | /* protect arrays of thresholds */ |
305 | struct mutex thresholds_lock; | |
306 | ||
307 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 308 | struct mem_cgroup_thresholds thresholds; |
907860ed | 309 | |
2e72b634 | 310 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 311 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 312 | |
9490ff27 KH |
313 | /* For oom notifier event fd */ |
314 | struct list_head oom_notify; | |
185efc0f | 315 | |
7dc74be0 DN |
316 | /* |
317 | * Should we move charges of a task when a task is moved into this | |
318 | * mem_cgroup ? And what type of charges should we move ? | |
319 | */ | |
320 | unsigned long move_charge_at_immigrate; | |
619d094b KH |
321 | /* |
322 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
323 | */ | |
324 | atomic_t moving_account; | |
312734c0 KH |
325 | /* taken only while moving_account > 0 */ |
326 | spinlock_t move_lock; | |
d52aa412 | 327 | /* |
c62b1a3b | 328 | * percpu counter. |
d52aa412 | 329 | */ |
3a7951b4 | 330 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
331 | /* |
332 | * used when a cpu is offlined or other synchronizations | |
333 | * See mem_cgroup_read_stat(). | |
334 | */ | |
335 | struct mem_cgroup_stat_cpu nocpu_base; | |
336 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 337 | |
4bd2c1ee | 338 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
339 | struct tcp_memcontrol tcp_mem; |
340 | #endif | |
2633d7a0 GC |
341 | #if defined(CONFIG_MEMCG_KMEM) |
342 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
343 | struct list_head memcg_slab_caches; | |
344 | /* Not a spinlock, we can take a lot of time walking the list */ | |
345 | struct mutex slab_caches_mutex; | |
346 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
347 | int kmemcg_id; | |
348 | #endif | |
45cf7ebd GC |
349 | |
350 | int last_scanned_node; | |
351 | #if MAX_NUMNODES > 1 | |
352 | nodemask_t scan_nodes; | |
353 | atomic_t numainfo_events; | |
354 | atomic_t numainfo_updating; | |
355 | #endif | |
356 | /* | |
357 | * Per cgroup active and inactive list, similar to the | |
358 | * per zone LRU lists. | |
359 | * | |
360 | * WARNING: This has to be the last element of the struct. Don't | |
361 | * add new fields after this point. | |
362 | */ | |
363 | struct mem_cgroup_lru_info info; | |
8cdea7c0 BS |
364 | }; |
365 | ||
45cf7ebd GC |
366 | static size_t memcg_size(void) |
367 | { | |
368 | return sizeof(struct mem_cgroup) + | |
369 | nr_node_ids * sizeof(struct mem_cgroup_per_node); | |
370 | } | |
371 | ||
510fc4e1 GC |
372 | /* internal only representation about the status of kmem accounting. */ |
373 | enum { | |
374 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
a8964b9b | 375 | KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */ |
7de37682 | 376 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
377 | }; |
378 | ||
a8964b9b GC |
379 | /* We account when limit is on, but only after call sites are patched */ |
380 | #define KMEM_ACCOUNTED_MASK \ | |
381 | ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED)) | |
510fc4e1 GC |
382 | |
383 | #ifdef CONFIG_MEMCG_KMEM | |
384 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
385 | { | |
386 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
387 | } | |
7de37682 GC |
388 | |
389 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
390 | { | |
391 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
392 | } | |
393 | ||
a8964b9b GC |
394 | static void memcg_kmem_set_activated(struct mem_cgroup *memcg) |
395 | { | |
396 | set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
397 | } | |
398 | ||
55007d84 GC |
399 | static void memcg_kmem_clear_activated(struct mem_cgroup *memcg) |
400 | { | |
401 | clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
402 | } | |
403 | ||
7de37682 GC |
404 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
405 | { | |
406 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) | |
407 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
408 | } | |
409 | ||
410 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
411 | { | |
412 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
413 | &memcg->kmem_account_flags); | |
414 | } | |
510fc4e1 GC |
415 | #endif |
416 | ||
7dc74be0 DN |
417 | /* Stuffs for move charges at task migration. */ |
418 | /* | |
ee5e8472 GC |
419 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
420 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
421 | */ |
422 | enum move_type { | |
4ffef5fe | 423 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 424 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
425 | NR_MOVE_TYPE, |
426 | }; | |
427 | ||
4ffef5fe DN |
428 | /* "mc" and its members are protected by cgroup_mutex */ |
429 | static struct move_charge_struct { | |
b1dd693e | 430 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
431 | struct mem_cgroup *from; |
432 | struct mem_cgroup *to; | |
ee5e8472 | 433 | unsigned long immigrate_flags; |
4ffef5fe | 434 | unsigned long precharge; |
854ffa8d | 435 | unsigned long moved_charge; |
483c30b5 | 436 | unsigned long moved_swap; |
8033b97c DN |
437 | struct task_struct *moving_task; /* a task moving charges */ |
438 | wait_queue_head_t waitq; /* a waitq for other context */ | |
439 | } mc = { | |
2bd9bb20 | 440 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
441 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
442 | }; | |
4ffef5fe | 443 | |
90254a65 DN |
444 | static bool move_anon(void) |
445 | { | |
ee5e8472 | 446 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
447 | } |
448 | ||
87946a72 DN |
449 | static bool move_file(void) |
450 | { | |
ee5e8472 | 451 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
452 | } |
453 | ||
4e416953 BS |
454 | /* |
455 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
456 | * limit reclaim to prevent infinite loops, if they ever occur. | |
457 | */ | |
a0db00fc KS |
458 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
459 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 | |
4e416953 | 460 | |
217bc319 KH |
461 | enum charge_type { |
462 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 463 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 464 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 465 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
466 | NR_CHARGE_TYPE, |
467 | }; | |
468 | ||
8c7c6e34 | 469 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
470 | enum res_type { |
471 | _MEM, | |
472 | _MEMSWAP, | |
473 | _OOM_TYPE, | |
510fc4e1 | 474 | _KMEM, |
86ae53e1 GC |
475 | }; |
476 | ||
a0db00fc KS |
477 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
478 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 479 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
480 | /* Used for OOM nofiier */ |
481 | #define OOM_CONTROL (0) | |
8c7c6e34 | 482 | |
75822b44 BS |
483 | /* |
484 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
485 | */ | |
486 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
487 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
488 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
489 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
490 | ||
0999821b GC |
491 | /* |
492 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
493 | * As a consequence, any change that needs to protect against new child cgroups | |
494 | * appearing has to hold it as well. | |
495 | */ | |
496 | static DEFINE_MUTEX(memcg_create_mutex); | |
497 | ||
c0ff4b85 R |
498 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
499 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 | 500 | |
b2145145 WL |
501 | static inline |
502 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) | |
503 | { | |
504 | return container_of(s, struct mem_cgroup, css); | |
505 | } | |
506 | ||
7ffc0edc MH |
507 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
508 | { | |
509 | return (memcg == root_mem_cgroup); | |
510 | } | |
511 | ||
e1aab161 | 512 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 513 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 514 | |
e1aab161 GC |
515 | void sock_update_memcg(struct sock *sk) |
516 | { | |
376be5ff | 517 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 518 | struct mem_cgroup *memcg; |
3f134619 | 519 | struct cg_proto *cg_proto; |
e1aab161 GC |
520 | |
521 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
522 | ||
f3f511e1 GC |
523 | /* Socket cloning can throw us here with sk_cgrp already |
524 | * filled. It won't however, necessarily happen from | |
525 | * process context. So the test for root memcg given | |
526 | * the current task's memcg won't help us in this case. | |
527 | * | |
528 | * Respecting the original socket's memcg is a better | |
529 | * decision in this case. | |
530 | */ | |
531 | if (sk->sk_cgrp) { | |
532 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
533 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
534 | return; | |
535 | } | |
536 | ||
e1aab161 GC |
537 | rcu_read_lock(); |
538 | memcg = mem_cgroup_from_task(current); | |
3f134619 GC |
539 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
540 | if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) { | |
e1aab161 | 541 | mem_cgroup_get(memcg); |
3f134619 | 542 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
543 | } |
544 | rcu_read_unlock(); | |
545 | } | |
546 | } | |
547 | EXPORT_SYMBOL(sock_update_memcg); | |
548 | ||
549 | void sock_release_memcg(struct sock *sk) | |
550 | { | |
376be5ff | 551 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
552 | struct mem_cgroup *memcg; |
553 | WARN_ON(!sk->sk_cgrp->memcg); | |
554 | memcg = sk->sk_cgrp->memcg; | |
555 | mem_cgroup_put(memcg); | |
556 | } | |
557 | } | |
d1a4c0b3 GC |
558 | |
559 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
560 | { | |
561 | if (!memcg || mem_cgroup_is_root(memcg)) | |
562 | return NULL; | |
563 | ||
564 | return &memcg->tcp_mem.cg_proto; | |
565 | } | |
566 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 567 | |
3f134619 GC |
568 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
569 | { | |
570 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
571 | return; | |
572 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
573 | } | |
574 | #else | |
575 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
576 | { | |
577 | } | |
578 | #endif | |
579 | ||
a8964b9b | 580 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
581 | /* |
582 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
583 | * There are two main reasons for not using the css_id for this: | |
584 | * 1) this works better in sparse environments, where we have a lot of memcgs, | |
585 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
586 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
587 | * 200 entry array for that. | |
588 | * | |
589 | * 2) In order not to violate the cgroup API, we would like to do all memory | |
590 | * allocation in ->create(). At that point, we haven't yet allocated the | |
591 | * css_id. Having a separate index prevents us from messing with the cgroup | |
592 | * core for this | |
593 | * | |
594 | * The current size of the caches array is stored in | |
595 | * memcg_limited_groups_array_size. It will double each time we have to | |
596 | * increase it. | |
597 | */ | |
598 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
599 | int memcg_limited_groups_array_size; |
600 | ||
55007d84 GC |
601 | /* |
602 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
603 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
604 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
605 | * tunable, but that is strictly not necessary. | |
606 | * | |
607 | * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get | |
608 | * this constant directly from cgroup, but it is understandable that this is | |
609 | * better kept as an internal representation in cgroup.c. In any case, the | |
610 | * css_id space is not getting any smaller, and we don't have to necessarily | |
611 | * increase ours as well if it increases. | |
612 | */ | |
613 | #define MEMCG_CACHES_MIN_SIZE 4 | |
614 | #define MEMCG_CACHES_MAX_SIZE 65535 | |
615 | ||
d7f25f8a GC |
616 | /* |
617 | * A lot of the calls to the cache allocation functions are expected to be | |
618 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
619 | * conditional to this static branch, we'll have to allow modules that does | |
620 | * kmem_cache_alloc and the such to see this symbol as well | |
621 | */ | |
a8964b9b | 622 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 623 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
624 | |
625 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
626 | { | |
55007d84 | 627 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 628 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
629 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
630 | } | |
bea207c8 GC |
631 | /* |
632 | * This check can't live in kmem destruction function, | |
633 | * since the charges will outlive the cgroup | |
634 | */ | |
635 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
636 | } |
637 | #else | |
638 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
639 | { | |
640 | } | |
641 | #endif /* CONFIG_MEMCG_KMEM */ | |
642 | ||
643 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
644 | { | |
645 | disarm_sock_keys(memcg); | |
646 | disarm_kmem_keys(memcg); | |
647 | } | |
648 | ||
c0ff4b85 | 649 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 650 | |
f64c3f54 | 651 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 652 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 653 | { |
45cf7ebd | 654 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
c0ff4b85 | 655 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
656 | } |
657 | ||
c0ff4b85 | 658 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 659 | { |
c0ff4b85 | 660 | return &memcg->css; |
d324236b WF |
661 | } |
662 | ||
f64c3f54 | 663 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 664 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 665 | { |
97a6c37b JW |
666 | int nid = page_to_nid(page); |
667 | int zid = page_zonenum(page); | |
f64c3f54 | 668 | |
c0ff4b85 | 669 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
670 | } |
671 | ||
672 | static struct mem_cgroup_tree_per_zone * | |
673 | soft_limit_tree_node_zone(int nid, int zid) | |
674 | { | |
675 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
676 | } | |
677 | ||
678 | static struct mem_cgroup_tree_per_zone * | |
679 | soft_limit_tree_from_page(struct page *page) | |
680 | { | |
681 | int nid = page_to_nid(page); | |
682 | int zid = page_zonenum(page); | |
683 | ||
684 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
685 | } | |
686 | ||
687 | static void | |
c0ff4b85 | 688 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 689 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
690 | struct mem_cgroup_tree_per_zone *mctz, |
691 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
692 | { |
693 | struct rb_node **p = &mctz->rb_root.rb_node; | |
694 | struct rb_node *parent = NULL; | |
695 | struct mem_cgroup_per_zone *mz_node; | |
696 | ||
697 | if (mz->on_tree) | |
698 | return; | |
699 | ||
ef8745c1 KH |
700 | mz->usage_in_excess = new_usage_in_excess; |
701 | if (!mz->usage_in_excess) | |
702 | return; | |
f64c3f54 BS |
703 | while (*p) { |
704 | parent = *p; | |
705 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
706 | tree_node); | |
707 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
708 | p = &(*p)->rb_left; | |
709 | /* | |
710 | * We can't avoid mem cgroups that are over their soft | |
711 | * limit by the same amount | |
712 | */ | |
713 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
714 | p = &(*p)->rb_right; | |
715 | } | |
716 | rb_link_node(&mz->tree_node, parent, p); | |
717 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
718 | mz->on_tree = true; | |
4e416953 BS |
719 | } |
720 | ||
721 | static void | |
c0ff4b85 | 722 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
723 | struct mem_cgroup_per_zone *mz, |
724 | struct mem_cgroup_tree_per_zone *mctz) | |
725 | { | |
726 | if (!mz->on_tree) | |
727 | return; | |
728 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
729 | mz->on_tree = false; | |
730 | } | |
731 | ||
f64c3f54 | 732 | static void |
c0ff4b85 | 733 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
734 | struct mem_cgroup_per_zone *mz, |
735 | struct mem_cgroup_tree_per_zone *mctz) | |
736 | { | |
737 | spin_lock(&mctz->lock); | |
c0ff4b85 | 738 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
739 | spin_unlock(&mctz->lock); |
740 | } | |
741 | ||
f64c3f54 | 742 | |
c0ff4b85 | 743 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 744 | { |
ef8745c1 | 745 | unsigned long long excess; |
f64c3f54 BS |
746 | struct mem_cgroup_per_zone *mz; |
747 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
748 | int nid = page_to_nid(page); |
749 | int zid = page_zonenum(page); | |
f64c3f54 BS |
750 | mctz = soft_limit_tree_from_page(page); |
751 | ||
752 | /* | |
4e649152 KH |
753 | * Necessary to update all ancestors when hierarchy is used. |
754 | * because their event counter is not touched. | |
f64c3f54 | 755 | */ |
c0ff4b85 R |
756 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
757 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
758 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
759 | /* |
760 | * We have to update the tree if mz is on RB-tree or | |
761 | * mem is over its softlimit. | |
762 | */ | |
ef8745c1 | 763 | if (excess || mz->on_tree) { |
4e649152 KH |
764 | spin_lock(&mctz->lock); |
765 | /* if on-tree, remove it */ | |
766 | if (mz->on_tree) | |
c0ff4b85 | 767 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 768 | /* |
ef8745c1 KH |
769 | * Insert again. mz->usage_in_excess will be updated. |
770 | * If excess is 0, no tree ops. | |
4e649152 | 771 | */ |
c0ff4b85 | 772 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
773 | spin_unlock(&mctz->lock); |
774 | } | |
f64c3f54 BS |
775 | } |
776 | } | |
777 | ||
c0ff4b85 | 778 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
779 | { |
780 | int node, zone; | |
781 | struct mem_cgroup_per_zone *mz; | |
782 | struct mem_cgroup_tree_per_zone *mctz; | |
783 | ||
3ed28fa1 | 784 | for_each_node(node) { |
f64c3f54 | 785 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 786 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 787 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 788 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
789 | } |
790 | } | |
791 | } | |
792 | ||
4e416953 BS |
793 | static struct mem_cgroup_per_zone * |
794 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
795 | { | |
796 | struct rb_node *rightmost = NULL; | |
26251eaf | 797 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
798 | |
799 | retry: | |
26251eaf | 800 | mz = NULL; |
4e416953 BS |
801 | rightmost = rb_last(&mctz->rb_root); |
802 | if (!rightmost) | |
803 | goto done; /* Nothing to reclaim from */ | |
804 | ||
805 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
806 | /* | |
807 | * Remove the node now but someone else can add it back, | |
808 | * we will to add it back at the end of reclaim to its correct | |
809 | * position in the tree. | |
810 | */ | |
d79154bb HD |
811 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
812 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
813 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
814 | goto retry; |
815 | done: | |
816 | return mz; | |
817 | } | |
818 | ||
819 | static struct mem_cgroup_per_zone * | |
820 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
821 | { | |
822 | struct mem_cgroup_per_zone *mz; | |
823 | ||
824 | spin_lock(&mctz->lock); | |
825 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
826 | spin_unlock(&mctz->lock); | |
827 | return mz; | |
828 | } | |
829 | ||
711d3d2c KH |
830 | /* |
831 | * Implementation Note: reading percpu statistics for memcg. | |
832 | * | |
833 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
834 | * synchronization to implement "quick" read. There are trade-off between | |
835 | * reading cost and precision of value. Then, we may have a chance to implement | |
836 | * a periodic synchronizion of counter in memcg's counter. | |
837 | * | |
838 | * But this _read() function is used for user interface now. The user accounts | |
839 | * memory usage by memory cgroup and he _always_ requires exact value because | |
840 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
841 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
842 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
843 | * | |
844 | * If there are kernel internal actions which can make use of some not-exact | |
845 | * value, and reading all cpu value can be performance bottleneck in some | |
846 | * common workload, threashold and synchonization as vmstat[] should be | |
847 | * implemented. | |
848 | */ | |
c0ff4b85 | 849 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 850 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 851 | { |
7a159cc9 | 852 | long val = 0; |
c62b1a3b | 853 | int cpu; |
c62b1a3b | 854 | |
711d3d2c KH |
855 | get_online_cpus(); |
856 | for_each_online_cpu(cpu) | |
c0ff4b85 | 857 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 858 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
859 | spin_lock(&memcg->pcp_counter_lock); |
860 | val += memcg->nocpu_base.count[idx]; | |
861 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
862 | #endif |
863 | put_online_cpus(); | |
c62b1a3b KH |
864 | return val; |
865 | } | |
866 | ||
c0ff4b85 | 867 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
868 | bool charge) |
869 | { | |
870 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 871 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
872 | } |
873 | ||
c0ff4b85 | 874 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
875 | enum mem_cgroup_events_index idx) |
876 | { | |
877 | unsigned long val = 0; | |
878 | int cpu; | |
879 | ||
880 | for_each_online_cpu(cpu) | |
c0ff4b85 | 881 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 882 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
883 | spin_lock(&memcg->pcp_counter_lock); |
884 | val += memcg->nocpu_base.events[idx]; | |
885 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
886 | #endif |
887 | return val; | |
888 | } | |
889 | ||
c0ff4b85 | 890 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 891 | bool anon, int nr_pages) |
d52aa412 | 892 | { |
c62b1a3b KH |
893 | preempt_disable(); |
894 | ||
b2402857 KH |
895 | /* |
896 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
897 | * counted as CACHE even if it's on ANON LRU. | |
898 | */ | |
899 | if (anon) | |
900 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 901 | nr_pages); |
d52aa412 | 902 | else |
b2402857 | 903 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 904 | nr_pages); |
55e462b0 | 905 | |
e401f176 KH |
906 | /* pagein of a big page is an event. So, ignore page size */ |
907 | if (nr_pages > 0) | |
c0ff4b85 | 908 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 909 | else { |
c0ff4b85 | 910 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
911 | nr_pages = -nr_pages; /* for event */ |
912 | } | |
e401f176 | 913 | |
13114716 | 914 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 915 | |
c62b1a3b | 916 | preempt_enable(); |
6d12e2d8 KH |
917 | } |
918 | ||
bb2a0de9 | 919 | unsigned long |
4d7dcca2 | 920 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
921 | { |
922 | struct mem_cgroup_per_zone *mz; | |
923 | ||
924 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
925 | return mz->lru_size[lru]; | |
926 | } | |
927 | ||
928 | static unsigned long | |
c0ff4b85 | 929 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 930 | unsigned int lru_mask) |
889976db YH |
931 | { |
932 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 933 | enum lru_list lru; |
bb2a0de9 KH |
934 | unsigned long ret = 0; |
935 | ||
c0ff4b85 | 936 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 937 | |
f156ab93 HD |
938 | for_each_lru(lru) { |
939 | if (BIT(lru) & lru_mask) | |
940 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
941 | } |
942 | return ret; | |
943 | } | |
944 | ||
945 | static unsigned long | |
c0ff4b85 | 946 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
947 | int nid, unsigned int lru_mask) |
948 | { | |
889976db YH |
949 | u64 total = 0; |
950 | int zid; | |
951 | ||
bb2a0de9 | 952 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
953 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
954 | nid, zid, lru_mask); | |
bb2a0de9 | 955 | |
889976db YH |
956 | return total; |
957 | } | |
bb2a0de9 | 958 | |
c0ff4b85 | 959 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 960 | unsigned int lru_mask) |
6d12e2d8 | 961 | { |
889976db | 962 | int nid; |
6d12e2d8 KH |
963 | u64 total = 0; |
964 | ||
31aaea4a | 965 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 966 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 967 | return total; |
d52aa412 KH |
968 | } |
969 | ||
f53d7ce3 JW |
970 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
971 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
972 | { |
973 | unsigned long val, next; | |
974 | ||
13114716 | 975 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 976 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 977 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
978 | if ((long)next - (long)val < 0) { |
979 | switch (target) { | |
980 | case MEM_CGROUP_TARGET_THRESH: | |
981 | next = val + THRESHOLDS_EVENTS_TARGET; | |
982 | break; | |
983 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
984 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
985 | break; | |
986 | case MEM_CGROUP_TARGET_NUMAINFO: | |
987 | next = val + NUMAINFO_EVENTS_TARGET; | |
988 | break; | |
989 | default: | |
990 | break; | |
991 | } | |
992 | __this_cpu_write(memcg->stat->targets[target], next); | |
993 | return true; | |
7a159cc9 | 994 | } |
f53d7ce3 | 995 | return false; |
d2265e6f KH |
996 | } |
997 | ||
998 | /* | |
999 | * Check events in order. | |
1000 | * | |
1001 | */ | |
c0ff4b85 | 1002 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 1003 | { |
4799401f | 1004 | preempt_disable(); |
d2265e6f | 1005 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1006 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1007 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
1008 | bool do_softlimit; |
1009 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
1010 | |
1011 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
1012 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
1013 | #if MAX_NUMNODES > 1 | |
1014 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1015 | MEM_CGROUP_TARGET_NUMAINFO); | |
1016 | #endif | |
1017 | preempt_enable(); | |
1018 | ||
c0ff4b85 | 1019 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 1020 | if (unlikely(do_softlimit)) |
c0ff4b85 | 1021 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 1022 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1023 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1024 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1025 | #endif |
f53d7ce3 JW |
1026 | } else |
1027 | preempt_enable(); | |
d2265e6f KH |
1028 | } |
1029 | ||
d1a4c0b3 | 1030 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 | 1031 | { |
b2145145 WL |
1032 | return mem_cgroup_from_css( |
1033 | cgroup_subsys_state(cont, mem_cgroup_subsys_id)); | |
8cdea7c0 BS |
1034 | } |
1035 | ||
cf475ad2 | 1036 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1037 | { |
31a78f23 BS |
1038 | /* |
1039 | * mm_update_next_owner() may clear mm->owner to NULL | |
1040 | * if it races with swapoff, page migration, etc. | |
1041 | * So this can be called with p == NULL. | |
1042 | */ | |
1043 | if (unlikely(!p)) | |
1044 | return NULL; | |
1045 | ||
b2145145 | 1046 | return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
1047 | } |
1048 | ||
a433658c | 1049 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1050 | { |
c0ff4b85 | 1051 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
1052 | |
1053 | if (!mm) | |
1054 | return NULL; | |
54595fe2 KH |
1055 | /* |
1056 | * Because we have no locks, mm->owner's may be being moved to other | |
1057 | * cgroup. We use css_tryget() here even if this looks | |
1058 | * pessimistic (rather than adding locks here). | |
1059 | */ | |
1060 | rcu_read_lock(); | |
1061 | do { | |
c0ff4b85 R |
1062 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1063 | if (unlikely(!memcg)) | |
54595fe2 | 1064 | break; |
c0ff4b85 | 1065 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1066 | rcu_read_unlock(); |
c0ff4b85 | 1067 | return memcg; |
54595fe2 KH |
1068 | } |
1069 | ||
5660048c JW |
1070 | /** |
1071 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1072 | * @root: hierarchy root | |
1073 | * @prev: previously returned memcg, NULL on first invocation | |
1074 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1075 | * | |
1076 | * Returns references to children of the hierarchy below @root, or | |
1077 | * @root itself, or %NULL after a full round-trip. | |
1078 | * | |
1079 | * Caller must pass the return value in @prev on subsequent | |
1080 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1081 | * to cancel a hierarchy walk before the round-trip is complete. | |
1082 | * | |
1083 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1084 | * divide up the memcgs in the hierarchy among all concurrent | |
1085 | * reclaimers operating on the same zone and priority. | |
1086 | */ | |
1087 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
1088 | struct mem_cgroup *prev, | |
1089 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 1090 | { |
9f3a0d09 JW |
1091 | struct mem_cgroup *memcg = NULL; |
1092 | int id = 0; | |
711d3d2c | 1093 | |
5660048c JW |
1094 | if (mem_cgroup_disabled()) |
1095 | return NULL; | |
1096 | ||
9f3a0d09 JW |
1097 | if (!root) |
1098 | root = root_mem_cgroup; | |
7d74b06f | 1099 | |
9f3a0d09 JW |
1100 | if (prev && !reclaim) |
1101 | id = css_id(&prev->css); | |
14067bb3 | 1102 | |
9f3a0d09 JW |
1103 | if (prev && prev != root) |
1104 | css_put(&prev->css); | |
14067bb3 | 1105 | |
9f3a0d09 JW |
1106 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1107 | if (prev) | |
1108 | return NULL; | |
1109 | return root; | |
1110 | } | |
14067bb3 | 1111 | |
9f3a0d09 | 1112 | while (!memcg) { |
527a5ec9 | 1113 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 1114 | struct cgroup_subsys_state *css; |
711d3d2c | 1115 | |
527a5ec9 JW |
1116 | if (reclaim) { |
1117 | int nid = zone_to_nid(reclaim->zone); | |
1118 | int zid = zone_idx(reclaim->zone); | |
1119 | struct mem_cgroup_per_zone *mz; | |
1120 | ||
1121 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1122 | iter = &mz->reclaim_iter[reclaim->priority]; | |
1123 | if (prev && reclaim->generation != iter->generation) | |
1124 | return NULL; | |
1125 | id = iter->position; | |
1126 | } | |
7d74b06f | 1127 | |
9f3a0d09 JW |
1128 | rcu_read_lock(); |
1129 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
1130 | if (css) { | |
1131 | if (css == &root->css || css_tryget(css)) | |
b2145145 | 1132 | memcg = mem_cgroup_from_css(css); |
9f3a0d09 JW |
1133 | } else |
1134 | id = 0; | |
14067bb3 | 1135 | rcu_read_unlock(); |
14067bb3 | 1136 | |
527a5ec9 JW |
1137 | if (reclaim) { |
1138 | iter->position = id; | |
1139 | if (!css) | |
1140 | iter->generation++; | |
1141 | else if (!prev && memcg) | |
1142 | reclaim->generation = iter->generation; | |
1143 | } | |
9f3a0d09 JW |
1144 | |
1145 | if (prev && !css) | |
1146 | return NULL; | |
1147 | } | |
1148 | return memcg; | |
14067bb3 | 1149 | } |
7d74b06f | 1150 | |
5660048c JW |
1151 | /** |
1152 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1153 | * @root: hierarchy root | |
1154 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1155 | */ | |
1156 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1157 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1158 | { |
1159 | if (!root) | |
1160 | root = root_mem_cgroup; | |
1161 | if (prev && prev != root) | |
1162 | css_put(&prev->css); | |
1163 | } | |
7d74b06f | 1164 | |
9f3a0d09 JW |
1165 | /* |
1166 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1167 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1168 | * be used for reference counting. | |
1169 | */ | |
1170 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1171 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1172 | iter != NULL; \ |
527a5ec9 | 1173 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1174 | |
9f3a0d09 | 1175 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1176 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1177 | iter != NULL; \ |
527a5ec9 | 1178 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1179 | |
68ae564b | 1180 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1181 | { |
c0ff4b85 | 1182 | struct mem_cgroup *memcg; |
456f998e | 1183 | |
456f998e | 1184 | rcu_read_lock(); |
c0ff4b85 R |
1185 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1186 | if (unlikely(!memcg)) | |
456f998e YH |
1187 | goto out; |
1188 | ||
1189 | switch (idx) { | |
456f998e | 1190 | case PGFAULT: |
0e574a93 JW |
1191 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1192 | break; | |
1193 | case PGMAJFAULT: | |
1194 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1195 | break; |
1196 | default: | |
1197 | BUG(); | |
1198 | } | |
1199 | out: | |
1200 | rcu_read_unlock(); | |
1201 | } | |
68ae564b | 1202 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1203 | |
925b7673 JW |
1204 | /** |
1205 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1206 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1207 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1208 | * |
1209 | * Returns the lru list vector holding pages for the given @zone and | |
1210 | * @mem. This can be the global zone lruvec, if the memory controller | |
1211 | * is disabled. | |
1212 | */ | |
1213 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1214 | struct mem_cgroup *memcg) | |
1215 | { | |
1216 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1217 | struct lruvec *lruvec; |
925b7673 | 1218 | |
bea8c150 HD |
1219 | if (mem_cgroup_disabled()) { |
1220 | lruvec = &zone->lruvec; | |
1221 | goto out; | |
1222 | } | |
925b7673 JW |
1223 | |
1224 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1225 | lruvec = &mz->lruvec; |
1226 | out: | |
1227 | /* | |
1228 | * Since a node can be onlined after the mem_cgroup was created, | |
1229 | * we have to be prepared to initialize lruvec->zone here; | |
1230 | * and if offlined then reonlined, we need to reinitialize it. | |
1231 | */ | |
1232 | if (unlikely(lruvec->zone != zone)) | |
1233 | lruvec->zone = zone; | |
1234 | return lruvec; | |
925b7673 JW |
1235 | } |
1236 | ||
08e552c6 KH |
1237 | /* |
1238 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1239 | * Operations are called by routine of global LRU independently from memcg. | |
1240 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1241 | * | |
1242 | * Changes to pc->mem_cgroup happens when | |
1243 | * 1. charge | |
1244 | * 2. moving account | |
1245 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1246 | * It is added to LRU before charge. | |
1247 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1248 | * When moving account, the page is not on LRU. It's isolated. | |
1249 | */ | |
4f98a2fe | 1250 | |
925b7673 | 1251 | /** |
fa9add64 | 1252 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1253 | * @page: the page |
fa9add64 | 1254 | * @zone: zone of the page |
925b7673 | 1255 | */ |
fa9add64 | 1256 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1257 | { |
08e552c6 | 1258 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1259 | struct mem_cgroup *memcg; |
1260 | struct page_cgroup *pc; | |
bea8c150 | 1261 | struct lruvec *lruvec; |
6d12e2d8 | 1262 | |
bea8c150 HD |
1263 | if (mem_cgroup_disabled()) { |
1264 | lruvec = &zone->lruvec; | |
1265 | goto out; | |
1266 | } | |
925b7673 | 1267 | |
08e552c6 | 1268 | pc = lookup_page_cgroup(page); |
38c5d72f | 1269 | memcg = pc->mem_cgroup; |
7512102c HD |
1270 | |
1271 | /* | |
fa9add64 | 1272 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1273 | * an uncharged page off lru does nothing to secure |
1274 | * its former mem_cgroup from sudden removal. | |
1275 | * | |
1276 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1277 | * under page_cgroup lock: between them, they make all uses | |
1278 | * of pc->mem_cgroup safe. | |
1279 | */ | |
fa9add64 | 1280 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1281 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1282 | ||
925b7673 | 1283 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1284 | lruvec = &mz->lruvec; |
1285 | out: | |
1286 | /* | |
1287 | * Since a node can be onlined after the mem_cgroup was created, | |
1288 | * we have to be prepared to initialize lruvec->zone here; | |
1289 | * and if offlined then reonlined, we need to reinitialize it. | |
1290 | */ | |
1291 | if (unlikely(lruvec->zone != zone)) | |
1292 | lruvec->zone = zone; | |
1293 | return lruvec; | |
08e552c6 | 1294 | } |
b69408e8 | 1295 | |
925b7673 | 1296 | /** |
fa9add64 HD |
1297 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1298 | * @lruvec: mem_cgroup per zone lru vector | |
1299 | * @lru: index of lru list the page is sitting on | |
1300 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1301 | * |
fa9add64 HD |
1302 | * This function must be called when a page is added to or removed from an |
1303 | * lru list. | |
3f58a829 | 1304 | */ |
fa9add64 HD |
1305 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1306 | int nr_pages) | |
3f58a829 MK |
1307 | { |
1308 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1309 | unsigned long *lru_size; |
3f58a829 MK |
1310 | |
1311 | if (mem_cgroup_disabled()) | |
1312 | return; | |
1313 | ||
fa9add64 HD |
1314 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1315 | lru_size = mz->lru_size + lru; | |
1316 | *lru_size += nr_pages; | |
1317 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1318 | } |
544122e5 | 1319 | |
3e92041d | 1320 | /* |
c0ff4b85 | 1321 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1322 | * hierarchy subtree |
1323 | */ | |
c3ac9a8a JW |
1324 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1325 | struct mem_cgroup *memcg) | |
3e92041d | 1326 | { |
91c63734 JW |
1327 | if (root_memcg == memcg) |
1328 | return true; | |
3a981f48 | 1329 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1330 | return false; |
c3ac9a8a JW |
1331 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1332 | } | |
1333 | ||
1334 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1335 | struct mem_cgroup *memcg) | |
1336 | { | |
1337 | bool ret; | |
1338 | ||
91c63734 | 1339 | rcu_read_lock(); |
c3ac9a8a | 1340 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1341 | rcu_read_unlock(); |
1342 | return ret; | |
3e92041d MH |
1343 | } |
1344 | ||
c0ff4b85 | 1345 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1346 | { |
1347 | int ret; | |
0b7f569e | 1348 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1349 | struct task_struct *p; |
4c4a2214 | 1350 | |
158e0a2d | 1351 | p = find_lock_task_mm(task); |
de077d22 DR |
1352 | if (p) { |
1353 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1354 | task_unlock(p); | |
1355 | } else { | |
1356 | /* | |
1357 | * All threads may have already detached their mm's, but the oom | |
1358 | * killer still needs to detect if they have already been oom | |
1359 | * killed to prevent needlessly killing additional tasks. | |
1360 | */ | |
1361 | task_lock(task); | |
1362 | curr = mem_cgroup_from_task(task); | |
1363 | if (curr) | |
1364 | css_get(&curr->css); | |
1365 | task_unlock(task); | |
1366 | } | |
0b7f569e KH |
1367 | if (!curr) |
1368 | return 0; | |
d31f56db | 1369 | /* |
c0ff4b85 | 1370 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1371 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1372 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1373 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1374 | */ |
c0ff4b85 | 1375 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1376 | css_put(&curr->css); |
4c4a2214 DR |
1377 | return ret; |
1378 | } | |
1379 | ||
c56d5c7d | 1380 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1381 | { |
9b272977 | 1382 | unsigned long inactive_ratio; |
14797e23 | 1383 | unsigned long inactive; |
9b272977 | 1384 | unsigned long active; |
c772be93 | 1385 | unsigned long gb; |
14797e23 | 1386 | |
4d7dcca2 HD |
1387 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1388 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1389 | |
c772be93 KM |
1390 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1391 | if (gb) | |
1392 | inactive_ratio = int_sqrt(10 * gb); | |
1393 | else | |
1394 | inactive_ratio = 1; | |
1395 | ||
9b272977 | 1396 | return inactive * inactive_ratio < active; |
14797e23 KM |
1397 | } |
1398 | ||
6d61ef40 BS |
1399 | #define mem_cgroup_from_res_counter(counter, member) \ |
1400 | container_of(counter, struct mem_cgroup, member) | |
1401 | ||
19942822 | 1402 | /** |
9d11ea9f | 1403 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1404 | * @memcg: the memory cgroup |
19942822 | 1405 | * |
9d11ea9f | 1406 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1407 | * pages. |
19942822 | 1408 | */ |
c0ff4b85 | 1409 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1410 | { |
9d11ea9f JW |
1411 | unsigned long long margin; |
1412 | ||
c0ff4b85 | 1413 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1414 | if (do_swap_account) |
c0ff4b85 | 1415 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1416 | return margin >> PAGE_SHIFT; |
19942822 JW |
1417 | } |
1418 | ||
1f4c025b | 1419 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1420 | { |
1421 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1422 | |
1423 | /* root ? */ | |
1424 | if (cgrp->parent == NULL) | |
1425 | return vm_swappiness; | |
1426 | ||
bf1ff263 | 1427 | return memcg->swappiness; |
a7885eb8 KM |
1428 | } |
1429 | ||
619d094b KH |
1430 | /* |
1431 | * memcg->moving_account is used for checking possibility that some thread is | |
1432 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1433 | * a memcg, other threads should check memcg->moving_account under | |
1434 | * rcu_read_lock(), like this: | |
1435 | * | |
1436 | * CPU-A CPU-B | |
1437 | * rcu_read_lock() | |
1438 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1439 | * take heavy locks. | |
1440 | * synchronize_rcu() update something. | |
1441 | * rcu_read_unlock() | |
1442 | * start move here. | |
1443 | */ | |
4331f7d3 KH |
1444 | |
1445 | /* for quick checking without looking up memcg */ | |
1446 | atomic_t memcg_moving __read_mostly; | |
1447 | ||
c0ff4b85 | 1448 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1449 | { |
4331f7d3 | 1450 | atomic_inc(&memcg_moving); |
619d094b | 1451 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1452 | synchronize_rcu(); |
1453 | } | |
1454 | ||
c0ff4b85 | 1455 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1456 | { |
619d094b KH |
1457 | /* |
1458 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1459 | * We check NULL in callee rather than caller. | |
1460 | */ | |
4331f7d3 KH |
1461 | if (memcg) { |
1462 | atomic_dec(&memcg_moving); | |
619d094b | 1463 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1464 | } |
32047e2a | 1465 | } |
619d094b | 1466 | |
32047e2a KH |
1467 | /* |
1468 | * 2 routines for checking "mem" is under move_account() or not. | |
1469 | * | |
13fd1dd9 AM |
1470 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1471 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1472 | * pc->mem_cgroup may be overwritten. |
1473 | * | |
1474 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1475 | * under hierarchy of moving cgroups. This is for | |
1476 | * waiting at hith-memory prressure caused by "move". | |
1477 | */ | |
1478 | ||
13fd1dd9 | 1479 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1480 | { |
1481 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1482 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1483 | } |
4b534334 | 1484 | |
c0ff4b85 | 1485 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1486 | { |
2bd9bb20 KH |
1487 | struct mem_cgroup *from; |
1488 | struct mem_cgroup *to; | |
4b534334 | 1489 | bool ret = false; |
2bd9bb20 KH |
1490 | /* |
1491 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1492 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1493 | */ | |
1494 | spin_lock(&mc.lock); | |
1495 | from = mc.from; | |
1496 | to = mc.to; | |
1497 | if (!from) | |
1498 | goto unlock; | |
3e92041d | 1499 | |
c0ff4b85 R |
1500 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1501 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1502 | unlock: |
1503 | spin_unlock(&mc.lock); | |
4b534334 KH |
1504 | return ret; |
1505 | } | |
1506 | ||
c0ff4b85 | 1507 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1508 | { |
1509 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1510 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1511 | DEFINE_WAIT(wait); |
1512 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1513 | /* moving charge context might have finished. */ | |
1514 | if (mc.moving_task) | |
1515 | schedule(); | |
1516 | finish_wait(&mc.waitq, &wait); | |
1517 | return true; | |
1518 | } | |
1519 | } | |
1520 | return false; | |
1521 | } | |
1522 | ||
312734c0 KH |
1523 | /* |
1524 | * Take this lock when | |
1525 | * - a code tries to modify page's memcg while it's USED. | |
1526 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1527 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1528 | */ |
1529 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1530 | unsigned long *flags) | |
1531 | { | |
1532 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1533 | } | |
1534 | ||
1535 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1536 | unsigned long *flags) | |
1537 | { | |
1538 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1539 | } | |
1540 | ||
58cf188e | 1541 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1542 | /** |
58cf188e | 1543 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1544 | * @memcg: The memory cgroup that went over limit |
1545 | * @p: Task that is going to be killed | |
1546 | * | |
1547 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1548 | * enabled | |
1549 | */ | |
1550 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1551 | { | |
1552 | struct cgroup *task_cgrp; | |
1553 | struct cgroup *mem_cgrp; | |
1554 | /* | |
1555 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1556 | * on the assumption that OOM is serialized for memory controller. | |
1557 | * If this assumption is broken, revisit this code. | |
1558 | */ | |
1559 | static char memcg_name[PATH_MAX]; | |
1560 | int ret; | |
58cf188e SZ |
1561 | struct mem_cgroup *iter; |
1562 | unsigned int i; | |
e222432b | 1563 | |
58cf188e | 1564 | if (!p) |
e222432b BS |
1565 | return; |
1566 | ||
e222432b BS |
1567 | rcu_read_lock(); |
1568 | ||
1569 | mem_cgrp = memcg->css.cgroup; | |
1570 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1571 | ||
1572 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1573 | if (ret < 0) { | |
1574 | /* | |
1575 | * Unfortunately, we are unable to convert to a useful name | |
1576 | * But we'll still print out the usage information | |
1577 | */ | |
1578 | rcu_read_unlock(); | |
1579 | goto done; | |
1580 | } | |
1581 | rcu_read_unlock(); | |
1582 | ||
d045197f | 1583 | pr_info("Task in %s killed", memcg_name); |
e222432b BS |
1584 | |
1585 | rcu_read_lock(); | |
1586 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1587 | if (ret < 0) { | |
1588 | rcu_read_unlock(); | |
1589 | goto done; | |
1590 | } | |
1591 | rcu_read_unlock(); | |
1592 | ||
1593 | /* | |
1594 | * Continues from above, so we don't need an KERN_ level | |
1595 | */ | |
d045197f | 1596 | pr_cont(" as a result of limit of %s\n", memcg_name); |
e222432b BS |
1597 | done: |
1598 | ||
d045197f | 1599 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1600 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1601 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1602 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1603 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1604 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1605 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1606 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1607 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1608 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1609 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1610 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1611 | |
1612 | for_each_mem_cgroup_tree(iter, memcg) { | |
1613 | pr_info("Memory cgroup stats"); | |
1614 | ||
1615 | rcu_read_lock(); | |
1616 | ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX); | |
1617 | if (!ret) | |
1618 | pr_cont(" for %s", memcg_name); | |
1619 | rcu_read_unlock(); | |
1620 | pr_cont(":"); | |
1621 | ||
1622 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1623 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1624 | continue; | |
1625 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1626 | K(mem_cgroup_read_stat(iter, i))); | |
1627 | } | |
1628 | ||
1629 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1630 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1631 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1632 | ||
1633 | pr_cont("\n"); | |
1634 | } | |
e222432b BS |
1635 | } |
1636 | ||
81d39c20 KH |
1637 | /* |
1638 | * This function returns the number of memcg under hierarchy tree. Returns | |
1639 | * 1(self count) if no children. | |
1640 | */ | |
c0ff4b85 | 1641 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1642 | { |
1643 | int num = 0; | |
7d74b06f KH |
1644 | struct mem_cgroup *iter; |
1645 | ||
c0ff4b85 | 1646 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1647 | num++; |
81d39c20 KH |
1648 | return num; |
1649 | } | |
1650 | ||
a63d83f4 DR |
1651 | /* |
1652 | * Return the memory (and swap, if configured) limit for a memcg. | |
1653 | */ | |
9cbb78bb | 1654 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1655 | { |
1656 | u64 limit; | |
a63d83f4 | 1657 | |
f3e8eb70 | 1658 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1659 | |
a63d83f4 | 1660 | /* |
9a5a8f19 | 1661 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1662 | */ |
9a5a8f19 MH |
1663 | if (mem_cgroup_swappiness(memcg)) { |
1664 | u64 memsw; | |
1665 | ||
1666 | limit += total_swap_pages << PAGE_SHIFT; | |
1667 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1668 | ||
1669 | /* | |
1670 | * If memsw is finite and limits the amount of swap space | |
1671 | * available to this memcg, return that limit. | |
1672 | */ | |
1673 | limit = min(limit, memsw); | |
1674 | } | |
1675 | ||
1676 | return limit; | |
a63d83f4 DR |
1677 | } |
1678 | ||
19965460 DR |
1679 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1680 | int order) | |
9cbb78bb DR |
1681 | { |
1682 | struct mem_cgroup *iter; | |
1683 | unsigned long chosen_points = 0; | |
1684 | unsigned long totalpages; | |
1685 | unsigned int points = 0; | |
1686 | struct task_struct *chosen = NULL; | |
1687 | ||
876aafbf DR |
1688 | /* |
1689 | * If current has a pending SIGKILL, then automatically select it. The | |
1690 | * goal is to allow it to allocate so that it may quickly exit and free | |
1691 | * its memory. | |
1692 | */ | |
1693 | if (fatal_signal_pending(current)) { | |
1694 | set_thread_flag(TIF_MEMDIE); | |
1695 | return; | |
1696 | } | |
1697 | ||
1698 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1699 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1700 | for_each_mem_cgroup_tree(iter, memcg) { | |
1701 | struct cgroup *cgroup = iter->css.cgroup; | |
1702 | struct cgroup_iter it; | |
1703 | struct task_struct *task; | |
1704 | ||
1705 | cgroup_iter_start(cgroup, &it); | |
1706 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
1707 | switch (oom_scan_process_thread(task, totalpages, NULL, | |
1708 | false)) { | |
1709 | case OOM_SCAN_SELECT: | |
1710 | if (chosen) | |
1711 | put_task_struct(chosen); | |
1712 | chosen = task; | |
1713 | chosen_points = ULONG_MAX; | |
1714 | get_task_struct(chosen); | |
1715 | /* fall through */ | |
1716 | case OOM_SCAN_CONTINUE: | |
1717 | continue; | |
1718 | case OOM_SCAN_ABORT: | |
1719 | cgroup_iter_end(cgroup, &it); | |
1720 | mem_cgroup_iter_break(memcg, iter); | |
1721 | if (chosen) | |
1722 | put_task_struct(chosen); | |
1723 | return; | |
1724 | case OOM_SCAN_OK: | |
1725 | break; | |
1726 | }; | |
1727 | points = oom_badness(task, memcg, NULL, totalpages); | |
1728 | if (points > chosen_points) { | |
1729 | if (chosen) | |
1730 | put_task_struct(chosen); | |
1731 | chosen = task; | |
1732 | chosen_points = points; | |
1733 | get_task_struct(chosen); | |
1734 | } | |
1735 | } | |
1736 | cgroup_iter_end(cgroup, &it); | |
1737 | } | |
1738 | ||
1739 | if (!chosen) | |
1740 | return; | |
1741 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1742 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1743 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1744 | } |
1745 | ||
5660048c JW |
1746 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1747 | gfp_t gfp_mask, | |
1748 | unsigned long flags) | |
1749 | { | |
1750 | unsigned long total = 0; | |
1751 | bool noswap = false; | |
1752 | int loop; | |
1753 | ||
1754 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1755 | noswap = true; | |
1756 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1757 | noswap = true; | |
1758 | ||
1759 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1760 | if (loop) | |
1761 | drain_all_stock_async(memcg); | |
1762 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1763 | /* | |
1764 | * Allow limit shrinkers, which are triggered directly | |
1765 | * by userspace, to catch signals and stop reclaim | |
1766 | * after minimal progress, regardless of the margin. | |
1767 | */ | |
1768 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1769 | break; | |
1770 | if (mem_cgroup_margin(memcg)) | |
1771 | break; | |
1772 | /* | |
1773 | * If nothing was reclaimed after two attempts, there | |
1774 | * may be no reclaimable pages in this hierarchy. | |
1775 | */ | |
1776 | if (loop && !total) | |
1777 | break; | |
1778 | } | |
1779 | return total; | |
1780 | } | |
1781 | ||
4d0c066d KH |
1782 | /** |
1783 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1784 | * @memcg: the target memcg |
4d0c066d KH |
1785 | * @nid: the node ID to be checked. |
1786 | * @noswap : specify true here if the user wants flle only information. | |
1787 | * | |
1788 | * This function returns whether the specified memcg contains any | |
1789 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1790 | * pages in the node. | |
1791 | */ | |
c0ff4b85 | 1792 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1793 | int nid, bool noswap) |
1794 | { | |
c0ff4b85 | 1795 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1796 | return true; |
1797 | if (noswap || !total_swap_pages) | |
1798 | return false; | |
c0ff4b85 | 1799 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1800 | return true; |
1801 | return false; | |
1802 | ||
1803 | } | |
889976db YH |
1804 | #if MAX_NUMNODES > 1 |
1805 | ||
1806 | /* | |
1807 | * Always updating the nodemask is not very good - even if we have an empty | |
1808 | * list or the wrong list here, we can start from some node and traverse all | |
1809 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1810 | * | |
1811 | */ | |
c0ff4b85 | 1812 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1813 | { |
1814 | int nid; | |
453a9bf3 KH |
1815 | /* |
1816 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1817 | * pagein/pageout changes since the last update. | |
1818 | */ | |
c0ff4b85 | 1819 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1820 | return; |
c0ff4b85 | 1821 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1822 | return; |
1823 | ||
889976db | 1824 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1825 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1826 | |
31aaea4a | 1827 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1828 | |
c0ff4b85 R |
1829 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1830 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1831 | } |
453a9bf3 | 1832 | |
c0ff4b85 R |
1833 | atomic_set(&memcg->numainfo_events, 0); |
1834 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1835 | } |
1836 | ||
1837 | /* | |
1838 | * Selecting a node where we start reclaim from. Because what we need is just | |
1839 | * reducing usage counter, start from anywhere is O,K. Considering | |
1840 | * memory reclaim from current node, there are pros. and cons. | |
1841 | * | |
1842 | * Freeing memory from current node means freeing memory from a node which | |
1843 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1844 | * hit limits, it will see a contention on a node. But freeing from remote | |
1845 | * node means more costs for memory reclaim because of memory latency. | |
1846 | * | |
1847 | * Now, we use round-robin. Better algorithm is welcomed. | |
1848 | */ | |
c0ff4b85 | 1849 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1850 | { |
1851 | int node; | |
1852 | ||
c0ff4b85 R |
1853 | mem_cgroup_may_update_nodemask(memcg); |
1854 | node = memcg->last_scanned_node; | |
889976db | 1855 | |
c0ff4b85 | 1856 | node = next_node(node, memcg->scan_nodes); |
889976db | 1857 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1858 | node = first_node(memcg->scan_nodes); |
889976db YH |
1859 | /* |
1860 | * We call this when we hit limit, not when pages are added to LRU. | |
1861 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1862 | * memcg is too small and all pages are not on LRU. In that case, | |
1863 | * we use curret node. | |
1864 | */ | |
1865 | if (unlikely(node == MAX_NUMNODES)) | |
1866 | node = numa_node_id(); | |
1867 | ||
c0ff4b85 | 1868 | memcg->last_scanned_node = node; |
889976db YH |
1869 | return node; |
1870 | } | |
1871 | ||
4d0c066d KH |
1872 | /* |
1873 | * Check all nodes whether it contains reclaimable pages or not. | |
1874 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1875 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1876 | * enough new information. We need to do double check. | |
1877 | */ | |
6bbda35c | 1878 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1879 | { |
1880 | int nid; | |
1881 | ||
1882 | /* | |
1883 | * quick check...making use of scan_node. | |
1884 | * We can skip unused nodes. | |
1885 | */ | |
c0ff4b85 R |
1886 | if (!nodes_empty(memcg->scan_nodes)) { |
1887 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1888 | nid < MAX_NUMNODES; |
c0ff4b85 | 1889 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1890 | |
c0ff4b85 | 1891 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1892 | return true; |
1893 | } | |
1894 | } | |
1895 | /* | |
1896 | * Check rest of nodes. | |
1897 | */ | |
31aaea4a | 1898 | for_each_node_state(nid, N_MEMORY) { |
c0ff4b85 | 1899 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1900 | continue; |
c0ff4b85 | 1901 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1902 | return true; |
1903 | } | |
1904 | return false; | |
1905 | } | |
1906 | ||
889976db | 1907 | #else |
c0ff4b85 | 1908 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1909 | { |
1910 | return 0; | |
1911 | } | |
4d0c066d | 1912 | |
6bbda35c | 1913 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1914 | { |
c0ff4b85 | 1915 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1916 | } |
889976db YH |
1917 | #endif |
1918 | ||
5660048c JW |
1919 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1920 | struct zone *zone, | |
1921 | gfp_t gfp_mask, | |
1922 | unsigned long *total_scanned) | |
6d61ef40 | 1923 | { |
9f3a0d09 | 1924 | struct mem_cgroup *victim = NULL; |
5660048c | 1925 | int total = 0; |
04046e1a | 1926 | int loop = 0; |
9d11ea9f | 1927 | unsigned long excess; |
185efc0f | 1928 | unsigned long nr_scanned; |
527a5ec9 JW |
1929 | struct mem_cgroup_reclaim_cookie reclaim = { |
1930 | .zone = zone, | |
1931 | .priority = 0, | |
1932 | }; | |
9d11ea9f | 1933 | |
c0ff4b85 | 1934 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1935 | |
4e416953 | 1936 | while (1) { |
527a5ec9 | 1937 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1938 | if (!victim) { |
04046e1a | 1939 | loop++; |
4e416953 BS |
1940 | if (loop >= 2) { |
1941 | /* | |
1942 | * If we have not been able to reclaim | |
1943 | * anything, it might because there are | |
1944 | * no reclaimable pages under this hierarchy | |
1945 | */ | |
5660048c | 1946 | if (!total) |
4e416953 | 1947 | break; |
4e416953 | 1948 | /* |
25985edc | 1949 | * We want to do more targeted reclaim. |
4e416953 BS |
1950 | * excess >> 2 is not to excessive so as to |
1951 | * reclaim too much, nor too less that we keep | |
1952 | * coming back to reclaim from this cgroup | |
1953 | */ | |
1954 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1955 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1956 | break; |
4e416953 | 1957 | } |
9f3a0d09 | 1958 | continue; |
4e416953 | 1959 | } |
5660048c | 1960 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1961 | continue; |
5660048c JW |
1962 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1963 | zone, &nr_scanned); | |
1964 | *total_scanned += nr_scanned; | |
1965 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1966 | break; |
6d61ef40 | 1967 | } |
9f3a0d09 | 1968 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1969 | return total; |
6d61ef40 BS |
1970 | } |
1971 | ||
867578cb KH |
1972 | /* |
1973 | * Check OOM-Killer is already running under our hierarchy. | |
1974 | * If someone is running, return false. | |
1af8efe9 | 1975 | * Has to be called with memcg_oom_lock |
867578cb | 1976 | */ |
c0ff4b85 | 1977 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1978 | { |
79dfdacc | 1979 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1980 | |
9f3a0d09 | 1981 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1982 | if (iter->oom_lock) { |
79dfdacc MH |
1983 | /* |
1984 | * this subtree of our hierarchy is already locked | |
1985 | * so we cannot give a lock. | |
1986 | */ | |
79dfdacc | 1987 | failed = iter; |
9f3a0d09 JW |
1988 | mem_cgroup_iter_break(memcg, iter); |
1989 | break; | |
23751be0 JW |
1990 | } else |
1991 | iter->oom_lock = true; | |
7d74b06f | 1992 | } |
867578cb | 1993 | |
79dfdacc | 1994 | if (!failed) |
23751be0 | 1995 | return true; |
79dfdacc MH |
1996 | |
1997 | /* | |
1998 | * OK, we failed to lock the whole subtree so we have to clean up | |
1999 | * what we set up to the failing subtree | |
2000 | */ | |
9f3a0d09 | 2001 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 2002 | if (iter == failed) { |
9f3a0d09 JW |
2003 | mem_cgroup_iter_break(memcg, iter); |
2004 | break; | |
79dfdacc MH |
2005 | } |
2006 | iter->oom_lock = false; | |
2007 | } | |
23751be0 | 2008 | return false; |
a636b327 | 2009 | } |
0b7f569e | 2010 | |
79dfdacc | 2011 | /* |
1af8efe9 | 2012 | * Has to be called with memcg_oom_lock |
79dfdacc | 2013 | */ |
c0ff4b85 | 2014 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2015 | { |
7d74b06f KH |
2016 | struct mem_cgroup *iter; |
2017 | ||
c0ff4b85 | 2018 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2019 | iter->oom_lock = false; |
2020 | return 0; | |
2021 | } | |
2022 | ||
c0ff4b85 | 2023 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2024 | { |
2025 | struct mem_cgroup *iter; | |
2026 | ||
c0ff4b85 | 2027 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2028 | atomic_inc(&iter->under_oom); |
2029 | } | |
2030 | ||
c0ff4b85 | 2031 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2032 | { |
2033 | struct mem_cgroup *iter; | |
2034 | ||
867578cb KH |
2035 | /* |
2036 | * When a new child is created while the hierarchy is under oom, | |
2037 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2038 | * atomic_add_unless() here. | |
2039 | */ | |
c0ff4b85 | 2040 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2041 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2042 | } |
2043 | ||
1af8efe9 | 2044 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
2045 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2046 | ||
dc98df5a | 2047 | struct oom_wait_info { |
d79154bb | 2048 | struct mem_cgroup *memcg; |
dc98df5a KH |
2049 | wait_queue_t wait; |
2050 | }; | |
2051 | ||
2052 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2053 | unsigned mode, int sync, void *arg) | |
2054 | { | |
d79154bb HD |
2055 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2056 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2057 | struct oom_wait_info *oom_wait_info; |
2058 | ||
2059 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2060 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2061 | |
dc98df5a | 2062 | /* |
d79154bb | 2063 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2064 | * Then we can use css_is_ancestor without taking care of RCU. |
2065 | */ | |
c0ff4b85 R |
2066 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2067 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2068 | return 0; |
dc98df5a KH |
2069 | return autoremove_wake_function(wait, mode, sync, arg); |
2070 | } | |
2071 | ||
c0ff4b85 | 2072 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2073 | { |
c0ff4b85 R |
2074 | /* for filtering, pass "memcg" as argument. */ |
2075 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2076 | } |
2077 | ||
c0ff4b85 | 2078 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2079 | { |
c0ff4b85 R |
2080 | if (memcg && atomic_read(&memcg->under_oom)) |
2081 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2082 | } |
2083 | ||
867578cb KH |
2084 | /* |
2085 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
2086 | */ | |
6bbda35c KS |
2087 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
2088 | int order) | |
0b7f569e | 2089 | { |
dc98df5a | 2090 | struct oom_wait_info owait; |
3c11ecf4 | 2091 | bool locked, need_to_kill; |
867578cb | 2092 | |
d79154bb | 2093 | owait.memcg = memcg; |
dc98df5a KH |
2094 | owait.wait.flags = 0; |
2095 | owait.wait.func = memcg_oom_wake_function; | |
2096 | owait.wait.private = current; | |
2097 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 2098 | need_to_kill = true; |
c0ff4b85 | 2099 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 2100 | |
c0ff4b85 | 2101 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 2102 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 2103 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
2104 | /* |
2105 | * Even if signal_pending(), we can't quit charge() loop without | |
2106 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
2107 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
2108 | */ | |
3c11ecf4 | 2109 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 2110 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
2111 | need_to_kill = false; |
2112 | if (locked) | |
c0ff4b85 | 2113 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 2114 | spin_unlock(&memcg_oom_lock); |
867578cb | 2115 | |
3c11ecf4 KH |
2116 | if (need_to_kill) { |
2117 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 2118 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 2119 | } else { |
867578cb | 2120 | schedule(); |
dc98df5a | 2121 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 2122 | } |
1af8efe9 | 2123 | spin_lock(&memcg_oom_lock); |
79dfdacc | 2124 | if (locked) |
c0ff4b85 R |
2125 | mem_cgroup_oom_unlock(memcg); |
2126 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 2127 | spin_unlock(&memcg_oom_lock); |
867578cb | 2128 | |
c0ff4b85 | 2129 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 2130 | |
867578cb KH |
2131 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
2132 | return false; | |
2133 | /* Give chance to dying process */ | |
715a5ee8 | 2134 | schedule_timeout_uninterruptible(1); |
867578cb | 2135 | return true; |
0b7f569e KH |
2136 | } |
2137 | ||
d69b042f BS |
2138 | /* |
2139 | * Currently used to update mapped file statistics, but the routine can be | |
2140 | * generalized to update other statistics as well. | |
32047e2a KH |
2141 | * |
2142 | * Notes: Race condition | |
2143 | * | |
2144 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2145 | * it tends to be costly. But considering some conditions, we doesn't need | |
2146 | * to do so _always_. | |
2147 | * | |
2148 | * Considering "charge", lock_page_cgroup() is not required because all | |
2149 | * file-stat operations happen after a page is attached to radix-tree. There | |
2150 | * are no race with "charge". | |
2151 | * | |
2152 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2153 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2154 | * if there are race with "uncharge". Statistics itself is properly handled | |
2155 | * by flags. | |
2156 | * | |
2157 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2158 | * small, we check mm->moving_account and detect there are possibility of race |
2159 | * If there is, we take a lock. | |
d69b042f | 2160 | */ |
26174efd | 2161 | |
89c06bd5 KH |
2162 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2163 | bool *locked, unsigned long *flags) | |
2164 | { | |
2165 | struct mem_cgroup *memcg; | |
2166 | struct page_cgroup *pc; | |
2167 | ||
2168 | pc = lookup_page_cgroup(page); | |
2169 | again: | |
2170 | memcg = pc->mem_cgroup; | |
2171 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2172 | return; | |
2173 | /* | |
2174 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2175 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2176 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2177 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2178 | */ |
13fd1dd9 | 2179 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2180 | return; |
2181 | ||
2182 | move_lock_mem_cgroup(memcg, flags); | |
2183 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2184 | move_unlock_mem_cgroup(memcg, flags); | |
2185 | goto again; | |
2186 | } | |
2187 | *locked = true; | |
2188 | } | |
2189 | ||
2190 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2191 | { | |
2192 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2193 | ||
2194 | /* | |
2195 | * It's guaranteed that pc->mem_cgroup never changes while | |
2196 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2197 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2198 | */ |
2199 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2200 | } | |
2201 | ||
2a7106f2 GT |
2202 | void mem_cgroup_update_page_stat(struct page *page, |
2203 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 2204 | { |
c0ff4b85 | 2205 | struct mem_cgroup *memcg; |
32047e2a | 2206 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2207 | unsigned long uninitialized_var(flags); |
d69b042f | 2208 | |
cfa44946 | 2209 | if (mem_cgroup_disabled()) |
d69b042f | 2210 | return; |
89c06bd5 | 2211 | |
c0ff4b85 R |
2212 | memcg = pc->mem_cgroup; |
2213 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2214 | return; |
26174efd | 2215 | |
26174efd | 2216 | switch (idx) { |
2a7106f2 | 2217 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2218 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2219 | break; |
2220 | default: | |
2221 | BUG(); | |
8725d541 | 2222 | } |
d69b042f | 2223 | |
c0ff4b85 | 2224 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2225 | } |
26174efd | 2226 | |
cdec2e42 KH |
2227 | /* |
2228 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2229 | * TODO: maybe necessary to use big numbers in big irons. | |
2230 | */ | |
7ec99d62 | 2231 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2232 | struct memcg_stock_pcp { |
2233 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2234 | unsigned int nr_pages; |
cdec2e42 | 2235 | struct work_struct work; |
26fe6168 | 2236 | unsigned long flags; |
a0db00fc | 2237 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2238 | }; |
2239 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2240 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2241 | |
a0956d54 SS |
2242 | /** |
2243 | * consume_stock: Try to consume stocked charge on this cpu. | |
2244 | * @memcg: memcg to consume from. | |
2245 | * @nr_pages: how many pages to charge. | |
2246 | * | |
2247 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2248 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2249 | * service an allocation will refill the stock. | |
2250 | * | |
2251 | * returns true if successful, false otherwise. | |
cdec2e42 | 2252 | */ |
a0956d54 | 2253 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2254 | { |
2255 | struct memcg_stock_pcp *stock; | |
2256 | bool ret = true; | |
2257 | ||
a0956d54 SS |
2258 | if (nr_pages > CHARGE_BATCH) |
2259 | return false; | |
2260 | ||
cdec2e42 | 2261 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2262 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2263 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2264 | else /* need to call res_counter_charge */ |
2265 | ret = false; | |
2266 | put_cpu_var(memcg_stock); | |
2267 | return ret; | |
2268 | } | |
2269 | ||
2270 | /* | |
2271 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2272 | */ | |
2273 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2274 | { | |
2275 | struct mem_cgroup *old = stock->cached; | |
2276 | ||
11c9ea4e JW |
2277 | if (stock->nr_pages) { |
2278 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2279 | ||
2280 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2281 | if (do_swap_account) |
11c9ea4e JW |
2282 | res_counter_uncharge(&old->memsw, bytes); |
2283 | stock->nr_pages = 0; | |
cdec2e42 KH |
2284 | } |
2285 | stock->cached = NULL; | |
cdec2e42 KH |
2286 | } |
2287 | ||
2288 | /* | |
2289 | * This must be called under preempt disabled or must be called by | |
2290 | * a thread which is pinned to local cpu. | |
2291 | */ | |
2292 | static void drain_local_stock(struct work_struct *dummy) | |
2293 | { | |
2294 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2295 | drain_stock(stock); | |
26fe6168 | 2296 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2297 | } |
2298 | ||
2299 | /* | |
2300 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2301 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2302 | */ |
c0ff4b85 | 2303 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2304 | { |
2305 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2306 | ||
c0ff4b85 | 2307 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2308 | drain_stock(stock); |
c0ff4b85 | 2309 | stock->cached = memcg; |
cdec2e42 | 2310 | } |
11c9ea4e | 2311 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2312 | put_cpu_var(memcg_stock); |
2313 | } | |
2314 | ||
2315 | /* | |
c0ff4b85 | 2316 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2317 | * of the hierarchy under it. sync flag says whether we should block |
2318 | * until the work is done. | |
cdec2e42 | 2319 | */ |
c0ff4b85 | 2320 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2321 | { |
26fe6168 | 2322 | int cpu, curcpu; |
d38144b7 | 2323 | |
cdec2e42 | 2324 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2325 | get_online_cpus(); |
5af12d0e | 2326 | curcpu = get_cpu(); |
cdec2e42 KH |
2327 | for_each_online_cpu(cpu) { |
2328 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2329 | struct mem_cgroup *memcg; |
26fe6168 | 2330 | |
c0ff4b85 R |
2331 | memcg = stock->cached; |
2332 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2333 | continue; |
c0ff4b85 | 2334 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2335 | continue; |
d1a05b69 MH |
2336 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2337 | if (cpu == curcpu) | |
2338 | drain_local_stock(&stock->work); | |
2339 | else | |
2340 | schedule_work_on(cpu, &stock->work); | |
2341 | } | |
cdec2e42 | 2342 | } |
5af12d0e | 2343 | put_cpu(); |
d38144b7 MH |
2344 | |
2345 | if (!sync) | |
2346 | goto out; | |
2347 | ||
2348 | for_each_online_cpu(cpu) { | |
2349 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2350 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2351 | flush_work(&stock->work); |
2352 | } | |
2353 | out: | |
cdec2e42 | 2354 | put_online_cpus(); |
d38144b7 MH |
2355 | } |
2356 | ||
2357 | /* | |
2358 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2359 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2360 | * expects some charges will be back to res_counter later but cannot wait for | |
2361 | * it. | |
2362 | */ | |
c0ff4b85 | 2363 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2364 | { |
9f50fad6 MH |
2365 | /* |
2366 | * If someone calls draining, avoid adding more kworker runs. | |
2367 | */ | |
2368 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2369 | return; | |
c0ff4b85 | 2370 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2371 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2372 | } |
2373 | ||
2374 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2375 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2376 | { |
2377 | /* called when force_empty is called */ | |
9f50fad6 | 2378 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2379 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2380 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2381 | } |
2382 | ||
711d3d2c KH |
2383 | /* |
2384 | * This function drains percpu counter value from DEAD cpu and | |
2385 | * move it to local cpu. Note that this function can be preempted. | |
2386 | */ | |
c0ff4b85 | 2387 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2388 | { |
2389 | int i; | |
2390 | ||
c0ff4b85 | 2391 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2392 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2393 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2394 | |
c0ff4b85 R |
2395 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2396 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2397 | } |
e9f8974f | 2398 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2399 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2400 | |
c0ff4b85 R |
2401 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2402 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2403 | } |
c0ff4b85 | 2404 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2405 | } |
2406 | ||
2407 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2408 | unsigned long action, |
2409 | void *hcpu) | |
2410 | { | |
2411 | int cpu = (unsigned long)hcpu; | |
2412 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2413 | struct mem_cgroup *iter; |
cdec2e42 | 2414 | |
619d094b | 2415 | if (action == CPU_ONLINE) |
1489ebad | 2416 | return NOTIFY_OK; |
1489ebad | 2417 | |
d833049b | 2418 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2419 | return NOTIFY_OK; |
711d3d2c | 2420 | |
9f3a0d09 | 2421 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2422 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2423 | ||
cdec2e42 KH |
2424 | stock = &per_cpu(memcg_stock, cpu); |
2425 | drain_stock(stock); | |
2426 | return NOTIFY_OK; | |
2427 | } | |
2428 | ||
4b534334 KH |
2429 | |
2430 | /* See __mem_cgroup_try_charge() for details */ | |
2431 | enum { | |
2432 | CHARGE_OK, /* success */ | |
2433 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2434 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2435 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2436 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2437 | }; | |
2438 | ||
c0ff4b85 | 2439 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 SS |
2440 | unsigned int nr_pages, unsigned int min_pages, |
2441 | bool oom_check) | |
4b534334 | 2442 | { |
7ec99d62 | 2443 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2444 | struct mem_cgroup *mem_over_limit; |
2445 | struct res_counter *fail_res; | |
2446 | unsigned long flags = 0; | |
2447 | int ret; | |
2448 | ||
c0ff4b85 | 2449 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2450 | |
2451 | if (likely(!ret)) { | |
2452 | if (!do_swap_account) | |
2453 | return CHARGE_OK; | |
c0ff4b85 | 2454 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2455 | if (likely(!ret)) |
2456 | return CHARGE_OK; | |
2457 | ||
c0ff4b85 | 2458 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2459 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2460 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2461 | } else | |
2462 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2463 | /* |
9221edb7 JW |
2464 | * Never reclaim on behalf of optional batching, retry with a |
2465 | * single page instead. | |
2466 | */ | |
4c9c5359 | 2467 | if (nr_pages > min_pages) |
4b534334 KH |
2468 | return CHARGE_RETRY; |
2469 | ||
2470 | if (!(gfp_mask & __GFP_WAIT)) | |
2471 | return CHARGE_WOULDBLOCK; | |
2472 | ||
4c9c5359 SS |
2473 | if (gfp_mask & __GFP_NORETRY) |
2474 | return CHARGE_NOMEM; | |
2475 | ||
5660048c | 2476 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2477 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2478 | return CHARGE_RETRY; |
4b534334 | 2479 | /* |
19942822 JW |
2480 | * Even though the limit is exceeded at this point, reclaim |
2481 | * may have been able to free some pages. Retry the charge | |
2482 | * before killing the task. | |
2483 | * | |
2484 | * Only for regular pages, though: huge pages are rather | |
2485 | * unlikely to succeed so close to the limit, and we fall back | |
2486 | * to regular pages anyway in case of failure. | |
4b534334 | 2487 | */ |
4c9c5359 | 2488 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2489 | return CHARGE_RETRY; |
2490 | ||
2491 | /* | |
2492 | * At task move, charge accounts can be doubly counted. So, it's | |
2493 | * better to wait until the end of task_move if something is going on. | |
2494 | */ | |
2495 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2496 | return CHARGE_RETRY; | |
2497 | ||
2498 | /* If we don't need to call oom-killer at el, return immediately */ | |
2499 | if (!oom_check) | |
2500 | return CHARGE_NOMEM; | |
2501 | /* check OOM */ | |
e845e199 | 2502 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2503 | return CHARGE_OOM_DIE; |
2504 | ||
2505 | return CHARGE_RETRY; | |
2506 | } | |
2507 | ||
f817ed48 | 2508 | /* |
38c5d72f KH |
2509 | * __mem_cgroup_try_charge() does |
2510 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2511 | * 2. update res_counter | |
2512 | * 3. call memory reclaim if necessary. | |
2513 | * | |
2514 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2515 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2516 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2517 | * as possible without any hazards. 2: all pages should have a valid | |
2518 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2519 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2520 | * | |
2521 | * So __mem_cgroup_try_charge() will return | |
2522 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2523 | * -ENOMEM ... charge failure because of resource limits. | |
2524 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2525 | * | |
2526 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2527 | * the oom-killer can be invoked. | |
8a9f3ccd | 2528 | */ |
f817ed48 | 2529 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2530 | gfp_t gfp_mask, |
7ec99d62 | 2531 | unsigned int nr_pages, |
c0ff4b85 | 2532 | struct mem_cgroup **ptr, |
7ec99d62 | 2533 | bool oom) |
8a9f3ccd | 2534 | { |
7ec99d62 | 2535 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2536 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2537 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2538 | int ret; |
a636b327 | 2539 | |
867578cb KH |
2540 | /* |
2541 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2542 | * in system level. So, allow to go ahead dying process in addition to | |
2543 | * MEMDIE process. | |
2544 | */ | |
2545 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2546 | || fatal_signal_pending(current))) | |
2547 | goto bypass; | |
a636b327 | 2548 | |
8a9f3ccd | 2549 | /* |
3be91277 HD |
2550 | * We always charge the cgroup the mm_struct belongs to. |
2551 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2552 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2553 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2554 | */ |
c0ff4b85 | 2555 | if (!*ptr && !mm) |
38c5d72f | 2556 | *ptr = root_mem_cgroup; |
f75ca962 | 2557 | again: |
c0ff4b85 R |
2558 | if (*ptr) { /* css should be a valid one */ |
2559 | memcg = *ptr; | |
c0ff4b85 | 2560 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2561 | goto done; |
a0956d54 | 2562 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2563 | goto done; |
c0ff4b85 | 2564 | css_get(&memcg->css); |
4b534334 | 2565 | } else { |
f75ca962 | 2566 | struct task_struct *p; |
54595fe2 | 2567 | |
f75ca962 KH |
2568 | rcu_read_lock(); |
2569 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2570 | /* |
ebb76ce1 | 2571 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2572 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2573 | * race with swapoff. Then, we have small risk of mis-accouning. |
2574 | * But such kind of mis-account by race always happens because | |
2575 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2576 | * small race, here. | |
2577 | * (*) swapoff at el will charge against mm-struct not against | |
2578 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2579 | */ |
c0ff4b85 | 2580 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2581 | if (!memcg) |
2582 | memcg = root_mem_cgroup; | |
2583 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2584 | rcu_read_unlock(); |
2585 | goto done; | |
2586 | } | |
a0956d54 | 2587 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2588 | /* |
2589 | * It seems dagerous to access memcg without css_get(). | |
2590 | * But considering how consume_stok works, it's not | |
2591 | * necessary. If consume_stock success, some charges | |
2592 | * from this memcg are cached on this cpu. So, we | |
2593 | * don't need to call css_get()/css_tryget() before | |
2594 | * calling consume_stock(). | |
2595 | */ | |
2596 | rcu_read_unlock(); | |
2597 | goto done; | |
2598 | } | |
2599 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2600 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2601 | rcu_read_unlock(); |
2602 | goto again; | |
2603 | } | |
2604 | rcu_read_unlock(); | |
2605 | } | |
8a9f3ccd | 2606 | |
4b534334 KH |
2607 | do { |
2608 | bool oom_check; | |
7a81b88c | 2609 | |
4b534334 | 2610 | /* If killed, bypass charge */ |
f75ca962 | 2611 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2612 | css_put(&memcg->css); |
4b534334 | 2613 | goto bypass; |
f75ca962 | 2614 | } |
6d61ef40 | 2615 | |
4b534334 KH |
2616 | oom_check = false; |
2617 | if (oom && !nr_oom_retries) { | |
2618 | oom_check = true; | |
2619 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2620 | } |
66e1707b | 2621 | |
4c9c5359 SS |
2622 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages, |
2623 | oom_check); | |
4b534334 KH |
2624 | switch (ret) { |
2625 | case CHARGE_OK: | |
2626 | break; | |
2627 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2628 | batch = nr_pages; |
c0ff4b85 R |
2629 | css_put(&memcg->css); |
2630 | memcg = NULL; | |
f75ca962 | 2631 | goto again; |
4b534334 | 2632 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2633 | css_put(&memcg->css); |
4b534334 KH |
2634 | goto nomem; |
2635 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2636 | if (!oom) { |
c0ff4b85 | 2637 | css_put(&memcg->css); |
867578cb | 2638 | goto nomem; |
f75ca962 | 2639 | } |
4b534334 KH |
2640 | /* If oom, we never return -ENOMEM */ |
2641 | nr_oom_retries--; | |
2642 | break; | |
2643 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2644 | css_put(&memcg->css); |
867578cb | 2645 | goto bypass; |
66e1707b | 2646 | } |
4b534334 KH |
2647 | } while (ret != CHARGE_OK); |
2648 | ||
7ec99d62 | 2649 | if (batch > nr_pages) |
c0ff4b85 R |
2650 | refill_stock(memcg, batch - nr_pages); |
2651 | css_put(&memcg->css); | |
0c3e73e8 | 2652 | done: |
c0ff4b85 | 2653 | *ptr = memcg; |
7a81b88c KH |
2654 | return 0; |
2655 | nomem: | |
c0ff4b85 | 2656 | *ptr = NULL; |
7a81b88c | 2657 | return -ENOMEM; |
867578cb | 2658 | bypass: |
38c5d72f KH |
2659 | *ptr = root_mem_cgroup; |
2660 | return -EINTR; | |
7a81b88c | 2661 | } |
8a9f3ccd | 2662 | |
a3032a2c DN |
2663 | /* |
2664 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2665 | * This function is for that and do uncharge, put css's refcnt. | |
2666 | * gotten by try_charge(). | |
2667 | */ | |
c0ff4b85 | 2668 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2669 | unsigned int nr_pages) |
a3032a2c | 2670 | { |
c0ff4b85 | 2671 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2672 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2673 | ||
c0ff4b85 | 2674 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2675 | if (do_swap_account) |
c0ff4b85 | 2676 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2677 | } |
854ffa8d DN |
2678 | } |
2679 | ||
d01dd17f KH |
2680 | /* |
2681 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2682 | * This is useful when moving usage to parent cgroup. | |
2683 | */ | |
2684 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2685 | unsigned int nr_pages) | |
2686 | { | |
2687 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2688 | ||
2689 | if (mem_cgroup_is_root(memcg)) | |
2690 | return; | |
2691 | ||
2692 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2693 | if (do_swap_account) | |
2694 | res_counter_uncharge_until(&memcg->memsw, | |
2695 | memcg->memsw.parent, bytes); | |
2696 | } | |
2697 | ||
a3b2d692 KH |
2698 | /* |
2699 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2700 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2701 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2702 | * called against removed memcg.) | |
a3b2d692 KH |
2703 | */ |
2704 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2705 | { | |
2706 | struct cgroup_subsys_state *css; | |
2707 | ||
2708 | /* ID 0 is unused ID */ | |
2709 | if (!id) | |
2710 | return NULL; | |
2711 | css = css_lookup(&mem_cgroup_subsys, id); | |
2712 | if (!css) | |
2713 | return NULL; | |
b2145145 | 2714 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2715 | } |
2716 | ||
e42d9d5d | 2717 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2718 | { |
c0ff4b85 | 2719 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2720 | struct page_cgroup *pc; |
a3b2d692 | 2721 | unsigned short id; |
b5a84319 KH |
2722 | swp_entry_t ent; |
2723 | ||
3c776e64 DN |
2724 | VM_BUG_ON(!PageLocked(page)); |
2725 | ||
3c776e64 | 2726 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2727 | lock_page_cgroup(pc); |
a3b2d692 | 2728 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2729 | memcg = pc->mem_cgroup; |
2730 | if (memcg && !css_tryget(&memcg->css)) | |
2731 | memcg = NULL; | |
e42d9d5d | 2732 | } else if (PageSwapCache(page)) { |
3c776e64 | 2733 | ent.val = page_private(page); |
9fb4b7cc | 2734 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2735 | rcu_read_lock(); |
c0ff4b85 R |
2736 | memcg = mem_cgroup_lookup(id); |
2737 | if (memcg && !css_tryget(&memcg->css)) | |
2738 | memcg = NULL; | |
a3b2d692 | 2739 | rcu_read_unlock(); |
3c776e64 | 2740 | } |
c0bd3f63 | 2741 | unlock_page_cgroup(pc); |
c0ff4b85 | 2742 | return memcg; |
b5a84319 KH |
2743 | } |
2744 | ||
c0ff4b85 | 2745 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2746 | struct page *page, |
7ec99d62 | 2747 | unsigned int nr_pages, |
9ce70c02 HD |
2748 | enum charge_type ctype, |
2749 | bool lrucare) | |
7a81b88c | 2750 | { |
ce587e65 | 2751 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2752 | struct zone *uninitialized_var(zone); |
fa9add64 | 2753 | struct lruvec *lruvec; |
9ce70c02 | 2754 | bool was_on_lru = false; |
b2402857 | 2755 | bool anon; |
9ce70c02 | 2756 | |
ca3e0214 | 2757 | lock_page_cgroup(pc); |
90deb788 | 2758 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2759 | /* |
2760 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2761 | * accessed by any other context at this point. | |
2762 | */ | |
9ce70c02 HD |
2763 | |
2764 | /* | |
2765 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2766 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2767 | */ | |
2768 | if (lrucare) { | |
2769 | zone = page_zone(page); | |
2770 | spin_lock_irq(&zone->lru_lock); | |
2771 | if (PageLRU(page)) { | |
fa9add64 | 2772 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2773 | ClearPageLRU(page); |
fa9add64 | 2774 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2775 | was_on_lru = true; |
2776 | } | |
2777 | } | |
2778 | ||
c0ff4b85 | 2779 | pc->mem_cgroup = memcg; |
261fb61a KH |
2780 | /* |
2781 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2782 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2783 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2784 | * before USED bit, we need memory barrier here. | |
2785 | * See mem_cgroup_add_lru_list(), etc. | |
2786 | */ | |
08e552c6 | 2787 | smp_wmb(); |
b2402857 | 2788 | SetPageCgroupUsed(pc); |
3be91277 | 2789 | |
9ce70c02 HD |
2790 | if (lrucare) { |
2791 | if (was_on_lru) { | |
fa9add64 | 2792 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2793 | VM_BUG_ON(PageLRU(page)); |
2794 | SetPageLRU(page); | |
fa9add64 | 2795 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2796 | } |
2797 | spin_unlock_irq(&zone->lru_lock); | |
2798 | } | |
2799 | ||
41326c17 | 2800 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2801 | anon = true; |
2802 | else | |
2803 | anon = false; | |
2804 | ||
2805 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2806 | unlock_page_cgroup(pc); |
9ce70c02 | 2807 | |
430e4863 KH |
2808 | /* |
2809 | * "charge_statistics" updated event counter. Then, check it. | |
2810 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2811 | * if they exceeds softlimit. | |
2812 | */ | |
c0ff4b85 | 2813 | memcg_check_events(memcg, page); |
7a81b88c | 2814 | } |
66e1707b | 2815 | |
7cf27982 GC |
2816 | static DEFINE_MUTEX(set_limit_mutex); |
2817 | ||
7ae1e1d0 GC |
2818 | #ifdef CONFIG_MEMCG_KMEM |
2819 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) | |
2820 | { | |
2821 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
2822 | (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK); | |
2823 | } | |
2824 | ||
1f458cbf GC |
2825 | /* |
2826 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2827 | * in the memcg_cache_params struct. | |
2828 | */ | |
2829 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2830 | { | |
2831 | struct kmem_cache *cachep; | |
2832 | ||
2833 | VM_BUG_ON(p->is_root_cache); | |
2834 | cachep = p->root_cache; | |
2835 | return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)]; | |
2836 | } | |
2837 | ||
749c5415 GC |
2838 | #ifdef CONFIG_SLABINFO |
2839 | static int mem_cgroup_slabinfo_read(struct cgroup *cont, struct cftype *cft, | |
2840 | struct seq_file *m) | |
2841 | { | |
2842 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); | |
2843 | struct memcg_cache_params *params; | |
2844 | ||
2845 | if (!memcg_can_account_kmem(memcg)) | |
2846 | return -EIO; | |
2847 | ||
2848 | print_slabinfo_header(m); | |
2849 | ||
2850 | mutex_lock(&memcg->slab_caches_mutex); | |
2851 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
2852 | cache_show(memcg_params_to_cache(params), m); | |
2853 | mutex_unlock(&memcg->slab_caches_mutex); | |
2854 | ||
2855 | return 0; | |
2856 | } | |
2857 | #endif | |
2858 | ||
7ae1e1d0 GC |
2859 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
2860 | { | |
2861 | struct res_counter *fail_res; | |
2862 | struct mem_cgroup *_memcg; | |
2863 | int ret = 0; | |
2864 | bool may_oom; | |
2865 | ||
2866 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2867 | if (ret) | |
2868 | return ret; | |
2869 | ||
2870 | /* | |
2871 | * Conditions under which we can wait for the oom_killer. Those are | |
2872 | * the same conditions tested by the core page allocator | |
2873 | */ | |
2874 | may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY); | |
2875 | ||
2876 | _memcg = memcg; | |
2877 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
2878 | &_memcg, may_oom); | |
2879 | ||
2880 | if (ret == -EINTR) { | |
2881 | /* | |
2882 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
2883 | * OOM kill or fatal signal. Since our only options are to | |
2884 | * either fail the allocation or charge it to this cgroup, do | |
2885 | * it as a temporary condition. But we can't fail. From a | |
2886 | * kmem/slab perspective, the cache has already been selected, | |
2887 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
2888 | * our minds. | |
2889 | * | |
2890 | * This condition will only trigger if the task entered | |
2891 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
2892 | * __mem_cgroup_try_charge() above. Tasks that were already | |
2893 | * dying when the allocation triggers should have been already | |
2894 | * directed to the root cgroup in memcontrol.h | |
2895 | */ | |
2896 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
2897 | if (do_swap_account) | |
2898 | res_counter_charge_nofail(&memcg->memsw, size, | |
2899 | &fail_res); | |
2900 | ret = 0; | |
2901 | } else if (ret) | |
2902 | res_counter_uncharge(&memcg->kmem, size); | |
2903 | ||
2904 | return ret; | |
2905 | } | |
2906 | ||
2907 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
2908 | { | |
7ae1e1d0 GC |
2909 | res_counter_uncharge(&memcg->res, size); |
2910 | if (do_swap_account) | |
2911 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
2912 | |
2913 | /* Not down to 0 */ | |
2914 | if (res_counter_uncharge(&memcg->kmem, size)) | |
2915 | return; | |
2916 | ||
2917 | if (memcg_kmem_test_and_clear_dead(memcg)) | |
2918 | mem_cgroup_put(memcg); | |
7ae1e1d0 GC |
2919 | } |
2920 | ||
2633d7a0 GC |
2921 | void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep) |
2922 | { | |
2923 | if (!memcg) | |
2924 | return; | |
2925 | ||
2926 | mutex_lock(&memcg->slab_caches_mutex); | |
2927 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); | |
2928 | mutex_unlock(&memcg->slab_caches_mutex); | |
2929 | } | |
2930 | ||
2931 | /* | |
2932 | * helper for acessing a memcg's index. It will be used as an index in the | |
2933 | * child cache array in kmem_cache, and also to derive its name. This function | |
2934 | * will return -1 when this is not a kmem-limited memcg. | |
2935 | */ | |
2936 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2937 | { | |
2938 | return memcg ? memcg->kmemcg_id : -1; | |
2939 | } | |
2940 | ||
55007d84 GC |
2941 | /* |
2942 | * This ends up being protected by the set_limit mutex, during normal | |
2943 | * operation, because that is its main call site. | |
2944 | * | |
2945 | * But when we create a new cache, we can call this as well if its parent | |
2946 | * is kmem-limited. That will have to hold set_limit_mutex as well. | |
2947 | */ | |
2948 | int memcg_update_cache_sizes(struct mem_cgroup *memcg) | |
2949 | { | |
2950 | int num, ret; | |
2951 | ||
2952 | num = ida_simple_get(&kmem_limited_groups, | |
2953 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2954 | if (num < 0) | |
2955 | return num; | |
2956 | /* | |
2957 | * After this point, kmem_accounted (that we test atomically in | |
2958 | * the beginning of this conditional), is no longer 0. This | |
2959 | * guarantees only one process will set the following boolean | |
2960 | * to true. We don't need test_and_set because we're protected | |
2961 | * by the set_limit_mutex anyway. | |
2962 | */ | |
2963 | memcg_kmem_set_activated(memcg); | |
2964 | ||
2965 | ret = memcg_update_all_caches(num+1); | |
2966 | if (ret) { | |
2967 | ida_simple_remove(&kmem_limited_groups, num); | |
2968 | memcg_kmem_clear_activated(memcg); | |
2969 | return ret; | |
2970 | } | |
2971 | ||
2972 | memcg->kmemcg_id = num; | |
2973 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
2974 | mutex_init(&memcg->slab_caches_mutex); | |
2975 | return 0; | |
2976 | } | |
2977 | ||
2978 | static size_t memcg_caches_array_size(int num_groups) | |
2979 | { | |
2980 | ssize_t size; | |
2981 | if (num_groups <= 0) | |
2982 | return 0; | |
2983 | ||
2984 | size = 2 * num_groups; | |
2985 | if (size < MEMCG_CACHES_MIN_SIZE) | |
2986 | size = MEMCG_CACHES_MIN_SIZE; | |
2987 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2988 | size = MEMCG_CACHES_MAX_SIZE; | |
2989 | ||
2990 | return size; | |
2991 | } | |
2992 | ||
2993 | /* | |
2994 | * We should update the current array size iff all caches updates succeed. This | |
2995 | * can only be done from the slab side. The slab mutex needs to be held when | |
2996 | * calling this. | |
2997 | */ | |
2998 | void memcg_update_array_size(int num) | |
2999 | { | |
3000 | if (num > memcg_limited_groups_array_size) | |
3001 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
3002 | } | |
3003 | ||
3004 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) | |
3005 | { | |
3006 | struct memcg_cache_params *cur_params = s->memcg_params; | |
3007 | ||
3008 | VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache); | |
3009 | ||
3010 | if (num_groups > memcg_limited_groups_array_size) { | |
3011 | int i; | |
3012 | ssize_t size = memcg_caches_array_size(num_groups); | |
3013 | ||
3014 | size *= sizeof(void *); | |
3015 | size += sizeof(struct memcg_cache_params); | |
3016 | ||
3017 | s->memcg_params = kzalloc(size, GFP_KERNEL); | |
3018 | if (!s->memcg_params) { | |
3019 | s->memcg_params = cur_params; | |
3020 | return -ENOMEM; | |
3021 | } | |
3022 | ||
3023 | s->memcg_params->is_root_cache = true; | |
3024 | ||
3025 | /* | |
3026 | * There is the chance it will be bigger than | |
3027 | * memcg_limited_groups_array_size, if we failed an allocation | |
3028 | * in a cache, in which case all caches updated before it, will | |
3029 | * have a bigger array. | |
3030 | * | |
3031 | * But if that is the case, the data after | |
3032 | * memcg_limited_groups_array_size is certainly unused | |
3033 | */ | |
3034 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3035 | if (!cur_params->memcg_caches[i]) | |
3036 | continue; | |
3037 | s->memcg_params->memcg_caches[i] = | |
3038 | cur_params->memcg_caches[i]; | |
3039 | } | |
3040 | ||
3041 | /* | |
3042 | * Ideally, we would wait until all caches succeed, and only | |
3043 | * then free the old one. But this is not worth the extra | |
3044 | * pointer per-cache we'd have to have for this. | |
3045 | * | |
3046 | * It is not a big deal if some caches are left with a size | |
3047 | * bigger than the others. And all updates will reset this | |
3048 | * anyway. | |
3049 | */ | |
3050 | kfree(cur_params); | |
3051 | } | |
3052 | return 0; | |
3053 | } | |
3054 | ||
943a451a GC |
3055 | int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s, |
3056 | struct kmem_cache *root_cache) | |
2633d7a0 GC |
3057 | { |
3058 | size_t size = sizeof(struct memcg_cache_params); | |
3059 | ||
3060 | if (!memcg_kmem_enabled()) | |
3061 | return 0; | |
3062 | ||
55007d84 GC |
3063 | if (!memcg) |
3064 | size += memcg_limited_groups_array_size * sizeof(void *); | |
3065 | ||
2633d7a0 GC |
3066 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3067 | if (!s->memcg_params) | |
3068 | return -ENOMEM; | |
3069 | ||
943a451a | 3070 | if (memcg) { |
2633d7a0 | 3071 | s->memcg_params->memcg = memcg; |
943a451a | 3072 | s->memcg_params->root_cache = root_cache; |
4ba902b5 GC |
3073 | } else |
3074 | s->memcg_params->is_root_cache = true; | |
3075 | ||
2633d7a0 GC |
3076 | return 0; |
3077 | } | |
3078 | ||
3079 | void memcg_release_cache(struct kmem_cache *s) | |
3080 | { | |
d7f25f8a GC |
3081 | struct kmem_cache *root; |
3082 | struct mem_cgroup *memcg; | |
3083 | int id; | |
3084 | ||
3085 | /* | |
3086 | * This happens, for instance, when a root cache goes away before we | |
3087 | * add any memcg. | |
3088 | */ | |
3089 | if (!s->memcg_params) | |
3090 | return; | |
3091 | ||
3092 | if (s->memcg_params->is_root_cache) | |
3093 | goto out; | |
3094 | ||
3095 | memcg = s->memcg_params->memcg; | |
3096 | id = memcg_cache_id(memcg); | |
3097 | ||
3098 | root = s->memcg_params->root_cache; | |
3099 | root->memcg_params->memcg_caches[id] = NULL; | |
3100 | mem_cgroup_put(memcg); | |
3101 | ||
3102 | mutex_lock(&memcg->slab_caches_mutex); | |
3103 | list_del(&s->memcg_params->list); | |
3104 | mutex_unlock(&memcg->slab_caches_mutex); | |
3105 | ||
3106 | out: | |
2633d7a0 GC |
3107 | kfree(s->memcg_params); |
3108 | } | |
3109 | ||
0e9d92f2 GC |
3110 | /* |
3111 | * During the creation a new cache, we need to disable our accounting mechanism | |
3112 | * altogether. This is true even if we are not creating, but rather just | |
3113 | * enqueing new caches to be created. | |
3114 | * | |
3115 | * This is because that process will trigger allocations; some visible, like | |
3116 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3117 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3118 | * objects during debug. | |
3119 | * | |
3120 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3121 | * to it. This may not be a bounded recursion: since the first cache creation | |
3122 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3123 | * cache again, failing at the same point. | |
3124 | * | |
3125 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3126 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3127 | * inside the following two functions. | |
3128 | */ | |
3129 | static inline void memcg_stop_kmem_account(void) | |
3130 | { | |
3131 | VM_BUG_ON(!current->mm); | |
3132 | current->memcg_kmem_skip_account++; | |
3133 | } | |
3134 | ||
3135 | static inline void memcg_resume_kmem_account(void) | |
3136 | { | |
3137 | VM_BUG_ON(!current->mm); | |
3138 | current->memcg_kmem_skip_account--; | |
3139 | } | |
3140 | ||
1f458cbf GC |
3141 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3142 | { | |
3143 | struct kmem_cache *cachep; | |
3144 | struct memcg_cache_params *p; | |
3145 | ||
3146 | p = container_of(w, struct memcg_cache_params, destroy); | |
3147 | ||
3148 | cachep = memcg_params_to_cache(p); | |
3149 | ||
22933152 GC |
3150 | /* |
3151 | * If we get down to 0 after shrink, we could delete right away. | |
3152 | * However, memcg_release_pages() already puts us back in the workqueue | |
3153 | * in that case. If we proceed deleting, we'll get a dangling | |
3154 | * reference, and removing the object from the workqueue in that case | |
3155 | * is unnecessary complication. We are not a fast path. | |
3156 | * | |
3157 | * Note that this case is fundamentally different from racing with | |
3158 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3159 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3160 | * into the queue, but doing so from inside the worker racing to | |
3161 | * destroy it. | |
3162 | * | |
3163 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3164 | * again | |
3165 | */ | |
3166 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) { | |
3167 | kmem_cache_shrink(cachep); | |
3168 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
3169 | return; | |
3170 | } else | |
1f458cbf GC |
3171 | kmem_cache_destroy(cachep); |
3172 | } | |
3173 | ||
3174 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3175 | { | |
3176 | if (!cachep->memcg_params->dead) | |
3177 | return; | |
3178 | ||
22933152 GC |
3179 | /* |
3180 | * There are many ways in which we can get here. | |
3181 | * | |
3182 | * We can get to a memory-pressure situation while the delayed work is | |
3183 | * still pending to run. The vmscan shrinkers can then release all | |
3184 | * cache memory and get us to destruction. If this is the case, we'll | |
3185 | * be executed twice, which is a bug (the second time will execute over | |
3186 | * bogus data). In this case, cancelling the work should be fine. | |
3187 | * | |
3188 | * But we can also get here from the worker itself, if | |
3189 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3190 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3191 | * cancel the work (the worker runs with an internal lock held, which | |
3192 | * is the same lock we would hold for cancel_work_sync().) | |
3193 | * | |
3194 | * Since we can't possibly know who got us here, just refrain from | |
3195 | * running if there is already work pending | |
3196 | */ | |
3197 | if (work_pending(&cachep->memcg_params->destroy)) | |
3198 | return; | |
1f458cbf GC |
3199 | /* |
3200 | * We have to defer the actual destroying to a workqueue, because | |
3201 | * we might currently be in a context that cannot sleep. | |
3202 | */ | |
3203 | schedule_work(&cachep->memcg_params->destroy); | |
3204 | } | |
3205 | ||
d7f25f8a GC |
3206 | static char *memcg_cache_name(struct mem_cgroup *memcg, struct kmem_cache *s) |
3207 | { | |
3208 | char *name; | |
3209 | struct dentry *dentry; | |
3210 | ||
3211 | rcu_read_lock(); | |
3212 | dentry = rcu_dereference(memcg->css.cgroup->dentry); | |
3213 | rcu_read_unlock(); | |
3214 | ||
3215 | BUG_ON(dentry == NULL); | |
3216 | ||
3217 | name = kasprintf(GFP_KERNEL, "%s(%d:%s)", s->name, | |
3218 | memcg_cache_id(memcg), dentry->d_name.name); | |
3219 | ||
3220 | return name; | |
3221 | } | |
3222 | ||
3223 | static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg, | |
3224 | struct kmem_cache *s) | |
3225 | { | |
3226 | char *name; | |
3227 | struct kmem_cache *new; | |
3228 | ||
3229 | name = memcg_cache_name(memcg, s); | |
3230 | if (!name) | |
3231 | return NULL; | |
3232 | ||
3233 | new = kmem_cache_create_memcg(memcg, name, s->object_size, s->align, | |
943a451a | 3234 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d7f25f8a | 3235 | |
d79923fa GC |
3236 | if (new) |
3237 | new->allocflags |= __GFP_KMEMCG; | |
3238 | ||
d7f25f8a GC |
3239 | kfree(name); |
3240 | return new; | |
3241 | } | |
3242 | ||
3243 | /* | |
3244 | * This lock protects updaters, not readers. We want readers to be as fast as | |
3245 | * they can, and they will either see NULL or a valid cache value. Our model | |
3246 | * allow them to see NULL, in which case the root memcg will be selected. | |
3247 | * | |
3248 | * We need this lock because multiple allocations to the same cache from a non | |
3249 | * will span more than one worker. Only one of them can create the cache. | |
3250 | */ | |
3251 | static DEFINE_MUTEX(memcg_cache_mutex); | |
3252 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, | |
3253 | struct kmem_cache *cachep) | |
3254 | { | |
3255 | struct kmem_cache *new_cachep; | |
3256 | int idx; | |
3257 | ||
3258 | BUG_ON(!memcg_can_account_kmem(memcg)); | |
3259 | ||
3260 | idx = memcg_cache_id(memcg); | |
3261 | ||
3262 | mutex_lock(&memcg_cache_mutex); | |
3263 | new_cachep = cachep->memcg_params->memcg_caches[idx]; | |
3264 | if (new_cachep) | |
3265 | goto out; | |
3266 | ||
3267 | new_cachep = kmem_cache_dup(memcg, cachep); | |
d7f25f8a GC |
3268 | if (new_cachep == NULL) { |
3269 | new_cachep = cachep; | |
3270 | goto out; | |
3271 | } | |
3272 | ||
3273 | mem_cgroup_get(memcg); | |
1f458cbf | 3274 | atomic_set(&new_cachep->memcg_params->nr_pages , 0); |
d7f25f8a GC |
3275 | |
3276 | cachep->memcg_params->memcg_caches[idx] = new_cachep; | |
3277 | /* | |
3278 | * the readers won't lock, make sure everybody sees the updated value, | |
3279 | * so they won't put stuff in the queue again for no reason | |
3280 | */ | |
3281 | wmb(); | |
3282 | out: | |
3283 | mutex_unlock(&memcg_cache_mutex); | |
3284 | return new_cachep; | |
3285 | } | |
3286 | ||
7cf27982 GC |
3287 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3288 | { | |
3289 | struct kmem_cache *c; | |
3290 | int i; | |
3291 | ||
3292 | if (!s->memcg_params) | |
3293 | return; | |
3294 | if (!s->memcg_params->is_root_cache) | |
3295 | return; | |
3296 | ||
3297 | /* | |
3298 | * If the cache is being destroyed, we trust that there is no one else | |
3299 | * requesting objects from it. Even if there are, the sanity checks in | |
3300 | * kmem_cache_destroy should caught this ill-case. | |
3301 | * | |
3302 | * Still, we don't want anyone else freeing memcg_caches under our | |
3303 | * noses, which can happen if a new memcg comes to life. As usual, | |
3304 | * we'll take the set_limit_mutex to protect ourselves against this. | |
3305 | */ | |
3306 | mutex_lock(&set_limit_mutex); | |
3307 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3308 | c = s->memcg_params->memcg_caches[i]; | |
3309 | if (!c) | |
3310 | continue; | |
3311 | ||
3312 | /* | |
3313 | * We will now manually delete the caches, so to avoid races | |
3314 | * we need to cancel all pending destruction workers and | |
3315 | * proceed with destruction ourselves. | |
3316 | * | |
3317 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3318 | * and that could spawn the workers again: it is likely that | |
3319 | * the cache still have active pages until this very moment. | |
3320 | * This would lead us back to mem_cgroup_destroy_cache. | |
3321 | * | |
3322 | * But that will not execute at all if the "dead" flag is not | |
3323 | * set, so flip it down to guarantee we are in control. | |
3324 | */ | |
3325 | c->memcg_params->dead = false; | |
22933152 | 3326 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3327 | kmem_cache_destroy(c); |
3328 | } | |
3329 | mutex_unlock(&set_limit_mutex); | |
3330 | } | |
3331 | ||
d7f25f8a GC |
3332 | struct create_work { |
3333 | struct mem_cgroup *memcg; | |
3334 | struct kmem_cache *cachep; | |
3335 | struct work_struct work; | |
3336 | }; | |
3337 | ||
1f458cbf GC |
3338 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3339 | { | |
3340 | struct kmem_cache *cachep; | |
3341 | struct memcg_cache_params *params; | |
3342 | ||
3343 | if (!memcg_kmem_is_active(memcg)) | |
3344 | return; | |
3345 | ||
3346 | mutex_lock(&memcg->slab_caches_mutex); | |
3347 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3348 | cachep = memcg_params_to_cache(params); | |
3349 | cachep->memcg_params->dead = true; | |
3350 | INIT_WORK(&cachep->memcg_params->destroy, | |
22933152 | 3351 | kmem_cache_destroy_work_func); |
1f458cbf GC |
3352 | schedule_work(&cachep->memcg_params->destroy); |
3353 | } | |
3354 | mutex_unlock(&memcg->slab_caches_mutex); | |
3355 | } | |
3356 | ||
d7f25f8a GC |
3357 | static void memcg_create_cache_work_func(struct work_struct *w) |
3358 | { | |
3359 | struct create_work *cw; | |
3360 | ||
3361 | cw = container_of(w, struct create_work, work); | |
3362 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
3363 | /* Drop the reference gotten when we enqueued. */ | |
3364 | css_put(&cw->memcg->css); | |
3365 | kfree(cw); | |
3366 | } | |
3367 | ||
3368 | /* | |
3369 | * Enqueue the creation of a per-memcg kmem_cache. | |
3370 | * Called with rcu_read_lock. | |
3371 | */ | |
0e9d92f2 GC |
3372 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3373 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3374 | { |
3375 | struct create_work *cw; | |
3376 | ||
3377 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
3378 | if (cw == NULL) | |
3379 | return; | |
3380 | ||
3381 | /* The corresponding put will be done in the workqueue. */ | |
3382 | if (!css_tryget(&memcg->css)) { | |
3383 | kfree(cw); | |
3384 | return; | |
3385 | } | |
3386 | ||
3387 | cw->memcg = memcg; | |
3388 | cw->cachep = cachep; | |
3389 | ||
3390 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3391 | schedule_work(&cw->work); | |
3392 | } | |
3393 | ||
0e9d92f2 GC |
3394 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3395 | struct kmem_cache *cachep) | |
3396 | { | |
3397 | /* | |
3398 | * We need to stop accounting when we kmalloc, because if the | |
3399 | * corresponding kmalloc cache is not yet created, the first allocation | |
3400 | * in __memcg_create_cache_enqueue will recurse. | |
3401 | * | |
3402 | * However, it is better to enclose the whole function. Depending on | |
3403 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3404 | * trigger an allocation. This too, will make us recurse. Because at | |
3405 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3406 | * the safest choice is to do it like this, wrapping the whole function. | |
3407 | */ | |
3408 | memcg_stop_kmem_account(); | |
3409 | __memcg_create_cache_enqueue(memcg, cachep); | |
3410 | memcg_resume_kmem_account(); | |
3411 | } | |
d7f25f8a GC |
3412 | /* |
3413 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3414 | * We try to use the current memcg's version of the cache. | |
3415 | * | |
3416 | * If the cache does not exist yet, if we are the first user of it, | |
3417 | * we either create it immediately, if possible, or create it asynchronously | |
3418 | * in a workqueue. | |
3419 | * In the latter case, we will let the current allocation go through with | |
3420 | * the original cache. | |
3421 | * | |
3422 | * Can't be called in interrupt context or from kernel threads. | |
3423 | * This function needs to be called with rcu_read_lock() held. | |
3424 | */ | |
3425 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3426 | gfp_t gfp) | |
3427 | { | |
3428 | struct mem_cgroup *memcg; | |
3429 | int idx; | |
3430 | ||
3431 | VM_BUG_ON(!cachep->memcg_params); | |
3432 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3433 | ||
0e9d92f2 GC |
3434 | if (!current->mm || current->memcg_kmem_skip_account) |
3435 | return cachep; | |
3436 | ||
d7f25f8a GC |
3437 | rcu_read_lock(); |
3438 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
3439 | rcu_read_unlock(); | |
3440 | ||
3441 | if (!memcg_can_account_kmem(memcg)) | |
3442 | return cachep; | |
3443 | ||
3444 | idx = memcg_cache_id(memcg); | |
3445 | ||
3446 | /* | |
3447 | * barrier to mare sure we're always seeing the up to date value. The | |
3448 | * code updating memcg_caches will issue a write barrier to match this. | |
3449 | */ | |
3450 | read_barrier_depends(); | |
3451 | if (unlikely(cachep->memcg_params->memcg_caches[idx] == NULL)) { | |
3452 | /* | |
3453 | * If we are in a safe context (can wait, and not in interrupt | |
3454 | * context), we could be be predictable and return right away. | |
3455 | * This would guarantee that the allocation being performed | |
3456 | * already belongs in the new cache. | |
3457 | * | |
3458 | * However, there are some clashes that can arrive from locking. | |
3459 | * For instance, because we acquire the slab_mutex while doing | |
3460 | * kmem_cache_dup, this means no further allocation could happen | |
3461 | * with the slab_mutex held. | |
3462 | * | |
3463 | * Also, because cache creation issue get_online_cpus(), this | |
3464 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3465 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3466 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3467 | * better to defer everything. | |
3468 | */ | |
3469 | memcg_create_cache_enqueue(memcg, cachep); | |
3470 | return cachep; | |
3471 | } | |
3472 | ||
3473 | return cachep->memcg_params->memcg_caches[idx]; | |
3474 | } | |
3475 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3476 | ||
7ae1e1d0 GC |
3477 | /* |
3478 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3479 | * possible for the given order. But the page is not allocated yet, so we'll | |
3480 | * need a further commit step to do the final arrangements. | |
3481 | * | |
3482 | * It is possible for the task to switch cgroups in this mean time, so at | |
3483 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3484 | * the handle argument to return to the caller which cgroup we should commit | |
3485 | * against. We could also return the memcg directly and avoid the pointer | |
3486 | * passing, but a boolean return value gives better semantics considering | |
3487 | * the compiled-out case as well. | |
3488 | * | |
3489 | * Returning true means the allocation is possible. | |
3490 | */ | |
3491 | bool | |
3492 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3493 | { | |
3494 | struct mem_cgroup *memcg; | |
3495 | int ret; | |
3496 | ||
3497 | *_memcg = NULL; | |
3498 | memcg = try_get_mem_cgroup_from_mm(current->mm); | |
3499 | ||
3500 | /* | |
3501 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3502 | * isn't much we can do without complicating this too much, and it would | |
3503 | * be gfp-dependent anyway. Just let it go | |
3504 | */ | |
3505 | if (unlikely(!memcg)) | |
3506 | return true; | |
3507 | ||
3508 | if (!memcg_can_account_kmem(memcg)) { | |
3509 | css_put(&memcg->css); | |
3510 | return true; | |
3511 | } | |
3512 | ||
7ae1e1d0 GC |
3513 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3514 | if (!ret) | |
3515 | *_memcg = memcg; | |
7ae1e1d0 GC |
3516 | |
3517 | css_put(&memcg->css); | |
3518 | return (ret == 0); | |
3519 | } | |
3520 | ||
3521 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3522 | int order) | |
3523 | { | |
3524 | struct page_cgroup *pc; | |
3525 | ||
3526 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3527 | ||
3528 | /* The page allocation failed. Revert */ | |
3529 | if (!page) { | |
3530 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3531 | return; |
3532 | } | |
3533 | ||
3534 | pc = lookup_page_cgroup(page); | |
3535 | lock_page_cgroup(pc); | |
3536 | pc->mem_cgroup = memcg; | |
3537 | SetPageCgroupUsed(pc); | |
3538 | unlock_page_cgroup(pc); | |
3539 | } | |
3540 | ||
3541 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3542 | { | |
3543 | struct mem_cgroup *memcg = NULL; | |
3544 | struct page_cgroup *pc; | |
3545 | ||
3546 | ||
3547 | pc = lookup_page_cgroup(page); | |
3548 | /* | |
3549 | * Fast unlocked return. Theoretically might have changed, have to | |
3550 | * check again after locking. | |
3551 | */ | |
3552 | if (!PageCgroupUsed(pc)) | |
3553 | return; | |
3554 | ||
3555 | lock_page_cgroup(pc); | |
3556 | if (PageCgroupUsed(pc)) { | |
3557 | memcg = pc->mem_cgroup; | |
3558 | ClearPageCgroupUsed(pc); | |
3559 | } | |
3560 | unlock_page_cgroup(pc); | |
3561 | ||
3562 | /* | |
3563 | * We trust that only if there is a memcg associated with the page, it | |
3564 | * is a valid allocation | |
3565 | */ | |
3566 | if (!memcg) | |
3567 | return; | |
3568 | ||
3569 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3570 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 | 3571 | } |
1f458cbf GC |
3572 | #else |
3573 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3574 | { | |
3575 | } | |
7ae1e1d0 GC |
3576 | #endif /* CONFIG_MEMCG_KMEM */ |
3577 | ||
ca3e0214 KH |
3578 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3579 | ||
a0db00fc | 3580 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3581 | /* |
3582 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3583 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3584 | * charge/uncharge will be never happen and move_account() is done under | |
3585 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3586 | */ |
e94c8a9c | 3587 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3588 | { |
3589 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
3590 | struct page_cgroup *pc; |
3591 | int i; | |
ca3e0214 | 3592 | |
3d37c4a9 KH |
3593 | if (mem_cgroup_disabled()) |
3594 | return; | |
e94c8a9c KH |
3595 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3596 | pc = head_pc + i; | |
3597 | pc->mem_cgroup = head_pc->mem_cgroup; | |
3598 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
3599 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3600 | } | |
ca3e0214 | 3601 | } |
12d27107 | 3602 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3603 | |
f817ed48 | 3604 | /** |
de3638d9 | 3605 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3606 | * @page: the page |
7ec99d62 | 3607 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3608 | * @pc: page_cgroup of the page. |
3609 | * @from: mem_cgroup which the page is moved from. | |
3610 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3611 | * | |
3612 | * The caller must confirm following. | |
08e552c6 | 3613 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3614 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3615 | * |
2f3479b1 KH |
3616 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3617 | * from old cgroup. | |
f817ed48 | 3618 | */ |
7ec99d62 JW |
3619 | static int mem_cgroup_move_account(struct page *page, |
3620 | unsigned int nr_pages, | |
3621 | struct page_cgroup *pc, | |
3622 | struct mem_cgroup *from, | |
2f3479b1 | 3623 | struct mem_cgroup *to) |
f817ed48 | 3624 | { |
de3638d9 JW |
3625 | unsigned long flags; |
3626 | int ret; | |
b2402857 | 3627 | bool anon = PageAnon(page); |
987eba66 | 3628 | |
f817ed48 | 3629 | VM_BUG_ON(from == to); |
5564e88b | 3630 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
3631 | /* |
3632 | * The page is isolated from LRU. So, collapse function | |
3633 | * will not handle this page. But page splitting can happen. | |
3634 | * Do this check under compound_page_lock(). The caller should | |
3635 | * hold it. | |
3636 | */ | |
3637 | ret = -EBUSY; | |
7ec99d62 | 3638 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3639 | goto out; |
3640 | ||
3641 | lock_page_cgroup(pc); | |
3642 | ||
3643 | ret = -EINVAL; | |
3644 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3645 | goto unlock; | |
3646 | ||
312734c0 | 3647 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3648 | |
2ff76f11 | 3649 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
3650 | /* Update mapped_file data for mem_cgroup */ |
3651 | preempt_disable(); | |
3652 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3653 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3654 | preempt_enable(); | |
d69b042f | 3655 | } |
b2402857 | 3656 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
d69b042f | 3657 | |
854ffa8d | 3658 | /* caller should have done css_get */ |
08e552c6 | 3659 | pc->mem_cgroup = to; |
b2402857 | 3660 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
312734c0 | 3661 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3662 | ret = 0; |
3663 | unlock: | |
57f9fd7d | 3664 | unlock_page_cgroup(pc); |
d2265e6f KH |
3665 | /* |
3666 | * check events | |
3667 | */ | |
5564e88b JW |
3668 | memcg_check_events(to, page); |
3669 | memcg_check_events(from, page); | |
de3638d9 | 3670 | out: |
f817ed48 KH |
3671 | return ret; |
3672 | } | |
3673 | ||
2ef37d3f MH |
3674 | /** |
3675 | * mem_cgroup_move_parent - moves page to the parent group | |
3676 | * @page: the page to move | |
3677 | * @pc: page_cgroup of the page | |
3678 | * @child: page's cgroup | |
3679 | * | |
3680 | * move charges to its parent or the root cgroup if the group has no | |
3681 | * parent (aka use_hierarchy==0). | |
3682 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3683 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3684 | * it signals a race with a page removal/uncharge or migration. In the | |
3685 | * first case the page is on the way out and it will vanish from the LRU | |
3686 | * on the next attempt and the call should be retried later. | |
3687 | * Isolation from the LRU fails only if page has been isolated from | |
3688 | * the LRU since we looked at it and that usually means either global | |
3689 | * reclaim or migration going on. The page will either get back to the | |
3690 | * LRU or vanish. | |
3691 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3692 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3693 | * disappear in the next attempt. | |
f817ed48 | 3694 | */ |
5564e88b JW |
3695 | static int mem_cgroup_move_parent(struct page *page, |
3696 | struct page_cgroup *pc, | |
6068bf01 | 3697 | struct mem_cgroup *child) |
f817ed48 | 3698 | { |
f817ed48 | 3699 | struct mem_cgroup *parent; |
7ec99d62 | 3700 | unsigned int nr_pages; |
4be4489f | 3701 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3702 | int ret; |
3703 | ||
d8423011 | 3704 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3705 | |
57f9fd7d DN |
3706 | ret = -EBUSY; |
3707 | if (!get_page_unless_zero(page)) | |
3708 | goto out; | |
3709 | if (isolate_lru_page(page)) | |
3710 | goto put; | |
52dbb905 | 3711 | |
7ec99d62 | 3712 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3713 | |
cc926f78 KH |
3714 | parent = parent_mem_cgroup(child); |
3715 | /* | |
3716 | * If no parent, move charges to root cgroup. | |
3717 | */ | |
3718 | if (!parent) | |
3719 | parent = root_mem_cgroup; | |
f817ed48 | 3720 | |
2ef37d3f MH |
3721 | if (nr_pages > 1) { |
3722 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 3723 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3724 | } |
987eba66 | 3725 | |
cc926f78 | 3726 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3727 | pc, child, parent); |
cc926f78 KH |
3728 | if (!ret) |
3729 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3730 | |
7ec99d62 | 3731 | if (nr_pages > 1) |
987eba66 | 3732 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3733 | putback_lru_page(page); |
57f9fd7d | 3734 | put: |
40d58138 | 3735 | put_page(page); |
57f9fd7d | 3736 | out: |
f817ed48 KH |
3737 | return ret; |
3738 | } | |
3739 | ||
7a81b88c KH |
3740 | /* |
3741 | * Charge the memory controller for page usage. | |
3742 | * Return | |
3743 | * 0 if the charge was successful | |
3744 | * < 0 if the cgroup is over its limit | |
3745 | */ | |
3746 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3747 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3748 | { |
c0ff4b85 | 3749 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3750 | unsigned int nr_pages = 1; |
8493ae43 | 3751 | bool oom = true; |
7a81b88c | 3752 | int ret; |
ec168510 | 3753 | |
37c2ac78 | 3754 | if (PageTransHuge(page)) { |
7ec99d62 | 3755 | nr_pages <<= compound_order(page); |
37c2ac78 | 3756 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
3757 | /* |
3758 | * Never OOM-kill a process for a huge page. The | |
3759 | * fault handler will fall back to regular pages. | |
3760 | */ | |
3761 | oom = false; | |
37c2ac78 | 3762 | } |
7a81b88c | 3763 | |
c0ff4b85 | 3764 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3765 | if (ret == -ENOMEM) |
7a81b88c | 3766 | return ret; |
ce587e65 | 3767 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3768 | return 0; |
8a9f3ccd BS |
3769 | } |
3770 | ||
7a81b88c KH |
3771 | int mem_cgroup_newpage_charge(struct page *page, |
3772 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3773 | { |
f8d66542 | 3774 | if (mem_cgroup_disabled()) |
cede86ac | 3775 | return 0; |
7a0524cf JW |
3776 | VM_BUG_ON(page_mapped(page)); |
3777 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
3778 | VM_BUG_ON(!mm); | |
217bc319 | 3779 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3780 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
3781 | } |
3782 | ||
54595fe2 KH |
3783 | /* |
3784 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3785 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3786 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3787 | * "commit()" or removed by "cancel()" |
3788 | */ | |
0435a2fd JW |
3789 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3790 | struct page *page, | |
3791 | gfp_t mask, | |
3792 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3793 | { |
c0ff4b85 | 3794 | struct mem_cgroup *memcg; |
90deb788 | 3795 | struct page_cgroup *pc; |
54595fe2 | 3796 | int ret; |
8c7c6e34 | 3797 | |
90deb788 JW |
3798 | pc = lookup_page_cgroup(page); |
3799 | /* | |
3800 | * Every swap fault against a single page tries to charge the | |
3801 | * page, bail as early as possible. shmem_unuse() encounters | |
3802 | * already charged pages, too. The USED bit is protected by | |
3803 | * the page lock, which serializes swap cache removal, which | |
3804 | * in turn serializes uncharging. | |
3805 | */ | |
3806 | if (PageCgroupUsed(pc)) | |
3807 | return 0; | |
8c7c6e34 KH |
3808 | if (!do_swap_account) |
3809 | goto charge_cur_mm; | |
c0ff4b85 R |
3810 | memcg = try_get_mem_cgroup_from_page(page); |
3811 | if (!memcg) | |
54595fe2 | 3812 | goto charge_cur_mm; |
72835c86 JW |
3813 | *memcgp = memcg; |
3814 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 3815 | css_put(&memcg->css); |
38c5d72f KH |
3816 | if (ret == -EINTR) |
3817 | ret = 0; | |
54595fe2 | 3818 | return ret; |
8c7c6e34 | 3819 | charge_cur_mm: |
38c5d72f KH |
3820 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
3821 | if (ret == -EINTR) | |
3822 | ret = 0; | |
3823 | return ret; | |
8c7c6e34 KH |
3824 | } |
3825 | ||
0435a2fd JW |
3826 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
3827 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
3828 | { | |
3829 | *memcgp = NULL; | |
3830 | if (mem_cgroup_disabled()) | |
3831 | return 0; | |
bdf4f4d2 JW |
3832 | /* |
3833 | * A racing thread's fault, or swapoff, may have already | |
3834 | * updated the pte, and even removed page from swap cache: in | |
3835 | * those cases unuse_pte()'s pte_same() test will fail; but | |
3836 | * there's also a KSM case which does need to charge the page. | |
3837 | */ | |
3838 | if (!PageSwapCache(page)) { | |
3839 | int ret; | |
3840 | ||
3841 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
3842 | if (ret == -EINTR) | |
3843 | ret = 0; | |
3844 | return ret; | |
3845 | } | |
0435a2fd JW |
3846 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
3847 | } | |
3848 | ||
827a03d2 JW |
3849 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
3850 | { | |
3851 | if (mem_cgroup_disabled()) | |
3852 | return; | |
3853 | if (!memcg) | |
3854 | return; | |
3855 | __mem_cgroup_cancel_charge(memcg, 1); | |
3856 | } | |
3857 | ||
83aae4c7 | 3858 | static void |
72835c86 | 3859 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 3860 | enum charge_type ctype) |
7a81b88c | 3861 | { |
f8d66542 | 3862 | if (mem_cgroup_disabled()) |
7a81b88c | 3863 | return; |
72835c86 | 3864 | if (!memcg) |
7a81b88c | 3865 | return; |
5a6475a4 | 3866 | |
ce587e65 | 3867 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
3868 | /* |
3869 | * Now swap is on-memory. This means this page may be | |
3870 | * counted both as mem and swap....double count. | |
03f3c433 KH |
3871 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
3872 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
3873 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 3874 | */ |
03f3c433 | 3875 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 3876 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 3877 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 3878 | } |
7a81b88c KH |
3879 | } |
3880 | ||
72835c86 JW |
3881 | void mem_cgroup_commit_charge_swapin(struct page *page, |
3882 | struct mem_cgroup *memcg) | |
83aae4c7 | 3883 | { |
72835c86 | 3884 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 3885 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
3886 | } |
3887 | ||
827a03d2 JW |
3888 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
3889 | gfp_t gfp_mask) | |
7a81b88c | 3890 | { |
827a03d2 JW |
3891 | struct mem_cgroup *memcg = NULL; |
3892 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3893 | int ret; | |
3894 | ||
f8d66542 | 3895 | if (mem_cgroup_disabled()) |
827a03d2 JW |
3896 | return 0; |
3897 | if (PageCompound(page)) | |
3898 | return 0; | |
3899 | ||
827a03d2 JW |
3900 | if (!PageSwapCache(page)) |
3901 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
3902 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
3903 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
3904 | gfp_mask, &memcg); | |
827a03d2 JW |
3905 | if (!ret) |
3906 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
3907 | } | |
3908 | return ret; | |
7a81b88c KH |
3909 | } |
3910 | ||
c0ff4b85 | 3911 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
3912 | unsigned int nr_pages, |
3913 | const enum charge_type ctype) | |
569b846d KH |
3914 | { |
3915 | struct memcg_batch_info *batch = NULL; | |
3916 | bool uncharge_memsw = true; | |
7ec99d62 | 3917 | |
569b846d KH |
3918 | /* If swapout, usage of swap doesn't decrease */ |
3919 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
3920 | uncharge_memsw = false; | |
569b846d KH |
3921 | |
3922 | batch = ¤t->memcg_batch; | |
3923 | /* | |
3924 | * In usual, we do css_get() when we remember memcg pointer. | |
3925 | * But in this case, we keep res->usage until end of a series of | |
3926 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
3927 | */ | |
3928 | if (!batch->memcg) | |
c0ff4b85 | 3929 | batch->memcg = memcg; |
3c11ecf4 KH |
3930 | /* |
3931 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 3932 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
3933 | * the same cgroup and we have chance to coalesce uncharges. |
3934 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
3935 | * because we want to do uncharge as soon as possible. | |
3936 | */ | |
3937 | ||
3938 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
3939 | goto direct_uncharge; | |
3940 | ||
7ec99d62 | 3941 | if (nr_pages > 1) |
ec168510 AA |
3942 | goto direct_uncharge; |
3943 | ||
569b846d KH |
3944 | /* |
3945 | * In typical case, batch->memcg == mem. This means we can | |
3946 | * merge a series of uncharges to an uncharge of res_counter. | |
3947 | * If not, we uncharge res_counter ony by one. | |
3948 | */ | |
c0ff4b85 | 3949 | if (batch->memcg != memcg) |
569b846d KH |
3950 | goto direct_uncharge; |
3951 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 3952 | batch->nr_pages++; |
569b846d | 3953 | if (uncharge_memsw) |
7ffd4ca7 | 3954 | batch->memsw_nr_pages++; |
569b846d KH |
3955 | return; |
3956 | direct_uncharge: | |
c0ff4b85 | 3957 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 3958 | if (uncharge_memsw) |
c0ff4b85 R |
3959 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
3960 | if (unlikely(batch->memcg != memcg)) | |
3961 | memcg_oom_recover(memcg); | |
569b846d | 3962 | } |
7a81b88c | 3963 | |
8a9f3ccd | 3964 | /* |
69029cd5 | 3965 | * uncharge if !page_mapped(page) |
8a9f3ccd | 3966 | */ |
8c7c6e34 | 3967 | static struct mem_cgroup * |
0030f535 JW |
3968 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
3969 | bool end_migration) | |
8a9f3ccd | 3970 | { |
c0ff4b85 | 3971 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
3972 | unsigned int nr_pages = 1; |
3973 | struct page_cgroup *pc; | |
b2402857 | 3974 | bool anon; |
8a9f3ccd | 3975 | |
f8d66542 | 3976 | if (mem_cgroup_disabled()) |
8c7c6e34 | 3977 | return NULL; |
4077960e | 3978 | |
0c59b89c | 3979 | VM_BUG_ON(PageSwapCache(page)); |
d13d1443 | 3980 | |
37c2ac78 | 3981 | if (PageTransHuge(page)) { |
7ec99d62 | 3982 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
3983 | VM_BUG_ON(!PageTransHuge(page)); |
3984 | } | |
8697d331 | 3985 | /* |
3c541e14 | 3986 | * Check if our page_cgroup is valid |
8697d331 | 3987 | */ |
52d4b9ac | 3988 | pc = lookup_page_cgroup(page); |
cfa44946 | 3989 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 3990 | return NULL; |
b9c565d5 | 3991 | |
52d4b9ac | 3992 | lock_page_cgroup(pc); |
d13d1443 | 3993 | |
c0ff4b85 | 3994 | memcg = pc->mem_cgroup; |
8c7c6e34 | 3995 | |
d13d1443 KH |
3996 | if (!PageCgroupUsed(pc)) |
3997 | goto unlock_out; | |
3998 | ||
b2402857 KH |
3999 | anon = PageAnon(page); |
4000 | ||
d13d1443 | 4001 | switch (ctype) { |
41326c17 | 4002 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
4003 | /* |
4004 | * Generally PageAnon tells if it's the anon statistics to be | |
4005 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4006 | * used before page reached the stage of being marked PageAnon. | |
4007 | */ | |
b2402857 KH |
4008 | anon = true; |
4009 | /* fallthrough */ | |
8a9478ca | 4010 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4011 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4012 | if (page_mapped(page)) |
4013 | goto unlock_out; | |
4014 | /* | |
4015 | * Pages under migration may not be uncharged. But | |
4016 | * end_migration() /must/ be the one uncharging the | |
4017 | * unused post-migration page and so it has to call | |
4018 | * here with the migration bit still set. See the | |
4019 | * res_counter handling below. | |
4020 | */ | |
4021 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4022 | goto unlock_out; |
4023 | break; | |
4024 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4025 | if (!PageAnon(page)) { /* Shared memory */ | |
4026 | if (page->mapping && !page_is_file_cache(page)) | |
4027 | goto unlock_out; | |
4028 | } else if (page_mapped(page)) /* Anon */ | |
4029 | goto unlock_out; | |
4030 | break; | |
4031 | default: | |
4032 | break; | |
52d4b9ac | 4033 | } |
d13d1443 | 4034 | |
b2402857 | 4035 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 4036 | |
52d4b9ac | 4037 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4038 | /* |
4039 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4040 | * freed from LRU. This is safe because uncharged page is expected not | |
4041 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4042 | * special functions. | |
4043 | */ | |
b9c565d5 | 4044 | |
52d4b9ac | 4045 | unlock_page_cgroup(pc); |
f75ca962 | 4046 | /* |
c0ff4b85 | 4047 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
4048 | * will never be freed. |
4049 | */ | |
c0ff4b85 | 4050 | memcg_check_events(memcg, page); |
f75ca962 | 4051 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
4052 | mem_cgroup_swap_statistics(memcg, true); |
4053 | mem_cgroup_get(memcg); | |
f75ca962 | 4054 | } |
0030f535 JW |
4055 | /* |
4056 | * Migration does not charge the res_counter for the | |
4057 | * replacement page, so leave it alone when phasing out the | |
4058 | * page that is unused after the migration. | |
4059 | */ | |
4060 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4061 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4062 | |
c0ff4b85 | 4063 | return memcg; |
d13d1443 KH |
4064 | |
4065 | unlock_out: | |
4066 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4067 | return NULL; |
3c541e14 BS |
4068 | } |
4069 | ||
69029cd5 KH |
4070 | void mem_cgroup_uncharge_page(struct page *page) |
4071 | { | |
52d4b9ac KH |
4072 | /* early check. */ |
4073 | if (page_mapped(page)) | |
4074 | return; | |
40f23a21 | 4075 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
0c59b89c JW |
4076 | if (PageSwapCache(page)) |
4077 | return; | |
0030f535 | 4078 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4079 | } |
4080 | ||
4081 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4082 | { | |
4083 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 4084 | VM_BUG_ON(page->mapping); |
0030f535 | 4085 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4086 | } |
4087 | ||
569b846d KH |
4088 | /* |
4089 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4090 | * In that cases, pages are freed continuously and we can expect pages | |
4091 | * are in the same memcg. All these calls itself limits the number of | |
4092 | * pages freed at once, then uncharge_start/end() is called properly. | |
4093 | * This may be called prural(2) times in a context, | |
4094 | */ | |
4095 | ||
4096 | void mem_cgroup_uncharge_start(void) | |
4097 | { | |
4098 | current->memcg_batch.do_batch++; | |
4099 | /* We can do nest. */ | |
4100 | if (current->memcg_batch.do_batch == 1) { | |
4101 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4102 | current->memcg_batch.nr_pages = 0; |
4103 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4104 | } |
4105 | } | |
4106 | ||
4107 | void mem_cgroup_uncharge_end(void) | |
4108 | { | |
4109 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4110 | ||
4111 | if (!batch->do_batch) | |
4112 | return; | |
4113 | ||
4114 | batch->do_batch--; | |
4115 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4116 | return; | |
4117 | ||
4118 | if (!batch->memcg) | |
4119 | return; | |
4120 | /* | |
4121 | * This "batch->memcg" is valid without any css_get/put etc... | |
4122 | * bacause we hide charges behind us. | |
4123 | */ | |
7ffd4ca7 JW |
4124 | if (batch->nr_pages) |
4125 | res_counter_uncharge(&batch->memcg->res, | |
4126 | batch->nr_pages * PAGE_SIZE); | |
4127 | if (batch->memsw_nr_pages) | |
4128 | res_counter_uncharge(&batch->memcg->memsw, | |
4129 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4130 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4131 | /* forget this pointer (for sanity check) */ |
4132 | batch->memcg = NULL; | |
4133 | } | |
4134 | ||
e767e056 | 4135 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4136 | /* |
e767e056 | 4137 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4138 | * memcg information is recorded to swap_cgroup of "ent" |
4139 | */ | |
8a9478ca KH |
4140 | void |
4141 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4142 | { |
4143 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4144 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4145 | ||
4146 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4147 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4148 | ||
0030f535 | 4149 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4150 | |
f75ca962 KH |
4151 | /* |
4152 | * record memcg information, if swapout && memcg != NULL, | |
4153 | * mem_cgroup_get() was called in uncharge(). | |
4154 | */ | |
4155 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 4156 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 4157 | } |
e767e056 | 4158 | #endif |
8c7c6e34 | 4159 | |
c255a458 | 4160 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4161 | /* |
4162 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4163 | * uncharge "memsw" account. | |
4164 | */ | |
4165 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4166 | { |
8c7c6e34 | 4167 | struct mem_cgroup *memcg; |
a3b2d692 | 4168 | unsigned short id; |
8c7c6e34 KH |
4169 | |
4170 | if (!do_swap_account) | |
4171 | return; | |
4172 | ||
a3b2d692 KH |
4173 | id = swap_cgroup_record(ent, 0); |
4174 | rcu_read_lock(); | |
4175 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4176 | if (memcg) { |
a3b2d692 KH |
4177 | /* |
4178 | * We uncharge this because swap is freed. | |
4179 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4180 | */ | |
0c3e73e8 | 4181 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4182 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4183 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
4184 | mem_cgroup_put(memcg); |
4185 | } | |
a3b2d692 | 4186 | rcu_read_unlock(); |
d13d1443 | 4187 | } |
02491447 DN |
4188 | |
4189 | /** | |
4190 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4191 | * @entry: swap entry to be moved | |
4192 | * @from: mem_cgroup which the entry is moved from | |
4193 | * @to: mem_cgroup which the entry is moved to | |
4194 | * | |
4195 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4196 | * as the mem_cgroup's id of @from. | |
4197 | * | |
4198 | * Returns 0 on success, -EINVAL on failure. | |
4199 | * | |
4200 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4201 | * both res and memsw, and called css_get(). | |
4202 | */ | |
4203 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4204 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4205 | { |
4206 | unsigned short old_id, new_id; | |
4207 | ||
4208 | old_id = css_id(&from->css); | |
4209 | new_id = css_id(&to->css); | |
4210 | ||
4211 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4212 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4213 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4214 | /* |
483c30b5 DN |
4215 | * This function is only called from task migration context now. |
4216 | * It postpones res_counter and refcount handling till the end | |
4217 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4218 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
4219 | * because if the process that has been moved to @to does | |
4220 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 4221 | */ |
02491447 | 4222 | mem_cgroup_get(to); |
02491447 DN |
4223 | return 0; |
4224 | } | |
4225 | return -EINVAL; | |
4226 | } | |
4227 | #else | |
4228 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4229 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4230 | { |
4231 | return -EINVAL; | |
4232 | } | |
8c7c6e34 | 4233 | #endif |
d13d1443 | 4234 | |
ae41be37 | 4235 | /* |
01b1ae63 KH |
4236 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4237 | * page belongs to. | |
ae41be37 | 4238 | */ |
0030f535 JW |
4239 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4240 | struct mem_cgroup **memcgp) | |
ae41be37 | 4241 | { |
c0ff4b85 | 4242 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4243 | unsigned int nr_pages = 1; |
7ec99d62 | 4244 | struct page_cgroup *pc; |
ac39cf8c | 4245 | enum charge_type ctype; |
8869b8f6 | 4246 | |
72835c86 | 4247 | *memcgp = NULL; |
56039efa | 4248 | |
f8d66542 | 4249 | if (mem_cgroup_disabled()) |
0030f535 | 4250 | return; |
4077960e | 4251 | |
b32967ff MG |
4252 | if (PageTransHuge(page)) |
4253 | nr_pages <<= compound_order(page); | |
4254 | ||
52d4b9ac KH |
4255 | pc = lookup_page_cgroup(page); |
4256 | lock_page_cgroup(pc); | |
4257 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4258 | memcg = pc->mem_cgroup; |
4259 | css_get(&memcg->css); | |
ac39cf8c | 4260 | /* |
4261 | * At migrating an anonymous page, its mapcount goes down | |
4262 | * to 0 and uncharge() will be called. But, even if it's fully | |
4263 | * unmapped, migration may fail and this page has to be | |
4264 | * charged again. We set MIGRATION flag here and delay uncharge | |
4265 | * until end_migration() is called | |
4266 | * | |
4267 | * Corner Case Thinking | |
4268 | * A) | |
4269 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4270 | * while migration was ongoing. | |
4271 | * If unmap finds the old page, uncharge() of it will be delayed | |
4272 | * until end_migration(). If unmap finds a new page, it's | |
4273 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4274 | * finds swap_migration_entry, the new page will not be mapped | |
4275 | * and end_migration() will find it(mapcount==0). | |
4276 | * | |
4277 | * B) | |
4278 | * When the old page was mapped but migraion fails, the kernel | |
4279 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4280 | * if mapcount goes down to 0. We can do remap successfully | |
4281 | * without charging it again. | |
4282 | * | |
4283 | * C) | |
4284 | * The "old" page is under lock_page() until the end of | |
4285 | * migration, so, the old page itself will not be swapped-out. | |
4286 | * If the new page is swapped out before end_migraton, our | |
4287 | * hook to usual swap-out path will catch the event. | |
4288 | */ | |
4289 | if (PageAnon(page)) | |
4290 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4291 | } |
52d4b9ac | 4292 | unlock_page_cgroup(pc); |
ac39cf8c | 4293 | /* |
4294 | * If the page is not charged at this point, | |
4295 | * we return here. | |
4296 | */ | |
c0ff4b85 | 4297 | if (!memcg) |
0030f535 | 4298 | return; |
01b1ae63 | 4299 | |
72835c86 | 4300 | *memcgp = memcg; |
ac39cf8c | 4301 | /* |
4302 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4303 | * is called before end_migration, we can catch all events on this new | |
4304 | * page. In the case new page is migrated but not remapped, new page's | |
4305 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4306 | */ | |
ac39cf8c | 4307 | if (PageAnon(page)) |
41326c17 | 4308 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4309 | else |
62ba7442 | 4310 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4311 | /* |
4312 | * The page is committed to the memcg, but it's not actually | |
4313 | * charged to the res_counter since we plan on replacing the | |
4314 | * old one and only one page is going to be left afterwards. | |
4315 | */ | |
b32967ff | 4316 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4317 | } |
8869b8f6 | 4318 | |
69029cd5 | 4319 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4320 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4321 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4322 | { |
ac39cf8c | 4323 | struct page *used, *unused; |
01b1ae63 | 4324 | struct page_cgroup *pc; |
b2402857 | 4325 | bool anon; |
01b1ae63 | 4326 | |
c0ff4b85 | 4327 | if (!memcg) |
01b1ae63 | 4328 | return; |
b25ed609 | 4329 | |
50de1dd9 | 4330 | if (!migration_ok) { |
ac39cf8c | 4331 | used = oldpage; |
4332 | unused = newpage; | |
01b1ae63 | 4333 | } else { |
ac39cf8c | 4334 | used = newpage; |
01b1ae63 KH |
4335 | unused = oldpage; |
4336 | } | |
0030f535 | 4337 | anon = PageAnon(used); |
7d188958 JW |
4338 | __mem_cgroup_uncharge_common(unused, |
4339 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4340 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4341 | true); | |
0030f535 | 4342 | css_put(&memcg->css); |
69029cd5 | 4343 | /* |
ac39cf8c | 4344 | * We disallowed uncharge of pages under migration because mapcount |
4345 | * of the page goes down to zero, temporarly. | |
4346 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4347 | */ |
ac39cf8c | 4348 | pc = lookup_page_cgroup(oldpage); |
4349 | lock_page_cgroup(pc); | |
4350 | ClearPageCgroupMigration(pc); | |
4351 | unlock_page_cgroup(pc); | |
ac39cf8c | 4352 | |
01b1ae63 | 4353 | /* |
ac39cf8c | 4354 | * If a page is a file cache, radix-tree replacement is very atomic |
4355 | * and we can skip this check. When it was an Anon page, its mapcount | |
4356 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4357 | * uncharged yet. There are several case but page->mapcount check | |
4358 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4359 | * check. (see prepare_charge() also) | |
69029cd5 | 4360 | */ |
b2402857 | 4361 | if (anon) |
ac39cf8c | 4362 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4363 | } |
78fb7466 | 4364 | |
ab936cbc KH |
4365 | /* |
4366 | * At replace page cache, newpage is not under any memcg but it's on | |
4367 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4368 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4369 | */ | |
4370 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4371 | struct page *newpage) | |
4372 | { | |
bde05d1c | 4373 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4374 | struct page_cgroup *pc; |
ab936cbc | 4375 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4376 | |
4377 | if (mem_cgroup_disabled()) | |
4378 | return; | |
4379 | ||
4380 | pc = lookup_page_cgroup(oldpage); | |
4381 | /* fix accounting on old pages */ | |
4382 | lock_page_cgroup(pc); | |
bde05d1c HD |
4383 | if (PageCgroupUsed(pc)) { |
4384 | memcg = pc->mem_cgroup; | |
4385 | mem_cgroup_charge_statistics(memcg, false, -1); | |
4386 | ClearPageCgroupUsed(pc); | |
4387 | } | |
ab936cbc KH |
4388 | unlock_page_cgroup(pc); |
4389 | ||
bde05d1c HD |
4390 | /* |
4391 | * When called from shmem_replace_page(), in some cases the | |
4392 | * oldpage has already been charged, and in some cases not. | |
4393 | */ | |
4394 | if (!memcg) | |
4395 | return; | |
ab936cbc KH |
4396 | /* |
4397 | * Even if newpage->mapping was NULL before starting replacement, | |
4398 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4399 | * LRU while we overwrite pc->mem_cgroup. | |
4400 | */ | |
ce587e65 | 4401 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4402 | } |
4403 | ||
f212ad7c DN |
4404 | #ifdef CONFIG_DEBUG_VM |
4405 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4406 | { | |
4407 | struct page_cgroup *pc; | |
4408 | ||
4409 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4410 | /* |
4411 | * Can be NULL while feeding pages into the page allocator for | |
4412 | * the first time, i.e. during boot or memory hotplug; | |
4413 | * or when mem_cgroup_disabled(). | |
4414 | */ | |
f212ad7c DN |
4415 | if (likely(pc) && PageCgroupUsed(pc)) |
4416 | return pc; | |
4417 | return NULL; | |
4418 | } | |
4419 | ||
4420 | bool mem_cgroup_bad_page_check(struct page *page) | |
4421 | { | |
4422 | if (mem_cgroup_disabled()) | |
4423 | return false; | |
4424 | ||
4425 | return lookup_page_cgroup_used(page) != NULL; | |
4426 | } | |
4427 | ||
4428 | void mem_cgroup_print_bad_page(struct page *page) | |
4429 | { | |
4430 | struct page_cgroup *pc; | |
4431 | ||
4432 | pc = lookup_page_cgroup_used(page); | |
4433 | if (pc) { | |
d045197f AM |
4434 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4435 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4436 | } |
4437 | } | |
4438 | #endif | |
4439 | ||
d38d2a75 | 4440 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4441 | unsigned long long val) |
628f4235 | 4442 | { |
81d39c20 | 4443 | int retry_count; |
3c11ecf4 | 4444 | u64 memswlimit, memlimit; |
628f4235 | 4445 | int ret = 0; |
81d39c20 KH |
4446 | int children = mem_cgroup_count_children(memcg); |
4447 | u64 curusage, oldusage; | |
3c11ecf4 | 4448 | int enlarge; |
81d39c20 KH |
4449 | |
4450 | /* | |
4451 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4452 | * is depends on callers. We set our retry-count to be function | |
4453 | * of # of children which we should visit in this loop. | |
4454 | */ | |
4455 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4456 | ||
4457 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4458 | |
3c11ecf4 | 4459 | enlarge = 0; |
8c7c6e34 | 4460 | while (retry_count) { |
628f4235 KH |
4461 | if (signal_pending(current)) { |
4462 | ret = -EINTR; | |
4463 | break; | |
4464 | } | |
8c7c6e34 KH |
4465 | /* |
4466 | * Rather than hide all in some function, I do this in | |
4467 | * open coded manner. You see what this really does. | |
aaad153e | 4468 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4469 | */ |
4470 | mutex_lock(&set_limit_mutex); | |
4471 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4472 | if (memswlimit < val) { | |
4473 | ret = -EINVAL; | |
4474 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4475 | break; |
4476 | } | |
3c11ecf4 KH |
4477 | |
4478 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4479 | if (memlimit < val) | |
4480 | enlarge = 1; | |
4481 | ||
8c7c6e34 | 4482 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4483 | if (!ret) { |
4484 | if (memswlimit == val) | |
4485 | memcg->memsw_is_minimum = true; | |
4486 | else | |
4487 | memcg->memsw_is_minimum = false; | |
4488 | } | |
8c7c6e34 KH |
4489 | mutex_unlock(&set_limit_mutex); |
4490 | ||
4491 | if (!ret) | |
4492 | break; | |
4493 | ||
5660048c JW |
4494 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4495 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4496 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4497 | /* Usage is reduced ? */ | |
4498 | if (curusage >= oldusage) | |
4499 | retry_count--; | |
4500 | else | |
4501 | oldusage = curusage; | |
8c7c6e34 | 4502 | } |
3c11ecf4 KH |
4503 | if (!ret && enlarge) |
4504 | memcg_oom_recover(memcg); | |
14797e23 | 4505 | |
8c7c6e34 KH |
4506 | return ret; |
4507 | } | |
4508 | ||
338c8431 LZ |
4509 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4510 | unsigned long long val) | |
8c7c6e34 | 4511 | { |
81d39c20 | 4512 | int retry_count; |
3c11ecf4 | 4513 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4514 | int children = mem_cgroup_count_children(memcg); |
4515 | int ret = -EBUSY; | |
3c11ecf4 | 4516 | int enlarge = 0; |
8c7c6e34 | 4517 | |
81d39c20 KH |
4518 | /* see mem_cgroup_resize_res_limit */ |
4519 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
4520 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
4521 | while (retry_count) { |
4522 | if (signal_pending(current)) { | |
4523 | ret = -EINTR; | |
4524 | break; | |
4525 | } | |
4526 | /* | |
4527 | * Rather than hide all in some function, I do this in | |
4528 | * open coded manner. You see what this really does. | |
aaad153e | 4529 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4530 | */ |
4531 | mutex_lock(&set_limit_mutex); | |
4532 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4533 | if (memlimit > val) { | |
4534 | ret = -EINVAL; | |
4535 | mutex_unlock(&set_limit_mutex); | |
4536 | break; | |
4537 | } | |
3c11ecf4 KH |
4538 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4539 | if (memswlimit < val) | |
4540 | enlarge = 1; | |
8c7c6e34 | 4541 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4542 | if (!ret) { |
4543 | if (memlimit == val) | |
4544 | memcg->memsw_is_minimum = true; | |
4545 | else | |
4546 | memcg->memsw_is_minimum = false; | |
4547 | } | |
8c7c6e34 KH |
4548 | mutex_unlock(&set_limit_mutex); |
4549 | ||
4550 | if (!ret) | |
4551 | break; | |
4552 | ||
5660048c JW |
4553 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4554 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4555 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4556 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4557 | /* Usage is reduced ? */ |
8c7c6e34 | 4558 | if (curusage >= oldusage) |
628f4235 | 4559 | retry_count--; |
81d39c20 KH |
4560 | else |
4561 | oldusage = curusage; | |
628f4235 | 4562 | } |
3c11ecf4 KH |
4563 | if (!ret && enlarge) |
4564 | memcg_oom_recover(memcg); | |
628f4235 KH |
4565 | return ret; |
4566 | } | |
4567 | ||
4e416953 | 4568 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
4569 | gfp_t gfp_mask, |
4570 | unsigned long *total_scanned) | |
4e416953 BS |
4571 | { |
4572 | unsigned long nr_reclaimed = 0; | |
4573 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
4574 | unsigned long reclaimed; | |
4575 | int loop = 0; | |
4576 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 4577 | unsigned long long excess; |
0ae5e89c | 4578 | unsigned long nr_scanned; |
4e416953 BS |
4579 | |
4580 | if (order > 0) | |
4581 | return 0; | |
4582 | ||
00918b6a | 4583 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
4584 | /* |
4585 | * This loop can run a while, specially if mem_cgroup's continuously | |
4586 | * keep exceeding their soft limit and putting the system under | |
4587 | * pressure | |
4588 | */ | |
4589 | do { | |
4590 | if (next_mz) | |
4591 | mz = next_mz; | |
4592 | else | |
4593 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
4594 | if (!mz) | |
4595 | break; | |
4596 | ||
0ae5e89c | 4597 | nr_scanned = 0; |
d79154bb | 4598 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 4599 | gfp_mask, &nr_scanned); |
4e416953 | 4600 | nr_reclaimed += reclaimed; |
0ae5e89c | 4601 | *total_scanned += nr_scanned; |
4e416953 BS |
4602 | spin_lock(&mctz->lock); |
4603 | ||
4604 | /* | |
4605 | * If we failed to reclaim anything from this memory cgroup | |
4606 | * it is time to move on to the next cgroup | |
4607 | */ | |
4608 | next_mz = NULL; | |
4609 | if (!reclaimed) { | |
4610 | do { | |
4611 | /* | |
4612 | * Loop until we find yet another one. | |
4613 | * | |
4614 | * By the time we get the soft_limit lock | |
4615 | * again, someone might have aded the | |
4616 | * group back on the RB tree. Iterate to | |
4617 | * make sure we get a different mem. | |
4618 | * mem_cgroup_largest_soft_limit_node returns | |
4619 | * NULL if no other cgroup is present on | |
4620 | * the tree | |
4621 | */ | |
4622 | next_mz = | |
4623 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 4624 | if (next_mz == mz) |
d79154bb | 4625 | css_put(&next_mz->memcg->css); |
39cc98f1 | 4626 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
4627 | break; |
4628 | } while (1); | |
4629 | } | |
d79154bb HD |
4630 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
4631 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
4632 | /* |
4633 | * One school of thought says that we should not add | |
4634 | * back the node to the tree if reclaim returns 0. | |
4635 | * But our reclaim could return 0, simply because due | |
4636 | * to priority we are exposing a smaller subset of | |
4637 | * memory to reclaim from. Consider this as a longer | |
4638 | * term TODO. | |
4639 | */ | |
ef8745c1 | 4640 | /* If excess == 0, no tree ops */ |
d79154bb | 4641 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 4642 | spin_unlock(&mctz->lock); |
d79154bb | 4643 | css_put(&mz->memcg->css); |
4e416953 BS |
4644 | loop++; |
4645 | /* | |
4646 | * Could not reclaim anything and there are no more | |
4647 | * mem cgroups to try or we seem to be looping without | |
4648 | * reclaiming anything. | |
4649 | */ | |
4650 | if (!nr_reclaimed && | |
4651 | (next_mz == NULL || | |
4652 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4653 | break; | |
4654 | } while (!nr_reclaimed); | |
4655 | if (next_mz) | |
d79154bb | 4656 | css_put(&next_mz->memcg->css); |
4e416953 BS |
4657 | return nr_reclaimed; |
4658 | } | |
4659 | ||
2ef37d3f MH |
4660 | /** |
4661 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4662 | * @memcg: group to clear | |
4663 | * @node: NUMA node | |
4664 | * @zid: zone id | |
4665 | * @lru: lru to to clear | |
4666 | * | |
3c935d18 | 4667 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4668 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4669 | * group. | |
cc847582 | 4670 | */ |
2ef37d3f | 4671 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4672 | int node, int zid, enum lru_list lru) |
cc847582 | 4673 | { |
bea8c150 | 4674 | struct lruvec *lruvec; |
2ef37d3f | 4675 | unsigned long flags; |
072c56c1 | 4676 | struct list_head *list; |
925b7673 JW |
4677 | struct page *busy; |
4678 | struct zone *zone; | |
072c56c1 | 4679 | |
08e552c6 | 4680 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4681 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4682 | list = &lruvec->lists[lru]; | |
cc847582 | 4683 | |
f817ed48 | 4684 | busy = NULL; |
2ef37d3f | 4685 | do { |
925b7673 | 4686 | struct page_cgroup *pc; |
5564e88b JW |
4687 | struct page *page; |
4688 | ||
08e552c6 | 4689 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4690 | if (list_empty(list)) { |
08e552c6 | 4691 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4692 | break; |
f817ed48 | 4693 | } |
925b7673 JW |
4694 | page = list_entry(list->prev, struct page, lru); |
4695 | if (busy == page) { | |
4696 | list_move(&page->lru, list); | |
648bcc77 | 4697 | busy = NULL; |
08e552c6 | 4698 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4699 | continue; |
4700 | } | |
08e552c6 | 4701 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4702 | |
925b7673 | 4703 | pc = lookup_page_cgroup(page); |
5564e88b | 4704 | |
3c935d18 | 4705 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4706 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4707 | busy = page; |
f817ed48 KH |
4708 | cond_resched(); |
4709 | } else | |
4710 | busy = NULL; | |
2ef37d3f | 4711 | } while (!list_empty(list)); |
cc847582 KH |
4712 | } |
4713 | ||
4714 | /* | |
c26251f9 MH |
4715 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4716 | * all the charges and pages to the parent. | |
cc847582 | 4717 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4718 | * |
4719 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4720 | */ |
ab5196c2 | 4721 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4722 | { |
c26251f9 | 4723 | int node, zid; |
bea207c8 | 4724 | u64 usage; |
f817ed48 | 4725 | |
fce66477 | 4726 | do { |
52d4b9ac KH |
4727 | /* This is for making all *used* pages to be on LRU. */ |
4728 | lru_add_drain_all(); | |
c0ff4b85 | 4729 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4730 | mem_cgroup_start_move(memcg); |
31aaea4a | 4731 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4732 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4733 | enum lru_list lru; |
4734 | for_each_lru(lru) { | |
2ef37d3f | 4735 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4736 | node, zid, lru); |
f817ed48 | 4737 | } |
1ecaab2b | 4738 | } |
f817ed48 | 4739 | } |
c0ff4b85 R |
4740 | mem_cgroup_end_move(memcg); |
4741 | memcg_oom_recover(memcg); | |
52d4b9ac | 4742 | cond_resched(); |
f817ed48 | 4743 | |
2ef37d3f | 4744 | /* |
bea207c8 GC |
4745 | * Kernel memory may not necessarily be trackable to a specific |
4746 | * process. So they are not migrated, and therefore we can't | |
4747 | * expect their value to drop to 0 here. | |
4748 | * Having res filled up with kmem only is enough. | |
4749 | * | |
2ef37d3f MH |
4750 | * This is a safety check because mem_cgroup_force_empty_list |
4751 | * could have raced with mem_cgroup_replace_page_cache callers | |
4752 | * so the lru seemed empty but the page could have been added | |
4753 | * right after the check. RES_USAGE should be safe as we always | |
4754 | * charge before adding to the LRU. | |
4755 | */ | |
bea207c8 GC |
4756 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4757 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4758 | } while (usage > 0); | |
c26251f9 MH |
4759 | } |
4760 | ||
b5f99b53 GC |
4761 | /* |
4762 | * This mainly exists for tests during the setting of set of use_hierarchy. | |
4763 | * Since this is the very setting we are changing, the current hierarchy value | |
4764 | * is meaningless | |
4765 | */ | |
4766 | static inline bool __memcg_has_children(struct mem_cgroup *memcg) | |
4767 | { | |
4768 | struct cgroup *pos; | |
4769 | ||
4770 | /* bounce at first found */ | |
4771 | cgroup_for_each_child(pos, memcg->css.cgroup) | |
4772 | return true; | |
4773 | return false; | |
4774 | } | |
4775 | ||
4776 | /* | |
0999821b GC |
4777 | * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed |
4778 | * to be already dead (as in mem_cgroup_force_empty, for instance). This is | |
b5f99b53 GC |
4779 | * from mem_cgroup_count_children(), in the sense that we don't really care how |
4780 | * many children we have; we only need to know if we have any. It also counts | |
4781 | * any memcg without hierarchy as infertile. | |
4782 | */ | |
4783 | static inline bool memcg_has_children(struct mem_cgroup *memcg) | |
4784 | { | |
4785 | return memcg->use_hierarchy && __memcg_has_children(memcg); | |
4786 | } | |
4787 | ||
c26251f9 MH |
4788 | /* |
4789 | * Reclaims as many pages from the given memcg as possible and moves | |
4790 | * the rest to the parent. | |
4791 | * | |
4792 | * Caller is responsible for holding css reference for memcg. | |
4793 | */ | |
4794 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4795 | { | |
4796 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
4797 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 4798 | |
c1e862c1 | 4799 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
4800 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
4801 | return -EBUSY; | |
4802 | ||
c1e862c1 KH |
4803 | /* we call try-to-free pages for make this cgroup empty */ |
4804 | lru_add_drain_all(); | |
f817ed48 | 4805 | /* try to free all pages in this cgroup */ |
569530fb | 4806 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4807 | int progress; |
c1e862c1 | 4808 | |
c26251f9 MH |
4809 | if (signal_pending(current)) |
4810 | return -EINTR; | |
4811 | ||
c0ff4b85 | 4812 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4813 | false); |
c1e862c1 | 4814 | if (!progress) { |
f817ed48 | 4815 | nr_retries--; |
c1e862c1 | 4816 | /* maybe some writeback is necessary */ |
8aa7e847 | 4817 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4818 | } |
f817ed48 KH |
4819 | |
4820 | } | |
08e552c6 | 4821 | lru_add_drain(); |
ab5196c2 MH |
4822 | mem_cgroup_reparent_charges(memcg); |
4823 | ||
4824 | return 0; | |
cc847582 KH |
4825 | } |
4826 | ||
6bbda35c | 4827 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 | 4828 | { |
c26251f9 MH |
4829 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
4830 | int ret; | |
4831 | ||
d8423011 MH |
4832 | if (mem_cgroup_is_root(memcg)) |
4833 | return -EINVAL; | |
c26251f9 MH |
4834 | css_get(&memcg->css); |
4835 | ret = mem_cgroup_force_empty(memcg); | |
4836 | css_put(&memcg->css); | |
4837 | ||
4838 | return ret; | |
c1e862c1 KH |
4839 | } |
4840 | ||
4841 | ||
18f59ea7 BS |
4842 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
4843 | { | |
4844 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
4845 | } | |
4846 | ||
4847 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
4848 | u64 val) | |
4849 | { | |
4850 | int retval = 0; | |
c0ff4b85 | 4851 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 4852 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 4853 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
4854 | |
4855 | if (parent) | |
c0ff4b85 | 4856 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 | 4857 | |
0999821b | 4858 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
4859 | |
4860 | if (memcg->use_hierarchy == val) | |
4861 | goto out; | |
4862 | ||
18f59ea7 | 4863 | /* |
af901ca1 | 4864 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
4865 | * in the child subtrees. If it is unset, then the change can |
4866 | * occur, provided the current cgroup has no children. | |
4867 | * | |
4868 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
4869 | * set if there are no children. | |
4870 | */ | |
c0ff4b85 | 4871 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 4872 | (val == 1 || val == 0)) { |
b5f99b53 | 4873 | if (!__memcg_has_children(memcg)) |
c0ff4b85 | 4874 | memcg->use_hierarchy = val; |
18f59ea7 BS |
4875 | else |
4876 | retval = -EBUSY; | |
4877 | } else | |
4878 | retval = -EINVAL; | |
567fb435 GC |
4879 | |
4880 | out: | |
0999821b | 4881 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
4882 | |
4883 | return retval; | |
4884 | } | |
4885 | ||
0c3e73e8 | 4886 | |
c0ff4b85 | 4887 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 4888 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 4889 | { |
7d74b06f | 4890 | struct mem_cgroup *iter; |
7a159cc9 | 4891 | long val = 0; |
0c3e73e8 | 4892 | |
7a159cc9 | 4893 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 4894 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
4895 | val += mem_cgroup_read_stat(iter, idx); |
4896 | ||
4897 | if (val < 0) /* race ? */ | |
4898 | val = 0; | |
4899 | return val; | |
0c3e73e8 BS |
4900 | } |
4901 | ||
c0ff4b85 | 4902 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 4903 | { |
7d74b06f | 4904 | u64 val; |
104f3928 | 4905 | |
c0ff4b85 | 4906 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 4907 | if (!swap) |
65c64ce8 | 4908 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 4909 | else |
65c64ce8 | 4910 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
4911 | } |
4912 | ||
c0ff4b85 R |
4913 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
4914 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 4915 | |
7d74b06f | 4916 | if (swap) |
bff6bb83 | 4917 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
4918 | |
4919 | return val << PAGE_SHIFT; | |
4920 | } | |
4921 | ||
af36f906 TH |
4922 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
4923 | struct file *file, char __user *buf, | |
4924 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 4925 | { |
c0ff4b85 | 4926 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 4927 | char str[64]; |
104f3928 | 4928 | u64 val; |
86ae53e1 GC |
4929 | int name, len; |
4930 | enum res_type type; | |
8c7c6e34 KH |
4931 | |
4932 | type = MEMFILE_TYPE(cft->private); | |
4933 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
4934 | |
4935 | if (!do_swap_account && type == _MEMSWAP) | |
4936 | return -EOPNOTSUPP; | |
4937 | ||
8c7c6e34 KH |
4938 | switch (type) { |
4939 | case _MEM: | |
104f3928 | 4940 | if (name == RES_USAGE) |
c0ff4b85 | 4941 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 4942 | else |
c0ff4b85 | 4943 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
4944 | break; |
4945 | case _MEMSWAP: | |
104f3928 | 4946 | if (name == RES_USAGE) |
c0ff4b85 | 4947 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 4948 | else |
c0ff4b85 | 4949 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 4950 | break; |
510fc4e1 GC |
4951 | case _KMEM: |
4952 | val = res_counter_read_u64(&memcg->kmem, name); | |
4953 | break; | |
8c7c6e34 KH |
4954 | default: |
4955 | BUG(); | |
8c7c6e34 | 4956 | } |
af36f906 TH |
4957 | |
4958 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
4959 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 4960 | } |
510fc4e1 GC |
4961 | |
4962 | static int memcg_update_kmem_limit(struct cgroup *cont, u64 val) | |
4963 | { | |
4964 | int ret = -EINVAL; | |
4965 | #ifdef CONFIG_MEMCG_KMEM | |
4966 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); | |
4967 | /* | |
4968 | * For simplicity, we won't allow this to be disabled. It also can't | |
4969 | * be changed if the cgroup has children already, or if tasks had | |
4970 | * already joined. | |
4971 | * | |
4972 | * If tasks join before we set the limit, a person looking at | |
4973 | * kmem.usage_in_bytes will have no way to determine when it took | |
4974 | * place, which makes the value quite meaningless. | |
4975 | * | |
4976 | * After it first became limited, changes in the value of the limit are | |
4977 | * of course permitted. | |
510fc4e1 | 4978 | */ |
0999821b | 4979 | mutex_lock(&memcg_create_mutex); |
510fc4e1 GC |
4980 | mutex_lock(&set_limit_mutex); |
4981 | if (!memcg->kmem_account_flags && val != RESOURCE_MAX) { | |
b5f99b53 | 4982 | if (cgroup_task_count(cont) || memcg_has_children(memcg)) { |
510fc4e1 GC |
4983 | ret = -EBUSY; |
4984 | goto out; | |
4985 | } | |
4986 | ret = res_counter_set_limit(&memcg->kmem, val); | |
4987 | VM_BUG_ON(ret); | |
4988 | ||
55007d84 GC |
4989 | ret = memcg_update_cache_sizes(memcg); |
4990 | if (ret) { | |
4991 | res_counter_set_limit(&memcg->kmem, RESOURCE_MAX); | |
4992 | goto out; | |
4993 | } | |
692e89ab GC |
4994 | static_key_slow_inc(&memcg_kmem_enabled_key); |
4995 | /* | |
4996 | * setting the active bit after the inc will guarantee no one | |
4997 | * starts accounting before all call sites are patched | |
4998 | */ | |
4999 | memcg_kmem_set_active(memcg); | |
5000 | ||
7de37682 GC |
5001 | /* |
5002 | * kmem charges can outlive the cgroup. In the case of slab | |
5003 | * pages, for instance, a page contain objects from various | |
5004 | * processes, so it is unfeasible to migrate them away. We | |
5005 | * need to reference count the memcg because of that. | |
5006 | */ | |
5007 | mem_cgroup_get(memcg); | |
510fc4e1 GC |
5008 | } else |
5009 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5010 | out: | |
5011 | mutex_unlock(&set_limit_mutex); | |
0999821b | 5012 | mutex_unlock(&memcg_create_mutex); |
510fc4e1 GC |
5013 | #endif |
5014 | return ret; | |
5015 | } | |
5016 | ||
55007d84 | 5017 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 5018 | { |
55007d84 | 5019 | int ret = 0; |
510fc4e1 GC |
5020 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
5021 | if (!parent) | |
55007d84 GC |
5022 | goto out; |
5023 | ||
510fc4e1 | 5024 | memcg->kmem_account_flags = parent->kmem_account_flags; |
7de37682 | 5025 | #ifdef CONFIG_MEMCG_KMEM |
a8964b9b GC |
5026 | /* |
5027 | * When that happen, we need to disable the static branch only on those | |
5028 | * memcgs that enabled it. To achieve this, we would be forced to | |
5029 | * complicate the code by keeping track of which memcgs were the ones | |
5030 | * that actually enabled limits, and which ones got it from its | |
5031 | * parents. | |
5032 | * | |
5033 | * It is a lot simpler just to do static_key_slow_inc() on every child | |
5034 | * that is accounted. | |
5035 | */ | |
55007d84 GC |
5036 | if (!memcg_kmem_is_active(memcg)) |
5037 | goto out; | |
5038 | ||
5039 | /* | |
5040 | * destroy(), called if we fail, will issue static_key_slow_inc() and | |
5041 | * mem_cgroup_put() if kmem is enabled. We have to either call them | |
5042 | * unconditionally, or clear the KMEM_ACTIVE flag. I personally find | |
5043 | * this more consistent, since it always leads to the same destroy path | |
5044 | */ | |
5045 | mem_cgroup_get(memcg); | |
5046 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
5047 | ||
5048 | mutex_lock(&set_limit_mutex); | |
5049 | ret = memcg_update_cache_sizes(memcg); | |
5050 | mutex_unlock(&set_limit_mutex); | |
7de37682 | 5051 | #endif |
55007d84 GC |
5052 | out: |
5053 | return ret; | |
510fc4e1 GC |
5054 | } |
5055 | ||
628f4235 KH |
5056 | /* |
5057 | * The user of this function is... | |
5058 | * RES_LIMIT. | |
5059 | */ | |
856c13aa PM |
5060 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
5061 | const char *buffer) | |
8cdea7c0 | 5062 | { |
628f4235 | 5063 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
5064 | enum res_type type; |
5065 | int name; | |
628f4235 KH |
5066 | unsigned long long val; |
5067 | int ret; | |
5068 | ||
8c7c6e34 KH |
5069 | type = MEMFILE_TYPE(cft->private); |
5070 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
5071 | |
5072 | if (!do_swap_account && type == _MEMSWAP) | |
5073 | return -EOPNOTSUPP; | |
5074 | ||
8c7c6e34 | 5075 | switch (name) { |
628f4235 | 5076 | case RES_LIMIT: |
4b3bde4c BS |
5077 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
5078 | ret = -EINVAL; | |
5079 | break; | |
5080 | } | |
628f4235 KH |
5081 | /* This function does all necessary parse...reuse it */ |
5082 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
5083 | if (ret) |
5084 | break; | |
5085 | if (type == _MEM) | |
628f4235 | 5086 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 5087 | else if (type == _MEMSWAP) |
8c7c6e34 | 5088 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 GC |
5089 | else if (type == _KMEM) |
5090 | ret = memcg_update_kmem_limit(cont, val); | |
5091 | else | |
5092 | return -EINVAL; | |
628f4235 | 5093 | break; |
296c81d8 BS |
5094 | case RES_SOFT_LIMIT: |
5095 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
5096 | if (ret) | |
5097 | break; | |
5098 | /* | |
5099 | * For memsw, soft limits are hard to implement in terms | |
5100 | * of semantics, for now, we support soft limits for | |
5101 | * control without swap | |
5102 | */ | |
5103 | if (type == _MEM) | |
5104 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5105 | else | |
5106 | ret = -EINVAL; | |
5107 | break; | |
628f4235 KH |
5108 | default: |
5109 | ret = -EINVAL; /* should be BUG() ? */ | |
5110 | break; | |
5111 | } | |
5112 | return ret; | |
8cdea7c0 BS |
5113 | } |
5114 | ||
fee7b548 KH |
5115 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5116 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5117 | { | |
5118 | struct cgroup *cgroup; | |
5119 | unsigned long long min_limit, min_memsw_limit, tmp; | |
5120 | ||
5121 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5122 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5123 | cgroup = memcg->css.cgroup; | |
5124 | if (!memcg->use_hierarchy) | |
5125 | goto out; | |
5126 | ||
5127 | while (cgroup->parent) { | |
5128 | cgroup = cgroup->parent; | |
5129 | memcg = mem_cgroup_from_cont(cgroup); | |
5130 | if (!memcg->use_hierarchy) | |
5131 | break; | |
5132 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5133 | min_limit = min(min_limit, tmp); | |
5134 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5135 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5136 | } | |
5137 | out: | |
5138 | *mem_limit = min_limit; | |
5139 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5140 | } |
5141 | ||
29f2a4da | 5142 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 5143 | { |
af36f906 | 5144 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
5145 | int name; |
5146 | enum res_type type; | |
c84872e1 | 5147 | |
8c7c6e34 KH |
5148 | type = MEMFILE_TYPE(event); |
5149 | name = MEMFILE_ATTR(event); | |
af36f906 TH |
5150 | |
5151 | if (!do_swap_account && type == _MEMSWAP) | |
5152 | return -EOPNOTSUPP; | |
5153 | ||
8c7c6e34 | 5154 | switch (name) { |
29f2a4da | 5155 | case RES_MAX_USAGE: |
8c7c6e34 | 5156 | if (type == _MEM) |
c0ff4b85 | 5157 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5158 | else if (type == _MEMSWAP) |
c0ff4b85 | 5159 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5160 | else if (type == _KMEM) |
5161 | res_counter_reset_max(&memcg->kmem); | |
5162 | else | |
5163 | return -EINVAL; | |
29f2a4da PE |
5164 | break; |
5165 | case RES_FAILCNT: | |
8c7c6e34 | 5166 | if (type == _MEM) |
c0ff4b85 | 5167 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5168 | else if (type == _MEMSWAP) |
c0ff4b85 | 5169 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5170 | else if (type == _KMEM) |
5171 | res_counter_reset_failcnt(&memcg->kmem); | |
5172 | else | |
5173 | return -EINVAL; | |
29f2a4da PE |
5174 | break; |
5175 | } | |
f64c3f54 | 5176 | |
85cc59db | 5177 | return 0; |
c84872e1 PE |
5178 | } |
5179 | ||
7dc74be0 DN |
5180 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
5181 | struct cftype *cft) | |
5182 | { | |
5183 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
5184 | } | |
5185 | ||
02491447 | 5186 | #ifdef CONFIG_MMU |
7dc74be0 DN |
5187 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
5188 | struct cftype *cft, u64 val) | |
5189 | { | |
c0ff4b85 | 5190 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
5191 | |
5192 | if (val >= (1 << NR_MOVE_TYPE)) | |
5193 | return -EINVAL; | |
ee5e8472 | 5194 | |
7dc74be0 | 5195 | /* |
ee5e8472 GC |
5196 | * No kind of locking is needed in here, because ->can_attach() will |
5197 | * check this value once in the beginning of the process, and then carry | |
5198 | * on with stale data. This means that changes to this value will only | |
5199 | * affect task migrations starting after the change. | |
7dc74be0 | 5200 | */ |
c0ff4b85 | 5201 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5202 | return 0; |
5203 | } | |
02491447 DN |
5204 | #else |
5205 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
5206 | struct cftype *cft, u64 val) | |
5207 | { | |
5208 | return -ENOSYS; | |
5209 | } | |
5210 | #endif | |
7dc74be0 | 5211 | |
406eb0c9 | 5212 | #ifdef CONFIG_NUMA |
ab215884 | 5213 | static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft, |
fada52ca | 5214 | struct seq_file *m) |
406eb0c9 YH |
5215 | { |
5216 | int nid; | |
5217 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
5218 | unsigned long node_nr; | |
d79154bb | 5219 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 5220 | |
d79154bb | 5221 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 5222 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 5223 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5224 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
5225 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5226 | } | |
5227 | seq_putc(m, '\n'); | |
5228 | ||
d79154bb | 5229 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 5230 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 5231 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5232 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5233 | LRU_ALL_FILE); |
406eb0c9 YH |
5234 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5235 | } | |
5236 | seq_putc(m, '\n'); | |
5237 | ||
d79154bb | 5238 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 5239 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 5240 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5241 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5242 | LRU_ALL_ANON); |
406eb0c9 YH |
5243 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5244 | } | |
5245 | seq_putc(m, '\n'); | |
5246 | ||
d79154bb | 5247 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 5248 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 5249 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5250 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5251 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
5252 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5253 | } | |
5254 | seq_putc(m, '\n'); | |
5255 | return 0; | |
5256 | } | |
5257 | #endif /* CONFIG_NUMA */ | |
5258 | ||
af7c4b0e JW |
5259 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5260 | { | |
5261 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5262 | } | |
5263 | ||
ab215884 | 5264 | static int memcg_stat_show(struct cgroup *cont, struct cftype *cft, |
78ccf5b5 | 5265 | struct seq_file *m) |
d2ceb9b7 | 5266 | { |
d79154bb | 5267 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af7c4b0e JW |
5268 | struct mem_cgroup *mi; |
5269 | unsigned int i; | |
406eb0c9 | 5270 | |
af7c4b0e | 5271 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5272 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5273 | continue; |
af7c4b0e JW |
5274 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5275 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5276 | } |
7b854121 | 5277 | |
af7c4b0e JW |
5278 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5279 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5280 | mem_cgroup_read_events(memcg, i)); | |
5281 | ||
5282 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5283 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5284 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5285 | ||
14067bb3 | 5286 | /* Hierarchical information */ |
fee7b548 KH |
5287 | { |
5288 | unsigned long long limit, memsw_limit; | |
d79154bb | 5289 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5290 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5291 | if (do_swap_account) |
78ccf5b5 JW |
5292 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5293 | memsw_limit); | |
fee7b548 | 5294 | } |
7f016ee8 | 5295 | |
af7c4b0e JW |
5296 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5297 | long long val = 0; | |
5298 | ||
bff6bb83 | 5299 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5300 | continue; |
af7c4b0e JW |
5301 | for_each_mem_cgroup_tree(mi, memcg) |
5302 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5303 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5304 | } | |
5305 | ||
5306 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5307 | unsigned long long val = 0; | |
5308 | ||
5309 | for_each_mem_cgroup_tree(mi, memcg) | |
5310 | val += mem_cgroup_read_events(mi, i); | |
5311 | seq_printf(m, "total_%s %llu\n", | |
5312 | mem_cgroup_events_names[i], val); | |
5313 | } | |
5314 | ||
5315 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5316 | unsigned long long val = 0; | |
5317 | ||
5318 | for_each_mem_cgroup_tree(mi, memcg) | |
5319 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5320 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5321 | } |
14067bb3 | 5322 | |
7f016ee8 | 5323 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5324 | { |
5325 | int nid, zid; | |
5326 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5327 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5328 | unsigned long recent_rotated[2] = {0, 0}; |
5329 | unsigned long recent_scanned[2] = {0, 0}; | |
5330 | ||
5331 | for_each_online_node(nid) | |
5332 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5333 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5334 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5335 | |
89abfab1 HD |
5336 | recent_rotated[0] += rstat->recent_rotated[0]; |
5337 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5338 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5339 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5340 | } |
78ccf5b5 JW |
5341 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5342 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5343 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5344 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5345 | } |
5346 | #endif | |
5347 | ||
d2ceb9b7 KH |
5348 | return 0; |
5349 | } | |
5350 | ||
a7885eb8 KM |
5351 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
5352 | { | |
5353 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
5354 | ||
1f4c025b | 5355 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5356 | } |
5357 | ||
5358 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
5359 | u64 val) | |
5360 | { | |
5361 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
5362 | struct mem_cgroup *parent; | |
068b38c1 | 5363 | |
a7885eb8 KM |
5364 | if (val > 100) |
5365 | return -EINVAL; | |
5366 | ||
5367 | if (cgrp->parent == NULL) | |
5368 | return -EINVAL; | |
5369 | ||
5370 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 | 5371 | |
0999821b | 5372 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5373 | |
a7885eb8 | 5374 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5375 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5376 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5377 | return -EINVAL; |
068b38c1 | 5378 | } |
a7885eb8 | 5379 | |
a7885eb8 | 5380 | memcg->swappiness = val; |
a7885eb8 | 5381 | |
0999821b | 5382 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5383 | |
a7885eb8 KM |
5384 | return 0; |
5385 | } | |
5386 | ||
2e72b634 KS |
5387 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5388 | { | |
5389 | struct mem_cgroup_threshold_ary *t; | |
5390 | u64 usage; | |
5391 | int i; | |
5392 | ||
5393 | rcu_read_lock(); | |
5394 | if (!swap) | |
2c488db2 | 5395 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5396 | else |
2c488db2 | 5397 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5398 | |
5399 | if (!t) | |
5400 | goto unlock; | |
5401 | ||
5402 | usage = mem_cgroup_usage(memcg, swap); | |
5403 | ||
5404 | /* | |
748dad36 | 5405 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5406 | * If it's not true, a threshold was crossed after last |
5407 | * call of __mem_cgroup_threshold(). | |
5408 | */ | |
5407a562 | 5409 | i = t->current_threshold; |
2e72b634 KS |
5410 | |
5411 | /* | |
5412 | * Iterate backward over array of thresholds starting from | |
5413 | * current_threshold and check if a threshold is crossed. | |
5414 | * If none of thresholds below usage is crossed, we read | |
5415 | * only one element of the array here. | |
5416 | */ | |
5417 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5418 | eventfd_signal(t->entries[i].eventfd, 1); | |
5419 | ||
5420 | /* i = current_threshold + 1 */ | |
5421 | i++; | |
5422 | ||
5423 | /* | |
5424 | * Iterate forward over array of thresholds starting from | |
5425 | * current_threshold+1 and check if a threshold is crossed. | |
5426 | * If none of thresholds above usage is crossed, we read | |
5427 | * only one element of the array here. | |
5428 | */ | |
5429 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5430 | eventfd_signal(t->entries[i].eventfd, 1); | |
5431 | ||
5432 | /* Update current_threshold */ | |
5407a562 | 5433 | t->current_threshold = i - 1; |
2e72b634 KS |
5434 | unlock: |
5435 | rcu_read_unlock(); | |
5436 | } | |
5437 | ||
5438 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5439 | { | |
ad4ca5f4 KS |
5440 | while (memcg) { |
5441 | __mem_cgroup_threshold(memcg, false); | |
5442 | if (do_swap_account) | |
5443 | __mem_cgroup_threshold(memcg, true); | |
5444 | ||
5445 | memcg = parent_mem_cgroup(memcg); | |
5446 | } | |
2e72b634 KS |
5447 | } |
5448 | ||
5449 | static int compare_thresholds(const void *a, const void *b) | |
5450 | { | |
5451 | const struct mem_cgroup_threshold *_a = a; | |
5452 | const struct mem_cgroup_threshold *_b = b; | |
5453 | ||
5454 | return _a->threshold - _b->threshold; | |
5455 | } | |
5456 | ||
c0ff4b85 | 5457 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5458 | { |
5459 | struct mem_cgroup_eventfd_list *ev; | |
5460 | ||
c0ff4b85 | 5461 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5462 | eventfd_signal(ev->eventfd, 1); |
5463 | return 0; | |
5464 | } | |
5465 | ||
c0ff4b85 | 5466 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5467 | { |
7d74b06f KH |
5468 | struct mem_cgroup *iter; |
5469 | ||
c0ff4b85 | 5470 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5471 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5472 | } |
5473 | ||
5474 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
5475 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
5476 | { |
5477 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
5478 | struct mem_cgroup_thresholds *thresholds; |
5479 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5480 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5481 | u64 threshold, usage; |
2c488db2 | 5482 | int i, size, ret; |
2e72b634 KS |
5483 | |
5484 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5485 | if (ret) | |
5486 | return ret; | |
5487 | ||
5488 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5489 | |
2e72b634 | 5490 | if (type == _MEM) |
2c488db2 | 5491 | thresholds = &memcg->thresholds; |
2e72b634 | 5492 | else if (type == _MEMSWAP) |
2c488db2 | 5493 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5494 | else |
5495 | BUG(); | |
5496 | ||
5497 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5498 | ||
5499 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5500 | if (thresholds->primary) |
2e72b634 KS |
5501 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5502 | ||
2c488db2 | 5503 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5504 | |
5505 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5506 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5507 | GFP_KERNEL); |
2c488db2 | 5508 | if (!new) { |
2e72b634 KS |
5509 | ret = -ENOMEM; |
5510 | goto unlock; | |
5511 | } | |
2c488db2 | 5512 | new->size = size; |
2e72b634 KS |
5513 | |
5514 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5515 | if (thresholds->primary) { |
5516 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5517 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5518 | } |
5519 | ||
2e72b634 | 5520 | /* Add new threshold */ |
2c488db2 KS |
5521 | new->entries[size - 1].eventfd = eventfd; |
5522 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5523 | |
5524 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5525 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5526 | compare_thresholds, NULL); |
5527 | ||
5528 | /* Find current threshold */ | |
2c488db2 | 5529 | new->current_threshold = -1; |
2e72b634 | 5530 | for (i = 0; i < size; i++) { |
748dad36 | 5531 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5532 | /* |
2c488db2 KS |
5533 | * new->current_threshold will not be used until |
5534 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5535 | * it here. |
5536 | */ | |
2c488db2 | 5537 | ++new->current_threshold; |
748dad36 SZ |
5538 | } else |
5539 | break; | |
2e72b634 KS |
5540 | } |
5541 | ||
2c488db2 KS |
5542 | /* Free old spare buffer and save old primary buffer as spare */ |
5543 | kfree(thresholds->spare); | |
5544 | thresholds->spare = thresholds->primary; | |
5545 | ||
5546 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5547 | |
907860ed | 5548 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5549 | synchronize_rcu(); |
5550 | ||
2e72b634 KS |
5551 | unlock: |
5552 | mutex_unlock(&memcg->thresholds_lock); | |
5553 | ||
5554 | return ret; | |
5555 | } | |
5556 | ||
907860ed | 5557 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 5558 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
5559 | { |
5560 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
5561 | struct mem_cgroup_thresholds *thresholds; |
5562 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5563 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5564 | u64 usage; |
2c488db2 | 5565 | int i, j, size; |
2e72b634 KS |
5566 | |
5567 | mutex_lock(&memcg->thresholds_lock); | |
5568 | if (type == _MEM) | |
2c488db2 | 5569 | thresholds = &memcg->thresholds; |
2e72b634 | 5570 | else if (type == _MEMSWAP) |
2c488db2 | 5571 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5572 | else |
5573 | BUG(); | |
5574 | ||
371528ca AV |
5575 | if (!thresholds->primary) |
5576 | goto unlock; | |
5577 | ||
2e72b634 KS |
5578 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5579 | ||
5580 | /* Check if a threshold crossed before removing */ | |
5581 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5582 | ||
5583 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5584 | size = 0; |
5585 | for (i = 0; i < thresholds->primary->size; i++) { | |
5586 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5587 | size++; |
5588 | } | |
5589 | ||
2c488db2 | 5590 | new = thresholds->spare; |
907860ed | 5591 | |
2e72b634 KS |
5592 | /* Set thresholds array to NULL if we don't have thresholds */ |
5593 | if (!size) { | |
2c488db2 KS |
5594 | kfree(new); |
5595 | new = NULL; | |
907860ed | 5596 | goto swap_buffers; |
2e72b634 KS |
5597 | } |
5598 | ||
2c488db2 | 5599 | new->size = size; |
2e72b634 KS |
5600 | |
5601 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5602 | new->current_threshold = -1; |
5603 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5604 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5605 | continue; |
5606 | ||
2c488db2 | 5607 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5608 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5609 | /* |
2c488db2 | 5610 | * new->current_threshold will not be used |
2e72b634 KS |
5611 | * until rcu_assign_pointer(), so it's safe to increment |
5612 | * it here. | |
5613 | */ | |
2c488db2 | 5614 | ++new->current_threshold; |
2e72b634 KS |
5615 | } |
5616 | j++; | |
5617 | } | |
5618 | ||
907860ed | 5619 | swap_buffers: |
2c488db2 KS |
5620 | /* Swap primary and spare array */ |
5621 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5622 | /* If all events are unregistered, free the spare array */ |
5623 | if (!new) { | |
5624 | kfree(thresholds->spare); | |
5625 | thresholds->spare = NULL; | |
5626 | } | |
5627 | ||
2c488db2 | 5628 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5629 | |
907860ed | 5630 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5631 | synchronize_rcu(); |
371528ca | 5632 | unlock: |
2e72b634 | 5633 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5634 | } |
c1e862c1 | 5635 | |
9490ff27 KH |
5636 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
5637 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
5638 | { | |
5639 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
5640 | struct mem_cgroup_eventfd_list *event; | |
86ae53e1 | 5641 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5642 | |
5643 | BUG_ON(type != _OOM_TYPE); | |
5644 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
5645 | if (!event) | |
5646 | return -ENOMEM; | |
5647 | ||
1af8efe9 | 5648 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5649 | |
5650 | event->eventfd = eventfd; | |
5651 | list_add(&event->list, &memcg->oom_notify); | |
5652 | ||
5653 | /* already in OOM ? */ | |
79dfdacc | 5654 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5655 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5656 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5657 | |
5658 | return 0; | |
5659 | } | |
5660 | ||
907860ed | 5661 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
5662 | struct cftype *cft, struct eventfd_ctx *eventfd) |
5663 | { | |
c0ff4b85 | 5664 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 | 5665 | struct mem_cgroup_eventfd_list *ev, *tmp; |
86ae53e1 | 5666 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5667 | |
5668 | BUG_ON(type != _OOM_TYPE); | |
5669 | ||
1af8efe9 | 5670 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5671 | |
c0ff4b85 | 5672 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5673 | if (ev->eventfd == eventfd) { |
5674 | list_del(&ev->list); | |
5675 | kfree(ev); | |
5676 | } | |
5677 | } | |
5678 | ||
1af8efe9 | 5679 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5680 | } |
5681 | ||
3c11ecf4 KH |
5682 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
5683 | struct cftype *cft, struct cgroup_map_cb *cb) | |
5684 | { | |
c0ff4b85 | 5685 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 5686 | |
c0ff4b85 | 5687 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 5688 | |
c0ff4b85 | 5689 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
5690 | cb->fill(cb, "under_oom", 1); |
5691 | else | |
5692 | cb->fill(cb, "under_oom", 0); | |
5693 | return 0; | |
5694 | } | |
5695 | ||
3c11ecf4 KH |
5696 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
5697 | struct cftype *cft, u64 val) | |
5698 | { | |
c0ff4b85 | 5699 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
5700 | struct mem_cgroup *parent; |
5701 | ||
5702 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
5703 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
5704 | return -EINVAL; | |
5705 | ||
5706 | parent = mem_cgroup_from_cont(cgrp->parent); | |
5707 | ||
0999821b | 5708 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5709 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5710 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5711 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5712 | return -EINVAL; |
5713 | } | |
c0ff4b85 | 5714 | memcg->oom_kill_disable = val; |
4d845ebf | 5715 | if (!val) |
c0ff4b85 | 5716 | memcg_oom_recover(memcg); |
0999821b | 5717 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5718 | return 0; |
5719 | } | |
5720 | ||
c255a458 | 5721 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5722 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5723 | { |
55007d84 GC |
5724 | int ret; |
5725 | ||
2633d7a0 | 5726 | memcg->kmemcg_id = -1; |
55007d84 GC |
5727 | ret = memcg_propagate_kmem(memcg); |
5728 | if (ret) | |
5729 | return ret; | |
2633d7a0 | 5730 | |
1d62e436 | 5731 | return mem_cgroup_sockets_init(memcg, ss); |
e5671dfa GC |
5732 | }; |
5733 | ||
1d62e436 | 5734 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 | 5735 | { |
1d62e436 | 5736 | mem_cgroup_sockets_destroy(memcg); |
7de37682 GC |
5737 | |
5738 | memcg_kmem_mark_dead(memcg); | |
5739 | ||
5740 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5741 | return; | |
5742 | ||
5743 | /* | |
5744 | * Charges already down to 0, undo mem_cgroup_get() done in the charge | |
5745 | * path here, being careful not to race with memcg_uncharge_kmem: it is | |
5746 | * possible that the charges went down to 0 between mark_dead and the | |
5747 | * res_counter read, so in that case, we don't need the put | |
5748 | */ | |
5749 | if (memcg_kmem_test_and_clear_dead(memcg)) | |
5750 | mem_cgroup_put(memcg); | |
d1a4c0b3 | 5751 | } |
e5671dfa | 5752 | #else |
cbe128e3 | 5753 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5754 | { |
5755 | return 0; | |
5756 | } | |
d1a4c0b3 | 5757 | |
1d62e436 | 5758 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 GC |
5759 | { |
5760 | } | |
e5671dfa GC |
5761 | #endif |
5762 | ||
8cdea7c0 BS |
5763 | static struct cftype mem_cgroup_files[] = { |
5764 | { | |
0eea1030 | 5765 | .name = "usage_in_bytes", |
8c7c6e34 | 5766 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 5767 | .read = mem_cgroup_read, |
9490ff27 KH |
5768 | .register_event = mem_cgroup_usage_register_event, |
5769 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 5770 | }, |
c84872e1 PE |
5771 | { |
5772 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5773 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 5774 | .trigger = mem_cgroup_reset, |
af36f906 | 5775 | .read = mem_cgroup_read, |
c84872e1 | 5776 | }, |
8cdea7c0 | 5777 | { |
0eea1030 | 5778 | .name = "limit_in_bytes", |
8c7c6e34 | 5779 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 5780 | .write_string = mem_cgroup_write, |
af36f906 | 5781 | .read = mem_cgroup_read, |
8cdea7c0 | 5782 | }, |
296c81d8 BS |
5783 | { |
5784 | .name = "soft_limit_in_bytes", | |
5785 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
5786 | .write_string = mem_cgroup_write, | |
af36f906 | 5787 | .read = mem_cgroup_read, |
296c81d8 | 5788 | }, |
8cdea7c0 BS |
5789 | { |
5790 | .name = "failcnt", | |
8c7c6e34 | 5791 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 5792 | .trigger = mem_cgroup_reset, |
af36f906 | 5793 | .read = mem_cgroup_read, |
8cdea7c0 | 5794 | }, |
d2ceb9b7 KH |
5795 | { |
5796 | .name = "stat", | |
ab215884 | 5797 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 5798 | }, |
c1e862c1 KH |
5799 | { |
5800 | .name = "force_empty", | |
5801 | .trigger = mem_cgroup_force_empty_write, | |
5802 | }, | |
18f59ea7 BS |
5803 | { |
5804 | .name = "use_hierarchy", | |
5805 | .write_u64 = mem_cgroup_hierarchy_write, | |
5806 | .read_u64 = mem_cgroup_hierarchy_read, | |
5807 | }, | |
a7885eb8 KM |
5808 | { |
5809 | .name = "swappiness", | |
5810 | .read_u64 = mem_cgroup_swappiness_read, | |
5811 | .write_u64 = mem_cgroup_swappiness_write, | |
5812 | }, | |
7dc74be0 DN |
5813 | { |
5814 | .name = "move_charge_at_immigrate", | |
5815 | .read_u64 = mem_cgroup_move_charge_read, | |
5816 | .write_u64 = mem_cgroup_move_charge_write, | |
5817 | }, | |
9490ff27 KH |
5818 | { |
5819 | .name = "oom_control", | |
3c11ecf4 KH |
5820 | .read_map = mem_cgroup_oom_control_read, |
5821 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
5822 | .register_event = mem_cgroup_oom_register_event, |
5823 | .unregister_event = mem_cgroup_oom_unregister_event, | |
5824 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
5825 | }, | |
406eb0c9 YH |
5826 | #ifdef CONFIG_NUMA |
5827 | { | |
5828 | .name = "numa_stat", | |
ab215884 | 5829 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
5830 | }, |
5831 | #endif | |
510fc4e1 GC |
5832 | #ifdef CONFIG_MEMCG_KMEM |
5833 | { | |
5834 | .name = "kmem.limit_in_bytes", | |
5835 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
5836 | .write_string = mem_cgroup_write, | |
5837 | .read = mem_cgroup_read, | |
5838 | }, | |
5839 | { | |
5840 | .name = "kmem.usage_in_bytes", | |
5841 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
5842 | .read = mem_cgroup_read, | |
5843 | }, | |
5844 | { | |
5845 | .name = "kmem.failcnt", | |
5846 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
5847 | .trigger = mem_cgroup_reset, | |
5848 | .read = mem_cgroup_read, | |
5849 | }, | |
5850 | { | |
5851 | .name = "kmem.max_usage_in_bytes", | |
5852 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
5853 | .trigger = mem_cgroup_reset, | |
5854 | .read = mem_cgroup_read, | |
5855 | }, | |
749c5415 GC |
5856 | #ifdef CONFIG_SLABINFO |
5857 | { | |
5858 | .name = "kmem.slabinfo", | |
5859 | .read_seq_string = mem_cgroup_slabinfo_read, | |
5860 | }, | |
5861 | #endif | |
8c7c6e34 | 5862 | #endif |
6bc10349 | 5863 | { }, /* terminate */ |
af36f906 | 5864 | }; |
8c7c6e34 | 5865 | |
2d11085e MH |
5866 | #ifdef CONFIG_MEMCG_SWAP |
5867 | static struct cftype memsw_cgroup_files[] = { | |
5868 | { | |
5869 | .name = "memsw.usage_in_bytes", | |
5870 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5871 | .read = mem_cgroup_read, | |
5872 | .register_event = mem_cgroup_usage_register_event, | |
5873 | .unregister_event = mem_cgroup_usage_unregister_event, | |
5874 | }, | |
5875 | { | |
5876 | .name = "memsw.max_usage_in_bytes", | |
5877 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5878 | .trigger = mem_cgroup_reset, | |
5879 | .read = mem_cgroup_read, | |
5880 | }, | |
5881 | { | |
5882 | .name = "memsw.limit_in_bytes", | |
5883 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5884 | .write_string = mem_cgroup_write, | |
5885 | .read = mem_cgroup_read, | |
5886 | }, | |
5887 | { | |
5888 | .name = "memsw.failcnt", | |
5889 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5890 | .trigger = mem_cgroup_reset, | |
5891 | .read = mem_cgroup_read, | |
5892 | }, | |
5893 | { }, /* terminate */ | |
5894 | }; | |
5895 | #endif | |
c0ff4b85 | 5896 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5897 | { |
5898 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 5899 | struct mem_cgroup_per_zone *mz; |
41e3355d | 5900 | int zone, tmp = node; |
1ecaab2b KH |
5901 | /* |
5902 | * This routine is called against possible nodes. | |
5903 | * But it's BUG to call kmalloc() against offline node. | |
5904 | * | |
5905 | * TODO: this routine can waste much memory for nodes which will | |
5906 | * never be onlined. It's better to use memory hotplug callback | |
5907 | * function. | |
5908 | */ | |
41e3355d KH |
5909 | if (!node_state(node, N_NORMAL_MEMORY)) |
5910 | tmp = -1; | |
17295c88 | 5911 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5912 | if (!pn) |
5913 | return 1; | |
1ecaab2b | 5914 | |
1ecaab2b KH |
5915 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
5916 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 5917 | lruvec_init(&mz->lruvec); |
f64c3f54 | 5918 | mz->usage_in_excess = 0; |
4e416953 | 5919 | mz->on_tree = false; |
d79154bb | 5920 | mz->memcg = memcg; |
1ecaab2b | 5921 | } |
0a619e58 | 5922 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
5923 | return 0; |
5924 | } | |
5925 | ||
c0ff4b85 | 5926 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5927 | { |
c0ff4b85 | 5928 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
5929 | } |
5930 | ||
33327948 KH |
5931 | static struct mem_cgroup *mem_cgroup_alloc(void) |
5932 | { | |
d79154bb | 5933 | struct mem_cgroup *memcg; |
45cf7ebd | 5934 | size_t size = memcg_size(); |
33327948 | 5935 | |
45cf7ebd | 5936 | /* Can be very big if nr_node_ids is very big */ |
c8dad2bb | 5937 | if (size < PAGE_SIZE) |
d79154bb | 5938 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 5939 | else |
d79154bb | 5940 | memcg = vzalloc(size); |
33327948 | 5941 | |
d79154bb | 5942 | if (!memcg) |
e7bbcdf3 DC |
5943 | return NULL; |
5944 | ||
d79154bb HD |
5945 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
5946 | if (!memcg->stat) | |
d2e61b8d | 5947 | goto out_free; |
d79154bb HD |
5948 | spin_lock_init(&memcg->pcp_counter_lock); |
5949 | return memcg; | |
d2e61b8d DC |
5950 | |
5951 | out_free: | |
5952 | if (size < PAGE_SIZE) | |
d79154bb | 5953 | kfree(memcg); |
d2e61b8d | 5954 | else |
d79154bb | 5955 | vfree(memcg); |
d2e61b8d | 5956 | return NULL; |
33327948 KH |
5957 | } |
5958 | ||
59927fb9 | 5959 | /* |
c8b2a36f GC |
5960 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
5961 | * (scanning all at force_empty is too costly...) | |
5962 | * | |
5963 | * Instead of clearing all references at force_empty, we remember | |
5964 | * the number of reference from swap_cgroup and free mem_cgroup when | |
5965 | * it goes down to 0. | |
5966 | * | |
5967 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 5968 | */ |
c8b2a36f GC |
5969 | |
5970 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 5971 | { |
c8b2a36f | 5972 | int node; |
45cf7ebd | 5973 | size_t size = memcg_size(); |
59927fb9 | 5974 | |
c8b2a36f GC |
5975 | mem_cgroup_remove_from_trees(memcg); |
5976 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
5977 | ||
5978 | for_each_node(node) | |
5979 | free_mem_cgroup_per_zone_info(memcg, node); | |
5980 | ||
5981 | free_percpu(memcg->stat); | |
5982 | ||
3f134619 GC |
5983 | /* |
5984 | * We need to make sure that (at least for now), the jump label | |
5985 | * destruction code runs outside of the cgroup lock. This is because | |
5986 | * get_online_cpus(), which is called from the static_branch update, | |
5987 | * can't be called inside the cgroup_lock. cpusets are the ones | |
5988 | * enforcing this dependency, so if they ever change, we might as well. | |
5989 | * | |
5990 | * schedule_work() will guarantee this happens. Be careful if you need | |
5991 | * to move this code around, and make sure it is outside | |
5992 | * the cgroup_lock. | |
5993 | */ | |
a8964b9b | 5994 | disarm_static_keys(memcg); |
3afe36b1 GC |
5995 | if (size < PAGE_SIZE) |
5996 | kfree(memcg); | |
5997 | else | |
5998 | vfree(memcg); | |
59927fb9 | 5999 | } |
3afe36b1 | 6000 | |
59927fb9 | 6001 | |
8c7c6e34 | 6002 | /* |
c8b2a36f GC |
6003 | * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU, |
6004 | * but in process context. The work_freeing structure is overlaid | |
6005 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
8c7c6e34 | 6006 | */ |
c8b2a36f | 6007 | static void free_work(struct work_struct *work) |
33327948 | 6008 | { |
c8b2a36f | 6009 | struct mem_cgroup *memcg; |
08e552c6 | 6010 | |
c8b2a36f GC |
6011 | memcg = container_of(work, struct mem_cgroup, work_freeing); |
6012 | __mem_cgroup_free(memcg); | |
6013 | } | |
04046e1a | 6014 | |
c8b2a36f GC |
6015 | static void free_rcu(struct rcu_head *rcu_head) |
6016 | { | |
6017 | struct mem_cgroup *memcg; | |
08e552c6 | 6018 | |
c8b2a36f GC |
6019 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); |
6020 | INIT_WORK(&memcg->work_freeing, free_work); | |
6021 | schedule_work(&memcg->work_freeing); | |
33327948 KH |
6022 | } |
6023 | ||
c0ff4b85 | 6024 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 6025 | { |
c0ff4b85 | 6026 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
6027 | } |
6028 | ||
c0ff4b85 | 6029 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 6030 | { |
c0ff4b85 R |
6031 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
6032 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
c8b2a36f | 6033 | call_rcu(&memcg->rcu_freeing, free_rcu); |
7bcc1bb1 DN |
6034 | if (parent) |
6035 | mem_cgroup_put(parent); | |
6036 | } | |
8c7c6e34 KH |
6037 | } |
6038 | ||
c0ff4b85 | 6039 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 6040 | { |
c0ff4b85 | 6041 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
6042 | } |
6043 | ||
7bcc1bb1 DN |
6044 | /* |
6045 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6046 | */ | |
e1aab161 | 6047 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6048 | { |
c0ff4b85 | 6049 | if (!memcg->res.parent) |
7bcc1bb1 | 6050 | return NULL; |
c0ff4b85 | 6051 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6052 | } |
e1aab161 | 6053 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6054 | |
8787a1df | 6055 | static void __init mem_cgroup_soft_limit_tree_init(void) |
f64c3f54 BS |
6056 | { |
6057 | struct mem_cgroup_tree_per_node *rtpn; | |
6058 | struct mem_cgroup_tree_per_zone *rtpz; | |
6059 | int tmp, node, zone; | |
6060 | ||
3ed28fa1 | 6061 | for_each_node(node) { |
f64c3f54 BS |
6062 | tmp = node; |
6063 | if (!node_state(node, N_NORMAL_MEMORY)) | |
6064 | tmp = -1; | |
6065 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
8787a1df | 6066 | BUG_ON(!rtpn); |
f64c3f54 BS |
6067 | |
6068 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
6069 | ||
6070 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
6071 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
6072 | rtpz->rb_root = RB_ROOT; | |
6073 | spin_lock_init(&rtpz->lock); | |
6074 | } | |
6075 | } | |
f64c3f54 BS |
6076 | } |
6077 | ||
0eb253e2 | 6078 | static struct cgroup_subsys_state * __ref |
92fb9748 | 6079 | mem_cgroup_css_alloc(struct cgroup *cont) |
8cdea7c0 | 6080 | { |
d142e3e6 | 6081 | struct mem_cgroup *memcg; |
04046e1a | 6082 | long error = -ENOMEM; |
6d12e2d8 | 6083 | int node; |
8cdea7c0 | 6084 | |
c0ff4b85 R |
6085 | memcg = mem_cgroup_alloc(); |
6086 | if (!memcg) | |
04046e1a | 6087 | return ERR_PTR(error); |
78fb7466 | 6088 | |
3ed28fa1 | 6089 | for_each_node(node) |
c0ff4b85 | 6090 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6091 | goto free_out; |
f64c3f54 | 6092 | |
c077719b | 6093 | /* root ? */ |
28dbc4b6 | 6094 | if (cont->parent == NULL) { |
cdec2e42 | 6095 | int cpu; |
d142e3e6 | 6096 | |
a41c58a6 | 6097 | root_mem_cgroup = memcg; |
cdec2e42 KH |
6098 | for_each_possible_cpu(cpu) { |
6099 | struct memcg_stock_pcp *stock = | |
6100 | &per_cpu(memcg_stock, cpu); | |
6101 | INIT_WORK(&stock->work, drain_local_stock); | |
6102 | } | |
d142e3e6 GC |
6103 | |
6104 | res_counter_init(&memcg->res, NULL); | |
6105 | res_counter_init(&memcg->memsw, NULL); | |
6106 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6107 | } |
28dbc4b6 | 6108 | |
d142e3e6 GC |
6109 | memcg->last_scanned_node = MAX_NUMNODES; |
6110 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6111 | atomic_set(&memcg->refcnt, 1); | |
6112 | memcg->move_charge_at_immigrate = 0; | |
6113 | mutex_init(&memcg->thresholds_lock); | |
6114 | spin_lock_init(&memcg->move_lock); | |
6115 | ||
6116 | return &memcg->css; | |
6117 | ||
6118 | free_out: | |
6119 | __mem_cgroup_free(memcg); | |
6120 | return ERR_PTR(error); | |
6121 | } | |
6122 | ||
6123 | static int | |
6124 | mem_cgroup_css_online(struct cgroup *cont) | |
6125 | { | |
6126 | struct mem_cgroup *memcg, *parent; | |
6127 | int error = 0; | |
6128 | ||
6129 | if (!cont->parent) | |
6130 | return 0; | |
6131 | ||
0999821b | 6132 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6133 | memcg = mem_cgroup_from_cont(cont); |
6134 | parent = mem_cgroup_from_cont(cont->parent); | |
6135 | ||
6136 | memcg->use_hierarchy = parent->use_hierarchy; | |
6137 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6138 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6139 | ||
6140 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6141 | res_counter_init(&memcg->res, &parent->res); |
6142 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6143 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6144 | |
7bcc1bb1 DN |
6145 | /* |
6146 | * We increment refcnt of the parent to ensure that we can | |
6147 | * safely access it on res_counter_charge/uncharge. | |
6148 | * This refcnt will be decremented when freeing this | |
6149 | * mem_cgroup(see mem_cgroup_put). | |
6150 | */ | |
6151 | mem_cgroup_get(parent); | |
18f59ea7 | 6152 | } else { |
c0ff4b85 R |
6153 | res_counter_init(&memcg->res, NULL); |
6154 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 6155 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
6156 | /* |
6157 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6158 | * much sense so let cgroup subsystem know about this | |
6159 | * unfortunate state in our controller. | |
6160 | */ | |
d142e3e6 | 6161 | if (parent != root_mem_cgroup) |
8c7f6edb | 6162 | mem_cgroup_subsys.broken_hierarchy = true; |
18f59ea7 | 6163 | } |
cbe128e3 GC |
6164 | |
6165 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
0999821b | 6166 | mutex_unlock(&memcg_create_mutex); |
cbe128e3 GC |
6167 | if (error) { |
6168 | /* | |
6169 | * We call put now because our (and parent's) refcnts | |
6170 | * are already in place. mem_cgroup_put() will internally | |
6171 | * call __mem_cgroup_free, so return directly | |
6172 | */ | |
6173 | mem_cgroup_put(memcg); | |
e4715f01 GC |
6174 | if (parent->use_hierarchy) |
6175 | mem_cgroup_put(parent); | |
cbe128e3 | 6176 | } |
d142e3e6 | 6177 | return error; |
8cdea7c0 BS |
6178 | } |
6179 | ||
92fb9748 | 6180 | static void mem_cgroup_css_offline(struct cgroup *cont) |
df878fb0 | 6181 | { |
c0ff4b85 | 6182 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 6183 | |
ab5196c2 | 6184 | mem_cgroup_reparent_charges(memcg); |
1f458cbf | 6185 | mem_cgroup_destroy_all_caches(memcg); |
df878fb0 KH |
6186 | } |
6187 | ||
92fb9748 | 6188 | static void mem_cgroup_css_free(struct cgroup *cont) |
8cdea7c0 | 6189 | { |
c0ff4b85 | 6190 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 6191 | |
1d62e436 | 6192 | kmem_cgroup_destroy(memcg); |
d1a4c0b3 | 6193 | |
c0ff4b85 | 6194 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
6195 | } |
6196 | ||
02491447 | 6197 | #ifdef CONFIG_MMU |
7dc74be0 | 6198 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6199 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6200 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6201 | { |
854ffa8d DN |
6202 | int ret = 0; |
6203 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6204 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6205 | |
c0ff4b85 | 6206 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6207 | mc.precharge += count; |
6208 | /* we don't need css_get for root */ | |
6209 | return ret; | |
6210 | } | |
6211 | /* try to charge at once */ | |
6212 | if (count > 1) { | |
6213 | struct res_counter *dummy; | |
6214 | /* | |
c0ff4b85 | 6215 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6216 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6217 | * are still under the same cgroup_mutex. So we can postpone | |
6218 | * css_get(). | |
6219 | */ | |
c0ff4b85 | 6220 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6221 | goto one_by_one; |
c0ff4b85 | 6222 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6223 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6224 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6225 | goto one_by_one; |
6226 | } | |
6227 | mc.precharge += count; | |
854ffa8d DN |
6228 | return ret; |
6229 | } | |
6230 | one_by_one: | |
6231 | /* fall back to one by one charge */ | |
6232 | while (count--) { | |
6233 | if (signal_pending(current)) { | |
6234 | ret = -EINTR; | |
6235 | break; | |
6236 | } | |
6237 | if (!batch_count--) { | |
6238 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6239 | cond_resched(); | |
6240 | } | |
c0ff4b85 R |
6241 | ret = __mem_cgroup_try_charge(NULL, |
6242 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6243 | if (ret) |
854ffa8d | 6244 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6245 | return ret; |
854ffa8d DN |
6246 | mc.precharge++; |
6247 | } | |
4ffef5fe DN |
6248 | return ret; |
6249 | } | |
6250 | ||
6251 | /** | |
8d32ff84 | 6252 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6253 | * @vma: the vma the pte to be checked belongs |
6254 | * @addr: the address corresponding to the pte to be checked | |
6255 | * @ptent: the pte to be checked | |
02491447 | 6256 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6257 | * |
6258 | * Returns | |
6259 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6260 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6261 | * move charge. if @target is not NULL, the page is stored in target->page | |
6262 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6263 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6264 | * target for charge migration. if @target is not NULL, the entry is stored | |
6265 | * in target->ent. | |
4ffef5fe DN |
6266 | * |
6267 | * Called with pte lock held. | |
6268 | */ | |
4ffef5fe DN |
6269 | union mc_target { |
6270 | struct page *page; | |
02491447 | 6271 | swp_entry_t ent; |
4ffef5fe DN |
6272 | }; |
6273 | ||
4ffef5fe | 6274 | enum mc_target_type { |
8d32ff84 | 6275 | MC_TARGET_NONE = 0, |
4ffef5fe | 6276 | MC_TARGET_PAGE, |
02491447 | 6277 | MC_TARGET_SWAP, |
4ffef5fe DN |
6278 | }; |
6279 | ||
90254a65 DN |
6280 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6281 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6282 | { |
90254a65 | 6283 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6284 | |
90254a65 DN |
6285 | if (!page || !page_mapped(page)) |
6286 | return NULL; | |
6287 | if (PageAnon(page)) { | |
6288 | /* we don't move shared anon */ | |
4b91355e | 6289 | if (!move_anon()) |
90254a65 | 6290 | return NULL; |
87946a72 DN |
6291 | } else if (!move_file()) |
6292 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6293 | return NULL; |
6294 | if (!get_page_unless_zero(page)) | |
6295 | return NULL; | |
6296 | ||
6297 | return page; | |
6298 | } | |
6299 | ||
4b91355e | 6300 | #ifdef CONFIG_SWAP |
90254a65 DN |
6301 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6302 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6303 | { | |
90254a65 DN |
6304 | struct page *page = NULL; |
6305 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6306 | ||
6307 | if (!move_anon() || non_swap_entry(ent)) | |
6308 | return NULL; | |
4b91355e KH |
6309 | /* |
6310 | * Because lookup_swap_cache() updates some statistics counter, | |
6311 | * we call find_get_page() with swapper_space directly. | |
6312 | */ | |
33806f06 | 6313 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6314 | if (do_swap_account) |
6315 | entry->val = ent.val; | |
6316 | ||
6317 | return page; | |
6318 | } | |
4b91355e KH |
6319 | #else |
6320 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6321 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6322 | { | |
6323 | return NULL; | |
6324 | } | |
6325 | #endif | |
90254a65 | 6326 | |
87946a72 DN |
6327 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6328 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6329 | { | |
6330 | struct page *page = NULL; | |
87946a72 DN |
6331 | struct address_space *mapping; |
6332 | pgoff_t pgoff; | |
6333 | ||
6334 | if (!vma->vm_file) /* anonymous vma */ | |
6335 | return NULL; | |
6336 | if (!move_file()) | |
6337 | return NULL; | |
6338 | ||
87946a72 DN |
6339 | mapping = vma->vm_file->f_mapping; |
6340 | if (pte_none(ptent)) | |
6341 | pgoff = linear_page_index(vma, addr); | |
6342 | else /* pte_file(ptent) is true */ | |
6343 | pgoff = pte_to_pgoff(ptent); | |
6344 | ||
6345 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6346 | page = find_get_page(mapping, pgoff); |
6347 | ||
6348 | #ifdef CONFIG_SWAP | |
6349 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6350 | if (radix_tree_exceptional_entry(page)) { | |
6351 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6352 | if (do_swap_account) |
aa3b1895 | 6353 | *entry = swap; |
33806f06 | 6354 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6355 | } |
aa3b1895 | 6356 | #endif |
87946a72 DN |
6357 | return page; |
6358 | } | |
6359 | ||
8d32ff84 | 6360 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6361 | unsigned long addr, pte_t ptent, union mc_target *target) |
6362 | { | |
6363 | struct page *page = NULL; | |
6364 | struct page_cgroup *pc; | |
8d32ff84 | 6365 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6366 | swp_entry_t ent = { .val = 0 }; |
6367 | ||
6368 | if (pte_present(ptent)) | |
6369 | page = mc_handle_present_pte(vma, addr, ptent); | |
6370 | else if (is_swap_pte(ptent)) | |
6371 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6372 | else if (pte_none(ptent) || pte_file(ptent)) |
6373 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6374 | |
6375 | if (!page && !ent.val) | |
8d32ff84 | 6376 | return ret; |
02491447 DN |
6377 | if (page) { |
6378 | pc = lookup_page_cgroup(page); | |
6379 | /* | |
6380 | * Do only loose check w/o page_cgroup lock. | |
6381 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6382 | * the lock. | |
6383 | */ | |
6384 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6385 | ret = MC_TARGET_PAGE; | |
6386 | if (target) | |
6387 | target->page = page; | |
6388 | } | |
6389 | if (!ret || !target) | |
6390 | put_page(page); | |
6391 | } | |
90254a65 DN |
6392 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6393 | if (ent.val && !ret && | |
9fb4b7cc | 6394 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6395 | ret = MC_TARGET_SWAP; |
6396 | if (target) | |
6397 | target->ent = ent; | |
4ffef5fe | 6398 | } |
4ffef5fe DN |
6399 | return ret; |
6400 | } | |
6401 | ||
12724850 NH |
6402 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6403 | /* | |
6404 | * We don't consider swapping or file mapped pages because THP does not | |
6405 | * support them for now. | |
6406 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6407 | */ | |
6408 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6409 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6410 | { | |
6411 | struct page *page = NULL; | |
6412 | struct page_cgroup *pc; | |
6413 | enum mc_target_type ret = MC_TARGET_NONE; | |
6414 | ||
6415 | page = pmd_page(pmd); | |
6416 | VM_BUG_ON(!page || !PageHead(page)); | |
6417 | if (!move_anon()) | |
6418 | return ret; | |
6419 | pc = lookup_page_cgroup(page); | |
6420 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6421 | ret = MC_TARGET_PAGE; | |
6422 | if (target) { | |
6423 | get_page(page); | |
6424 | target->page = page; | |
6425 | } | |
6426 | } | |
6427 | return ret; | |
6428 | } | |
6429 | #else | |
6430 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6431 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6432 | { | |
6433 | return MC_TARGET_NONE; | |
6434 | } | |
6435 | #endif | |
6436 | ||
4ffef5fe DN |
6437 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6438 | unsigned long addr, unsigned long end, | |
6439 | struct mm_walk *walk) | |
6440 | { | |
6441 | struct vm_area_struct *vma = walk->private; | |
6442 | pte_t *pte; | |
6443 | spinlock_t *ptl; | |
6444 | ||
12724850 NH |
6445 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
6446 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
6447 | mc.precharge += HPAGE_PMD_NR; | |
6448 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6449 | return 0; |
12724850 | 6450 | } |
03319327 | 6451 | |
45f83cef AA |
6452 | if (pmd_trans_unstable(pmd)) |
6453 | return 0; | |
4ffef5fe DN |
6454 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6455 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6456 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6457 | mc.precharge++; /* increment precharge temporarily */ |
6458 | pte_unmap_unlock(pte - 1, ptl); | |
6459 | cond_resched(); | |
6460 | ||
7dc74be0 DN |
6461 | return 0; |
6462 | } | |
6463 | ||
4ffef5fe DN |
6464 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6465 | { | |
6466 | unsigned long precharge; | |
6467 | struct vm_area_struct *vma; | |
6468 | ||
dfe076b0 | 6469 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6470 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6471 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6472 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6473 | .mm = mm, | |
6474 | .private = vma, | |
6475 | }; | |
6476 | if (is_vm_hugetlb_page(vma)) | |
6477 | continue; | |
4ffef5fe DN |
6478 | walk_page_range(vma->vm_start, vma->vm_end, |
6479 | &mem_cgroup_count_precharge_walk); | |
6480 | } | |
dfe076b0 | 6481 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6482 | |
6483 | precharge = mc.precharge; | |
6484 | mc.precharge = 0; | |
6485 | ||
6486 | return precharge; | |
6487 | } | |
6488 | ||
4ffef5fe DN |
6489 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6490 | { | |
dfe076b0 DN |
6491 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6492 | ||
6493 | VM_BUG_ON(mc.moving_task); | |
6494 | mc.moving_task = current; | |
6495 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6496 | } |
6497 | ||
dfe076b0 DN |
6498 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6499 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6500 | { |
2bd9bb20 KH |
6501 | struct mem_cgroup *from = mc.from; |
6502 | struct mem_cgroup *to = mc.to; | |
6503 | ||
4ffef5fe | 6504 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6505 | if (mc.precharge) { |
6506 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6507 | mc.precharge = 0; | |
6508 | } | |
6509 | /* | |
6510 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6511 | * we must uncharge here. | |
6512 | */ | |
6513 | if (mc.moved_charge) { | |
6514 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6515 | mc.moved_charge = 0; | |
4ffef5fe | 6516 | } |
483c30b5 DN |
6517 | /* we must fixup refcnts and charges */ |
6518 | if (mc.moved_swap) { | |
483c30b5 DN |
6519 | /* uncharge swap account from the old cgroup */ |
6520 | if (!mem_cgroup_is_root(mc.from)) | |
6521 | res_counter_uncharge(&mc.from->memsw, | |
6522 | PAGE_SIZE * mc.moved_swap); | |
6523 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
6524 | ||
6525 | if (!mem_cgroup_is_root(mc.to)) { | |
6526 | /* | |
6527 | * we charged both to->res and to->memsw, so we should | |
6528 | * uncharge to->res. | |
6529 | */ | |
6530 | res_counter_uncharge(&mc.to->res, | |
6531 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
6532 | } |
6533 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
6534 | mc.moved_swap = 0; |
6535 | } | |
dfe076b0 DN |
6536 | memcg_oom_recover(from); |
6537 | memcg_oom_recover(to); | |
6538 | wake_up_all(&mc.waitq); | |
6539 | } | |
6540 | ||
6541 | static void mem_cgroup_clear_mc(void) | |
6542 | { | |
6543 | struct mem_cgroup *from = mc.from; | |
6544 | ||
6545 | /* | |
6546 | * we must clear moving_task before waking up waiters at the end of | |
6547 | * task migration. | |
6548 | */ | |
6549 | mc.moving_task = NULL; | |
6550 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6551 | spin_lock(&mc.lock); |
4ffef5fe DN |
6552 | mc.from = NULL; |
6553 | mc.to = NULL; | |
2bd9bb20 | 6554 | spin_unlock(&mc.lock); |
32047e2a | 6555 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6556 | } |
6557 | ||
761b3ef5 LZ |
6558 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
6559 | struct cgroup_taskset *tset) | |
7dc74be0 | 6560 | { |
2f7ee569 | 6561 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6562 | int ret = 0; |
c0ff4b85 | 6563 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
ee5e8472 | 6564 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6565 | |
ee5e8472 GC |
6566 | /* |
6567 | * We are now commited to this value whatever it is. Changes in this | |
6568 | * tunable will only affect upcoming migrations, not the current one. | |
6569 | * So we need to save it, and keep it going. | |
6570 | */ | |
6571 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
6572 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
6573 | struct mm_struct *mm; |
6574 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
6575 | ||
c0ff4b85 | 6576 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
6577 | |
6578 | mm = get_task_mm(p); | |
6579 | if (!mm) | |
6580 | return 0; | |
7dc74be0 | 6581 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
6582 | if (mm->owner == p) { |
6583 | VM_BUG_ON(mc.from); | |
6584 | VM_BUG_ON(mc.to); | |
6585 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 6586 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 6587 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 6588 | mem_cgroup_start_move(from); |
2bd9bb20 | 6589 | spin_lock(&mc.lock); |
4ffef5fe | 6590 | mc.from = from; |
c0ff4b85 | 6591 | mc.to = memcg; |
ee5e8472 | 6592 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 6593 | spin_unlock(&mc.lock); |
dfe076b0 | 6594 | /* We set mc.moving_task later */ |
4ffef5fe DN |
6595 | |
6596 | ret = mem_cgroup_precharge_mc(mm); | |
6597 | if (ret) | |
6598 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
6599 | } |
6600 | mmput(mm); | |
7dc74be0 DN |
6601 | } |
6602 | return ret; | |
6603 | } | |
6604 | ||
761b3ef5 LZ |
6605 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
6606 | struct cgroup_taskset *tset) | |
7dc74be0 | 6607 | { |
4ffef5fe | 6608 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
6609 | } |
6610 | ||
4ffef5fe DN |
6611 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6612 | unsigned long addr, unsigned long end, | |
6613 | struct mm_walk *walk) | |
7dc74be0 | 6614 | { |
4ffef5fe DN |
6615 | int ret = 0; |
6616 | struct vm_area_struct *vma = walk->private; | |
6617 | pte_t *pte; | |
6618 | spinlock_t *ptl; | |
12724850 NH |
6619 | enum mc_target_type target_type; |
6620 | union mc_target target; | |
6621 | struct page *page; | |
6622 | struct page_cgroup *pc; | |
4ffef5fe | 6623 | |
12724850 NH |
6624 | /* |
6625 | * We don't take compound_lock() here but no race with splitting thp | |
6626 | * happens because: | |
6627 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
6628 | * under splitting, which means there's no concurrent thp split, | |
6629 | * - if another thread runs into split_huge_page() just after we | |
6630 | * entered this if-block, the thread must wait for page table lock | |
6631 | * to be unlocked in __split_huge_page_splitting(), where the main | |
6632 | * part of thp split is not executed yet. | |
6633 | */ | |
6634 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 6635 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
6636 | spin_unlock(&vma->vm_mm->page_table_lock); |
6637 | return 0; | |
6638 | } | |
6639 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6640 | if (target_type == MC_TARGET_PAGE) { | |
6641 | page = target.page; | |
6642 | if (!isolate_lru_page(page)) { | |
6643 | pc = lookup_page_cgroup(page); | |
6644 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6645 | pc, mc.from, mc.to)) { |
12724850 NH |
6646 | mc.precharge -= HPAGE_PMD_NR; |
6647 | mc.moved_charge += HPAGE_PMD_NR; | |
6648 | } | |
6649 | putback_lru_page(page); | |
6650 | } | |
6651 | put_page(page); | |
6652 | } | |
6653 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6654 | return 0; |
12724850 NH |
6655 | } |
6656 | ||
45f83cef AA |
6657 | if (pmd_trans_unstable(pmd)) |
6658 | return 0; | |
4ffef5fe DN |
6659 | retry: |
6660 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6661 | for (; addr != end; addr += PAGE_SIZE) { | |
6662 | pte_t ptent = *(pte++); | |
02491447 | 6663 | swp_entry_t ent; |
4ffef5fe DN |
6664 | |
6665 | if (!mc.precharge) | |
6666 | break; | |
6667 | ||
8d32ff84 | 6668 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6669 | case MC_TARGET_PAGE: |
6670 | page = target.page; | |
6671 | if (isolate_lru_page(page)) | |
6672 | goto put; | |
6673 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6674 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6675 | mc.from, mc.to)) { |
4ffef5fe | 6676 | mc.precharge--; |
854ffa8d DN |
6677 | /* we uncharge from mc.from later. */ |
6678 | mc.moved_charge++; | |
4ffef5fe DN |
6679 | } |
6680 | putback_lru_page(page); | |
8d32ff84 | 6681 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6682 | put_page(page); |
6683 | break; | |
02491447 DN |
6684 | case MC_TARGET_SWAP: |
6685 | ent = target.ent; | |
e91cbb42 | 6686 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6687 | mc.precharge--; |
483c30b5 DN |
6688 | /* we fixup refcnts and charges later. */ |
6689 | mc.moved_swap++; | |
6690 | } | |
02491447 | 6691 | break; |
4ffef5fe DN |
6692 | default: |
6693 | break; | |
6694 | } | |
6695 | } | |
6696 | pte_unmap_unlock(pte - 1, ptl); | |
6697 | cond_resched(); | |
6698 | ||
6699 | if (addr != end) { | |
6700 | /* | |
6701 | * We have consumed all precharges we got in can_attach(). | |
6702 | * We try charge one by one, but don't do any additional | |
6703 | * charges to mc.to if we have failed in charge once in attach() | |
6704 | * phase. | |
6705 | */ | |
854ffa8d | 6706 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6707 | if (!ret) |
6708 | goto retry; | |
6709 | } | |
6710 | ||
6711 | return ret; | |
6712 | } | |
6713 | ||
6714 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6715 | { | |
6716 | struct vm_area_struct *vma; | |
6717 | ||
6718 | lru_add_drain_all(); | |
dfe076b0 DN |
6719 | retry: |
6720 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6721 | /* | |
6722 | * Someone who are holding the mmap_sem might be waiting in | |
6723 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6724 | * and retry. Because we cancel precharges, we might not be able | |
6725 | * to move enough charges, but moving charge is a best-effort | |
6726 | * feature anyway, so it wouldn't be a big problem. | |
6727 | */ | |
6728 | __mem_cgroup_clear_mc(); | |
6729 | cond_resched(); | |
6730 | goto retry; | |
6731 | } | |
4ffef5fe DN |
6732 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6733 | int ret; | |
6734 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6735 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6736 | .mm = mm, | |
6737 | .private = vma, | |
6738 | }; | |
6739 | if (is_vm_hugetlb_page(vma)) | |
6740 | continue; | |
4ffef5fe DN |
6741 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6742 | &mem_cgroup_move_charge_walk); | |
6743 | if (ret) | |
6744 | /* | |
6745 | * means we have consumed all precharges and failed in | |
6746 | * doing additional charge. Just abandon here. | |
6747 | */ | |
6748 | break; | |
6749 | } | |
dfe076b0 | 6750 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6751 | } |
6752 | ||
761b3ef5 LZ |
6753 | static void mem_cgroup_move_task(struct cgroup *cont, |
6754 | struct cgroup_taskset *tset) | |
67e465a7 | 6755 | { |
2f7ee569 | 6756 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6757 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6758 | |
dfe076b0 | 6759 | if (mm) { |
a433658c KM |
6760 | if (mc.to) |
6761 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6762 | mmput(mm); |
6763 | } | |
a433658c KM |
6764 | if (mc.to) |
6765 | mem_cgroup_clear_mc(); | |
67e465a7 | 6766 | } |
5cfb80a7 | 6767 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
6768 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
6769 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6770 | { |
6771 | return 0; | |
6772 | } | |
761b3ef5 LZ |
6773 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
6774 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6775 | { |
6776 | } | |
761b3ef5 LZ |
6777 | static void mem_cgroup_move_task(struct cgroup *cont, |
6778 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6779 | { |
6780 | } | |
6781 | #endif | |
67e465a7 | 6782 | |
8cdea7c0 BS |
6783 | struct cgroup_subsys mem_cgroup_subsys = { |
6784 | .name = "memory", | |
6785 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 | 6786 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6787 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6788 | .css_offline = mem_cgroup_css_offline, |
6789 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
6790 | .can_attach = mem_cgroup_can_attach, |
6791 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6792 | .attach = mem_cgroup_move_task, |
6bc10349 | 6793 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 6794 | .early_init = 0, |
04046e1a | 6795 | .use_id = 1, |
8cdea7c0 | 6796 | }; |
c077719b | 6797 | |
c255a458 | 6798 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6799 | static int __init enable_swap_account(char *s) |
6800 | { | |
6801 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 6802 | if (!strcmp(s, "1")) |
a42c390c | 6803 | really_do_swap_account = 1; |
a2c8990a | 6804 | else if (!strcmp(s, "0")) |
a42c390c MH |
6805 | really_do_swap_account = 0; |
6806 | return 1; | |
6807 | } | |
a2c8990a | 6808 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6809 | |
2d11085e MH |
6810 | static void __init memsw_file_init(void) |
6811 | { | |
6acc8b02 MH |
6812 | WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files)); |
6813 | } | |
6814 | ||
6815 | static void __init enable_swap_cgroup(void) | |
6816 | { | |
6817 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6818 | do_swap_account = 1; | |
6819 | memsw_file_init(); | |
6820 | } | |
2d11085e | 6821 | } |
6acc8b02 | 6822 | |
2d11085e | 6823 | #else |
6acc8b02 | 6824 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6825 | { |
6826 | } | |
c077719b | 6827 | #endif |
2d11085e MH |
6828 | |
6829 | /* | |
6830 | * The rest of init is performed during ->css_alloc() for root css which | |
6831 | * happens before initcalls. hotcpu_notifier() can't be done together as | |
6832 | * it would introduce circular locking by adding cgroup_lock -> cpu hotplug | |
6833 | * dependency. Do it from a subsys_initcall(). | |
6834 | */ | |
6835 | static int __init mem_cgroup_init(void) | |
6836 | { | |
6837 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6838 | enable_swap_cgroup(); |
8787a1df | 6839 | mem_cgroup_soft_limit_tree_init(); |
2d11085e MH |
6840 | return 0; |
6841 | } | |
6842 | subsys_initcall(mem_cgroup_init); |