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 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
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> |
bb4cc1a8 | 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 | 47 | #include <linux/eventfd.h> |
79bd9814 | 48 | #include <linux/poll.h> |
2e72b634 | 49 | #include <linux/sort.h> |
66e1707b | 50 | #include <linux/fs.h> |
d2ceb9b7 | 51 | #include <linux/seq_file.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
52d4b9ac | 54 | #include <linux/page_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
79bd9814 | 58 | #include <linux/file.h> |
08e552c6 | 59 | #include "internal.h" |
d1a4c0b3 | 60 | #include <net/sock.h> |
4bd2c1ee | 61 | #include <net/ip.h> |
d1a4c0b3 | 62 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 63 | #include "slab.h" |
8cdea7c0 | 64 | |
8697d331 BS |
65 | #include <asm/uaccess.h> |
66 | ||
cc8e970c KM |
67 | #include <trace/events/vmscan.h> |
68 | ||
073219e9 TH |
69 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
70 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 71 | |
a181b0e8 | 72 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 73 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 74 | |
c255a458 | 75 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 76 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 77 | int do_swap_account __read_mostly; |
a42c390c MH |
78 | |
79 | /* for remember boot option*/ | |
c255a458 | 80 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
81 | static int really_do_swap_account __initdata = 1; |
82 | #else | |
ada4ba59 | 83 | static int really_do_swap_account __initdata; |
a42c390c MH |
84 | #endif |
85 | ||
c077719b | 86 | #else |
a0db00fc | 87 | #define do_swap_account 0 |
c077719b KH |
88 | #endif |
89 | ||
90 | ||
af7c4b0e JW |
91 | static const char * const mem_cgroup_stat_names[] = { |
92 | "cache", | |
93 | "rss", | |
b070e65c | 94 | "rss_huge", |
af7c4b0e | 95 | "mapped_file", |
3ea67d06 | 96 | "writeback", |
af7c4b0e JW |
97 | "swap", |
98 | }; | |
99 | ||
e9f8974f JW |
100 | enum mem_cgroup_events_index { |
101 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
102 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
103 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
104 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
105 | MEM_CGROUP_EVENTS_NSTATS, |
106 | }; | |
af7c4b0e JW |
107 | |
108 | static const char * const mem_cgroup_events_names[] = { | |
109 | "pgpgin", | |
110 | "pgpgout", | |
111 | "pgfault", | |
112 | "pgmajfault", | |
113 | }; | |
114 | ||
58cf188e SZ |
115 | static const char * const mem_cgroup_lru_names[] = { |
116 | "inactive_anon", | |
117 | "active_anon", | |
118 | "inactive_file", | |
119 | "active_file", | |
120 | "unevictable", | |
121 | }; | |
122 | ||
7a159cc9 JW |
123 | /* |
124 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
125 | * it will be incremated by the number of pages. This counter is used for | |
126 | * for trigger some periodic events. This is straightforward and better | |
127 | * than using jiffies etc. to handle periodic memcg event. | |
128 | */ | |
129 | enum mem_cgroup_events_target { | |
130 | MEM_CGROUP_TARGET_THRESH, | |
bb4cc1a8 | 131 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 132 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
133 | MEM_CGROUP_NTARGETS, |
134 | }; | |
a0db00fc KS |
135 | #define THRESHOLDS_EVENTS_TARGET 128 |
136 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
137 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 138 | |
d52aa412 | 139 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 140 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 141 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 142 | unsigned long nr_page_events; |
7a159cc9 | 143 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
144 | }; |
145 | ||
527a5ec9 | 146 | struct mem_cgroup_reclaim_iter { |
5f578161 MH |
147 | /* |
148 | * last scanned hierarchy member. Valid only if last_dead_count | |
149 | * matches memcg->dead_count of the hierarchy root group. | |
150 | */ | |
542f85f9 | 151 | struct mem_cgroup *last_visited; |
d2ab70aa | 152 | int last_dead_count; |
5f578161 | 153 | |
527a5ec9 JW |
154 | /* scan generation, increased every round-trip */ |
155 | unsigned int generation; | |
156 | }; | |
157 | ||
6d12e2d8 KH |
158 | /* |
159 | * per-zone information in memory controller. | |
160 | */ | |
6d12e2d8 | 161 | struct mem_cgroup_per_zone { |
6290df54 | 162 | struct lruvec lruvec; |
1eb49272 | 163 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 164 | |
527a5ec9 JW |
165 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
166 | ||
bb4cc1a8 | 167 | struct rb_node tree_node; /* RB tree node */ |
3e32cb2e | 168 | unsigned long usage_in_excess;/* Set to the value by which */ |
bb4cc1a8 AM |
169 | /* the soft limit is exceeded*/ |
170 | bool on_tree; | |
d79154bb | 171 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 172 | /* use container_of */ |
6d12e2d8 | 173 | }; |
6d12e2d8 KH |
174 | |
175 | struct mem_cgroup_per_node { | |
176 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
177 | }; | |
178 | ||
bb4cc1a8 AM |
179 | /* |
180 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
181 | * their hierarchy representation | |
182 | */ | |
183 | ||
184 | struct mem_cgroup_tree_per_zone { | |
185 | struct rb_root rb_root; | |
186 | spinlock_t lock; | |
187 | }; | |
188 | ||
189 | struct mem_cgroup_tree_per_node { | |
190 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
191 | }; | |
192 | ||
193 | struct mem_cgroup_tree { | |
194 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
195 | }; | |
196 | ||
197 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
198 | ||
2e72b634 KS |
199 | struct mem_cgroup_threshold { |
200 | struct eventfd_ctx *eventfd; | |
3e32cb2e | 201 | unsigned long threshold; |
2e72b634 KS |
202 | }; |
203 | ||
9490ff27 | 204 | /* For threshold */ |
2e72b634 | 205 | struct mem_cgroup_threshold_ary { |
748dad36 | 206 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 207 | int current_threshold; |
2e72b634 KS |
208 | /* Size of entries[] */ |
209 | unsigned int size; | |
210 | /* Array of thresholds */ | |
211 | struct mem_cgroup_threshold entries[0]; | |
212 | }; | |
2c488db2 KS |
213 | |
214 | struct mem_cgroup_thresholds { | |
215 | /* Primary thresholds array */ | |
216 | struct mem_cgroup_threshold_ary *primary; | |
217 | /* | |
218 | * Spare threshold array. | |
219 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
220 | * It must be able to store at least primary->size - 1 entries. | |
221 | */ | |
222 | struct mem_cgroup_threshold_ary *spare; | |
223 | }; | |
224 | ||
9490ff27 KH |
225 | /* for OOM */ |
226 | struct mem_cgroup_eventfd_list { | |
227 | struct list_head list; | |
228 | struct eventfd_ctx *eventfd; | |
229 | }; | |
2e72b634 | 230 | |
79bd9814 TH |
231 | /* |
232 | * cgroup_event represents events which userspace want to receive. | |
233 | */ | |
3bc942f3 | 234 | struct mem_cgroup_event { |
79bd9814 | 235 | /* |
59b6f873 | 236 | * memcg which the event belongs to. |
79bd9814 | 237 | */ |
59b6f873 | 238 | struct mem_cgroup *memcg; |
79bd9814 TH |
239 | /* |
240 | * eventfd to signal userspace about the event. | |
241 | */ | |
242 | struct eventfd_ctx *eventfd; | |
243 | /* | |
244 | * Each of these stored in a list by the cgroup. | |
245 | */ | |
246 | struct list_head list; | |
fba94807 TH |
247 | /* |
248 | * register_event() callback will be used to add new userspace | |
249 | * waiter for changes related to this event. Use eventfd_signal() | |
250 | * on eventfd to send notification to userspace. | |
251 | */ | |
59b6f873 | 252 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 253 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
254 | /* |
255 | * unregister_event() callback will be called when userspace closes | |
256 | * the eventfd or on cgroup removing. This callback must be set, | |
257 | * if you want provide notification functionality. | |
258 | */ | |
59b6f873 | 259 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 260 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
261 | /* |
262 | * All fields below needed to unregister event when | |
263 | * userspace closes eventfd. | |
264 | */ | |
265 | poll_table pt; | |
266 | wait_queue_head_t *wqh; | |
267 | wait_queue_t wait; | |
268 | struct work_struct remove; | |
269 | }; | |
270 | ||
c0ff4b85 R |
271 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
272 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 273 | |
8cdea7c0 BS |
274 | /* |
275 | * The memory controller data structure. The memory controller controls both | |
276 | * page cache and RSS per cgroup. We would eventually like to provide | |
277 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
278 | * to help the administrator determine what knobs to tune. | |
279 | * | |
280 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
281 | * we hit the water mark. May be even add a low water mark, such that |
282 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
283 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
284 | */ |
285 | struct mem_cgroup { | |
286 | struct cgroup_subsys_state css; | |
3e32cb2e JW |
287 | |
288 | /* Accounted resources */ | |
289 | struct page_counter memory; | |
290 | struct page_counter memsw; | |
291 | struct page_counter kmem; | |
292 | ||
293 | unsigned long soft_limit; | |
59927fb9 | 294 | |
70ddf637 AV |
295 | /* vmpressure notifications */ |
296 | struct vmpressure vmpressure; | |
297 | ||
2f7dd7a4 JW |
298 | /* css_online() has been completed */ |
299 | int initialized; | |
300 | ||
18f59ea7 BS |
301 | /* |
302 | * Should the accounting and control be hierarchical, per subtree? | |
303 | */ | |
304 | bool use_hierarchy; | |
510fc4e1 | 305 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
306 | |
307 | bool oom_lock; | |
308 | atomic_t under_oom; | |
3812c8c8 | 309 | atomic_t oom_wakeups; |
79dfdacc | 310 | |
1f4c025b | 311 | int swappiness; |
3c11ecf4 KH |
312 | /* OOM-Killer disable */ |
313 | int oom_kill_disable; | |
a7885eb8 | 314 | |
2e72b634 KS |
315 | /* protect arrays of thresholds */ |
316 | struct mutex thresholds_lock; | |
317 | ||
318 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 319 | struct mem_cgroup_thresholds thresholds; |
907860ed | 320 | |
2e72b634 | 321 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 322 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 323 | |
9490ff27 KH |
324 | /* For oom notifier event fd */ |
325 | struct list_head oom_notify; | |
185efc0f | 326 | |
7dc74be0 DN |
327 | /* |
328 | * Should we move charges of a task when a task is moved into this | |
329 | * mem_cgroup ? And what type of charges should we move ? | |
330 | */ | |
f894ffa8 | 331 | unsigned long move_charge_at_immigrate; |
619d094b KH |
332 | /* |
333 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
334 | */ | |
335 | atomic_t moving_account; | |
312734c0 KH |
336 | /* taken only while moving_account > 0 */ |
337 | spinlock_t move_lock; | |
d52aa412 | 338 | /* |
c62b1a3b | 339 | * percpu counter. |
d52aa412 | 340 | */ |
3a7951b4 | 341 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
342 | /* |
343 | * used when a cpu is offlined or other synchronizations | |
344 | * See mem_cgroup_read_stat(). | |
345 | */ | |
346 | struct mem_cgroup_stat_cpu nocpu_base; | |
347 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 348 | |
5f578161 | 349 | atomic_t dead_count; |
4bd2c1ee | 350 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
2e685cad | 351 | struct cg_proto tcp_mem; |
d1a4c0b3 | 352 | #endif |
2633d7a0 | 353 | #if defined(CONFIG_MEMCG_KMEM) |
bd673145 VD |
354 | /* analogous to slab_common's slab_caches list, but per-memcg; |
355 | * protected by memcg_slab_mutex */ | |
2633d7a0 | 356 | struct list_head memcg_slab_caches; |
2633d7a0 GC |
357 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ |
358 | int kmemcg_id; | |
359 | #endif | |
45cf7ebd GC |
360 | |
361 | int last_scanned_node; | |
362 | #if MAX_NUMNODES > 1 | |
363 | nodemask_t scan_nodes; | |
364 | atomic_t numainfo_events; | |
365 | atomic_t numainfo_updating; | |
366 | #endif | |
70ddf637 | 367 | |
fba94807 TH |
368 | /* List of events which userspace want to receive */ |
369 | struct list_head event_list; | |
370 | spinlock_t event_list_lock; | |
371 | ||
54f72fe0 JW |
372 | struct mem_cgroup_per_node *nodeinfo[0]; |
373 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
374 | }; |
375 | ||
510fc4e1 GC |
376 | /* internal only representation about the status of kmem accounting. */ |
377 | enum { | |
6de64beb | 378 | KMEM_ACCOUNTED_ACTIVE, /* accounted by this cgroup itself */ |
7de37682 | 379 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
380 | }; |
381 | ||
510fc4e1 GC |
382 | #ifdef CONFIG_MEMCG_KMEM |
383 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
384 | { | |
385 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
386 | } | |
7de37682 GC |
387 | |
388 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
389 | { | |
390 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
391 | } | |
392 | ||
393 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) | |
394 | { | |
10d5ebf4 LZ |
395 | /* |
396 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
397 | * will call css_put() if it sees the memcg is dead. | |
398 | */ | |
399 | smp_wmb(); | |
7de37682 GC |
400 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
401 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
402 | } | |
403 | ||
404 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
405 | { | |
406 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
407 | &memcg->kmem_account_flags); | |
408 | } | |
510fc4e1 GC |
409 | #endif |
410 | ||
7dc74be0 DN |
411 | /* Stuffs for move charges at task migration. */ |
412 | /* | |
ee5e8472 GC |
413 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
414 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
415 | */ |
416 | enum move_type { | |
4ffef5fe | 417 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 418 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
419 | NR_MOVE_TYPE, |
420 | }; | |
421 | ||
4ffef5fe DN |
422 | /* "mc" and its members are protected by cgroup_mutex */ |
423 | static struct move_charge_struct { | |
b1dd693e | 424 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
425 | struct mem_cgroup *from; |
426 | struct mem_cgroup *to; | |
ee5e8472 | 427 | unsigned long immigrate_flags; |
4ffef5fe | 428 | unsigned long precharge; |
854ffa8d | 429 | unsigned long moved_charge; |
483c30b5 | 430 | unsigned long moved_swap; |
8033b97c DN |
431 | struct task_struct *moving_task; /* a task moving charges */ |
432 | wait_queue_head_t waitq; /* a waitq for other context */ | |
433 | } mc = { | |
2bd9bb20 | 434 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
435 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
436 | }; | |
4ffef5fe | 437 | |
90254a65 DN |
438 | static bool move_anon(void) |
439 | { | |
ee5e8472 | 440 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
441 | } |
442 | ||
87946a72 DN |
443 | static bool move_file(void) |
444 | { | |
ee5e8472 | 445 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
446 | } |
447 | ||
4e416953 BS |
448 | /* |
449 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
450 | * limit reclaim to prevent infinite loops, if they ever occur. | |
451 | */ | |
a0db00fc | 452 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 453 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 454 | |
217bc319 KH |
455 | enum charge_type { |
456 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 457 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 458 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 459 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
460 | NR_CHARGE_TYPE, |
461 | }; | |
462 | ||
8c7c6e34 | 463 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
464 | enum res_type { |
465 | _MEM, | |
466 | _MEMSWAP, | |
467 | _OOM_TYPE, | |
510fc4e1 | 468 | _KMEM, |
86ae53e1 GC |
469 | }; |
470 | ||
a0db00fc KS |
471 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
472 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 473 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
474 | /* Used for OOM nofiier */ |
475 | #define OOM_CONTROL (0) | |
8c7c6e34 | 476 | |
0999821b GC |
477 | /* |
478 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
479 | * As a consequence, any change that needs to protect against new child cgroups | |
480 | * appearing has to hold it as well. | |
481 | */ | |
482 | static DEFINE_MUTEX(memcg_create_mutex); | |
483 | ||
b2145145 WL |
484 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
485 | { | |
a7c6d554 | 486 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
487 | } |
488 | ||
70ddf637 AV |
489 | /* Some nice accessors for the vmpressure. */ |
490 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
491 | { | |
492 | if (!memcg) | |
493 | memcg = root_mem_cgroup; | |
494 | return &memcg->vmpressure; | |
495 | } | |
496 | ||
497 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
498 | { | |
499 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
500 | } | |
501 | ||
7ffc0edc MH |
502 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
503 | { | |
504 | return (memcg == root_mem_cgroup); | |
505 | } | |
506 | ||
4219b2da LZ |
507 | /* |
508 | * We restrict the id in the range of [1, 65535], so it can fit into | |
509 | * an unsigned short. | |
510 | */ | |
511 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
512 | ||
34c00c31 LZ |
513 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
514 | { | |
15a4c835 | 515 | return memcg->css.id; |
34c00c31 LZ |
516 | } |
517 | ||
518 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
519 | { | |
520 | struct cgroup_subsys_state *css; | |
521 | ||
7d699ddb | 522 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
523 | return mem_cgroup_from_css(css); |
524 | } | |
525 | ||
e1aab161 | 526 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 527 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 528 | |
e1aab161 GC |
529 | void sock_update_memcg(struct sock *sk) |
530 | { | |
376be5ff | 531 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 532 | struct mem_cgroup *memcg; |
3f134619 | 533 | struct cg_proto *cg_proto; |
e1aab161 GC |
534 | |
535 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
536 | ||
f3f511e1 GC |
537 | /* Socket cloning can throw us here with sk_cgrp already |
538 | * filled. It won't however, necessarily happen from | |
539 | * process context. So the test for root memcg given | |
540 | * the current task's memcg won't help us in this case. | |
541 | * | |
542 | * Respecting the original socket's memcg is a better | |
543 | * decision in this case. | |
544 | */ | |
545 | if (sk->sk_cgrp) { | |
546 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 547 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
548 | return; |
549 | } | |
550 | ||
e1aab161 GC |
551 | rcu_read_lock(); |
552 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 553 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 554 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
555 | memcg_proto_active(cg_proto) && |
556 | css_tryget_online(&memcg->css)) { | |
3f134619 | 557 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
558 | } |
559 | rcu_read_unlock(); | |
560 | } | |
561 | } | |
562 | EXPORT_SYMBOL(sock_update_memcg); | |
563 | ||
564 | void sock_release_memcg(struct sock *sk) | |
565 | { | |
376be5ff | 566 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
567 | struct mem_cgroup *memcg; |
568 | WARN_ON(!sk->sk_cgrp->memcg); | |
569 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 570 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
571 | } |
572 | } | |
d1a4c0b3 GC |
573 | |
574 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
575 | { | |
576 | if (!memcg || mem_cgroup_is_root(memcg)) | |
577 | return NULL; | |
578 | ||
2e685cad | 579 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
580 | } |
581 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 582 | |
3f134619 GC |
583 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
584 | { | |
2e685cad | 585 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
586 | return; |
587 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
588 | } | |
589 | #else | |
590 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
591 | { | |
592 | } | |
593 | #endif | |
594 | ||
a8964b9b | 595 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
596 | /* |
597 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
b8627835 LZ |
598 | * The main reason for not using cgroup id for this: |
599 | * this works better in sparse environments, where we have a lot of memcgs, | |
600 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
601 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
602 | * 200 entry array for that. | |
55007d84 GC |
603 | * |
604 | * The current size of the caches array is stored in | |
605 | * memcg_limited_groups_array_size. It will double each time we have to | |
606 | * increase it. | |
607 | */ | |
608 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
609 | int memcg_limited_groups_array_size; |
610 | ||
55007d84 GC |
611 | /* |
612 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
613 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
614 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
615 | * tunable, but that is strictly not necessary. | |
616 | * | |
b8627835 | 617 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
618 | * this constant directly from cgroup, but it is understandable that this is |
619 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 620 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
621 | * increase ours as well if it increases. |
622 | */ | |
623 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 624 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 625 | |
d7f25f8a GC |
626 | /* |
627 | * A lot of the calls to the cache allocation functions are expected to be | |
628 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
629 | * conditional to this static branch, we'll have to allow modules that does | |
630 | * kmem_cache_alloc and the such to see this symbol as well | |
631 | */ | |
a8964b9b | 632 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 633 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 634 | |
f3bb3043 VD |
635 | static void memcg_free_cache_id(int id); |
636 | ||
a8964b9b GC |
637 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
638 | { | |
55007d84 | 639 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 640 | static_key_slow_dec(&memcg_kmem_enabled_key); |
f3bb3043 | 641 | memcg_free_cache_id(memcg->kmemcg_id); |
55007d84 | 642 | } |
bea207c8 GC |
643 | /* |
644 | * This check can't live in kmem destruction function, | |
645 | * since the charges will outlive the cgroup | |
646 | */ | |
3e32cb2e | 647 | WARN_ON(page_counter_read(&memcg->kmem)); |
a8964b9b GC |
648 | } |
649 | #else | |
650 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
651 | { | |
652 | } | |
653 | #endif /* CONFIG_MEMCG_KMEM */ | |
654 | ||
655 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
656 | { | |
657 | disarm_sock_keys(memcg); | |
658 | disarm_kmem_keys(memcg); | |
659 | } | |
660 | ||
c0ff4b85 | 661 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 662 | |
f64c3f54 | 663 | static struct mem_cgroup_per_zone * |
e231875b | 664 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 665 | { |
e231875b JZ |
666 | int nid = zone_to_nid(zone); |
667 | int zid = zone_idx(zone); | |
668 | ||
54f72fe0 | 669 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
670 | } |
671 | ||
c0ff4b85 | 672 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 673 | { |
c0ff4b85 | 674 | return &memcg->css; |
d324236b WF |
675 | } |
676 | ||
f64c3f54 | 677 | static struct mem_cgroup_per_zone * |
e231875b | 678 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 679 | { |
97a6c37b JW |
680 | int nid = page_to_nid(page); |
681 | int zid = page_zonenum(page); | |
f64c3f54 | 682 | |
e231875b | 683 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
684 | } |
685 | ||
bb4cc1a8 AM |
686 | static struct mem_cgroup_tree_per_zone * |
687 | soft_limit_tree_node_zone(int nid, int zid) | |
688 | { | |
689 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
690 | } | |
691 | ||
692 | static struct mem_cgroup_tree_per_zone * | |
693 | soft_limit_tree_from_page(struct page *page) | |
694 | { | |
695 | int nid = page_to_nid(page); | |
696 | int zid = page_zonenum(page); | |
697 | ||
698 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
699 | } | |
700 | ||
cf2c8127 JW |
701 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
702 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 703 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
704 | { |
705 | struct rb_node **p = &mctz->rb_root.rb_node; | |
706 | struct rb_node *parent = NULL; | |
707 | struct mem_cgroup_per_zone *mz_node; | |
708 | ||
709 | if (mz->on_tree) | |
710 | return; | |
711 | ||
712 | mz->usage_in_excess = new_usage_in_excess; | |
713 | if (!mz->usage_in_excess) | |
714 | return; | |
715 | while (*p) { | |
716 | parent = *p; | |
717 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
718 | tree_node); | |
719 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
720 | p = &(*p)->rb_left; | |
721 | /* | |
722 | * We can't avoid mem cgroups that are over their soft | |
723 | * limit by the same amount | |
724 | */ | |
725 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
726 | p = &(*p)->rb_right; | |
727 | } | |
728 | rb_link_node(&mz->tree_node, parent, p); | |
729 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
730 | mz->on_tree = true; | |
731 | } | |
732 | ||
cf2c8127 JW |
733 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
734 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
735 | { |
736 | if (!mz->on_tree) | |
737 | return; | |
738 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
739 | mz->on_tree = false; | |
740 | } | |
741 | ||
cf2c8127 JW |
742 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
743 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 744 | { |
0a31bc97 JW |
745 | unsigned long flags; |
746 | ||
747 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 748 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 749 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
750 | } |
751 | ||
3e32cb2e JW |
752 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
753 | { | |
754 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
755 | unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit); | |
756 | unsigned long excess = 0; | |
757 | ||
758 | if (nr_pages > soft_limit) | |
759 | excess = nr_pages - soft_limit; | |
760 | ||
761 | return excess; | |
762 | } | |
bb4cc1a8 AM |
763 | |
764 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
765 | { | |
3e32cb2e | 766 | unsigned long excess; |
bb4cc1a8 AM |
767 | struct mem_cgroup_per_zone *mz; |
768 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 769 | |
e231875b | 770 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
771 | /* |
772 | * Necessary to update all ancestors when hierarchy is used. | |
773 | * because their event counter is not touched. | |
774 | */ | |
775 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 776 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 777 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
778 | /* |
779 | * We have to update the tree if mz is on RB-tree or | |
780 | * mem is over its softlimit. | |
781 | */ | |
782 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
783 | unsigned long flags; |
784 | ||
785 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
786 | /* if on-tree, remove it */ |
787 | if (mz->on_tree) | |
cf2c8127 | 788 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
789 | /* |
790 | * Insert again. mz->usage_in_excess will be updated. | |
791 | * If excess is 0, no tree ops. | |
792 | */ | |
cf2c8127 | 793 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 794 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
795 | } |
796 | } | |
797 | } | |
798 | ||
799 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
800 | { | |
bb4cc1a8 | 801 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
802 | struct mem_cgroup_per_zone *mz; |
803 | int nid, zid; | |
bb4cc1a8 | 804 | |
e231875b JZ |
805 | for_each_node(nid) { |
806 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
807 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
808 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 809 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
810 | } |
811 | } | |
812 | } | |
813 | ||
814 | static struct mem_cgroup_per_zone * | |
815 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
816 | { | |
817 | struct rb_node *rightmost = NULL; | |
818 | struct mem_cgroup_per_zone *mz; | |
819 | ||
820 | retry: | |
821 | mz = NULL; | |
822 | rightmost = rb_last(&mctz->rb_root); | |
823 | if (!rightmost) | |
824 | goto done; /* Nothing to reclaim from */ | |
825 | ||
826 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
827 | /* | |
828 | * Remove the node now but someone else can add it back, | |
829 | * we will to add it back at the end of reclaim to its correct | |
830 | * position in the tree. | |
831 | */ | |
cf2c8127 | 832 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 833 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 834 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
835 | goto retry; |
836 | done: | |
837 | return mz; | |
838 | } | |
839 | ||
840 | static struct mem_cgroup_per_zone * | |
841 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
842 | { | |
843 | struct mem_cgroup_per_zone *mz; | |
844 | ||
0a31bc97 | 845 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 846 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 847 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
848 | return mz; |
849 | } | |
850 | ||
711d3d2c KH |
851 | /* |
852 | * Implementation Note: reading percpu statistics for memcg. | |
853 | * | |
854 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
855 | * synchronization to implement "quick" read. There are trade-off between | |
856 | * reading cost and precision of value. Then, we may have a chance to implement | |
857 | * a periodic synchronizion of counter in memcg's counter. | |
858 | * | |
859 | * But this _read() function is used for user interface now. The user accounts | |
860 | * memory usage by memory cgroup and he _always_ requires exact value because | |
861 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
862 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
863 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
864 | * | |
865 | * If there are kernel internal actions which can make use of some not-exact | |
866 | * value, and reading all cpu value can be performance bottleneck in some | |
867 | * common workload, threashold and synchonization as vmstat[] should be | |
868 | * implemented. | |
869 | */ | |
c0ff4b85 | 870 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 871 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 872 | { |
7a159cc9 | 873 | long val = 0; |
c62b1a3b | 874 | int cpu; |
c62b1a3b | 875 | |
711d3d2c KH |
876 | get_online_cpus(); |
877 | for_each_online_cpu(cpu) | |
c0ff4b85 | 878 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 879 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
880 | spin_lock(&memcg->pcp_counter_lock); |
881 | val += memcg->nocpu_base.count[idx]; | |
882 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
883 | #endif |
884 | put_online_cpus(); | |
c62b1a3b KH |
885 | return val; |
886 | } | |
887 | ||
c0ff4b85 | 888 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
889 | enum mem_cgroup_events_index idx) |
890 | { | |
891 | unsigned long val = 0; | |
892 | int cpu; | |
893 | ||
9c567512 | 894 | get_online_cpus(); |
e9f8974f | 895 | for_each_online_cpu(cpu) |
c0ff4b85 | 896 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 897 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
898 | spin_lock(&memcg->pcp_counter_lock); |
899 | val += memcg->nocpu_base.events[idx]; | |
900 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 901 | #endif |
9c567512 | 902 | put_online_cpus(); |
e9f8974f JW |
903 | return val; |
904 | } | |
905 | ||
c0ff4b85 | 906 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 907 | struct page *page, |
0a31bc97 | 908 | int nr_pages) |
d52aa412 | 909 | { |
b2402857 KH |
910 | /* |
911 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
912 | * counted as CACHE even if it's on ANON LRU. | |
913 | */ | |
0a31bc97 | 914 | if (PageAnon(page)) |
b2402857 | 915 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 916 | nr_pages); |
d52aa412 | 917 | else |
b2402857 | 918 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 919 | nr_pages); |
55e462b0 | 920 | |
b070e65c DR |
921 | if (PageTransHuge(page)) |
922 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
923 | nr_pages); | |
924 | ||
e401f176 KH |
925 | /* pagein of a big page is an event. So, ignore page size */ |
926 | if (nr_pages > 0) | |
c0ff4b85 | 927 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 928 | else { |
c0ff4b85 | 929 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
930 | nr_pages = -nr_pages; /* for event */ |
931 | } | |
e401f176 | 932 | |
13114716 | 933 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
934 | } |
935 | ||
e231875b | 936 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
937 | { |
938 | struct mem_cgroup_per_zone *mz; | |
939 | ||
940 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
941 | return mz->lru_size[lru]; | |
942 | } | |
943 | ||
e231875b JZ |
944 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
945 | int nid, | |
946 | unsigned int lru_mask) | |
bb2a0de9 | 947 | { |
e231875b | 948 | unsigned long nr = 0; |
889976db YH |
949 | int zid; |
950 | ||
e231875b | 951 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 952 | |
e231875b JZ |
953 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
954 | struct mem_cgroup_per_zone *mz; | |
955 | enum lru_list lru; | |
956 | ||
957 | for_each_lru(lru) { | |
958 | if (!(BIT(lru) & lru_mask)) | |
959 | continue; | |
960 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
961 | nr += mz->lru_size[lru]; | |
962 | } | |
963 | } | |
964 | return nr; | |
889976db | 965 | } |
bb2a0de9 | 966 | |
c0ff4b85 | 967 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 968 | unsigned int lru_mask) |
6d12e2d8 | 969 | { |
e231875b | 970 | unsigned long nr = 0; |
889976db | 971 | int nid; |
6d12e2d8 | 972 | |
31aaea4a | 973 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
974 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
975 | return nr; | |
d52aa412 KH |
976 | } |
977 | ||
f53d7ce3 JW |
978 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
979 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
980 | { |
981 | unsigned long val, next; | |
982 | ||
13114716 | 983 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 984 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 985 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
986 | if ((long)next - (long)val < 0) { |
987 | switch (target) { | |
988 | case MEM_CGROUP_TARGET_THRESH: | |
989 | next = val + THRESHOLDS_EVENTS_TARGET; | |
990 | break; | |
bb4cc1a8 AM |
991 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
992 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
993 | break; | |
f53d7ce3 JW |
994 | case MEM_CGROUP_TARGET_NUMAINFO: |
995 | next = val + NUMAINFO_EVENTS_TARGET; | |
996 | break; | |
997 | default: | |
998 | break; | |
999 | } | |
1000 | __this_cpu_write(memcg->stat->targets[target], next); | |
1001 | return true; | |
7a159cc9 | 1002 | } |
f53d7ce3 | 1003 | return false; |
d2265e6f KH |
1004 | } |
1005 | ||
1006 | /* | |
1007 | * Check events in order. | |
1008 | * | |
1009 | */ | |
c0ff4b85 | 1010 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
1011 | { |
1012 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
1013 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1014 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1015 | bool do_softlimit; |
82b3f2a7 | 1016 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1017 | |
bb4cc1a8 AM |
1018 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1019 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1020 | #if MAX_NUMNODES > 1 |
1021 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1022 | MEM_CGROUP_TARGET_NUMAINFO); | |
1023 | #endif | |
c0ff4b85 | 1024 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1025 | if (unlikely(do_softlimit)) |
1026 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1027 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1028 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1029 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1030 | #endif |
0a31bc97 | 1031 | } |
d2265e6f KH |
1032 | } |
1033 | ||
cf475ad2 | 1034 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1035 | { |
31a78f23 BS |
1036 | /* |
1037 | * mm_update_next_owner() may clear mm->owner to NULL | |
1038 | * if it races with swapoff, page migration, etc. | |
1039 | * So this can be called with p == NULL. | |
1040 | */ | |
1041 | if (unlikely(!p)) | |
1042 | return NULL; | |
1043 | ||
073219e9 | 1044 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1045 | } |
1046 | ||
df381975 | 1047 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1048 | { |
c0ff4b85 | 1049 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 1050 | |
54595fe2 KH |
1051 | rcu_read_lock(); |
1052 | do { | |
6f6acb00 MH |
1053 | /* |
1054 | * Page cache insertions can happen withou an | |
1055 | * actual mm context, e.g. during disk probing | |
1056 | * on boot, loopback IO, acct() writes etc. | |
1057 | */ | |
1058 | if (unlikely(!mm)) | |
df381975 | 1059 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1060 | else { |
1061 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1062 | if (unlikely(!memcg)) | |
1063 | memcg = root_mem_cgroup; | |
1064 | } | |
ec903c0c | 1065 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1066 | rcu_read_unlock(); |
c0ff4b85 | 1067 | return memcg; |
54595fe2 KH |
1068 | } |
1069 | ||
16248d8f MH |
1070 | /* |
1071 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1072 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1073 | * | |
1074 | * helper function to be used by mem_cgroup_iter | |
1075 | */ | |
1076 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
694fbc0f | 1077 | struct mem_cgroup *last_visited) |
16248d8f | 1078 | { |
492eb21b | 1079 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 1080 | |
bd8815a6 | 1081 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 1082 | skip_node: |
492eb21b | 1083 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
1084 | |
1085 | /* | |
1086 | * Even if we found a group we have to make sure it is | |
1087 | * alive. css && !memcg means that the groups should be | |
1088 | * skipped and we should continue the tree walk. | |
1089 | * last_visited css is safe to use because it is | |
1090 | * protected by css_get and the tree walk is rcu safe. | |
0eef6156 MH |
1091 | * |
1092 | * We do not take a reference on the root of the tree walk | |
1093 | * because we might race with the root removal when it would | |
1094 | * be the only node in the iterated hierarchy and mem_cgroup_iter | |
1095 | * would end up in an endless loop because it expects that at | |
1096 | * least one valid node will be returned. Root cannot disappear | |
1097 | * because caller of the iterator should hold it already so | |
1098 | * skipping css reference should be safe. | |
16248d8f | 1099 | */ |
492eb21b | 1100 | if (next_css) { |
2f7dd7a4 JW |
1101 | struct mem_cgroup *memcg = mem_cgroup_from_css(next_css); |
1102 | ||
1103 | if (next_css == &root->css) | |
1104 | return memcg; | |
1105 | ||
1106 | if (css_tryget_online(next_css)) { | |
1107 | /* | |
1108 | * Make sure the memcg is initialized: | |
1109 | * mem_cgroup_css_online() orders the the | |
1110 | * initialization against setting the flag. | |
1111 | */ | |
1112 | if (smp_load_acquire(&memcg->initialized)) | |
1113 | return memcg; | |
1114 | css_put(next_css); | |
1115 | } | |
0eef6156 MH |
1116 | |
1117 | prev_css = next_css; | |
1118 | goto skip_node; | |
16248d8f MH |
1119 | } |
1120 | ||
1121 | return NULL; | |
1122 | } | |
1123 | ||
519ebea3 JW |
1124 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1125 | { | |
1126 | /* | |
1127 | * When a group in the hierarchy below root is destroyed, the | |
1128 | * hierarchy iterator can no longer be trusted since it might | |
1129 | * have pointed to the destroyed group. Invalidate it. | |
1130 | */ | |
1131 | atomic_inc(&root->dead_count); | |
1132 | } | |
1133 | ||
1134 | static struct mem_cgroup * | |
1135 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1136 | struct mem_cgroup *root, | |
1137 | int *sequence) | |
1138 | { | |
1139 | struct mem_cgroup *position = NULL; | |
1140 | /* | |
1141 | * A cgroup destruction happens in two stages: offlining and | |
1142 | * release. They are separated by a RCU grace period. | |
1143 | * | |
1144 | * If the iterator is valid, we may still race with an | |
1145 | * offlining. The RCU lock ensures the object won't be | |
1146 | * released, tryget will fail if we lost the race. | |
1147 | */ | |
1148 | *sequence = atomic_read(&root->dead_count); | |
1149 | if (iter->last_dead_count == *sequence) { | |
1150 | smp_rmb(); | |
1151 | position = iter->last_visited; | |
ecc736fc MH |
1152 | |
1153 | /* | |
1154 | * We cannot take a reference to root because we might race | |
1155 | * with root removal and returning NULL would end up in | |
1156 | * an endless loop on the iterator user level when root | |
1157 | * would be returned all the time. | |
1158 | */ | |
1159 | if (position && position != root && | |
ec903c0c | 1160 | !css_tryget_online(&position->css)) |
519ebea3 JW |
1161 | position = NULL; |
1162 | } | |
1163 | return position; | |
1164 | } | |
1165 | ||
1166 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1167 | struct mem_cgroup *last_visited, | |
1168 | struct mem_cgroup *new_position, | |
ecc736fc | 1169 | struct mem_cgroup *root, |
519ebea3 JW |
1170 | int sequence) |
1171 | { | |
ecc736fc MH |
1172 | /* root reference counting symmetric to mem_cgroup_iter_load */ |
1173 | if (last_visited && last_visited != root) | |
519ebea3 JW |
1174 | css_put(&last_visited->css); |
1175 | /* | |
1176 | * We store the sequence count from the time @last_visited was | |
1177 | * loaded successfully instead of rereading it here so that we | |
1178 | * don't lose destruction events in between. We could have | |
1179 | * raced with the destruction of @new_position after all. | |
1180 | */ | |
1181 | iter->last_visited = new_position; | |
1182 | smp_wmb(); | |
1183 | iter->last_dead_count = sequence; | |
1184 | } | |
1185 | ||
5660048c JW |
1186 | /** |
1187 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1188 | * @root: hierarchy root | |
1189 | * @prev: previously returned memcg, NULL on first invocation | |
1190 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1191 | * | |
1192 | * Returns references to children of the hierarchy below @root, or | |
1193 | * @root itself, or %NULL after a full round-trip. | |
1194 | * | |
1195 | * Caller must pass the return value in @prev on subsequent | |
1196 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1197 | * to cancel a hierarchy walk before the round-trip is complete. | |
1198 | * | |
1199 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1200 | * divide up the memcgs in the hierarchy among all concurrent | |
1201 | * reclaimers operating on the same zone and priority. | |
1202 | */ | |
694fbc0f | 1203 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1204 | struct mem_cgroup *prev, |
694fbc0f | 1205 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1206 | { |
9f3a0d09 | 1207 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1208 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1209 | |
694fbc0f AM |
1210 | if (mem_cgroup_disabled()) |
1211 | return NULL; | |
5660048c | 1212 | |
9f3a0d09 JW |
1213 | if (!root) |
1214 | root = root_mem_cgroup; | |
7d74b06f | 1215 | |
9f3a0d09 | 1216 | if (prev && !reclaim) |
542f85f9 | 1217 | last_visited = prev; |
14067bb3 | 1218 | |
9f3a0d09 JW |
1219 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1220 | if (prev) | |
c40046f3 | 1221 | goto out_css_put; |
694fbc0f | 1222 | return root; |
9f3a0d09 | 1223 | } |
14067bb3 | 1224 | |
542f85f9 | 1225 | rcu_read_lock(); |
9f3a0d09 | 1226 | while (!memcg) { |
527a5ec9 | 1227 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1228 | int uninitialized_var(seq); |
711d3d2c | 1229 | |
527a5ec9 | 1230 | if (reclaim) { |
527a5ec9 JW |
1231 | struct mem_cgroup_per_zone *mz; |
1232 | ||
e231875b | 1233 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); |
527a5ec9 | 1234 | iter = &mz->reclaim_iter[reclaim->priority]; |
542f85f9 | 1235 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1236 | iter->last_visited = NULL; |
542f85f9 MH |
1237 | goto out_unlock; |
1238 | } | |
5f578161 | 1239 | |
519ebea3 | 1240 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1241 | } |
7d74b06f | 1242 | |
694fbc0f | 1243 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1244 | |
527a5ec9 | 1245 | if (reclaim) { |
ecc736fc MH |
1246 | mem_cgroup_iter_update(iter, last_visited, memcg, root, |
1247 | seq); | |
542f85f9 | 1248 | |
19f39402 | 1249 | if (!memcg) |
527a5ec9 JW |
1250 | iter->generation++; |
1251 | else if (!prev && memcg) | |
1252 | reclaim->generation = iter->generation; | |
1253 | } | |
9f3a0d09 | 1254 | |
694fbc0f | 1255 | if (prev && !memcg) |
542f85f9 | 1256 | goto out_unlock; |
9f3a0d09 | 1257 | } |
542f85f9 MH |
1258 | out_unlock: |
1259 | rcu_read_unlock(); | |
c40046f3 MH |
1260 | out_css_put: |
1261 | if (prev && prev != root) | |
1262 | css_put(&prev->css); | |
1263 | ||
9f3a0d09 | 1264 | return memcg; |
14067bb3 | 1265 | } |
7d74b06f | 1266 | |
5660048c JW |
1267 | /** |
1268 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1269 | * @root: hierarchy root | |
1270 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1271 | */ | |
1272 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1273 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1274 | { |
1275 | if (!root) | |
1276 | root = root_mem_cgroup; | |
1277 | if (prev && prev != root) | |
1278 | css_put(&prev->css); | |
1279 | } | |
7d74b06f | 1280 | |
9f3a0d09 JW |
1281 | /* |
1282 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1283 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1284 | * be used for reference counting. | |
1285 | */ | |
1286 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1287 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1288 | iter != NULL; \ |
527a5ec9 | 1289 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1290 | |
9f3a0d09 | 1291 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1292 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1293 | iter != NULL; \ |
527a5ec9 | 1294 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1295 | |
68ae564b | 1296 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1297 | { |
c0ff4b85 | 1298 | struct mem_cgroup *memcg; |
456f998e | 1299 | |
456f998e | 1300 | rcu_read_lock(); |
c0ff4b85 R |
1301 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1302 | if (unlikely(!memcg)) | |
456f998e YH |
1303 | goto out; |
1304 | ||
1305 | switch (idx) { | |
456f998e | 1306 | case PGFAULT: |
0e574a93 JW |
1307 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1308 | break; | |
1309 | case PGMAJFAULT: | |
1310 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1311 | break; |
1312 | default: | |
1313 | BUG(); | |
1314 | } | |
1315 | out: | |
1316 | rcu_read_unlock(); | |
1317 | } | |
68ae564b | 1318 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1319 | |
925b7673 JW |
1320 | /** |
1321 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1322 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1323 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1324 | * |
1325 | * Returns the lru list vector holding pages for the given @zone and | |
1326 | * @mem. This can be the global zone lruvec, if the memory controller | |
1327 | * is disabled. | |
1328 | */ | |
1329 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1330 | struct mem_cgroup *memcg) | |
1331 | { | |
1332 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1333 | struct lruvec *lruvec; |
925b7673 | 1334 | |
bea8c150 HD |
1335 | if (mem_cgroup_disabled()) { |
1336 | lruvec = &zone->lruvec; | |
1337 | goto out; | |
1338 | } | |
925b7673 | 1339 | |
e231875b | 1340 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1341 | lruvec = &mz->lruvec; |
1342 | out: | |
1343 | /* | |
1344 | * Since a node can be onlined after the mem_cgroup was created, | |
1345 | * we have to be prepared to initialize lruvec->zone here; | |
1346 | * and if offlined then reonlined, we need to reinitialize it. | |
1347 | */ | |
1348 | if (unlikely(lruvec->zone != zone)) | |
1349 | lruvec->zone = zone; | |
1350 | return lruvec; | |
925b7673 JW |
1351 | } |
1352 | ||
925b7673 | 1353 | /** |
fa9add64 | 1354 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1355 | * @page: the page |
fa9add64 | 1356 | * @zone: zone of the page |
925b7673 | 1357 | */ |
fa9add64 | 1358 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1359 | { |
08e552c6 | 1360 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1361 | struct mem_cgroup *memcg; |
1362 | struct page_cgroup *pc; | |
bea8c150 | 1363 | struct lruvec *lruvec; |
6d12e2d8 | 1364 | |
bea8c150 HD |
1365 | if (mem_cgroup_disabled()) { |
1366 | lruvec = &zone->lruvec; | |
1367 | goto out; | |
1368 | } | |
925b7673 | 1369 | |
08e552c6 | 1370 | pc = lookup_page_cgroup(page); |
38c5d72f | 1371 | memcg = pc->mem_cgroup; |
7512102c HD |
1372 | |
1373 | /* | |
fa9add64 | 1374 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1375 | * an uncharged page off lru does nothing to secure |
1376 | * its former mem_cgroup from sudden removal. | |
1377 | * | |
1378 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1379 | * under page_cgroup lock: between them, they make all uses | |
1380 | * of pc->mem_cgroup safe. | |
1381 | */ | |
fa9add64 | 1382 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1383 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1384 | ||
e231875b | 1385 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1386 | lruvec = &mz->lruvec; |
1387 | out: | |
1388 | /* | |
1389 | * Since a node can be onlined after the mem_cgroup was created, | |
1390 | * we have to be prepared to initialize lruvec->zone here; | |
1391 | * and if offlined then reonlined, we need to reinitialize it. | |
1392 | */ | |
1393 | if (unlikely(lruvec->zone != zone)) | |
1394 | lruvec->zone = zone; | |
1395 | return lruvec; | |
08e552c6 | 1396 | } |
b69408e8 | 1397 | |
925b7673 | 1398 | /** |
fa9add64 HD |
1399 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1400 | * @lruvec: mem_cgroup per zone lru vector | |
1401 | * @lru: index of lru list the page is sitting on | |
1402 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1403 | * |
fa9add64 HD |
1404 | * This function must be called when a page is added to or removed from an |
1405 | * lru list. | |
3f58a829 | 1406 | */ |
fa9add64 HD |
1407 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1408 | int nr_pages) | |
3f58a829 MK |
1409 | { |
1410 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1411 | unsigned long *lru_size; |
3f58a829 MK |
1412 | |
1413 | if (mem_cgroup_disabled()) | |
1414 | return; | |
1415 | ||
fa9add64 HD |
1416 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1417 | lru_size = mz->lru_size + lru; | |
1418 | *lru_size += nr_pages; | |
1419 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1420 | } |
544122e5 | 1421 | |
3e92041d | 1422 | /* |
c0ff4b85 | 1423 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1424 | * hierarchy subtree |
1425 | */ | |
c3ac9a8a JW |
1426 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1427 | struct mem_cgroup *memcg) | |
3e92041d | 1428 | { |
91c63734 JW |
1429 | if (root_memcg == memcg) |
1430 | return true; | |
3a981f48 | 1431 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1432 | return false; |
b47f77b5 | 1433 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1434 | } |
1435 | ||
1436 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1437 | struct mem_cgroup *memcg) | |
1438 | { | |
1439 | bool ret; | |
1440 | ||
91c63734 | 1441 | rcu_read_lock(); |
c3ac9a8a | 1442 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1443 | rcu_read_unlock(); |
1444 | return ret; | |
3e92041d MH |
1445 | } |
1446 | ||
ffbdccf5 DR |
1447 | bool task_in_mem_cgroup(struct task_struct *task, |
1448 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1449 | { |
0b7f569e | 1450 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1451 | struct task_struct *p; |
ffbdccf5 | 1452 | bool ret; |
4c4a2214 | 1453 | |
158e0a2d | 1454 | p = find_lock_task_mm(task); |
de077d22 | 1455 | if (p) { |
df381975 | 1456 | curr = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1457 | task_unlock(p); |
1458 | } else { | |
1459 | /* | |
1460 | * All threads may have already detached their mm's, but the oom | |
1461 | * killer still needs to detect if they have already been oom | |
1462 | * killed to prevent needlessly killing additional tasks. | |
1463 | */ | |
ffbdccf5 | 1464 | rcu_read_lock(); |
de077d22 DR |
1465 | curr = mem_cgroup_from_task(task); |
1466 | if (curr) | |
1467 | css_get(&curr->css); | |
ffbdccf5 | 1468 | rcu_read_unlock(); |
de077d22 | 1469 | } |
d31f56db | 1470 | /* |
c0ff4b85 | 1471 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1472 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1473 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1474 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1475 | */ |
c0ff4b85 | 1476 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1477 | css_put(&curr->css); |
4c4a2214 DR |
1478 | return ret; |
1479 | } | |
1480 | ||
c56d5c7d | 1481 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1482 | { |
9b272977 | 1483 | unsigned long inactive_ratio; |
14797e23 | 1484 | unsigned long inactive; |
9b272977 | 1485 | unsigned long active; |
c772be93 | 1486 | unsigned long gb; |
14797e23 | 1487 | |
4d7dcca2 HD |
1488 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1489 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1490 | |
c772be93 KM |
1491 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1492 | if (gb) | |
1493 | inactive_ratio = int_sqrt(10 * gb); | |
1494 | else | |
1495 | inactive_ratio = 1; | |
1496 | ||
9b272977 | 1497 | return inactive * inactive_ratio < active; |
14797e23 KM |
1498 | } |
1499 | ||
3e32cb2e | 1500 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1501 | container_of(counter, struct mem_cgroup, member) |
1502 | ||
19942822 | 1503 | /** |
9d11ea9f | 1504 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1505 | * @memcg: the memory cgroup |
19942822 | 1506 | * |
9d11ea9f | 1507 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1508 | * pages. |
19942822 | 1509 | */ |
c0ff4b85 | 1510 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1511 | { |
3e32cb2e JW |
1512 | unsigned long margin = 0; |
1513 | unsigned long count; | |
1514 | unsigned long limit; | |
9d11ea9f | 1515 | |
3e32cb2e JW |
1516 | count = page_counter_read(&memcg->memory); |
1517 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1518 | if (count < limit) | |
1519 | margin = limit - count; | |
1520 | ||
1521 | if (do_swap_account) { | |
1522 | count = page_counter_read(&memcg->memsw); | |
1523 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1524 | if (count <= limit) | |
1525 | margin = min(margin, limit - count); | |
1526 | } | |
1527 | ||
1528 | return margin; | |
19942822 JW |
1529 | } |
1530 | ||
1f4c025b | 1531 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1532 | { |
a7885eb8 | 1533 | /* root ? */ |
14208b0e | 1534 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1535 | return vm_swappiness; |
1536 | ||
bf1ff263 | 1537 | return memcg->swappiness; |
a7885eb8 KM |
1538 | } |
1539 | ||
619d094b KH |
1540 | /* |
1541 | * memcg->moving_account is used for checking possibility that some thread is | |
1542 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1543 | * a memcg, other threads should check memcg->moving_account under | |
1544 | * rcu_read_lock(), like this: | |
1545 | * | |
1546 | * CPU-A CPU-B | |
1547 | * rcu_read_lock() | |
1548 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1549 | * take heavy locks. | |
1550 | * synchronize_rcu() update something. | |
1551 | * rcu_read_unlock() | |
1552 | * start move here. | |
1553 | */ | |
4331f7d3 | 1554 | |
c0ff4b85 | 1555 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1556 | { |
619d094b | 1557 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1558 | synchronize_rcu(); |
1559 | } | |
1560 | ||
c0ff4b85 | 1561 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1562 | { |
619d094b KH |
1563 | /* |
1564 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1565 | * We check NULL in callee rather than caller. | |
1566 | */ | |
d7365e78 | 1567 | if (memcg) |
619d094b | 1568 | atomic_dec(&memcg->moving_account); |
32047e2a | 1569 | } |
619d094b | 1570 | |
32047e2a | 1571 | /* |
bdcbb659 | 1572 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1573 | * |
bdcbb659 QH |
1574 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1575 | * moving cgroups. This is for waiting at high-memory pressure | |
1576 | * caused by "move". | |
32047e2a | 1577 | */ |
c0ff4b85 | 1578 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1579 | { |
2bd9bb20 KH |
1580 | struct mem_cgroup *from; |
1581 | struct mem_cgroup *to; | |
4b534334 | 1582 | bool ret = false; |
2bd9bb20 KH |
1583 | /* |
1584 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1585 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1586 | */ | |
1587 | spin_lock(&mc.lock); | |
1588 | from = mc.from; | |
1589 | to = mc.to; | |
1590 | if (!from) | |
1591 | goto unlock; | |
3e92041d | 1592 | |
c0ff4b85 R |
1593 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1594 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1595 | unlock: |
1596 | spin_unlock(&mc.lock); | |
4b534334 KH |
1597 | return ret; |
1598 | } | |
1599 | ||
c0ff4b85 | 1600 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1601 | { |
1602 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1603 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1604 | DEFINE_WAIT(wait); |
1605 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1606 | /* moving charge context might have finished. */ | |
1607 | if (mc.moving_task) | |
1608 | schedule(); | |
1609 | finish_wait(&mc.waitq, &wait); | |
1610 | return true; | |
1611 | } | |
1612 | } | |
1613 | return false; | |
1614 | } | |
1615 | ||
312734c0 KH |
1616 | /* |
1617 | * Take this lock when | |
1618 | * - a code tries to modify page's memcg while it's USED. | |
1619 | * - a code tries to modify page state accounting in a memcg. | |
312734c0 KH |
1620 | */ |
1621 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1622 | unsigned long *flags) | |
1623 | { | |
1624 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1625 | } | |
1626 | ||
1627 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1628 | unsigned long *flags) | |
1629 | { | |
1630 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1631 | } | |
1632 | ||
58cf188e | 1633 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1634 | /** |
58cf188e | 1635 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1636 | * @memcg: The memory cgroup that went over limit |
1637 | * @p: Task that is going to be killed | |
1638 | * | |
1639 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1640 | * enabled | |
1641 | */ | |
1642 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1643 | { | |
e61734c5 | 1644 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1645 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1646 | struct mem_cgroup *iter; |
1647 | unsigned int i; | |
e222432b | 1648 | |
58cf188e | 1649 | if (!p) |
e222432b BS |
1650 | return; |
1651 | ||
08088cb9 | 1652 | mutex_lock(&oom_info_lock); |
e222432b BS |
1653 | rcu_read_lock(); |
1654 | ||
e61734c5 TH |
1655 | pr_info("Task in "); |
1656 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1657 | pr_info(" killed as a result of limit of "); | |
1658 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1659 | pr_info("\n"); | |
e222432b | 1660 | |
e222432b BS |
1661 | rcu_read_unlock(); |
1662 | ||
3e32cb2e JW |
1663 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1664 | K((u64)page_counter_read(&memcg->memory)), | |
1665 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1666 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1667 | K((u64)page_counter_read(&memcg->memsw)), | |
1668 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1669 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1670 | K((u64)page_counter_read(&memcg->kmem)), | |
1671 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1672 | |
1673 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1674 | pr_info("Memory cgroup stats for "); |
1675 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1676 | pr_cont(":"); |
1677 | ||
1678 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1679 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1680 | continue; | |
1681 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1682 | K(mem_cgroup_read_stat(iter, i))); | |
1683 | } | |
1684 | ||
1685 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1686 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1687 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1688 | ||
1689 | pr_cont("\n"); | |
1690 | } | |
08088cb9 | 1691 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1692 | } |
1693 | ||
81d39c20 KH |
1694 | /* |
1695 | * This function returns the number of memcg under hierarchy tree. Returns | |
1696 | * 1(self count) if no children. | |
1697 | */ | |
c0ff4b85 | 1698 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1699 | { |
1700 | int num = 0; | |
7d74b06f KH |
1701 | struct mem_cgroup *iter; |
1702 | ||
c0ff4b85 | 1703 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1704 | num++; |
81d39c20 KH |
1705 | return num; |
1706 | } | |
1707 | ||
a63d83f4 DR |
1708 | /* |
1709 | * Return the memory (and swap, if configured) limit for a memcg. | |
1710 | */ | |
3e32cb2e | 1711 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1712 | { |
3e32cb2e | 1713 | unsigned long limit; |
a63d83f4 | 1714 | |
3e32cb2e | 1715 | limit = memcg->memory.limit; |
9a5a8f19 | 1716 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1717 | unsigned long memsw_limit; |
9a5a8f19 | 1718 | |
3e32cb2e JW |
1719 | memsw_limit = memcg->memsw.limit; |
1720 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1721 | } |
9a5a8f19 | 1722 | return limit; |
a63d83f4 DR |
1723 | } |
1724 | ||
19965460 DR |
1725 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1726 | int order) | |
9cbb78bb DR |
1727 | { |
1728 | struct mem_cgroup *iter; | |
1729 | unsigned long chosen_points = 0; | |
1730 | unsigned long totalpages; | |
1731 | unsigned int points = 0; | |
1732 | struct task_struct *chosen = NULL; | |
1733 | ||
876aafbf | 1734 | /* |
465adcf1 DR |
1735 | * If current has a pending SIGKILL or is exiting, then automatically |
1736 | * select it. The goal is to allow it to allocate so that it may | |
1737 | * quickly exit and free its memory. | |
876aafbf | 1738 | */ |
465adcf1 | 1739 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1740 | set_thread_flag(TIF_MEMDIE); |
1741 | return; | |
1742 | } | |
1743 | ||
1744 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1745 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1746 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1747 | struct css_task_iter it; |
9cbb78bb DR |
1748 | struct task_struct *task; |
1749 | ||
72ec7029 TH |
1750 | css_task_iter_start(&iter->css, &it); |
1751 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1752 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1753 | false)) { | |
1754 | case OOM_SCAN_SELECT: | |
1755 | if (chosen) | |
1756 | put_task_struct(chosen); | |
1757 | chosen = task; | |
1758 | chosen_points = ULONG_MAX; | |
1759 | get_task_struct(chosen); | |
1760 | /* fall through */ | |
1761 | case OOM_SCAN_CONTINUE: | |
1762 | continue; | |
1763 | case OOM_SCAN_ABORT: | |
72ec7029 | 1764 | css_task_iter_end(&it); |
9cbb78bb DR |
1765 | mem_cgroup_iter_break(memcg, iter); |
1766 | if (chosen) | |
1767 | put_task_struct(chosen); | |
1768 | return; | |
1769 | case OOM_SCAN_OK: | |
1770 | break; | |
1771 | }; | |
1772 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1773 | if (!points || points < chosen_points) |
1774 | continue; | |
1775 | /* Prefer thread group leaders for display purposes */ | |
1776 | if (points == chosen_points && | |
1777 | thread_group_leader(chosen)) | |
1778 | continue; | |
1779 | ||
1780 | if (chosen) | |
1781 | put_task_struct(chosen); | |
1782 | chosen = task; | |
1783 | chosen_points = points; | |
1784 | get_task_struct(chosen); | |
9cbb78bb | 1785 | } |
72ec7029 | 1786 | css_task_iter_end(&it); |
9cbb78bb DR |
1787 | } |
1788 | ||
1789 | if (!chosen) | |
1790 | return; | |
1791 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1792 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1793 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1794 | } |
1795 | ||
4d0c066d KH |
1796 | /** |
1797 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1798 | * @memcg: the target memcg |
4d0c066d KH |
1799 | * @nid: the node ID to be checked. |
1800 | * @noswap : specify true here if the user wants flle only information. | |
1801 | * | |
1802 | * This function returns whether the specified memcg contains any | |
1803 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1804 | * pages in the node. | |
1805 | */ | |
c0ff4b85 | 1806 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1807 | int nid, bool noswap) |
1808 | { | |
c0ff4b85 | 1809 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1810 | return true; |
1811 | if (noswap || !total_swap_pages) | |
1812 | return false; | |
c0ff4b85 | 1813 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1814 | return true; |
1815 | return false; | |
1816 | ||
1817 | } | |
bb4cc1a8 | 1818 | #if MAX_NUMNODES > 1 |
889976db YH |
1819 | |
1820 | /* | |
1821 | * Always updating the nodemask is not very good - even if we have an empty | |
1822 | * list or the wrong list here, we can start from some node and traverse all | |
1823 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1824 | * | |
1825 | */ | |
c0ff4b85 | 1826 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1827 | { |
1828 | int nid; | |
453a9bf3 KH |
1829 | /* |
1830 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1831 | * pagein/pageout changes since the last update. | |
1832 | */ | |
c0ff4b85 | 1833 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1834 | return; |
c0ff4b85 | 1835 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1836 | return; |
1837 | ||
889976db | 1838 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1839 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1840 | |
31aaea4a | 1841 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1842 | |
c0ff4b85 R |
1843 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1844 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1845 | } |
453a9bf3 | 1846 | |
c0ff4b85 R |
1847 | atomic_set(&memcg->numainfo_events, 0); |
1848 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1849 | } |
1850 | ||
1851 | /* | |
1852 | * Selecting a node where we start reclaim from. Because what we need is just | |
1853 | * reducing usage counter, start from anywhere is O,K. Considering | |
1854 | * memory reclaim from current node, there are pros. and cons. | |
1855 | * | |
1856 | * Freeing memory from current node means freeing memory from a node which | |
1857 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1858 | * hit limits, it will see a contention on a node. But freeing from remote | |
1859 | * node means more costs for memory reclaim because of memory latency. | |
1860 | * | |
1861 | * Now, we use round-robin. Better algorithm is welcomed. | |
1862 | */ | |
c0ff4b85 | 1863 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1864 | { |
1865 | int node; | |
1866 | ||
c0ff4b85 R |
1867 | mem_cgroup_may_update_nodemask(memcg); |
1868 | node = memcg->last_scanned_node; | |
889976db | 1869 | |
c0ff4b85 | 1870 | node = next_node(node, memcg->scan_nodes); |
889976db | 1871 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1872 | node = first_node(memcg->scan_nodes); |
889976db YH |
1873 | /* |
1874 | * We call this when we hit limit, not when pages are added to LRU. | |
1875 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1876 | * memcg is too small and all pages are not on LRU. In that case, | |
1877 | * we use curret node. | |
1878 | */ | |
1879 | if (unlikely(node == MAX_NUMNODES)) | |
1880 | node = numa_node_id(); | |
1881 | ||
c0ff4b85 | 1882 | memcg->last_scanned_node = node; |
889976db YH |
1883 | return node; |
1884 | } | |
1885 | ||
bb4cc1a8 AM |
1886 | /* |
1887 | * Check all nodes whether it contains reclaimable pages or not. | |
1888 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1889 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1890 | * enough new information. We need to do double check. | |
1891 | */ | |
1892 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1893 | { | |
1894 | int nid; | |
1895 | ||
1896 | /* | |
1897 | * quick check...making use of scan_node. | |
1898 | * We can skip unused nodes. | |
1899 | */ | |
1900 | if (!nodes_empty(memcg->scan_nodes)) { | |
1901 | for (nid = first_node(memcg->scan_nodes); | |
1902 | nid < MAX_NUMNODES; | |
1903 | nid = next_node(nid, memcg->scan_nodes)) { | |
1904 | ||
1905 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1906 | return true; | |
1907 | } | |
1908 | } | |
1909 | /* | |
1910 | * Check rest of nodes. | |
1911 | */ | |
1912 | for_each_node_state(nid, N_MEMORY) { | |
1913 | if (node_isset(nid, memcg->scan_nodes)) | |
1914 | continue; | |
1915 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1916 | return true; | |
1917 | } | |
1918 | return false; | |
1919 | } | |
1920 | ||
889976db | 1921 | #else |
c0ff4b85 | 1922 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1923 | { |
1924 | return 0; | |
1925 | } | |
4d0c066d | 1926 | |
bb4cc1a8 AM |
1927 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
1928 | { | |
1929 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
1930 | } | |
889976db YH |
1931 | #endif |
1932 | ||
0608f43d AM |
1933 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1934 | struct zone *zone, | |
1935 | gfp_t gfp_mask, | |
1936 | unsigned long *total_scanned) | |
1937 | { | |
1938 | struct mem_cgroup *victim = NULL; | |
1939 | int total = 0; | |
1940 | int loop = 0; | |
1941 | unsigned long excess; | |
1942 | unsigned long nr_scanned; | |
1943 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1944 | .zone = zone, | |
1945 | .priority = 0, | |
1946 | }; | |
1947 | ||
3e32cb2e | 1948 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1949 | |
1950 | while (1) { | |
1951 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1952 | if (!victim) { | |
1953 | loop++; | |
1954 | if (loop >= 2) { | |
1955 | /* | |
1956 | * If we have not been able to reclaim | |
1957 | * anything, it might because there are | |
1958 | * no reclaimable pages under this hierarchy | |
1959 | */ | |
1960 | if (!total) | |
1961 | break; | |
1962 | /* | |
1963 | * We want to do more targeted reclaim. | |
1964 | * excess >> 2 is not to excessive so as to | |
1965 | * reclaim too much, nor too less that we keep | |
1966 | * coming back to reclaim from this cgroup | |
1967 | */ | |
1968 | if (total >= (excess >> 2) || | |
1969 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1970 | break; | |
1971 | } | |
1972 | continue; | |
1973 | } | |
1974 | if (!mem_cgroup_reclaimable(victim, false)) | |
1975 | continue; | |
1976 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
1977 | zone, &nr_scanned); | |
1978 | *total_scanned += nr_scanned; | |
3e32cb2e | 1979 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1980 | break; |
6d61ef40 | 1981 | } |
0608f43d AM |
1982 | mem_cgroup_iter_break(root_memcg, victim); |
1983 | return total; | |
6d61ef40 BS |
1984 | } |
1985 | ||
0056f4e6 JW |
1986 | #ifdef CONFIG_LOCKDEP |
1987 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1988 | .name = "memcg_oom_lock", | |
1989 | }; | |
1990 | #endif | |
1991 | ||
fb2a6fc5 JW |
1992 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1993 | ||
867578cb KH |
1994 | /* |
1995 | * Check OOM-Killer is already running under our hierarchy. | |
1996 | * If someone is running, return false. | |
1997 | */ | |
fb2a6fc5 | 1998 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1999 | { |
79dfdacc | 2000 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2001 | |
fb2a6fc5 JW |
2002 | spin_lock(&memcg_oom_lock); |
2003 | ||
9f3a0d09 | 2004 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2005 | if (iter->oom_lock) { |
79dfdacc MH |
2006 | /* |
2007 | * this subtree of our hierarchy is already locked | |
2008 | * so we cannot give a lock. | |
2009 | */ | |
79dfdacc | 2010 | failed = iter; |
9f3a0d09 JW |
2011 | mem_cgroup_iter_break(memcg, iter); |
2012 | break; | |
23751be0 JW |
2013 | } else |
2014 | iter->oom_lock = true; | |
7d74b06f | 2015 | } |
867578cb | 2016 | |
fb2a6fc5 JW |
2017 | if (failed) { |
2018 | /* | |
2019 | * OK, we failed to lock the whole subtree so we have | |
2020 | * to clean up what we set up to the failing subtree | |
2021 | */ | |
2022 | for_each_mem_cgroup_tree(iter, memcg) { | |
2023 | if (iter == failed) { | |
2024 | mem_cgroup_iter_break(memcg, iter); | |
2025 | break; | |
2026 | } | |
2027 | iter->oom_lock = false; | |
79dfdacc | 2028 | } |
0056f4e6 JW |
2029 | } else |
2030 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2031 | |
2032 | spin_unlock(&memcg_oom_lock); | |
2033 | ||
2034 | return !failed; | |
a636b327 | 2035 | } |
0b7f569e | 2036 | |
fb2a6fc5 | 2037 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2038 | { |
7d74b06f KH |
2039 | struct mem_cgroup *iter; |
2040 | ||
fb2a6fc5 | 2041 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 2042 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 2043 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2044 | iter->oom_lock = false; |
fb2a6fc5 | 2045 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2046 | } |
2047 | ||
c0ff4b85 | 2048 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2049 | { |
2050 | struct mem_cgroup *iter; | |
2051 | ||
c0ff4b85 | 2052 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2053 | atomic_inc(&iter->under_oom); |
2054 | } | |
2055 | ||
c0ff4b85 | 2056 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2057 | { |
2058 | struct mem_cgroup *iter; | |
2059 | ||
867578cb KH |
2060 | /* |
2061 | * When a new child is created while the hierarchy is under oom, | |
2062 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2063 | * atomic_add_unless() here. | |
2064 | */ | |
c0ff4b85 | 2065 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2066 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2067 | } |
2068 | ||
867578cb KH |
2069 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2070 | ||
dc98df5a | 2071 | struct oom_wait_info { |
d79154bb | 2072 | struct mem_cgroup *memcg; |
dc98df5a KH |
2073 | wait_queue_t wait; |
2074 | }; | |
2075 | ||
2076 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2077 | unsigned mode, int sync, void *arg) | |
2078 | { | |
d79154bb HD |
2079 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2080 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2081 | struct oom_wait_info *oom_wait_info; |
2082 | ||
2083 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2084 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2085 | |
dc98df5a | 2086 | /* |
d79154bb | 2087 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2088 | * Then we can use css_is_ancestor without taking care of RCU. |
2089 | */ | |
c0ff4b85 R |
2090 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2091 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2092 | return 0; |
dc98df5a KH |
2093 | return autoremove_wake_function(wait, mode, sync, arg); |
2094 | } | |
2095 | ||
c0ff4b85 | 2096 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2097 | { |
3812c8c8 | 2098 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2099 | /* for filtering, pass "memcg" as argument. */ |
2100 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2101 | } |
2102 | ||
c0ff4b85 | 2103 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2104 | { |
c0ff4b85 R |
2105 | if (memcg && atomic_read(&memcg->under_oom)) |
2106 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2107 | } |
2108 | ||
3812c8c8 | 2109 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2110 | { |
3812c8c8 JW |
2111 | if (!current->memcg_oom.may_oom) |
2112 | return; | |
867578cb | 2113 | /* |
49426420 JW |
2114 | * We are in the middle of the charge context here, so we |
2115 | * don't want to block when potentially sitting on a callstack | |
2116 | * that holds all kinds of filesystem and mm locks. | |
2117 | * | |
2118 | * Also, the caller may handle a failed allocation gracefully | |
2119 | * (like optional page cache readahead) and so an OOM killer | |
2120 | * invocation might not even be necessary. | |
2121 | * | |
2122 | * That's why we don't do anything here except remember the | |
2123 | * OOM context and then deal with it at the end of the page | |
2124 | * fault when the stack is unwound, the locks are released, | |
2125 | * and when we know whether the fault was overall successful. | |
867578cb | 2126 | */ |
49426420 JW |
2127 | css_get(&memcg->css); |
2128 | current->memcg_oom.memcg = memcg; | |
2129 | current->memcg_oom.gfp_mask = mask; | |
2130 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2131 | } |
2132 | ||
2133 | /** | |
2134 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2135 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2136 | * |
49426420 JW |
2137 | * This has to be called at the end of a page fault if the memcg OOM |
2138 | * handler was enabled. | |
3812c8c8 | 2139 | * |
49426420 | 2140 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2141 | * sleep on a waitqueue until the userspace task resolves the |
2142 | * situation. Sleeping directly in the charge context with all kinds | |
2143 | * of locks held is not a good idea, instead we remember an OOM state | |
2144 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2145 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2146 | * |
2147 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2148 | * completed, %false otherwise. |
3812c8c8 | 2149 | */ |
49426420 | 2150 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2151 | { |
49426420 | 2152 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2153 | struct oom_wait_info owait; |
49426420 | 2154 | bool locked; |
3812c8c8 JW |
2155 | |
2156 | /* OOM is global, do not handle */ | |
3812c8c8 | 2157 | if (!memcg) |
49426420 | 2158 | return false; |
3812c8c8 | 2159 | |
49426420 JW |
2160 | if (!handle) |
2161 | goto cleanup; | |
3812c8c8 JW |
2162 | |
2163 | owait.memcg = memcg; | |
2164 | owait.wait.flags = 0; | |
2165 | owait.wait.func = memcg_oom_wake_function; | |
2166 | owait.wait.private = current; | |
2167 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2168 | |
3812c8c8 | 2169 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2170 | mem_cgroup_mark_under_oom(memcg); |
2171 | ||
2172 | locked = mem_cgroup_oom_trylock(memcg); | |
2173 | ||
2174 | if (locked) | |
2175 | mem_cgroup_oom_notify(memcg); | |
2176 | ||
2177 | if (locked && !memcg->oom_kill_disable) { | |
2178 | mem_cgroup_unmark_under_oom(memcg); | |
2179 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2180 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2181 | current->memcg_oom.order); | |
2182 | } else { | |
3812c8c8 | 2183 | schedule(); |
49426420 JW |
2184 | mem_cgroup_unmark_under_oom(memcg); |
2185 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2186 | } | |
2187 | ||
2188 | if (locked) { | |
fb2a6fc5 JW |
2189 | mem_cgroup_oom_unlock(memcg); |
2190 | /* | |
2191 | * There is no guarantee that an OOM-lock contender | |
2192 | * sees the wakeups triggered by the OOM kill | |
2193 | * uncharges. Wake any sleepers explicitely. | |
2194 | */ | |
2195 | memcg_oom_recover(memcg); | |
2196 | } | |
49426420 JW |
2197 | cleanup: |
2198 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2199 | css_put(&memcg->css); |
867578cb | 2200 | return true; |
0b7f569e KH |
2201 | } |
2202 | ||
d7365e78 JW |
2203 | /** |
2204 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
2205 | * @page: page that is going to change accounted state | |
2206 | * @locked: &memcg->move_lock slowpath was taken | |
2207 | * @flags: IRQ-state flags for &memcg->move_lock | |
32047e2a | 2208 | * |
d7365e78 JW |
2209 | * This function must mark the beginning of an accounted page state |
2210 | * change to prevent double accounting when the page is concurrently | |
2211 | * being moved to another memcg: | |
32047e2a | 2212 | * |
d7365e78 JW |
2213 | * memcg = mem_cgroup_begin_page_stat(page, &locked, &flags); |
2214 | * if (TestClearPageState(page)) | |
2215 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
2216 | * mem_cgroup_end_page_stat(memcg, locked, flags); | |
32047e2a | 2217 | * |
d7365e78 JW |
2218 | * The RCU lock is held throughout the transaction. The fast path can |
2219 | * get away without acquiring the memcg->move_lock (@locked is false) | |
2220 | * because page moving starts with an RCU grace period. | |
32047e2a | 2221 | * |
d7365e78 JW |
2222 | * The RCU lock also protects the memcg from being freed when the page |
2223 | * state that is going to change is the only thing preventing the page | |
2224 | * from being uncharged. E.g. end-writeback clearing PageWriteback(), | |
2225 | * which allows migration to go ahead and uncharge the page before the | |
2226 | * account transaction might be complete. | |
d69b042f | 2227 | */ |
d7365e78 JW |
2228 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page, |
2229 | bool *locked, | |
2230 | unsigned long *flags) | |
89c06bd5 KH |
2231 | { |
2232 | struct mem_cgroup *memcg; | |
2233 | struct page_cgroup *pc; | |
2234 | ||
d7365e78 JW |
2235 | rcu_read_lock(); |
2236 | ||
2237 | if (mem_cgroup_disabled()) | |
2238 | return NULL; | |
2239 | ||
89c06bd5 KH |
2240 | pc = lookup_page_cgroup(page); |
2241 | again: | |
2242 | memcg = pc->mem_cgroup; | |
2243 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
d7365e78 JW |
2244 | return NULL; |
2245 | ||
2246 | *locked = false; | |
bdcbb659 | 2247 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2248 | return memcg; |
89c06bd5 KH |
2249 | |
2250 | move_lock_mem_cgroup(memcg, flags); | |
2251 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2252 | move_unlock_mem_cgroup(memcg, flags); | |
2253 | goto again; | |
2254 | } | |
2255 | *locked = true; | |
d7365e78 JW |
2256 | |
2257 | return memcg; | |
89c06bd5 KH |
2258 | } |
2259 | ||
d7365e78 JW |
2260 | /** |
2261 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2262 | * @memcg: the memcg that was accounted against | |
2263 | * @locked: value received from mem_cgroup_begin_page_stat() | |
2264 | * @flags: value received from mem_cgroup_begin_page_stat() | |
2265 | */ | |
2266 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg, bool locked, | |
2267 | unsigned long flags) | |
89c06bd5 | 2268 | { |
d7365e78 JW |
2269 | if (memcg && locked) |
2270 | move_unlock_mem_cgroup(memcg, &flags); | |
89c06bd5 | 2271 | |
d7365e78 | 2272 | rcu_read_unlock(); |
89c06bd5 KH |
2273 | } |
2274 | ||
d7365e78 JW |
2275 | /** |
2276 | * mem_cgroup_update_page_stat - update page state statistics | |
2277 | * @memcg: memcg to account against | |
2278 | * @idx: page state item to account | |
2279 | * @val: number of pages (positive or negative) | |
2280 | * | |
2281 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2282 | */ | |
2283 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2284 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2285 | { |
658b72c5 | 2286 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2287 | |
d7365e78 JW |
2288 | if (memcg) |
2289 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2290 | } |
26174efd | 2291 | |
cdec2e42 KH |
2292 | /* |
2293 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2294 | * TODO: maybe necessary to use big numbers in big irons. | |
2295 | */ | |
7ec99d62 | 2296 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2297 | struct memcg_stock_pcp { |
2298 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2299 | unsigned int nr_pages; |
cdec2e42 | 2300 | struct work_struct work; |
26fe6168 | 2301 | unsigned long flags; |
a0db00fc | 2302 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2303 | }; |
2304 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2305 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2306 | |
a0956d54 SS |
2307 | /** |
2308 | * consume_stock: Try to consume stocked charge on this cpu. | |
2309 | * @memcg: memcg to consume from. | |
2310 | * @nr_pages: how many pages to charge. | |
2311 | * | |
2312 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2313 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2314 | * service an allocation will refill the stock. | |
2315 | * | |
2316 | * returns true if successful, false otherwise. | |
cdec2e42 | 2317 | */ |
a0956d54 | 2318 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2319 | { |
2320 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2321 | bool ret = false; |
cdec2e42 | 2322 | |
a0956d54 | 2323 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2324 | return ret; |
a0956d54 | 2325 | |
cdec2e42 | 2326 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2327 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2328 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2329 | ret = true; |
2330 | } | |
cdec2e42 KH |
2331 | put_cpu_var(memcg_stock); |
2332 | return ret; | |
2333 | } | |
2334 | ||
2335 | /* | |
3e32cb2e | 2336 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2337 | */ |
2338 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2339 | { | |
2340 | struct mem_cgroup *old = stock->cached; | |
2341 | ||
11c9ea4e | 2342 | if (stock->nr_pages) { |
3e32cb2e | 2343 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2344 | if (do_swap_account) |
3e32cb2e | 2345 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2346 | stock->nr_pages = 0; |
cdec2e42 KH |
2347 | } |
2348 | stock->cached = NULL; | |
cdec2e42 KH |
2349 | } |
2350 | ||
2351 | /* | |
2352 | * This must be called under preempt disabled or must be called by | |
2353 | * a thread which is pinned to local cpu. | |
2354 | */ | |
2355 | static void drain_local_stock(struct work_struct *dummy) | |
2356 | { | |
7c8e0181 | 2357 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2358 | drain_stock(stock); |
26fe6168 | 2359 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2360 | } |
2361 | ||
e4777496 MH |
2362 | static void __init memcg_stock_init(void) |
2363 | { | |
2364 | int cpu; | |
2365 | ||
2366 | for_each_possible_cpu(cpu) { | |
2367 | struct memcg_stock_pcp *stock = | |
2368 | &per_cpu(memcg_stock, cpu); | |
2369 | INIT_WORK(&stock->work, drain_local_stock); | |
2370 | } | |
2371 | } | |
2372 | ||
cdec2e42 | 2373 | /* |
3e32cb2e | 2374 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2375 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2376 | */ |
c0ff4b85 | 2377 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2378 | { |
2379 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2380 | ||
c0ff4b85 | 2381 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2382 | drain_stock(stock); |
c0ff4b85 | 2383 | stock->cached = memcg; |
cdec2e42 | 2384 | } |
11c9ea4e | 2385 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2386 | put_cpu_var(memcg_stock); |
2387 | } | |
2388 | ||
2389 | /* | |
c0ff4b85 | 2390 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2391 | * of the hierarchy under it. sync flag says whether we should block |
2392 | * until the work is done. | |
cdec2e42 | 2393 | */ |
c0ff4b85 | 2394 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2395 | { |
26fe6168 | 2396 | int cpu, curcpu; |
d38144b7 | 2397 | |
cdec2e42 | 2398 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2399 | get_online_cpus(); |
5af12d0e | 2400 | curcpu = get_cpu(); |
cdec2e42 KH |
2401 | for_each_online_cpu(cpu) { |
2402 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2403 | struct mem_cgroup *memcg; |
26fe6168 | 2404 | |
c0ff4b85 R |
2405 | memcg = stock->cached; |
2406 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2407 | continue; |
c0ff4b85 | 2408 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2409 | continue; |
d1a05b69 MH |
2410 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2411 | if (cpu == curcpu) | |
2412 | drain_local_stock(&stock->work); | |
2413 | else | |
2414 | schedule_work_on(cpu, &stock->work); | |
2415 | } | |
cdec2e42 | 2416 | } |
5af12d0e | 2417 | put_cpu(); |
d38144b7 MH |
2418 | |
2419 | if (!sync) | |
2420 | goto out; | |
2421 | ||
2422 | for_each_online_cpu(cpu) { | |
2423 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2424 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2425 | flush_work(&stock->work); |
2426 | } | |
2427 | out: | |
f894ffa8 | 2428 | put_online_cpus(); |
d38144b7 MH |
2429 | } |
2430 | ||
2431 | /* | |
2432 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2433 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
3e32cb2e | 2434 | * expects some charges will be back later but cannot wait for it. |
d38144b7 | 2435 | */ |
c0ff4b85 | 2436 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2437 | { |
9f50fad6 MH |
2438 | /* |
2439 | * If someone calls draining, avoid adding more kworker runs. | |
2440 | */ | |
2441 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2442 | return; | |
c0ff4b85 | 2443 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2444 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2445 | } |
2446 | ||
2447 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2448 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2449 | { |
2450 | /* called when force_empty is called */ | |
9f50fad6 | 2451 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2452 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2453 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2454 | } |
2455 | ||
711d3d2c KH |
2456 | /* |
2457 | * This function drains percpu counter value from DEAD cpu and | |
2458 | * move it to local cpu. Note that this function can be preempted. | |
2459 | */ | |
c0ff4b85 | 2460 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2461 | { |
2462 | int i; | |
2463 | ||
c0ff4b85 | 2464 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2465 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2466 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2467 | |
c0ff4b85 R |
2468 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2469 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2470 | } |
e9f8974f | 2471 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2472 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2473 | |
c0ff4b85 R |
2474 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2475 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2476 | } |
c0ff4b85 | 2477 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2478 | } |
2479 | ||
0db0628d | 2480 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2481 | unsigned long action, |
2482 | void *hcpu) | |
2483 | { | |
2484 | int cpu = (unsigned long)hcpu; | |
2485 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2486 | struct mem_cgroup *iter; |
cdec2e42 | 2487 | |
619d094b | 2488 | if (action == CPU_ONLINE) |
1489ebad | 2489 | return NOTIFY_OK; |
1489ebad | 2490 | |
d833049b | 2491 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2492 | return NOTIFY_OK; |
711d3d2c | 2493 | |
9f3a0d09 | 2494 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2495 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2496 | ||
cdec2e42 KH |
2497 | stock = &per_cpu(memcg_stock, cpu); |
2498 | drain_stock(stock); | |
2499 | return NOTIFY_OK; | |
2500 | } | |
2501 | ||
00501b53 JW |
2502 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2503 | unsigned int nr_pages) | |
8a9f3ccd | 2504 | { |
7ec99d62 | 2505 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2506 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2507 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2508 | struct page_counter *counter; |
6539cc05 | 2509 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2510 | bool may_swap = true; |
2511 | bool drained = false; | |
05b84301 | 2512 | int ret = 0; |
a636b327 | 2513 | |
ce00a967 JW |
2514 | if (mem_cgroup_is_root(memcg)) |
2515 | goto done; | |
6539cc05 | 2516 | retry: |
b6b6cc72 MH |
2517 | if (consume_stock(memcg, nr_pages)) |
2518 | goto done; | |
8a9f3ccd | 2519 | |
3fbe7244 | 2520 | if (!do_swap_account || |
3e32cb2e JW |
2521 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2522 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2523 | goto done_restock; |
3fbe7244 | 2524 | if (do_swap_account) |
3e32cb2e JW |
2525 | page_counter_uncharge(&memcg->memsw, batch); |
2526 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2527 | } else { |
3e32cb2e | 2528 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2529 | may_swap = false; |
3fbe7244 | 2530 | } |
7a81b88c | 2531 | |
6539cc05 JW |
2532 | if (batch > nr_pages) { |
2533 | batch = nr_pages; | |
2534 | goto retry; | |
2535 | } | |
6d61ef40 | 2536 | |
06b078fc JW |
2537 | /* |
2538 | * Unlike in global OOM situations, memcg is not in a physical | |
2539 | * memory shortage. Allow dying and OOM-killed tasks to | |
2540 | * bypass the last charges so that they can exit quickly and | |
2541 | * free their memory. | |
2542 | */ | |
2543 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2544 | fatal_signal_pending(current) || | |
2545 | current->flags & PF_EXITING)) | |
2546 | goto bypass; | |
2547 | ||
2548 | if (unlikely(task_in_memcg_oom(current))) | |
2549 | goto nomem; | |
2550 | ||
6539cc05 JW |
2551 | if (!(gfp_mask & __GFP_WAIT)) |
2552 | goto nomem; | |
4b534334 | 2553 | |
b70a2a21 JW |
2554 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2555 | gfp_mask, may_swap); | |
6539cc05 | 2556 | |
61e02c74 | 2557 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2558 | goto retry; |
28c34c29 | 2559 | |
b70a2a21 JW |
2560 | if (!drained) { |
2561 | drain_all_stock_async(mem_over_limit); | |
2562 | drained = true; | |
2563 | goto retry; | |
2564 | } | |
2565 | ||
28c34c29 JW |
2566 | if (gfp_mask & __GFP_NORETRY) |
2567 | goto nomem; | |
6539cc05 JW |
2568 | /* |
2569 | * Even though the limit is exceeded at this point, reclaim | |
2570 | * may have been able to free some pages. Retry the charge | |
2571 | * before killing the task. | |
2572 | * | |
2573 | * Only for regular pages, though: huge pages are rather | |
2574 | * unlikely to succeed so close to the limit, and we fall back | |
2575 | * to regular pages anyway in case of failure. | |
2576 | */ | |
61e02c74 | 2577 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2578 | goto retry; |
2579 | /* | |
2580 | * At task move, charge accounts can be doubly counted. So, it's | |
2581 | * better to wait until the end of task_move if something is going on. | |
2582 | */ | |
2583 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2584 | goto retry; | |
2585 | ||
9b130619 JW |
2586 | if (nr_retries--) |
2587 | goto retry; | |
2588 | ||
06b078fc JW |
2589 | if (gfp_mask & __GFP_NOFAIL) |
2590 | goto bypass; | |
2591 | ||
6539cc05 JW |
2592 | if (fatal_signal_pending(current)) |
2593 | goto bypass; | |
2594 | ||
61e02c74 | 2595 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2596 | nomem: |
6d1fdc48 | 2597 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2598 | return -ENOMEM; |
867578cb | 2599 | bypass: |
ce00a967 | 2600 | return -EINTR; |
6539cc05 JW |
2601 | |
2602 | done_restock: | |
2603 | if (batch > nr_pages) | |
2604 | refill_stock(memcg, batch - nr_pages); | |
2605 | done: | |
05b84301 | 2606 | return ret; |
7a81b88c | 2607 | } |
8a9f3ccd | 2608 | |
00501b53 | 2609 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2610 | { |
ce00a967 JW |
2611 | if (mem_cgroup_is_root(memcg)) |
2612 | return; | |
2613 | ||
3e32cb2e | 2614 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2615 | if (do_swap_account) |
3e32cb2e | 2616 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f KH |
2617 | } |
2618 | ||
a3b2d692 KH |
2619 | /* |
2620 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2621 | * rcu_read_lock(). The caller is responsible for calling |
2622 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2623 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2624 | */ |
2625 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2626 | { | |
a3b2d692 KH |
2627 | /* ID 0 is unused ID */ |
2628 | if (!id) | |
2629 | return NULL; | |
34c00c31 | 2630 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2631 | } |
2632 | ||
0a31bc97 JW |
2633 | /* |
2634 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2635 | * @page: the page | |
2636 | * | |
2637 | * Look up, get a css reference, and return the memcg that owns @page. | |
2638 | * | |
2639 | * The page must be locked to prevent racing with swap-in and page | |
2640 | * cache charges. If coming from an unlocked page table, the caller | |
2641 | * must ensure the page is on the LRU or this can race with charging. | |
2642 | */ | |
e42d9d5d | 2643 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2644 | { |
c0ff4b85 | 2645 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2646 | struct page_cgroup *pc; |
a3b2d692 | 2647 | unsigned short id; |
b5a84319 KH |
2648 | swp_entry_t ent; |
2649 | ||
309381fe | 2650 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2651 | |
3c776e64 | 2652 | pc = lookup_page_cgroup(page); |
a3b2d692 | 2653 | if (PageCgroupUsed(pc)) { |
c0ff4b85 | 2654 | memcg = pc->mem_cgroup; |
ec903c0c | 2655 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2656 | memcg = NULL; |
e42d9d5d | 2657 | } else if (PageSwapCache(page)) { |
3c776e64 | 2658 | ent.val = page_private(page); |
9fb4b7cc | 2659 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2660 | rcu_read_lock(); |
c0ff4b85 | 2661 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2662 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2663 | memcg = NULL; |
a3b2d692 | 2664 | rcu_read_unlock(); |
3c776e64 | 2665 | } |
c0ff4b85 | 2666 | return memcg; |
b5a84319 KH |
2667 | } |
2668 | ||
0a31bc97 JW |
2669 | static void lock_page_lru(struct page *page, int *isolated) |
2670 | { | |
2671 | struct zone *zone = page_zone(page); | |
2672 | ||
2673 | spin_lock_irq(&zone->lru_lock); | |
2674 | if (PageLRU(page)) { | |
2675 | struct lruvec *lruvec; | |
2676 | ||
2677 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2678 | ClearPageLRU(page); | |
2679 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2680 | *isolated = 1; | |
2681 | } else | |
2682 | *isolated = 0; | |
2683 | } | |
2684 | ||
2685 | static void unlock_page_lru(struct page *page, int isolated) | |
2686 | { | |
2687 | struct zone *zone = page_zone(page); | |
2688 | ||
2689 | if (isolated) { | |
2690 | struct lruvec *lruvec; | |
2691 | ||
2692 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2693 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2694 | SetPageLRU(page); | |
2695 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2696 | } | |
2697 | spin_unlock_irq(&zone->lru_lock); | |
2698 | } | |
2699 | ||
00501b53 | 2700 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2701 | bool lrucare) |
7a81b88c | 2702 | { |
ce587e65 | 2703 | struct page_cgroup *pc = lookup_page_cgroup(page); |
0a31bc97 | 2704 | int isolated; |
9ce70c02 | 2705 | |
309381fe | 2706 | VM_BUG_ON_PAGE(PageCgroupUsed(pc), page); |
ca3e0214 KH |
2707 | /* |
2708 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2709 | * accessed by any other context at this point. | |
2710 | */ | |
9ce70c02 HD |
2711 | |
2712 | /* | |
2713 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2714 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2715 | */ | |
0a31bc97 JW |
2716 | if (lrucare) |
2717 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2718 | |
0a31bc97 JW |
2719 | /* |
2720 | * Nobody should be changing or seriously looking at | |
2721 | * pc->mem_cgroup and pc->flags at this point: | |
2722 | * | |
2723 | * - the page is uncharged | |
2724 | * | |
2725 | * - the page is off-LRU | |
2726 | * | |
2727 | * - an anonymous fault has exclusive page access, except for | |
2728 | * a locked page table | |
2729 | * | |
2730 | * - a page cache insertion, a swapin fault, or a migration | |
2731 | * have the page locked | |
2732 | */ | |
c0ff4b85 | 2733 | pc->mem_cgroup = memcg; |
0a31bc97 | 2734 | pc->flags = PCG_USED | PCG_MEM | (do_swap_account ? PCG_MEMSW : 0); |
9ce70c02 | 2735 | |
0a31bc97 JW |
2736 | if (lrucare) |
2737 | unlock_page_lru(page, isolated); | |
7a81b88c | 2738 | } |
66e1707b | 2739 | |
7ae1e1d0 | 2740 | #ifdef CONFIG_MEMCG_KMEM |
bd673145 VD |
2741 | /* |
2742 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or | |
2743 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. | |
2744 | */ | |
2745 | static DEFINE_MUTEX(memcg_slab_mutex); | |
2746 | ||
d6441637 VD |
2747 | static DEFINE_MUTEX(activate_kmem_mutex); |
2748 | ||
1f458cbf GC |
2749 | /* |
2750 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2751 | * in the memcg_cache_params struct. | |
2752 | */ | |
2753 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2754 | { | |
2755 | struct kmem_cache *cachep; | |
2756 | ||
2757 | VM_BUG_ON(p->is_root_cache); | |
2758 | cachep = p->root_cache; | |
7a67d7ab | 2759 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2760 | } |
2761 | ||
749c5415 | 2762 | #ifdef CONFIG_SLABINFO |
2da8ca82 | 2763 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
749c5415 | 2764 | { |
2da8ca82 | 2765 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
749c5415 GC |
2766 | struct memcg_cache_params *params; |
2767 | ||
cf2b8fbf | 2768 | if (!memcg_kmem_is_active(memcg)) |
749c5415 GC |
2769 | return -EIO; |
2770 | ||
2771 | print_slabinfo_header(m); | |
2772 | ||
bd673145 | 2773 | mutex_lock(&memcg_slab_mutex); |
749c5415 GC |
2774 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) |
2775 | cache_show(memcg_params_to_cache(params), m); | |
bd673145 | 2776 | mutex_unlock(&memcg_slab_mutex); |
749c5415 GC |
2777 | |
2778 | return 0; | |
2779 | } | |
2780 | #endif | |
2781 | ||
3e32cb2e JW |
2782 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2783 | unsigned long nr_pages) | |
7ae1e1d0 | 2784 | { |
3e32cb2e | 2785 | struct page_counter *counter; |
7ae1e1d0 | 2786 | int ret = 0; |
7ae1e1d0 | 2787 | |
3e32cb2e JW |
2788 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2789 | if (ret < 0) | |
7ae1e1d0 GC |
2790 | return ret; |
2791 | ||
3e32cb2e | 2792 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2793 | if (ret == -EINTR) { |
2794 | /* | |
00501b53 JW |
2795 | * try_charge() chose to bypass to root due to OOM kill or |
2796 | * fatal signal. Since our only options are to either fail | |
2797 | * the allocation or charge it to this cgroup, do it as a | |
2798 | * temporary condition. But we can't fail. From a kmem/slab | |
2799 | * perspective, the cache has already been selected, by | |
2800 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2801 | * our minds. |
2802 | * | |
2803 | * This condition will only trigger if the task entered | |
00501b53 JW |
2804 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2805 | * during try_charge() above. Tasks that were already dying | |
2806 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2807 | * directed to the root cgroup in memcontrol.h |
2808 | */ | |
3e32cb2e | 2809 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2810 | if (do_swap_account) |
3e32cb2e | 2811 | page_counter_charge(&memcg->memsw, nr_pages); |
7ae1e1d0 GC |
2812 | ret = 0; |
2813 | } else if (ret) | |
3e32cb2e | 2814 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2815 | |
2816 | return ret; | |
2817 | } | |
2818 | ||
3e32cb2e JW |
2819 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, |
2820 | unsigned long nr_pages) | |
7ae1e1d0 | 2821 | { |
3e32cb2e | 2822 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2823 | if (do_swap_account) |
3e32cb2e | 2824 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 GC |
2825 | |
2826 | /* Not down to 0 */ | |
3e32cb2e | 2827 | if (page_counter_uncharge(&memcg->kmem, nr_pages)) |
7de37682 GC |
2828 | return; |
2829 | ||
10d5ebf4 LZ |
2830 | /* |
2831 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
2832 | * this last uncharge is racing with the offlining code or it is | |
2833 | * outliving the memcg existence. | |
2834 | * | |
2835 | * The memory barrier imposed by test&clear is paired with the | |
2836 | * explicit one in memcg_kmem_mark_dead(). | |
2837 | */ | |
7de37682 | 2838 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 2839 | css_put(&memcg->css); |
7ae1e1d0 GC |
2840 | } |
2841 | ||
2633d7a0 GC |
2842 | /* |
2843 | * helper for acessing a memcg's index. It will be used as an index in the | |
2844 | * child cache array in kmem_cache, and also to derive its name. This function | |
2845 | * will return -1 when this is not a kmem-limited memcg. | |
2846 | */ | |
2847 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2848 | { | |
2849 | return memcg ? memcg->kmemcg_id : -1; | |
2850 | } | |
2851 | ||
f3bb3043 | 2852 | static int memcg_alloc_cache_id(void) |
55007d84 | 2853 | { |
f3bb3043 VD |
2854 | int id, size; |
2855 | int err; | |
2856 | ||
2857 | id = ida_simple_get(&kmem_limited_groups, | |
2858 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2859 | if (id < 0) | |
2860 | return id; | |
55007d84 | 2861 | |
f3bb3043 VD |
2862 | if (id < memcg_limited_groups_array_size) |
2863 | return id; | |
2864 | ||
2865 | /* | |
2866 | * There's no space for the new id in memcg_caches arrays, | |
2867 | * so we have to grow them. | |
2868 | */ | |
2869 | ||
2870 | size = 2 * (id + 1); | |
55007d84 GC |
2871 | if (size < MEMCG_CACHES_MIN_SIZE) |
2872 | size = MEMCG_CACHES_MIN_SIZE; | |
2873 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2874 | size = MEMCG_CACHES_MAX_SIZE; | |
2875 | ||
f3bb3043 VD |
2876 | mutex_lock(&memcg_slab_mutex); |
2877 | err = memcg_update_all_caches(size); | |
2878 | mutex_unlock(&memcg_slab_mutex); | |
2879 | ||
2880 | if (err) { | |
2881 | ida_simple_remove(&kmem_limited_groups, id); | |
2882 | return err; | |
2883 | } | |
2884 | return id; | |
2885 | } | |
2886 | ||
2887 | static void memcg_free_cache_id(int id) | |
2888 | { | |
2889 | ida_simple_remove(&kmem_limited_groups, id); | |
55007d84 GC |
2890 | } |
2891 | ||
2892 | /* | |
2893 | * We should update the current array size iff all caches updates succeed. This | |
2894 | * can only be done from the slab side. The slab mutex needs to be held when | |
2895 | * calling this. | |
2896 | */ | |
2897 | void memcg_update_array_size(int num) | |
2898 | { | |
f3bb3043 | 2899 | memcg_limited_groups_array_size = num; |
55007d84 GC |
2900 | } |
2901 | ||
776ed0f0 VD |
2902 | static void memcg_register_cache(struct mem_cgroup *memcg, |
2903 | struct kmem_cache *root_cache) | |
2633d7a0 | 2904 | { |
93f39eea VD |
2905 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by |
2906 | memcg_slab_mutex */ | |
bd673145 | 2907 | struct kmem_cache *cachep; |
d7f25f8a GC |
2908 | int id; |
2909 | ||
bd673145 VD |
2910 | lockdep_assert_held(&memcg_slab_mutex); |
2911 | ||
2912 | id = memcg_cache_id(memcg); | |
2913 | ||
2914 | /* | |
2915 | * Since per-memcg caches are created asynchronously on first | |
2916 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
2917 | * create the same cache, but only one of them may succeed. | |
2918 | */ | |
2919 | if (cache_from_memcg_idx(root_cache, id)) | |
1aa13254 VD |
2920 | return; |
2921 | ||
073ee1c6 | 2922 | cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1); |
776ed0f0 | 2923 | cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf); |
2edefe11 | 2924 | /* |
bd673145 VD |
2925 | * If we could not create a memcg cache, do not complain, because |
2926 | * that's not critical at all as we can always proceed with the root | |
2927 | * cache. | |
2edefe11 | 2928 | */ |
bd673145 VD |
2929 | if (!cachep) |
2930 | return; | |
2edefe11 | 2931 | |
33a690c4 | 2932 | css_get(&memcg->css); |
bd673145 | 2933 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
1aa13254 | 2934 | |
d7f25f8a | 2935 | /* |
959c8963 VD |
2936 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
2937 | * barrier here to ensure nobody will see the kmem_cache partially | |
2938 | * initialized. | |
d7f25f8a | 2939 | */ |
959c8963 VD |
2940 | smp_wmb(); |
2941 | ||
bd673145 VD |
2942 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
2943 | root_cache->memcg_params->memcg_caches[id] = cachep; | |
1aa13254 | 2944 | } |
d7f25f8a | 2945 | |
776ed0f0 | 2946 | static void memcg_unregister_cache(struct kmem_cache *cachep) |
1aa13254 | 2947 | { |
bd673145 | 2948 | struct kmem_cache *root_cache; |
1aa13254 VD |
2949 | struct mem_cgroup *memcg; |
2950 | int id; | |
2951 | ||
bd673145 | 2952 | lockdep_assert_held(&memcg_slab_mutex); |
d7f25f8a | 2953 | |
bd673145 | 2954 | BUG_ON(is_root_cache(cachep)); |
2edefe11 | 2955 | |
bd673145 VD |
2956 | root_cache = cachep->memcg_params->root_cache; |
2957 | memcg = cachep->memcg_params->memcg; | |
96403da2 | 2958 | id = memcg_cache_id(memcg); |
d7f25f8a | 2959 | |
bd673145 VD |
2960 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
2961 | root_cache->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a | 2962 | |
bd673145 VD |
2963 | list_del(&cachep->memcg_params->list); |
2964 | ||
2965 | kmem_cache_destroy(cachep); | |
33a690c4 VD |
2966 | |
2967 | /* drop the reference taken in memcg_register_cache */ | |
2968 | css_put(&memcg->css); | |
2633d7a0 GC |
2969 | } |
2970 | ||
0e9d92f2 GC |
2971 | /* |
2972 | * During the creation a new cache, we need to disable our accounting mechanism | |
2973 | * altogether. This is true even if we are not creating, but rather just | |
2974 | * enqueing new caches to be created. | |
2975 | * | |
2976 | * This is because that process will trigger allocations; some visible, like | |
2977 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
2978 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
2979 | * objects during debug. | |
2980 | * | |
2981 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
2982 | * to it. This may not be a bounded recursion: since the first cache creation | |
2983 | * failed to complete (waiting on the allocation), we'll just try to create the | |
2984 | * cache again, failing at the same point. | |
2985 | * | |
2986 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
2987 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
2988 | * inside the following two functions. | |
2989 | */ | |
2990 | static inline void memcg_stop_kmem_account(void) | |
2991 | { | |
2992 | VM_BUG_ON(!current->mm); | |
2993 | current->memcg_kmem_skip_account++; | |
2994 | } | |
2995 | ||
2996 | static inline void memcg_resume_kmem_account(void) | |
2997 | { | |
2998 | VM_BUG_ON(!current->mm); | |
2999 | current->memcg_kmem_skip_account--; | |
3000 | } | |
3001 | ||
776ed0f0 | 3002 | int __memcg_cleanup_cache_params(struct kmem_cache *s) |
7cf27982 GC |
3003 | { |
3004 | struct kmem_cache *c; | |
b8529907 | 3005 | int i, failed = 0; |
7cf27982 | 3006 | |
bd673145 | 3007 | mutex_lock(&memcg_slab_mutex); |
7a67d7ab QH |
3008 | for_each_memcg_cache_index(i) { |
3009 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
3010 | if (!c) |
3011 | continue; | |
3012 | ||
776ed0f0 | 3013 | memcg_unregister_cache(c); |
b8529907 VD |
3014 | |
3015 | if (cache_from_memcg_idx(s, i)) | |
3016 | failed++; | |
7cf27982 | 3017 | } |
bd673145 | 3018 | mutex_unlock(&memcg_slab_mutex); |
b8529907 | 3019 | return failed; |
7cf27982 GC |
3020 | } |
3021 | ||
776ed0f0 | 3022 | static void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3023 | { |
3024 | struct kmem_cache *cachep; | |
bd673145 | 3025 | struct memcg_cache_params *params, *tmp; |
1f458cbf GC |
3026 | |
3027 | if (!memcg_kmem_is_active(memcg)) | |
3028 | return; | |
3029 | ||
bd673145 VD |
3030 | mutex_lock(&memcg_slab_mutex); |
3031 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { | |
1f458cbf | 3032 | cachep = memcg_params_to_cache(params); |
bd673145 VD |
3033 | kmem_cache_shrink(cachep); |
3034 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
776ed0f0 | 3035 | memcg_unregister_cache(cachep); |
1f458cbf | 3036 | } |
bd673145 | 3037 | mutex_unlock(&memcg_slab_mutex); |
1f458cbf GC |
3038 | } |
3039 | ||
776ed0f0 | 3040 | struct memcg_register_cache_work { |
5722d094 VD |
3041 | struct mem_cgroup *memcg; |
3042 | struct kmem_cache *cachep; | |
3043 | struct work_struct work; | |
3044 | }; | |
3045 | ||
776ed0f0 | 3046 | static void memcg_register_cache_func(struct work_struct *w) |
d7f25f8a | 3047 | { |
776ed0f0 VD |
3048 | struct memcg_register_cache_work *cw = |
3049 | container_of(w, struct memcg_register_cache_work, work); | |
5722d094 VD |
3050 | struct mem_cgroup *memcg = cw->memcg; |
3051 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 3052 | |
bd673145 | 3053 | mutex_lock(&memcg_slab_mutex); |
776ed0f0 | 3054 | memcg_register_cache(memcg, cachep); |
bd673145 VD |
3055 | mutex_unlock(&memcg_slab_mutex); |
3056 | ||
5722d094 | 3057 | css_put(&memcg->css); |
d7f25f8a GC |
3058 | kfree(cw); |
3059 | } | |
3060 | ||
3061 | /* | |
3062 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3063 | */ |
776ed0f0 VD |
3064 | static void __memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3065 | struct kmem_cache *cachep) | |
d7f25f8a | 3066 | { |
776ed0f0 | 3067 | struct memcg_register_cache_work *cw; |
d7f25f8a | 3068 | |
776ed0f0 | 3069 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
ca0dde97 LZ |
3070 | if (cw == NULL) { |
3071 | css_put(&memcg->css); | |
d7f25f8a GC |
3072 | return; |
3073 | } | |
3074 | ||
3075 | cw->memcg = memcg; | |
3076 | cw->cachep = cachep; | |
3077 | ||
776ed0f0 | 3078 | INIT_WORK(&cw->work, memcg_register_cache_func); |
d7f25f8a GC |
3079 | schedule_work(&cw->work); |
3080 | } | |
3081 | ||
776ed0f0 VD |
3082 | static void memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3083 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
3084 | { |
3085 | /* | |
3086 | * We need to stop accounting when we kmalloc, because if the | |
3087 | * corresponding kmalloc cache is not yet created, the first allocation | |
776ed0f0 | 3088 | * in __memcg_schedule_register_cache will recurse. |
0e9d92f2 GC |
3089 | * |
3090 | * However, it is better to enclose the whole function. Depending on | |
3091 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3092 | * trigger an allocation. This too, will make us recurse. Because at | |
3093 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3094 | * the safest choice is to do it like this, wrapping the whole function. | |
3095 | */ | |
3096 | memcg_stop_kmem_account(); | |
776ed0f0 | 3097 | __memcg_schedule_register_cache(memcg, cachep); |
0e9d92f2 GC |
3098 | memcg_resume_kmem_account(); |
3099 | } | |
c67a8a68 VD |
3100 | |
3101 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) | |
3102 | { | |
3e32cb2e | 3103 | unsigned int nr_pages = 1 << order; |
c67a8a68 VD |
3104 | int res; |
3105 | ||
3e32cb2e | 3106 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, nr_pages); |
c67a8a68 | 3107 | if (!res) |
3e32cb2e | 3108 | atomic_add(nr_pages, &cachep->memcg_params->nr_pages); |
c67a8a68 VD |
3109 | return res; |
3110 | } | |
3111 | ||
3112 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) | |
3113 | { | |
3e32cb2e JW |
3114 | unsigned int nr_pages = 1 << order; |
3115 | ||
3116 | memcg_uncharge_kmem(cachep->memcg_params->memcg, nr_pages); | |
3117 | atomic_sub(nr_pages, &cachep->memcg_params->nr_pages); | |
c67a8a68 VD |
3118 | } |
3119 | ||
d7f25f8a GC |
3120 | /* |
3121 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3122 | * We try to use the current memcg's version of the cache. | |
3123 | * | |
3124 | * If the cache does not exist yet, if we are the first user of it, | |
3125 | * we either create it immediately, if possible, or create it asynchronously | |
3126 | * in a workqueue. | |
3127 | * In the latter case, we will let the current allocation go through with | |
3128 | * the original cache. | |
3129 | * | |
3130 | * Can't be called in interrupt context or from kernel threads. | |
3131 | * This function needs to be called with rcu_read_lock() held. | |
3132 | */ | |
3133 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3134 | gfp_t gfp) | |
3135 | { | |
3136 | struct mem_cgroup *memcg; | |
959c8963 | 3137 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
3138 | |
3139 | VM_BUG_ON(!cachep->memcg_params); | |
3140 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3141 | ||
0e9d92f2 GC |
3142 | if (!current->mm || current->memcg_kmem_skip_account) |
3143 | return cachep; | |
3144 | ||
d7f25f8a GC |
3145 | rcu_read_lock(); |
3146 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a | 3147 | |
cf2b8fbf | 3148 | if (!memcg_kmem_is_active(memcg)) |
ca0dde97 | 3149 | goto out; |
d7f25f8a | 3150 | |
959c8963 VD |
3151 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
3152 | if (likely(memcg_cachep)) { | |
3153 | cachep = memcg_cachep; | |
ca0dde97 | 3154 | goto out; |
d7f25f8a GC |
3155 | } |
3156 | ||
ca0dde97 | 3157 | /* The corresponding put will be done in the workqueue. */ |
ec903c0c | 3158 | if (!css_tryget_online(&memcg->css)) |
ca0dde97 LZ |
3159 | goto out; |
3160 | rcu_read_unlock(); | |
3161 | ||
3162 | /* | |
3163 | * If we are in a safe context (can wait, and not in interrupt | |
3164 | * context), we could be be predictable and return right away. | |
3165 | * This would guarantee that the allocation being performed | |
3166 | * already belongs in the new cache. | |
3167 | * | |
3168 | * However, there are some clashes that can arrive from locking. | |
3169 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
3170 | * memcg_create_kmem_cache, this means no further allocation |
3171 | * could happen with the slab_mutex held. So it's better to | |
3172 | * defer everything. | |
ca0dde97 | 3173 | */ |
776ed0f0 | 3174 | memcg_schedule_register_cache(memcg, cachep); |
ca0dde97 LZ |
3175 | return cachep; |
3176 | out: | |
3177 | rcu_read_unlock(); | |
3178 | return cachep; | |
d7f25f8a | 3179 | } |
d7f25f8a | 3180 | |
7ae1e1d0 GC |
3181 | /* |
3182 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3183 | * possible for the given order. But the page is not allocated yet, so we'll | |
3184 | * need a further commit step to do the final arrangements. | |
3185 | * | |
3186 | * It is possible for the task to switch cgroups in this mean time, so at | |
3187 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3188 | * the handle argument to return to the caller which cgroup we should commit | |
3189 | * against. We could also return the memcg directly and avoid the pointer | |
3190 | * passing, but a boolean return value gives better semantics considering | |
3191 | * the compiled-out case as well. | |
3192 | * | |
3193 | * Returning true means the allocation is possible. | |
3194 | */ | |
3195 | bool | |
3196 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3197 | { | |
3198 | struct mem_cgroup *memcg; | |
3199 | int ret; | |
3200 | ||
3201 | *_memcg = NULL; | |
6d42c232 GC |
3202 | |
3203 | /* | |
3204 | * Disabling accounting is only relevant for some specific memcg | |
3205 | * internal allocations. Therefore we would initially not have such | |
52383431 VD |
3206 | * check here, since direct calls to the page allocator that are |
3207 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen | |
3208 | * outside memcg core. We are mostly concerned with cache allocations, | |
3209 | * and by having this test at memcg_kmem_get_cache, we are already able | |
3210 | * to relay the allocation to the root cache and bypass the memcg cache | |
3211 | * altogether. | |
6d42c232 GC |
3212 | * |
3213 | * There is one exception, though: the SLUB allocator does not create | |
3214 | * large order caches, but rather service large kmallocs directly from | |
3215 | * the page allocator. Therefore, the following sequence when backed by | |
3216 | * the SLUB allocator: | |
3217 | * | |
f894ffa8 AM |
3218 | * memcg_stop_kmem_account(); |
3219 | * kmalloc(<large_number>) | |
3220 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3221 | * |
3222 | * would effectively ignore the fact that we should skip accounting, | |
3223 | * since it will drive us directly to this function without passing | |
3224 | * through the cache selector memcg_kmem_get_cache. Such large | |
3225 | * allocations are extremely rare but can happen, for instance, for the | |
3226 | * cache arrays. We bring this test here. | |
3227 | */ | |
3228 | if (!current->mm || current->memcg_kmem_skip_account) | |
3229 | return true; | |
3230 | ||
df381975 | 3231 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 3232 | |
cf2b8fbf | 3233 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
3234 | css_put(&memcg->css); |
3235 | return true; | |
3236 | } | |
3237 | ||
3e32cb2e | 3238 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
3239 | if (!ret) |
3240 | *_memcg = memcg; | |
7ae1e1d0 GC |
3241 | |
3242 | css_put(&memcg->css); | |
3243 | return (ret == 0); | |
3244 | } | |
3245 | ||
3246 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3247 | int order) | |
3248 | { | |
3249 | struct page_cgroup *pc; | |
3250 | ||
3251 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3252 | ||
3253 | /* The page allocation failed. Revert */ | |
3254 | if (!page) { | |
3e32cb2e | 3255 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
3256 | return; |
3257 | } | |
a840cda6 JW |
3258 | /* |
3259 | * The page is freshly allocated and not visible to any | |
3260 | * outside callers yet. Set up pc non-atomically. | |
3261 | */ | |
7ae1e1d0 | 3262 | pc = lookup_page_cgroup(page); |
7ae1e1d0 | 3263 | pc->mem_cgroup = memcg; |
a840cda6 | 3264 | pc->flags = PCG_USED; |
7ae1e1d0 GC |
3265 | } |
3266 | ||
3267 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3268 | { | |
3269 | struct mem_cgroup *memcg = NULL; | |
3270 | struct page_cgroup *pc; | |
3271 | ||
3272 | ||
3273 | pc = lookup_page_cgroup(page); | |
7ae1e1d0 GC |
3274 | if (!PageCgroupUsed(pc)) |
3275 | return; | |
3276 | ||
a840cda6 JW |
3277 | memcg = pc->mem_cgroup; |
3278 | pc->flags = 0; | |
7ae1e1d0 GC |
3279 | |
3280 | /* | |
3281 | * We trust that only if there is a memcg associated with the page, it | |
3282 | * is a valid allocation | |
3283 | */ | |
3284 | if (!memcg) | |
3285 | return; | |
3286 | ||
309381fe | 3287 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
3e32cb2e | 3288 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 | 3289 | } |
1f458cbf | 3290 | #else |
776ed0f0 | 3291 | static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3292 | { |
3293 | } | |
7ae1e1d0 GC |
3294 | #endif /* CONFIG_MEMCG_KMEM */ |
3295 | ||
ca3e0214 KH |
3296 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3297 | ||
ca3e0214 KH |
3298 | /* |
3299 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3300 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3301 | * charge/uncharge will be never happen and move_account() is done under | |
3302 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3303 | */ |
e94c8a9c | 3304 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3305 | { |
3306 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3307 | struct page_cgroup *pc; |
b070e65c | 3308 | struct mem_cgroup *memcg; |
e94c8a9c | 3309 | int i; |
ca3e0214 | 3310 | |
3d37c4a9 KH |
3311 | if (mem_cgroup_disabled()) |
3312 | return; | |
b070e65c DR |
3313 | |
3314 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3315 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3316 | pc = head_pc + i; | |
b070e65c | 3317 | pc->mem_cgroup = memcg; |
0a31bc97 | 3318 | pc->flags = head_pc->flags; |
e94c8a9c | 3319 | } |
b070e65c DR |
3320 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3321 | HPAGE_PMD_NR); | |
ca3e0214 | 3322 | } |
12d27107 | 3323 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3324 | |
f817ed48 | 3325 | /** |
de3638d9 | 3326 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3327 | * @page: the page |
7ec99d62 | 3328 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3329 | * @pc: page_cgroup of the page. |
3330 | * @from: mem_cgroup which the page is moved from. | |
3331 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3332 | * | |
3333 | * The caller must confirm following. | |
08e552c6 | 3334 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3335 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3336 | * |
2f3479b1 KH |
3337 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3338 | * from old cgroup. | |
f817ed48 | 3339 | */ |
7ec99d62 JW |
3340 | static int mem_cgroup_move_account(struct page *page, |
3341 | unsigned int nr_pages, | |
3342 | struct page_cgroup *pc, | |
3343 | struct mem_cgroup *from, | |
2f3479b1 | 3344 | struct mem_cgroup *to) |
f817ed48 | 3345 | { |
de3638d9 JW |
3346 | unsigned long flags; |
3347 | int ret; | |
987eba66 | 3348 | |
f817ed48 | 3349 | VM_BUG_ON(from == to); |
309381fe | 3350 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3351 | /* |
3352 | * The page is isolated from LRU. So, collapse function | |
3353 | * will not handle this page. But page splitting can happen. | |
3354 | * Do this check under compound_page_lock(). The caller should | |
3355 | * hold it. | |
3356 | */ | |
3357 | ret = -EBUSY; | |
7ec99d62 | 3358 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3359 | goto out; |
3360 | ||
0a31bc97 JW |
3361 | /* |
3362 | * Prevent mem_cgroup_migrate() from looking at pc->mem_cgroup | |
3363 | * of its source page while we change it: page migration takes | |
3364 | * both pages off the LRU, but page cache replacement doesn't. | |
3365 | */ | |
3366 | if (!trylock_page(page)) | |
3367 | goto out; | |
de3638d9 JW |
3368 | |
3369 | ret = -EINVAL; | |
3370 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
0a31bc97 | 3371 | goto out_unlock; |
de3638d9 | 3372 | |
312734c0 | 3373 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3374 | |
0a31bc97 | 3375 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
3376 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
3377 | nr_pages); | |
3378 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3379 | nr_pages); | |
3380 | } | |
3ea67d06 | 3381 | |
59d1d256 JW |
3382 | if (PageWriteback(page)) { |
3383 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3384 | nr_pages); | |
3385 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3386 | nr_pages); | |
3387 | } | |
3ea67d06 | 3388 | |
0a31bc97 JW |
3389 | /* |
3390 | * It is safe to change pc->mem_cgroup here because the page | |
3391 | * is referenced, charged, and isolated - we can't race with | |
3392 | * uncharging, charging, migration, or LRU putback. | |
3393 | */ | |
d69b042f | 3394 | |
854ffa8d | 3395 | /* caller should have done css_get */ |
08e552c6 | 3396 | pc->mem_cgroup = to; |
312734c0 | 3397 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 | 3398 | ret = 0; |
0a31bc97 JW |
3399 | |
3400 | local_irq_disable(); | |
3401 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 3402 | memcg_check_events(to, page); |
0a31bc97 | 3403 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 3404 | memcg_check_events(from, page); |
0a31bc97 JW |
3405 | local_irq_enable(); |
3406 | out_unlock: | |
3407 | unlock_page(page); | |
de3638d9 | 3408 | out: |
f817ed48 KH |
3409 | return ret; |
3410 | } | |
3411 | ||
2ef37d3f MH |
3412 | /** |
3413 | * mem_cgroup_move_parent - moves page to the parent group | |
3414 | * @page: the page to move | |
3415 | * @pc: page_cgroup of the page | |
3416 | * @child: page's cgroup | |
3417 | * | |
3418 | * move charges to its parent or the root cgroup if the group has no | |
3419 | * parent (aka use_hierarchy==0). | |
3420 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3421 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3422 | * it signals a race with a page removal/uncharge or migration. In the | |
3423 | * first case the page is on the way out and it will vanish from the LRU | |
3424 | * on the next attempt and the call should be retried later. | |
3425 | * Isolation from the LRU fails only if page has been isolated from | |
3426 | * the LRU since we looked at it and that usually means either global | |
3427 | * reclaim or migration going on. The page will either get back to the | |
3428 | * LRU or vanish. | |
3429 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3430 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3431 | * disappear in the next attempt. | |
f817ed48 | 3432 | */ |
5564e88b JW |
3433 | static int mem_cgroup_move_parent(struct page *page, |
3434 | struct page_cgroup *pc, | |
6068bf01 | 3435 | struct mem_cgroup *child) |
f817ed48 | 3436 | { |
f817ed48 | 3437 | struct mem_cgroup *parent; |
7ec99d62 | 3438 | unsigned int nr_pages; |
4be4489f | 3439 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3440 | int ret; |
3441 | ||
d8423011 | 3442 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3443 | |
57f9fd7d DN |
3444 | ret = -EBUSY; |
3445 | if (!get_page_unless_zero(page)) | |
3446 | goto out; | |
3447 | if (isolate_lru_page(page)) | |
3448 | goto put; | |
52dbb905 | 3449 | |
7ec99d62 | 3450 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3451 | |
cc926f78 KH |
3452 | parent = parent_mem_cgroup(child); |
3453 | /* | |
3454 | * If no parent, move charges to root cgroup. | |
3455 | */ | |
3456 | if (!parent) | |
3457 | parent = root_mem_cgroup; | |
f817ed48 | 3458 | |
2ef37d3f | 3459 | if (nr_pages > 1) { |
309381fe | 3460 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
987eba66 | 3461 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3462 | } |
987eba66 | 3463 | |
cc926f78 | 3464 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3465 | pc, child, parent); |
3e32cb2e JW |
3466 | if (!ret) { |
3467 | /* Take charge off the local counters */ | |
3468 | page_counter_cancel(&child->memory, nr_pages); | |
3469 | if (do_swap_account) | |
3470 | page_counter_cancel(&child->memsw, nr_pages); | |
3471 | } | |
8dba474f | 3472 | |
7ec99d62 | 3473 | if (nr_pages > 1) |
987eba66 | 3474 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3475 | putback_lru_page(page); |
57f9fd7d | 3476 | put: |
40d58138 | 3477 | put_page(page); |
57f9fd7d | 3478 | out: |
f817ed48 KH |
3479 | return ret; |
3480 | } | |
3481 | ||
c255a458 | 3482 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
3483 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
3484 | bool charge) | |
d13d1443 | 3485 | { |
0a31bc97 JW |
3486 | int val = (charge) ? 1 : -1; |
3487 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 3488 | } |
02491447 DN |
3489 | |
3490 | /** | |
3491 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3492 | * @entry: swap entry to be moved | |
3493 | * @from: mem_cgroup which the entry is moved from | |
3494 | * @to: mem_cgroup which the entry is moved to | |
3495 | * | |
3496 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3497 | * as the mem_cgroup's id of @from. | |
3498 | * | |
3499 | * Returns 0 on success, -EINVAL on failure. | |
3500 | * | |
3e32cb2e | 3501 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3502 | * both res and memsw, and called css_get(). |
3503 | */ | |
3504 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3505 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3506 | { |
3507 | unsigned short old_id, new_id; | |
3508 | ||
34c00c31 LZ |
3509 | old_id = mem_cgroup_id(from); |
3510 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3511 | |
3512 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3513 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3514 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3515 | /* |
483c30b5 | 3516 | * This function is only called from task migration context now. |
3e32cb2e | 3517 | * It postpones page_counter and refcount handling till the end |
483c30b5 | 3518 | * of task migration(mem_cgroup_clear_mc()) for performance |
4050377b LZ |
3519 | * improvement. But we cannot postpone css_get(to) because if |
3520 | * the process that has been moved to @to does swap-in, the | |
3521 | * refcount of @to might be decreased to 0. | |
3522 | * | |
3523 | * We are in attach() phase, so the cgroup is guaranteed to be | |
3524 | * alive, so we can just call css_get(). | |
02491447 | 3525 | */ |
4050377b | 3526 | css_get(&to->css); |
02491447 DN |
3527 | return 0; |
3528 | } | |
3529 | return -EINVAL; | |
3530 | } | |
3531 | #else | |
3532 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3533 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3534 | { |
3535 | return -EINVAL; | |
3536 | } | |
8c7c6e34 | 3537 | #endif |
d13d1443 | 3538 | |
f212ad7c DN |
3539 | #ifdef CONFIG_DEBUG_VM |
3540 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3541 | { | |
3542 | struct page_cgroup *pc; | |
3543 | ||
3544 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3545 | /* |
3546 | * Can be NULL while feeding pages into the page allocator for | |
3547 | * the first time, i.e. during boot or memory hotplug; | |
3548 | * or when mem_cgroup_disabled(). | |
3549 | */ | |
f212ad7c DN |
3550 | if (likely(pc) && PageCgroupUsed(pc)) |
3551 | return pc; | |
3552 | return NULL; | |
3553 | } | |
3554 | ||
3555 | bool mem_cgroup_bad_page_check(struct page *page) | |
3556 | { | |
3557 | if (mem_cgroup_disabled()) | |
3558 | return false; | |
3559 | ||
3560 | return lookup_page_cgroup_used(page) != NULL; | |
3561 | } | |
3562 | ||
3563 | void mem_cgroup_print_bad_page(struct page *page) | |
3564 | { | |
3565 | struct page_cgroup *pc; | |
3566 | ||
3567 | pc = lookup_page_cgroup_used(page); | |
3568 | if (pc) { | |
d045197f AM |
3569 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
3570 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
3571 | } |
3572 | } | |
3573 | #endif | |
3574 | ||
3e32cb2e JW |
3575 | static DEFINE_MUTEX(memcg_limit_mutex); |
3576 | ||
d38d2a75 | 3577 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3578 | unsigned long limit) |
628f4235 | 3579 | { |
3e32cb2e JW |
3580 | unsigned long curusage; |
3581 | unsigned long oldusage; | |
3582 | bool enlarge = false; | |
81d39c20 | 3583 | int retry_count; |
3e32cb2e | 3584 | int ret; |
81d39c20 KH |
3585 | |
3586 | /* | |
3587 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3588 | * is depends on callers. We set our retry-count to be function | |
3589 | * of # of children which we should visit in this loop. | |
3590 | */ | |
3e32cb2e JW |
3591 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3592 | mem_cgroup_count_children(memcg); | |
81d39c20 | 3593 | |
3e32cb2e | 3594 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 3595 | |
3e32cb2e | 3596 | do { |
628f4235 KH |
3597 | if (signal_pending(current)) { |
3598 | ret = -EINTR; | |
3599 | break; | |
3600 | } | |
3e32cb2e JW |
3601 | |
3602 | mutex_lock(&memcg_limit_mutex); | |
3603 | if (limit > memcg->memsw.limit) { | |
3604 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3605 | ret = -EINVAL; |
628f4235 KH |
3606 | break; |
3607 | } | |
3e32cb2e JW |
3608 | if (limit > memcg->memory.limit) |
3609 | enlarge = true; | |
3610 | ret = page_counter_limit(&memcg->memory, limit); | |
3611 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3612 | |
3613 | if (!ret) | |
3614 | break; | |
3615 | ||
b70a2a21 JW |
3616 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
3617 | ||
3e32cb2e | 3618 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 3619 | /* Usage is reduced ? */ |
f894ffa8 | 3620 | if (curusage >= oldusage) |
81d39c20 KH |
3621 | retry_count--; |
3622 | else | |
3623 | oldusage = curusage; | |
3e32cb2e JW |
3624 | } while (retry_count); |
3625 | ||
3c11ecf4 KH |
3626 | if (!ret && enlarge) |
3627 | memcg_oom_recover(memcg); | |
14797e23 | 3628 | |
8c7c6e34 KH |
3629 | return ret; |
3630 | } | |
3631 | ||
338c8431 | 3632 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3633 | unsigned long limit) |
8c7c6e34 | 3634 | { |
3e32cb2e JW |
3635 | unsigned long curusage; |
3636 | unsigned long oldusage; | |
3637 | bool enlarge = false; | |
81d39c20 | 3638 | int retry_count; |
3e32cb2e | 3639 | int ret; |
8c7c6e34 | 3640 | |
81d39c20 | 3641 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
3642 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3643 | mem_cgroup_count_children(memcg); | |
3644 | ||
3645 | oldusage = page_counter_read(&memcg->memsw); | |
3646 | ||
3647 | do { | |
8c7c6e34 KH |
3648 | if (signal_pending(current)) { |
3649 | ret = -EINTR; | |
3650 | break; | |
3651 | } | |
3e32cb2e JW |
3652 | |
3653 | mutex_lock(&memcg_limit_mutex); | |
3654 | if (limit < memcg->memory.limit) { | |
3655 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3656 | ret = -EINVAL; |
8c7c6e34 KH |
3657 | break; |
3658 | } | |
3e32cb2e JW |
3659 | if (limit > memcg->memsw.limit) |
3660 | enlarge = true; | |
3661 | ret = page_counter_limit(&memcg->memsw, limit); | |
3662 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3663 | |
3664 | if (!ret) | |
3665 | break; | |
3666 | ||
b70a2a21 JW |
3667 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3668 | ||
3e32cb2e | 3669 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3670 | /* Usage is reduced ? */ |
8c7c6e34 | 3671 | if (curusage >= oldusage) |
628f4235 | 3672 | retry_count--; |
81d39c20 KH |
3673 | else |
3674 | oldusage = curusage; | |
3e32cb2e JW |
3675 | } while (retry_count); |
3676 | ||
3c11ecf4 KH |
3677 | if (!ret && enlarge) |
3678 | memcg_oom_recover(memcg); | |
3e32cb2e | 3679 | |
628f4235 KH |
3680 | return ret; |
3681 | } | |
3682 | ||
0608f43d AM |
3683 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3684 | gfp_t gfp_mask, | |
3685 | unsigned long *total_scanned) | |
3686 | { | |
3687 | unsigned long nr_reclaimed = 0; | |
3688 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3689 | unsigned long reclaimed; | |
3690 | int loop = 0; | |
3691 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3692 | unsigned long excess; |
0608f43d AM |
3693 | unsigned long nr_scanned; |
3694 | ||
3695 | if (order > 0) | |
3696 | return 0; | |
3697 | ||
3698 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3699 | /* | |
3700 | * This loop can run a while, specially if mem_cgroup's continuously | |
3701 | * keep exceeding their soft limit and putting the system under | |
3702 | * pressure | |
3703 | */ | |
3704 | do { | |
3705 | if (next_mz) | |
3706 | mz = next_mz; | |
3707 | else | |
3708 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3709 | if (!mz) | |
3710 | break; | |
3711 | ||
3712 | nr_scanned = 0; | |
3713 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3714 | gfp_mask, &nr_scanned); | |
3715 | nr_reclaimed += reclaimed; | |
3716 | *total_scanned += nr_scanned; | |
0a31bc97 | 3717 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3718 | |
3719 | /* | |
3720 | * If we failed to reclaim anything from this memory cgroup | |
3721 | * it is time to move on to the next cgroup | |
3722 | */ | |
3723 | next_mz = NULL; | |
3724 | if (!reclaimed) { | |
3725 | do { | |
3726 | /* | |
3727 | * Loop until we find yet another one. | |
3728 | * | |
3729 | * By the time we get the soft_limit lock | |
3730 | * again, someone might have aded the | |
3731 | * group back on the RB tree. Iterate to | |
3732 | * make sure we get a different mem. | |
3733 | * mem_cgroup_largest_soft_limit_node returns | |
3734 | * NULL if no other cgroup is present on | |
3735 | * the tree | |
3736 | */ | |
3737 | next_mz = | |
3738 | __mem_cgroup_largest_soft_limit_node(mctz); | |
3739 | if (next_mz == mz) | |
3740 | css_put(&next_mz->memcg->css); | |
3741 | else /* next_mz == NULL or other memcg */ | |
3742 | break; | |
3743 | } while (1); | |
3744 | } | |
cf2c8127 | 3745 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 3746 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3747 | /* |
3748 | * One school of thought says that we should not add | |
3749 | * back the node to the tree if reclaim returns 0. | |
3750 | * But our reclaim could return 0, simply because due | |
3751 | * to priority we are exposing a smaller subset of | |
3752 | * memory to reclaim from. Consider this as a longer | |
3753 | * term TODO. | |
3754 | */ | |
3755 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3756 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3757 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3758 | css_put(&mz->memcg->css); |
3759 | loop++; | |
3760 | /* | |
3761 | * Could not reclaim anything and there are no more | |
3762 | * mem cgroups to try or we seem to be looping without | |
3763 | * reclaiming anything. | |
3764 | */ | |
3765 | if (!nr_reclaimed && | |
3766 | (next_mz == NULL || | |
3767 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3768 | break; | |
3769 | } while (!nr_reclaimed); | |
3770 | if (next_mz) | |
3771 | css_put(&next_mz->memcg->css); | |
3772 | return nr_reclaimed; | |
3773 | } | |
3774 | ||
2ef37d3f MH |
3775 | /** |
3776 | * mem_cgroup_force_empty_list - clears LRU of a group | |
3777 | * @memcg: group to clear | |
3778 | * @node: NUMA node | |
3779 | * @zid: zone id | |
3780 | * @lru: lru to to clear | |
3781 | * | |
3c935d18 | 3782 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
3783 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
3784 | * group. | |
cc847582 | 3785 | */ |
2ef37d3f | 3786 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3787 | int node, int zid, enum lru_list lru) |
cc847582 | 3788 | { |
bea8c150 | 3789 | struct lruvec *lruvec; |
2ef37d3f | 3790 | unsigned long flags; |
072c56c1 | 3791 | struct list_head *list; |
925b7673 JW |
3792 | struct page *busy; |
3793 | struct zone *zone; | |
072c56c1 | 3794 | |
08e552c6 | 3795 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
3796 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
3797 | list = &lruvec->lists[lru]; | |
cc847582 | 3798 | |
f817ed48 | 3799 | busy = NULL; |
2ef37d3f | 3800 | do { |
925b7673 | 3801 | struct page_cgroup *pc; |
5564e88b JW |
3802 | struct page *page; |
3803 | ||
08e552c6 | 3804 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3805 | if (list_empty(list)) { |
08e552c6 | 3806 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3807 | break; |
f817ed48 | 3808 | } |
925b7673 JW |
3809 | page = list_entry(list->prev, struct page, lru); |
3810 | if (busy == page) { | |
3811 | list_move(&page->lru, list); | |
648bcc77 | 3812 | busy = NULL; |
08e552c6 | 3813 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3814 | continue; |
3815 | } | |
08e552c6 | 3816 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3817 | |
925b7673 | 3818 | pc = lookup_page_cgroup(page); |
5564e88b | 3819 | |
3c935d18 | 3820 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 3821 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 3822 | busy = page; |
f817ed48 KH |
3823 | } else |
3824 | busy = NULL; | |
2a7a0e0f | 3825 | cond_resched(); |
2ef37d3f | 3826 | } while (!list_empty(list)); |
cc847582 KH |
3827 | } |
3828 | ||
3829 | /* | |
c26251f9 MH |
3830 | * make mem_cgroup's charge to be 0 if there is no task by moving |
3831 | * all the charges and pages to the parent. | |
cc847582 | 3832 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
3833 | * |
3834 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 3835 | */ |
ab5196c2 | 3836 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 3837 | { |
c26251f9 | 3838 | int node, zid; |
f817ed48 | 3839 | |
fce66477 | 3840 | do { |
52d4b9ac KH |
3841 | /* This is for making all *used* pages to be on LRU. */ |
3842 | lru_add_drain_all(); | |
c0ff4b85 | 3843 | drain_all_stock_sync(memcg); |
c0ff4b85 | 3844 | mem_cgroup_start_move(memcg); |
31aaea4a | 3845 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 3846 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3847 | enum lru_list lru; |
3848 | for_each_lru(lru) { | |
2ef37d3f | 3849 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3850 | node, zid, lru); |
f817ed48 | 3851 | } |
1ecaab2b | 3852 | } |
f817ed48 | 3853 | } |
c0ff4b85 R |
3854 | mem_cgroup_end_move(memcg); |
3855 | memcg_oom_recover(memcg); | |
52d4b9ac | 3856 | cond_resched(); |
f817ed48 | 3857 | |
2ef37d3f | 3858 | /* |
bea207c8 GC |
3859 | * Kernel memory may not necessarily be trackable to a specific |
3860 | * process. So they are not migrated, and therefore we can't | |
3861 | * expect their value to drop to 0 here. | |
3862 | * Having res filled up with kmem only is enough. | |
3863 | * | |
2ef37d3f MH |
3864 | * This is a safety check because mem_cgroup_force_empty_list |
3865 | * could have raced with mem_cgroup_replace_page_cache callers | |
3866 | * so the lru seemed empty but the page could have been added | |
3867 | * right after the check. RES_USAGE should be safe as we always | |
3868 | * charge before adding to the LRU. | |
3869 | */ | |
3e32cb2e JW |
3870 | } while (page_counter_read(&memcg->memory) - |
3871 | page_counter_read(&memcg->kmem) > 0); | |
c26251f9 MH |
3872 | } |
3873 | ||
ea280e7b TH |
3874 | /* |
3875 | * Test whether @memcg has children, dead or alive. Note that this | |
3876 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3877 | * returns %true if there are child csses according to the cgroup | |
3878 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3879 | */ | |
b5f99b53 GC |
3880 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3881 | { | |
ea280e7b TH |
3882 | bool ret; |
3883 | ||
696ac172 | 3884 | /* |
ea280e7b TH |
3885 | * The lock does not prevent addition or deletion of children, but |
3886 | * it prevents a new child from being initialized based on this | |
3887 | * parent in css_online(), so it's enough to decide whether | |
3888 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3889 | */ |
ea280e7b TH |
3890 | lockdep_assert_held(&memcg_create_mutex); |
3891 | ||
3892 | rcu_read_lock(); | |
3893 | ret = css_next_child(NULL, &memcg->css); | |
3894 | rcu_read_unlock(); | |
3895 | return ret; | |
b5f99b53 GC |
3896 | } |
3897 | ||
c26251f9 MH |
3898 | /* |
3899 | * Reclaims as many pages from the given memcg as possible and moves | |
3900 | * the rest to the parent. | |
3901 | * | |
3902 | * Caller is responsible for holding css reference for memcg. | |
3903 | */ | |
3904 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3905 | { | |
3906 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3907 | |
c1e862c1 KH |
3908 | /* we call try-to-free pages for make this cgroup empty */ |
3909 | lru_add_drain_all(); | |
f817ed48 | 3910 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3911 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3912 | int progress; |
c1e862c1 | 3913 | |
c26251f9 MH |
3914 | if (signal_pending(current)) |
3915 | return -EINTR; | |
3916 | ||
b70a2a21 JW |
3917 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3918 | GFP_KERNEL, true); | |
c1e862c1 | 3919 | if (!progress) { |
f817ed48 | 3920 | nr_retries--; |
c1e862c1 | 3921 | /* maybe some writeback is necessary */ |
8aa7e847 | 3922 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3923 | } |
f817ed48 KH |
3924 | |
3925 | } | |
ab5196c2 MH |
3926 | |
3927 | return 0; | |
cc847582 KH |
3928 | } |
3929 | ||
6770c64e TH |
3930 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3931 | char *buf, size_t nbytes, | |
3932 | loff_t off) | |
c1e862c1 | 3933 | { |
6770c64e | 3934 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3935 | |
d8423011 MH |
3936 | if (mem_cgroup_is_root(memcg)) |
3937 | return -EINVAL; | |
6770c64e | 3938 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3939 | } |
3940 | ||
182446d0 TH |
3941 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3942 | struct cftype *cft) | |
18f59ea7 | 3943 | { |
182446d0 | 3944 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3945 | } |
3946 | ||
182446d0 TH |
3947 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3948 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3949 | { |
3950 | int retval = 0; | |
182446d0 | 3951 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3952 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3953 | |
0999821b | 3954 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3955 | |
3956 | if (memcg->use_hierarchy == val) | |
3957 | goto out; | |
3958 | ||
18f59ea7 | 3959 | /* |
af901ca1 | 3960 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3961 | * in the child subtrees. If it is unset, then the change can |
3962 | * occur, provided the current cgroup has no children. | |
3963 | * | |
3964 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3965 | * set if there are no children. | |
3966 | */ | |
c0ff4b85 | 3967 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3968 | (val == 1 || val == 0)) { |
ea280e7b | 3969 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3970 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3971 | else |
3972 | retval = -EBUSY; | |
3973 | } else | |
3974 | retval = -EINVAL; | |
567fb435 GC |
3975 | |
3976 | out: | |
0999821b | 3977 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3978 | |
3979 | return retval; | |
3980 | } | |
3981 | ||
3e32cb2e JW |
3982 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3983 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3984 | { |
3985 | struct mem_cgroup *iter; | |
3986 | long val = 0; | |
3987 | ||
3988 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3989 | for_each_mem_cgroup_tree(iter, memcg) | |
3990 | val += mem_cgroup_read_stat(iter, idx); | |
3991 | ||
3992 | if (val < 0) /* race ? */ | |
3993 | val = 0; | |
3994 | return val; | |
3995 | } | |
3996 | ||
3997 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3998 | { | |
3999 | u64 val; | |
4000 | ||
3e32cb2e JW |
4001 | if (mem_cgroup_is_root(memcg)) { |
4002 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
4003 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
4004 | if (swap) | |
4005 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
4006 | } else { | |
ce00a967 | 4007 | if (!swap) |
3e32cb2e | 4008 | val = page_counter_read(&memcg->memory); |
ce00a967 | 4009 | else |
3e32cb2e | 4010 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 4011 | } |
ce00a967 JW |
4012 | return val << PAGE_SHIFT; |
4013 | } | |
4014 | ||
3e32cb2e JW |
4015 | enum { |
4016 | RES_USAGE, | |
4017 | RES_LIMIT, | |
4018 | RES_MAX_USAGE, | |
4019 | RES_FAILCNT, | |
4020 | RES_SOFT_LIMIT, | |
4021 | }; | |
ce00a967 | 4022 | |
791badbd | 4023 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 4024 | struct cftype *cft) |
8cdea7c0 | 4025 | { |
182446d0 | 4026 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 4027 | struct page_counter *counter; |
af36f906 | 4028 | |
3e32cb2e | 4029 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 4030 | case _MEM: |
3e32cb2e JW |
4031 | counter = &memcg->memory; |
4032 | break; | |
8c7c6e34 | 4033 | case _MEMSWAP: |
3e32cb2e JW |
4034 | counter = &memcg->memsw; |
4035 | break; | |
510fc4e1 | 4036 | case _KMEM: |
3e32cb2e | 4037 | counter = &memcg->kmem; |
510fc4e1 | 4038 | break; |
8c7c6e34 KH |
4039 | default: |
4040 | BUG(); | |
8c7c6e34 | 4041 | } |
3e32cb2e JW |
4042 | |
4043 | switch (MEMFILE_ATTR(cft->private)) { | |
4044 | case RES_USAGE: | |
4045 | if (counter == &memcg->memory) | |
4046 | return mem_cgroup_usage(memcg, false); | |
4047 | if (counter == &memcg->memsw) | |
4048 | return mem_cgroup_usage(memcg, true); | |
4049 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
4050 | case RES_LIMIT: | |
4051 | return (u64)counter->limit * PAGE_SIZE; | |
4052 | case RES_MAX_USAGE: | |
4053 | return (u64)counter->watermark * PAGE_SIZE; | |
4054 | case RES_FAILCNT: | |
4055 | return counter->failcnt; | |
4056 | case RES_SOFT_LIMIT: | |
4057 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
4058 | default: | |
4059 | BUG(); | |
4060 | } | |
8cdea7c0 | 4061 | } |
510fc4e1 | 4062 | |
510fc4e1 | 4063 | #ifdef CONFIG_MEMCG_KMEM |
d6441637 VD |
4064 | /* should be called with activate_kmem_mutex held */ |
4065 | static int __memcg_activate_kmem(struct mem_cgroup *memcg, | |
3e32cb2e | 4066 | unsigned long nr_pages) |
d6441637 VD |
4067 | { |
4068 | int err = 0; | |
4069 | int memcg_id; | |
4070 | ||
4071 | if (memcg_kmem_is_active(memcg)) | |
4072 | return 0; | |
4073 | ||
4074 | /* | |
4075 | * We are going to allocate memory for data shared by all memory | |
4076 | * cgroups so let's stop accounting here. | |
4077 | */ | |
4078 | memcg_stop_kmem_account(); | |
4079 | ||
510fc4e1 GC |
4080 | /* |
4081 | * For simplicity, we won't allow this to be disabled. It also can't | |
4082 | * be changed if the cgroup has children already, or if tasks had | |
4083 | * already joined. | |
4084 | * | |
4085 | * If tasks join before we set the limit, a person looking at | |
4086 | * kmem.usage_in_bytes will have no way to determine when it took | |
4087 | * place, which makes the value quite meaningless. | |
4088 | * | |
4089 | * After it first became limited, changes in the value of the limit are | |
4090 | * of course permitted. | |
510fc4e1 | 4091 | */ |
0999821b | 4092 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
4093 | if (cgroup_has_tasks(memcg->css.cgroup) || |
4094 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
4095 | err = -EBUSY; |
4096 | mutex_unlock(&memcg_create_mutex); | |
4097 | if (err) | |
4098 | goto out; | |
510fc4e1 | 4099 | |
f3bb3043 | 4100 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
4101 | if (memcg_id < 0) { |
4102 | err = memcg_id; | |
4103 | goto out; | |
4104 | } | |
4105 | ||
d6441637 VD |
4106 | memcg->kmemcg_id = memcg_id; |
4107 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
d6441637 VD |
4108 | |
4109 | /* | |
4110 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
4111 | * active bit set yet, so this should succeed. | |
4112 | */ | |
3e32cb2e | 4113 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
4114 | VM_BUG_ON(err); |
4115 | ||
4116 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
4117 | /* | |
4118 | * Setting the active bit after enabling static branching will | |
4119 | * guarantee no one starts accounting before all call sites are | |
4120 | * patched. | |
4121 | */ | |
4122 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 4123 | out: |
d6441637 VD |
4124 | memcg_resume_kmem_account(); |
4125 | return err; | |
d6441637 VD |
4126 | } |
4127 | ||
4128 | static int memcg_activate_kmem(struct mem_cgroup *memcg, | |
3e32cb2e | 4129 | unsigned long nr_pages) |
d6441637 VD |
4130 | { |
4131 | int ret; | |
4132 | ||
4133 | mutex_lock(&activate_kmem_mutex); | |
3e32cb2e | 4134 | ret = __memcg_activate_kmem(memcg, nr_pages); |
d6441637 VD |
4135 | mutex_unlock(&activate_kmem_mutex); |
4136 | return ret; | |
4137 | } | |
4138 | ||
4139 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 4140 | unsigned long limit) |
d6441637 VD |
4141 | { |
4142 | int ret; | |
4143 | ||
3e32cb2e | 4144 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 4145 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 4146 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 4147 | else |
3e32cb2e JW |
4148 | ret = page_counter_limit(&memcg->kmem, limit); |
4149 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
4150 | return ret; |
4151 | } | |
4152 | ||
55007d84 | 4153 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 4154 | { |
55007d84 | 4155 | int ret = 0; |
510fc4e1 | 4156 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 4157 | |
d6441637 VD |
4158 | if (!parent) |
4159 | return 0; | |
55007d84 | 4160 | |
d6441637 | 4161 | mutex_lock(&activate_kmem_mutex); |
55007d84 | 4162 | /* |
d6441637 VD |
4163 | * If the parent cgroup is not kmem-active now, it cannot be activated |
4164 | * after this point, because it has at least one child already. | |
55007d84 | 4165 | */ |
d6441637 | 4166 | if (memcg_kmem_is_active(parent)) |
3e32cb2e | 4167 | ret = __memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
d6441637 | 4168 | mutex_unlock(&activate_kmem_mutex); |
55007d84 | 4169 | return ret; |
510fc4e1 | 4170 | } |
d6441637 VD |
4171 | #else |
4172 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 4173 | unsigned long limit) |
d6441637 VD |
4174 | { |
4175 | return -EINVAL; | |
4176 | } | |
6d043990 | 4177 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 4178 | |
628f4235 KH |
4179 | /* |
4180 | * The user of this function is... | |
4181 | * RES_LIMIT. | |
4182 | */ | |
451af504 TH |
4183 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
4184 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 4185 | { |
451af504 | 4186 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4187 | unsigned long nr_pages; |
628f4235 KH |
4188 | int ret; |
4189 | ||
451af504 | 4190 | buf = strstrip(buf); |
3e32cb2e JW |
4191 | ret = page_counter_memparse(buf, &nr_pages); |
4192 | if (ret) | |
4193 | return ret; | |
af36f906 | 4194 | |
3e32cb2e | 4195 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 4196 | case RES_LIMIT: |
4b3bde4c BS |
4197 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4198 | ret = -EINVAL; | |
4199 | break; | |
4200 | } | |
3e32cb2e JW |
4201 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4202 | case _MEM: | |
4203 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 4204 | break; |
3e32cb2e JW |
4205 | case _MEMSWAP: |
4206 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 4207 | break; |
3e32cb2e JW |
4208 | case _KMEM: |
4209 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
4210 | break; | |
4211 | } | |
296c81d8 | 4212 | break; |
3e32cb2e JW |
4213 | case RES_SOFT_LIMIT: |
4214 | memcg->soft_limit = nr_pages; | |
4215 | ret = 0; | |
628f4235 KH |
4216 | break; |
4217 | } | |
451af504 | 4218 | return ret ?: nbytes; |
8cdea7c0 BS |
4219 | } |
4220 | ||
6770c64e TH |
4221 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
4222 | size_t nbytes, loff_t off) | |
c84872e1 | 4223 | { |
6770c64e | 4224 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4225 | struct page_counter *counter; |
c84872e1 | 4226 | |
3e32cb2e JW |
4227 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4228 | case _MEM: | |
4229 | counter = &memcg->memory; | |
4230 | break; | |
4231 | case _MEMSWAP: | |
4232 | counter = &memcg->memsw; | |
4233 | break; | |
4234 | case _KMEM: | |
4235 | counter = &memcg->kmem; | |
4236 | break; | |
4237 | default: | |
4238 | BUG(); | |
4239 | } | |
af36f906 | 4240 | |
3e32cb2e | 4241 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 4242 | case RES_MAX_USAGE: |
3e32cb2e | 4243 | page_counter_reset_watermark(counter); |
29f2a4da PE |
4244 | break; |
4245 | case RES_FAILCNT: | |
3e32cb2e | 4246 | counter->failcnt = 0; |
29f2a4da | 4247 | break; |
3e32cb2e JW |
4248 | default: |
4249 | BUG(); | |
29f2a4da | 4250 | } |
f64c3f54 | 4251 | |
6770c64e | 4252 | return nbytes; |
c84872e1 PE |
4253 | } |
4254 | ||
182446d0 | 4255 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4256 | struct cftype *cft) |
4257 | { | |
182446d0 | 4258 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
4259 | } |
4260 | ||
02491447 | 4261 | #ifdef CONFIG_MMU |
182446d0 | 4262 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4263 | struct cftype *cft, u64 val) |
4264 | { | |
182446d0 | 4265 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
4266 | |
4267 | if (val >= (1 << NR_MOVE_TYPE)) | |
4268 | return -EINVAL; | |
ee5e8472 | 4269 | |
7dc74be0 | 4270 | /* |
ee5e8472 GC |
4271 | * No kind of locking is needed in here, because ->can_attach() will |
4272 | * check this value once in the beginning of the process, and then carry | |
4273 | * on with stale data. This means that changes to this value will only | |
4274 | * affect task migrations starting after the change. | |
7dc74be0 | 4275 | */ |
c0ff4b85 | 4276 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4277 | return 0; |
4278 | } | |
02491447 | 4279 | #else |
182446d0 | 4280 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
4281 | struct cftype *cft, u64 val) |
4282 | { | |
4283 | return -ENOSYS; | |
4284 | } | |
4285 | #endif | |
7dc74be0 | 4286 | |
406eb0c9 | 4287 | #ifdef CONFIG_NUMA |
2da8ca82 | 4288 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 4289 | { |
25485de6 GT |
4290 | struct numa_stat { |
4291 | const char *name; | |
4292 | unsigned int lru_mask; | |
4293 | }; | |
4294 | ||
4295 | static const struct numa_stat stats[] = { | |
4296 | { "total", LRU_ALL }, | |
4297 | { "file", LRU_ALL_FILE }, | |
4298 | { "anon", LRU_ALL_ANON }, | |
4299 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4300 | }; | |
4301 | const struct numa_stat *stat; | |
406eb0c9 | 4302 | int nid; |
25485de6 | 4303 | unsigned long nr; |
2da8ca82 | 4304 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 4305 | |
25485de6 GT |
4306 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
4307 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
4308 | seq_printf(m, "%s=%lu", stat->name, nr); | |
4309 | for_each_node_state(nid, N_MEMORY) { | |
4310 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4311 | stat->lru_mask); | |
4312 | seq_printf(m, " N%d=%lu", nid, nr); | |
4313 | } | |
4314 | seq_putc(m, '\n'); | |
406eb0c9 | 4315 | } |
406eb0c9 | 4316 | |
071aee13 YH |
4317 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
4318 | struct mem_cgroup *iter; | |
4319 | ||
4320 | nr = 0; | |
4321 | for_each_mem_cgroup_tree(iter, memcg) | |
4322 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
4323 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
4324 | for_each_node_state(nid, N_MEMORY) { | |
4325 | nr = 0; | |
4326 | for_each_mem_cgroup_tree(iter, memcg) | |
4327 | nr += mem_cgroup_node_nr_lru_pages( | |
4328 | iter, nid, stat->lru_mask); | |
4329 | seq_printf(m, " N%d=%lu", nid, nr); | |
4330 | } | |
4331 | seq_putc(m, '\n'); | |
406eb0c9 | 4332 | } |
406eb0c9 | 4333 | |
406eb0c9 YH |
4334 | return 0; |
4335 | } | |
4336 | #endif /* CONFIG_NUMA */ | |
4337 | ||
af7c4b0e JW |
4338 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
4339 | { | |
4340 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4341 | } | |
4342 | ||
2da8ca82 | 4343 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4344 | { |
2da8ca82 | 4345 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 4346 | unsigned long memory, memsw; |
af7c4b0e JW |
4347 | struct mem_cgroup *mi; |
4348 | unsigned int i; | |
406eb0c9 | 4349 | |
af7c4b0e | 4350 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4351 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4352 | continue; |
af7c4b0e JW |
4353 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4354 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4355 | } |
7b854121 | 4356 | |
af7c4b0e JW |
4357 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4358 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4359 | mem_cgroup_read_events(memcg, i)); | |
4360 | ||
4361 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4362 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4363 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4364 | ||
14067bb3 | 4365 | /* Hierarchical information */ |
3e32cb2e JW |
4366 | memory = memsw = PAGE_COUNTER_MAX; |
4367 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
4368 | memory = min(memory, mi->memory.limit); | |
4369 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 4370 | } |
3e32cb2e JW |
4371 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4372 | (u64)memory * PAGE_SIZE); | |
4373 | if (do_swap_account) | |
4374 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
4375 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4376 | |
af7c4b0e JW |
4377 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4378 | long long val = 0; | |
4379 | ||
bff6bb83 | 4380 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4381 | continue; |
af7c4b0e JW |
4382 | for_each_mem_cgroup_tree(mi, memcg) |
4383 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4384 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4385 | } | |
4386 | ||
4387 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4388 | unsigned long long val = 0; | |
4389 | ||
4390 | for_each_mem_cgroup_tree(mi, memcg) | |
4391 | val += mem_cgroup_read_events(mi, i); | |
4392 | seq_printf(m, "total_%s %llu\n", | |
4393 | mem_cgroup_events_names[i], val); | |
4394 | } | |
4395 | ||
4396 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4397 | unsigned long long val = 0; | |
4398 | ||
4399 | for_each_mem_cgroup_tree(mi, memcg) | |
4400 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4401 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4402 | } |
14067bb3 | 4403 | |
7f016ee8 | 4404 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4405 | { |
4406 | int nid, zid; | |
4407 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4408 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4409 | unsigned long recent_rotated[2] = {0, 0}; |
4410 | unsigned long recent_scanned[2] = {0, 0}; | |
4411 | ||
4412 | for_each_online_node(nid) | |
4413 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 4414 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 4415 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4416 | |
89abfab1 HD |
4417 | recent_rotated[0] += rstat->recent_rotated[0]; |
4418 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4419 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4420 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4421 | } |
78ccf5b5 JW |
4422 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4423 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4424 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4425 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4426 | } |
4427 | #endif | |
4428 | ||
d2ceb9b7 KH |
4429 | return 0; |
4430 | } | |
4431 | ||
182446d0 TH |
4432 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4433 | struct cftype *cft) | |
a7885eb8 | 4434 | { |
182446d0 | 4435 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4436 | |
1f4c025b | 4437 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4438 | } |
4439 | ||
182446d0 TH |
4440 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4441 | struct cftype *cft, u64 val) | |
a7885eb8 | 4442 | { |
182446d0 | 4443 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4444 | |
3dae7fec | 4445 | if (val > 100) |
a7885eb8 KM |
4446 | return -EINVAL; |
4447 | ||
14208b0e | 4448 | if (css->parent) |
3dae7fec JW |
4449 | memcg->swappiness = val; |
4450 | else | |
4451 | vm_swappiness = val; | |
068b38c1 | 4452 | |
a7885eb8 KM |
4453 | return 0; |
4454 | } | |
4455 | ||
2e72b634 KS |
4456 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4457 | { | |
4458 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4459 | unsigned long usage; |
2e72b634 KS |
4460 | int i; |
4461 | ||
4462 | rcu_read_lock(); | |
4463 | if (!swap) | |
2c488db2 | 4464 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4465 | else |
2c488db2 | 4466 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4467 | |
4468 | if (!t) | |
4469 | goto unlock; | |
4470 | ||
ce00a967 | 4471 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4472 | |
4473 | /* | |
748dad36 | 4474 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4475 | * If it's not true, a threshold was crossed after last |
4476 | * call of __mem_cgroup_threshold(). | |
4477 | */ | |
5407a562 | 4478 | i = t->current_threshold; |
2e72b634 KS |
4479 | |
4480 | /* | |
4481 | * Iterate backward over array of thresholds starting from | |
4482 | * current_threshold and check if a threshold is crossed. | |
4483 | * If none of thresholds below usage is crossed, we read | |
4484 | * only one element of the array here. | |
4485 | */ | |
4486 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4487 | eventfd_signal(t->entries[i].eventfd, 1); | |
4488 | ||
4489 | /* i = current_threshold + 1 */ | |
4490 | i++; | |
4491 | ||
4492 | /* | |
4493 | * Iterate forward over array of thresholds starting from | |
4494 | * current_threshold+1 and check if a threshold is crossed. | |
4495 | * If none of thresholds above usage is crossed, we read | |
4496 | * only one element of the array here. | |
4497 | */ | |
4498 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4499 | eventfd_signal(t->entries[i].eventfd, 1); | |
4500 | ||
4501 | /* Update current_threshold */ | |
5407a562 | 4502 | t->current_threshold = i - 1; |
2e72b634 KS |
4503 | unlock: |
4504 | rcu_read_unlock(); | |
4505 | } | |
4506 | ||
4507 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4508 | { | |
ad4ca5f4 KS |
4509 | while (memcg) { |
4510 | __mem_cgroup_threshold(memcg, false); | |
4511 | if (do_swap_account) | |
4512 | __mem_cgroup_threshold(memcg, true); | |
4513 | ||
4514 | memcg = parent_mem_cgroup(memcg); | |
4515 | } | |
2e72b634 KS |
4516 | } |
4517 | ||
4518 | static int compare_thresholds(const void *a, const void *b) | |
4519 | { | |
4520 | const struct mem_cgroup_threshold *_a = a; | |
4521 | const struct mem_cgroup_threshold *_b = b; | |
4522 | ||
2bff24a3 GT |
4523 | if (_a->threshold > _b->threshold) |
4524 | return 1; | |
4525 | ||
4526 | if (_a->threshold < _b->threshold) | |
4527 | return -1; | |
4528 | ||
4529 | return 0; | |
2e72b634 KS |
4530 | } |
4531 | ||
c0ff4b85 | 4532 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4533 | { |
4534 | struct mem_cgroup_eventfd_list *ev; | |
4535 | ||
2bcf2e92 MH |
4536 | spin_lock(&memcg_oom_lock); |
4537 | ||
c0ff4b85 | 4538 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4539 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4540 | |
4541 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4542 | return 0; |
4543 | } | |
4544 | ||
c0ff4b85 | 4545 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4546 | { |
7d74b06f KH |
4547 | struct mem_cgroup *iter; |
4548 | ||
c0ff4b85 | 4549 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4550 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4551 | } |
4552 | ||
59b6f873 | 4553 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4554 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4555 | { |
2c488db2 KS |
4556 | struct mem_cgroup_thresholds *thresholds; |
4557 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4558 | unsigned long threshold; |
4559 | unsigned long usage; | |
2c488db2 | 4560 | int i, size, ret; |
2e72b634 | 4561 | |
3e32cb2e | 4562 | ret = page_counter_memparse(args, &threshold); |
2e72b634 KS |
4563 | if (ret) |
4564 | return ret; | |
4565 | ||
4566 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4567 | |
05b84301 | 4568 | if (type == _MEM) { |
2c488db2 | 4569 | thresholds = &memcg->thresholds; |
ce00a967 | 4570 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4571 | } else if (type == _MEMSWAP) { |
2c488db2 | 4572 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4573 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4574 | } else |
2e72b634 KS |
4575 | BUG(); |
4576 | ||
2e72b634 | 4577 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4578 | if (thresholds->primary) |
2e72b634 KS |
4579 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4580 | ||
2c488db2 | 4581 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4582 | |
4583 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4584 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4585 | GFP_KERNEL); |
2c488db2 | 4586 | if (!new) { |
2e72b634 KS |
4587 | ret = -ENOMEM; |
4588 | goto unlock; | |
4589 | } | |
2c488db2 | 4590 | new->size = size; |
2e72b634 KS |
4591 | |
4592 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4593 | if (thresholds->primary) { |
4594 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4595 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4596 | } |
4597 | ||
2e72b634 | 4598 | /* Add new threshold */ |
2c488db2 KS |
4599 | new->entries[size - 1].eventfd = eventfd; |
4600 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4601 | |
4602 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4603 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4604 | compare_thresholds, NULL); |
4605 | ||
4606 | /* Find current threshold */ | |
2c488db2 | 4607 | new->current_threshold = -1; |
2e72b634 | 4608 | for (i = 0; i < size; i++) { |
748dad36 | 4609 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4610 | /* |
2c488db2 KS |
4611 | * new->current_threshold will not be used until |
4612 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4613 | * it here. |
4614 | */ | |
2c488db2 | 4615 | ++new->current_threshold; |
748dad36 SZ |
4616 | } else |
4617 | break; | |
2e72b634 KS |
4618 | } |
4619 | ||
2c488db2 KS |
4620 | /* Free old spare buffer and save old primary buffer as spare */ |
4621 | kfree(thresholds->spare); | |
4622 | thresholds->spare = thresholds->primary; | |
4623 | ||
4624 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4625 | |
907860ed | 4626 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4627 | synchronize_rcu(); |
4628 | ||
2e72b634 KS |
4629 | unlock: |
4630 | mutex_unlock(&memcg->thresholds_lock); | |
4631 | ||
4632 | return ret; | |
4633 | } | |
4634 | ||
59b6f873 | 4635 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4636 | struct eventfd_ctx *eventfd, const char *args) |
4637 | { | |
59b6f873 | 4638 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4639 | } |
4640 | ||
59b6f873 | 4641 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4642 | struct eventfd_ctx *eventfd, const char *args) |
4643 | { | |
59b6f873 | 4644 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4645 | } |
4646 | ||
59b6f873 | 4647 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4648 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4649 | { |
2c488db2 KS |
4650 | struct mem_cgroup_thresholds *thresholds; |
4651 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4652 | unsigned long usage; |
2c488db2 | 4653 | int i, j, size; |
2e72b634 KS |
4654 | |
4655 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4656 | |
4657 | if (type == _MEM) { | |
2c488db2 | 4658 | thresholds = &memcg->thresholds; |
ce00a967 | 4659 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4660 | } else if (type == _MEMSWAP) { |
2c488db2 | 4661 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4662 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4663 | } else |
2e72b634 KS |
4664 | BUG(); |
4665 | ||
371528ca AV |
4666 | if (!thresholds->primary) |
4667 | goto unlock; | |
4668 | ||
2e72b634 KS |
4669 | /* Check if a threshold crossed before removing */ |
4670 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4671 | ||
4672 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4673 | size = 0; |
4674 | for (i = 0; i < thresholds->primary->size; i++) { | |
4675 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4676 | size++; |
4677 | } | |
4678 | ||
2c488db2 | 4679 | new = thresholds->spare; |
907860ed | 4680 | |
2e72b634 KS |
4681 | /* Set thresholds array to NULL if we don't have thresholds */ |
4682 | if (!size) { | |
2c488db2 KS |
4683 | kfree(new); |
4684 | new = NULL; | |
907860ed | 4685 | goto swap_buffers; |
2e72b634 KS |
4686 | } |
4687 | ||
2c488db2 | 4688 | new->size = size; |
2e72b634 KS |
4689 | |
4690 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4691 | new->current_threshold = -1; |
4692 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4693 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4694 | continue; |
4695 | ||
2c488db2 | 4696 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4697 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4698 | /* |
2c488db2 | 4699 | * new->current_threshold will not be used |
2e72b634 KS |
4700 | * until rcu_assign_pointer(), so it's safe to increment |
4701 | * it here. | |
4702 | */ | |
2c488db2 | 4703 | ++new->current_threshold; |
2e72b634 KS |
4704 | } |
4705 | j++; | |
4706 | } | |
4707 | ||
907860ed | 4708 | swap_buffers: |
2c488db2 KS |
4709 | /* Swap primary and spare array */ |
4710 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4711 | /* If all events are unregistered, free the spare array */ |
4712 | if (!new) { | |
4713 | kfree(thresholds->spare); | |
4714 | thresholds->spare = NULL; | |
4715 | } | |
4716 | ||
2c488db2 | 4717 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4718 | |
907860ed | 4719 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4720 | synchronize_rcu(); |
371528ca | 4721 | unlock: |
2e72b634 | 4722 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4723 | } |
c1e862c1 | 4724 | |
59b6f873 | 4725 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4726 | struct eventfd_ctx *eventfd) |
4727 | { | |
59b6f873 | 4728 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4729 | } |
4730 | ||
59b6f873 | 4731 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4732 | struct eventfd_ctx *eventfd) |
4733 | { | |
59b6f873 | 4734 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4735 | } |
4736 | ||
59b6f873 | 4737 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4738 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4739 | { |
9490ff27 | 4740 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4741 | |
9490ff27 KH |
4742 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4743 | if (!event) | |
4744 | return -ENOMEM; | |
4745 | ||
1af8efe9 | 4746 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4747 | |
4748 | event->eventfd = eventfd; | |
4749 | list_add(&event->list, &memcg->oom_notify); | |
4750 | ||
4751 | /* already in OOM ? */ | |
79dfdacc | 4752 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4753 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4754 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4755 | |
4756 | return 0; | |
4757 | } | |
4758 | ||
59b6f873 | 4759 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4760 | struct eventfd_ctx *eventfd) |
9490ff27 | 4761 | { |
9490ff27 | 4762 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4763 | |
1af8efe9 | 4764 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4765 | |
c0ff4b85 | 4766 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4767 | if (ev->eventfd == eventfd) { |
4768 | list_del(&ev->list); | |
4769 | kfree(ev); | |
4770 | } | |
4771 | } | |
4772 | ||
1af8efe9 | 4773 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4774 | } |
4775 | ||
2da8ca82 | 4776 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4777 | { |
2da8ca82 | 4778 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4779 | |
791badbd TH |
4780 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4781 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4782 | return 0; |
4783 | } | |
4784 | ||
182446d0 | 4785 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4786 | struct cftype *cft, u64 val) |
4787 | { | |
182446d0 | 4788 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4789 | |
4790 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4791 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4792 | return -EINVAL; |
4793 | ||
c0ff4b85 | 4794 | memcg->oom_kill_disable = val; |
4d845ebf | 4795 | if (!val) |
c0ff4b85 | 4796 | memcg_oom_recover(memcg); |
3dae7fec | 4797 | |
3c11ecf4 KH |
4798 | return 0; |
4799 | } | |
4800 | ||
c255a458 | 4801 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4802 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4803 | { |
55007d84 GC |
4804 | int ret; |
4805 | ||
2633d7a0 | 4806 | memcg->kmemcg_id = -1; |
55007d84 GC |
4807 | ret = memcg_propagate_kmem(memcg); |
4808 | if (ret) | |
4809 | return ret; | |
2633d7a0 | 4810 | |
1d62e436 | 4811 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4812 | } |
e5671dfa | 4813 | |
10d5ebf4 | 4814 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4815 | { |
1d62e436 | 4816 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
4817 | } |
4818 | ||
4819 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
4820 | { | |
4821 | if (!memcg_kmem_is_active(memcg)) | |
4822 | return; | |
4823 | ||
4824 | /* | |
4825 | * kmem charges can outlive the cgroup. In the case of slab | |
4826 | * pages, for instance, a page contain objects from various | |
4827 | * processes. As we prevent from taking a reference for every | |
4828 | * such allocation we have to be careful when doing uncharge | |
4829 | * (see memcg_uncharge_kmem) and here during offlining. | |
4830 | * | |
4831 | * The idea is that that only the _last_ uncharge which sees | |
4832 | * the dead memcg will drop the last reference. An additional | |
4833 | * reference is taken here before the group is marked dead | |
4834 | * which is then paired with css_put during uncharge resp. here. | |
4835 | * | |
4836 | * Although this might sound strange as this path is called from | |
ec903c0c TH |
4837 | * css_offline() when the referencemight have dropped down to 0 and |
4838 | * shouldn't be incremented anymore (css_tryget_online() would | |
4839 | * fail) we do not have other options because of the kmem | |
4840 | * allocations lifetime. | |
10d5ebf4 LZ |
4841 | */ |
4842 | css_get(&memcg->css); | |
7de37682 GC |
4843 | |
4844 | memcg_kmem_mark_dead(memcg); | |
4845 | ||
3e32cb2e | 4846 | if (page_counter_read(&memcg->kmem)) |
7de37682 GC |
4847 | return; |
4848 | ||
7de37682 | 4849 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 4850 | css_put(&memcg->css); |
d1a4c0b3 | 4851 | } |
e5671dfa | 4852 | #else |
cbe128e3 | 4853 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4854 | { |
4855 | return 0; | |
4856 | } | |
d1a4c0b3 | 4857 | |
10d5ebf4 LZ |
4858 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4859 | { | |
4860 | } | |
4861 | ||
4862 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
4863 | { |
4864 | } | |
e5671dfa GC |
4865 | #endif |
4866 | ||
3bc942f3 TH |
4867 | /* |
4868 | * DO NOT USE IN NEW FILES. | |
4869 | * | |
4870 | * "cgroup.event_control" implementation. | |
4871 | * | |
4872 | * This is way over-engineered. It tries to support fully configurable | |
4873 | * events for each user. Such level of flexibility is completely | |
4874 | * unnecessary especially in the light of the planned unified hierarchy. | |
4875 | * | |
4876 | * Please deprecate this and replace with something simpler if at all | |
4877 | * possible. | |
4878 | */ | |
4879 | ||
79bd9814 TH |
4880 | /* |
4881 | * Unregister event and free resources. | |
4882 | * | |
4883 | * Gets called from workqueue. | |
4884 | */ | |
3bc942f3 | 4885 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4886 | { |
3bc942f3 TH |
4887 | struct mem_cgroup_event *event = |
4888 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4889 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4890 | |
4891 | remove_wait_queue(event->wqh, &event->wait); | |
4892 | ||
59b6f873 | 4893 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4894 | |
4895 | /* Notify userspace the event is going away. */ | |
4896 | eventfd_signal(event->eventfd, 1); | |
4897 | ||
4898 | eventfd_ctx_put(event->eventfd); | |
4899 | kfree(event); | |
59b6f873 | 4900 | css_put(&memcg->css); |
79bd9814 TH |
4901 | } |
4902 | ||
4903 | /* | |
4904 | * Gets called on POLLHUP on eventfd when user closes it. | |
4905 | * | |
4906 | * Called with wqh->lock held and interrupts disabled. | |
4907 | */ | |
3bc942f3 TH |
4908 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4909 | int sync, void *key) | |
79bd9814 | 4910 | { |
3bc942f3 TH |
4911 | struct mem_cgroup_event *event = |
4912 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4913 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4914 | unsigned long flags = (unsigned long)key; |
4915 | ||
4916 | if (flags & POLLHUP) { | |
4917 | /* | |
4918 | * If the event has been detached at cgroup removal, we | |
4919 | * can simply return knowing the other side will cleanup | |
4920 | * for us. | |
4921 | * | |
4922 | * We can't race against event freeing since the other | |
4923 | * side will require wqh->lock via remove_wait_queue(), | |
4924 | * which we hold. | |
4925 | */ | |
fba94807 | 4926 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4927 | if (!list_empty(&event->list)) { |
4928 | list_del_init(&event->list); | |
4929 | /* | |
4930 | * We are in atomic context, but cgroup_event_remove() | |
4931 | * may sleep, so we have to call it in workqueue. | |
4932 | */ | |
4933 | schedule_work(&event->remove); | |
4934 | } | |
fba94807 | 4935 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4936 | } |
4937 | ||
4938 | return 0; | |
4939 | } | |
4940 | ||
3bc942f3 | 4941 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4942 | wait_queue_head_t *wqh, poll_table *pt) |
4943 | { | |
3bc942f3 TH |
4944 | struct mem_cgroup_event *event = |
4945 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4946 | |
4947 | event->wqh = wqh; | |
4948 | add_wait_queue(wqh, &event->wait); | |
4949 | } | |
4950 | ||
4951 | /* | |
3bc942f3 TH |
4952 | * DO NOT USE IN NEW FILES. |
4953 | * | |
79bd9814 TH |
4954 | * Parse input and register new cgroup event handler. |
4955 | * | |
4956 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4957 | * Interpretation of args is defined by control file implementation. | |
4958 | */ | |
451af504 TH |
4959 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4960 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4961 | { |
451af504 | 4962 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4963 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4964 | struct mem_cgroup_event *event; |
79bd9814 TH |
4965 | struct cgroup_subsys_state *cfile_css; |
4966 | unsigned int efd, cfd; | |
4967 | struct fd efile; | |
4968 | struct fd cfile; | |
fba94807 | 4969 | const char *name; |
79bd9814 TH |
4970 | char *endp; |
4971 | int ret; | |
4972 | ||
451af504 TH |
4973 | buf = strstrip(buf); |
4974 | ||
4975 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4976 | if (*endp != ' ') |
4977 | return -EINVAL; | |
451af504 | 4978 | buf = endp + 1; |
79bd9814 | 4979 | |
451af504 | 4980 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4981 | if ((*endp != ' ') && (*endp != '\0')) |
4982 | return -EINVAL; | |
451af504 | 4983 | buf = endp + 1; |
79bd9814 TH |
4984 | |
4985 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4986 | if (!event) | |
4987 | return -ENOMEM; | |
4988 | ||
59b6f873 | 4989 | event->memcg = memcg; |
79bd9814 | 4990 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4991 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4992 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4993 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4994 | |
4995 | efile = fdget(efd); | |
4996 | if (!efile.file) { | |
4997 | ret = -EBADF; | |
4998 | goto out_kfree; | |
4999 | } | |
5000 | ||
5001 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
5002 | if (IS_ERR(event->eventfd)) { | |
5003 | ret = PTR_ERR(event->eventfd); | |
5004 | goto out_put_efile; | |
5005 | } | |
5006 | ||
5007 | cfile = fdget(cfd); | |
5008 | if (!cfile.file) { | |
5009 | ret = -EBADF; | |
5010 | goto out_put_eventfd; | |
5011 | } | |
5012 | ||
5013 | /* the process need read permission on control file */ | |
5014 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
5015 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
5016 | if (ret < 0) | |
5017 | goto out_put_cfile; | |
5018 | ||
fba94807 TH |
5019 | /* |
5020 | * Determine the event callbacks and set them in @event. This used | |
5021 | * to be done via struct cftype but cgroup core no longer knows | |
5022 | * about these events. The following is crude but the whole thing | |
5023 | * is for compatibility anyway. | |
3bc942f3 TH |
5024 | * |
5025 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
5026 | */ |
5027 | name = cfile.file->f_dentry->d_name.name; | |
5028 | ||
5029 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
5030 | event->register_event = mem_cgroup_usage_register_event; | |
5031 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
5032 | } else if (!strcmp(name, "memory.oom_control")) { | |
5033 | event->register_event = mem_cgroup_oom_register_event; | |
5034 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
5035 | } else if (!strcmp(name, "memory.pressure_level")) { | |
5036 | event->register_event = vmpressure_register_event; | |
5037 | event->unregister_event = vmpressure_unregister_event; | |
5038 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
5039 | event->register_event = memsw_cgroup_usage_register_event; |
5040 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
5041 | } else { |
5042 | ret = -EINVAL; | |
5043 | goto out_put_cfile; | |
5044 | } | |
5045 | ||
79bd9814 | 5046 | /* |
b5557c4c TH |
5047 | * Verify @cfile should belong to @css. Also, remaining events are |
5048 | * automatically removed on cgroup destruction but the removal is | |
5049 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 5050 | */ |
ec903c0c TH |
5051 | cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent, |
5052 | &memory_cgrp_subsys); | |
79bd9814 | 5053 | ret = -EINVAL; |
5a17f543 | 5054 | if (IS_ERR(cfile_css)) |
79bd9814 | 5055 | goto out_put_cfile; |
5a17f543 TH |
5056 | if (cfile_css != css) { |
5057 | css_put(cfile_css); | |
79bd9814 | 5058 | goto out_put_cfile; |
5a17f543 | 5059 | } |
79bd9814 | 5060 | |
451af504 | 5061 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
5062 | if (ret) |
5063 | goto out_put_css; | |
5064 | ||
5065 | efile.file->f_op->poll(efile.file, &event->pt); | |
5066 | ||
fba94807 TH |
5067 | spin_lock(&memcg->event_list_lock); |
5068 | list_add(&event->list, &memcg->event_list); | |
5069 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
5070 | |
5071 | fdput(cfile); | |
5072 | fdput(efile); | |
5073 | ||
451af504 | 5074 | return nbytes; |
79bd9814 TH |
5075 | |
5076 | out_put_css: | |
b5557c4c | 5077 | css_put(css); |
79bd9814 TH |
5078 | out_put_cfile: |
5079 | fdput(cfile); | |
5080 | out_put_eventfd: | |
5081 | eventfd_ctx_put(event->eventfd); | |
5082 | out_put_efile: | |
5083 | fdput(efile); | |
5084 | out_kfree: | |
5085 | kfree(event); | |
5086 | ||
5087 | return ret; | |
5088 | } | |
5089 | ||
8cdea7c0 BS |
5090 | static struct cftype mem_cgroup_files[] = { |
5091 | { | |
0eea1030 | 5092 | .name = "usage_in_bytes", |
8c7c6e34 | 5093 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 5094 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5095 | }, |
c84872e1 PE |
5096 | { |
5097 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5098 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5099 | .write = mem_cgroup_reset, |
791badbd | 5100 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5101 | }, |
8cdea7c0 | 5102 | { |
0eea1030 | 5103 | .name = "limit_in_bytes", |
8c7c6e34 | 5104 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5105 | .write = mem_cgroup_write, |
791badbd | 5106 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5107 | }, |
296c81d8 BS |
5108 | { |
5109 | .name = "soft_limit_in_bytes", | |
5110 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5111 | .write = mem_cgroup_write, |
791badbd | 5112 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5113 | }, |
8cdea7c0 BS |
5114 | { |
5115 | .name = "failcnt", | |
8c7c6e34 | 5116 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5117 | .write = mem_cgroup_reset, |
791badbd | 5118 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5119 | }, |
d2ceb9b7 KH |
5120 | { |
5121 | .name = "stat", | |
2da8ca82 | 5122 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5123 | }, |
c1e862c1 KH |
5124 | { |
5125 | .name = "force_empty", | |
6770c64e | 5126 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5127 | }, |
18f59ea7 BS |
5128 | { |
5129 | .name = "use_hierarchy", | |
5130 | .write_u64 = mem_cgroup_hierarchy_write, | |
5131 | .read_u64 = mem_cgroup_hierarchy_read, | |
5132 | }, | |
79bd9814 | 5133 | { |
3bc942f3 | 5134 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5135 | .write = memcg_write_event_control, |
79bd9814 TH |
5136 | .flags = CFTYPE_NO_PREFIX, |
5137 | .mode = S_IWUGO, | |
5138 | }, | |
a7885eb8 KM |
5139 | { |
5140 | .name = "swappiness", | |
5141 | .read_u64 = mem_cgroup_swappiness_read, | |
5142 | .write_u64 = mem_cgroup_swappiness_write, | |
5143 | }, | |
7dc74be0 DN |
5144 | { |
5145 | .name = "move_charge_at_immigrate", | |
5146 | .read_u64 = mem_cgroup_move_charge_read, | |
5147 | .write_u64 = mem_cgroup_move_charge_write, | |
5148 | }, | |
9490ff27 KH |
5149 | { |
5150 | .name = "oom_control", | |
2da8ca82 | 5151 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5152 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5153 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5154 | }, | |
70ddf637 AV |
5155 | { |
5156 | .name = "pressure_level", | |
70ddf637 | 5157 | }, |
406eb0c9 YH |
5158 | #ifdef CONFIG_NUMA |
5159 | { | |
5160 | .name = "numa_stat", | |
2da8ca82 | 5161 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5162 | }, |
5163 | #endif | |
510fc4e1 GC |
5164 | #ifdef CONFIG_MEMCG_KMEM |
5165 | { | |
5166 | .name = "kmem.limit_in_bytes", | |
5167 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5168 | .write = mem_cgroup_write, |
791badbd | 5169 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5170 | }, |
5171 | { | |
5172 | .name = "kmem.usage_in_bytes", | |
5173 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5174 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5175 | }, |
5176 | { | |
5177 | .name = "kmem.failcnt", | |
5178 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5179 | .write = mem_cgroup_reset, |
791badbd | 5180 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5181 | }, |
5182 | { | |
5183 | .name = "kmem.max_usage_in_bytes", | |
5184 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5185 | .write = mem_cgroup_reset, |
791badbd | 5186 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5187 | }, |
749c5415 GC |
5188 | #ifdef CONFIG_SLABINFO |
5189 | { | |
5190 | .name = "kmem.slabinfo", | |
2da8ca82 | 5191 | .seq_show = mem_cgroup_slabinfo_read, |
749c5415 GC |
5192 | }, |
5193 | #endif | |
8c7c6e34 | 5194 | #endif |
6bc10349 | 5195 | { }, /* terminate */ |
af36f906 | 5196 | }; |
8c7c6e34 | 5197 | |
2d11085e MH |
5198 | #ifdef CONFIG_MEMCG_SWAP |
5199 | static struct cftype memsw_cgroup_files[] = { | |
5200 | { | |
5201 | .name = "memsw.usage_in_bytes", | |
5202 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 5203 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5204 | }, |
5205 | { | |
5206 | .name = "memsw.max_usage_in_bytes", | |
5207 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 5208 | .write = mem_cgroup_reset, |
791badbd | 5209 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5210 | }, |
5211 | { | |
5212 | .name = "memsw.limit_in_bytes", | |
5213 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 5214 | .write = mem_cgroup_write, |
791badbd | 5215 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5216 | }, |
5217 | { | |
5218 | .name = "memsw.failcnt", | |
5219 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 5220 | .write = mem_cgroup_reset, |
791badbd | 5221 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5222 | }, |
5223 | { }, /* terminate */ | |
5224 | }; | |
5225 | #endif | |
c0ff4b85 | 5226 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5227 | { |
5228 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 5229 | struct mem_cgroup_per_zone *mz; |
41e3355d | 5230 | int zone, tmp = node; |
1ecaab2b KH |
5231 | /* |
5232 | * This routine is called against possible nodes. | |
5233 | * But it's BUG to call kmalloc() against offline node. | |
5234 | * | |
5235 | * TODO: this routine can waste much memory for nodes which will | |
5236 | * never be onlined. It's better to use memory hotplug callback | |
5237 | * function. | |
5238 | */ | |
41e3355d KH |
5239 | if (!node_state(node, N_NORMAL_MEMORY)) |
5240 | tmp = -1; | |
17295c88 | 5241 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5242 | if (!pn) |
5243 | return 1; | |
1ecaab2b | 5244 | |
1ecaab2b KH |
5245 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
5246 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 5247 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
5248 | mz->usage_in_excess = 0; |
5249 | mz->on_tree = false; | |
d79154bb | 5250 | mz->memcg = memcg; |
1ecaab2b | 5251 | } |
54f72fe0 | 5252 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5253 | return 0; |
5254 | } | |
5255 | ||
c0ff4b85 | 5256 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5257 | { |
54f72fe0 | 5258 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
5259 | } |
5260 | ||
33327948 KH |
5261 | static struct mem_cgroup *mem_cgroup_alloc(void) |
5262 | { | |
d79154bb | 5263 | struct mem_cgroup *memcg; |
8ff69e2c | 5264 | size_t size; |
33327948 | 5265 | |
8ff69e2c VD |
5266 | size = sizeof(struct mem_cgroup); |
5267 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 5268 | |
8ff69e2c | 5269 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 5270 | if (!memcg) |
e7bbcdf3 DC |
5271 | return NULL; |
5272 | ||
d79154bb HD |
5273 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
5274 | if (!memcg->stat) | |
d2e61b8d | 5275 | goto out_free; |
d79154bb HD |
5276 | spin_lock_init(&memcg->pcp_counter_lock); |
5277 | return memcg; | |
d2e61b8d DC |
5278 | |
5279 | out_free: | |
8ff69e2c | 5280 | kfree(memcg); |
d2e61b8d | 5281 | return NULL; |
33327948 KH |
5282 | } |
5283 | ||
59927fb9 | 5284 | /* |
c8b2a36f GC |
5285 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
5286 | * (scanning all at force_empty is too costly...) | |
5287 | * | |
5288 | * Instead of clearing all references at force_empty, we remember | |
5289 | * the number of reference from swap_cgroup and free mem_cgroup when | |
5290 | * it goes down to 0. | |
5291 | * | |
5292 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 5293 | */ |
c8b2a36f GC |
5294 | |
5295 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 5296 | { |
c8b2a36f | 5297 | int node; |
59927fb9 | 5298 | |
bb4cc1a8 | 5299 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
5300 | |
5301 | for_each_node(node) | |
5302 | free_mem_cgroup_per_zone_info(memcg, node); | |
5303 | ||
5304 | free_percpu(memcg->stat); | |
5305 | ||
3f134619 GC |
5306 | /* |
5307 | * We need to make sure that (at least for now), the jump label | |
5308 | * destruction code runs outside of the cgroup lock. This is because | |
5309 | * get_online_cpus(), which is called from the static_branch update, | |
5310 | * can't be called inside the cgroup_lock. cpusets are the ones | |
5311 | * enforcing this dependency, so if they ever change, we might as well. | |
5312 | * | |
5313 | * schedule_work() will guarantee this happens. Be careful if you need | |
5314 | * to move this code around, and make sure it is outside | |
5315 | * the cgroup_lock. | |
5316 | */ | |
a8964b9b | 5317 | disarm_static_keys(memcg); |
8ff69e2c | 5318 | kfree(memcg); |
59927fb9 | 5319 | } |
3afe36b1 | 5320 | |
7bcc1bb1 DN |
5321 | /* |
5322 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
5323 | */ | |
e1aab161 | 5324 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 5325 | { |
3e32cb2e | 5326 | if (!memcg->memory.parent) |
7bcc1bb1 | 5327 | return NULL; |
3e32cb2e | 5328 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 5329 | } |
e1aab161 | 5330 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 5331 | |
bb4cc1a8 AM |
5332 | static void __init mem_cgroup_soft_limit_tree_init(void) |
5333 | { | |
5334 | struct mem_cgroup_tree_per_node *rtpn; | |
5335 | struct mem_cgroup_tree_per_zone *rtpz; | |
5336 | int tmp, node, zone; | |
5337 | ||
5338 | for_each_node(node) { | |
5339 | tmp = node; | |
5340 | if (!node_state(node, N_NORMAL_MEMORY)) | |
5341 | tmp = -1; | |
5342 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
5343 | BUG_ON(!rtpn); | |
5344 | ||
5345 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5346 | ||
5347 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5348 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5349 | rtpz->rb_root = RB_ROOT; | |
5350 | spin_lock_init(&rtpz->lock); | |
5351 | } | |
5352 | } | |
5353 | } | |
5354 | ||
0eb253e2 | 5355 | static struct cgroup_subsys_state * __ref |
eb95419b | 5356 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 5357 | { |
d142e3e6 | 5358 | struct mem_cgroup *memcg; |
04046e1a | 5359 | long error = -ENOMEM; |
6d12e2d8 | 5360 | int node; |
8cdea7c0 | 5361 | |
c0ff4b85 R |
5362 | memcg = mem_cgroup_alloc(); |
5363 | if (!memcg) | |
04046e1a | 5364 | return ERR_PTR(error); |
78fb7466 | 5365 | |
3ed28fa1 | 5366 | for_each_node(node) |
c0ff4b85 | 5367 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5368 | goto free_out; |
f64c3f54 | 5369 | |
c077719b | 5370 | /* root ? */ |
eb95419b | 5371 | if (parent_css == NULL) { |
a41c58a6 | 5372 | root_mem_cgroup = memcg; |
3e32cb2e JW |
5373 | page_counter_init(&memcg->memory, NULL); |
5374 | page_counter_init(&memcg->memsw, NULL); | |
5375 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 5376 | } |
28dbc4b6 | 5377 | |
d142e3e6 GC |
5378 | memcg->last_scanned_node = MAX_NUMNODES; |
5379 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
5380 | memcg->move_charge_at_immigrate = 0; |
5381 | mutex_init(&memcg->thresholds_lock); | |
5382 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5383 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5384 | INIT_LIST_HEAD(&memcg->event_list); |
5385 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
5386 | |
5387 | return &memcg->css; | |
5388 | ||
5389 | free_out: | |
5390 | __mem_cgroup_free(memcg); | |
5391 | return ERR_PTR(error); | |
5392 | } | |
5393 | ||
5394 | static int | |
eb95419b | 5395 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 5396 | { |
eb95419b | 5397 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 5398 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 5399 | int ret; |
d142e3e6 | 5400 | |
15a4c835 | 5401 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
5402 | return -ENOSPC; |
5403 | ||
63876986 | 5404 | if (!parent) |
d142e3e6 GC |
5405 | return 0; |
5406 | ||
0999821b | 5407 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
5408 | |
5409 | memcg->use_hierarchy = parent->use_hierarchy; | |
5410 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5411 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5412 | ||
5413 | if (parent->use_hierarchy) { | |
3e32cb2e JW |
5414 | page_counter_init(&memcg->memory, &parent->memory); |
5415 | page_counter_init(&memcg->memsw, &parent->memsw); | |
5416 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 5417 | |
7bcc1bb1 | 5418 | /* |
8d76a979 LZ |
5419 | * No need to take a reference to the parent because cgroup |
5420 | * core guarantees its existence. | |
7bcc1bb1 | 5421 | */ |
18f59ea7 | 5422 | } else { |
3e32cb2e JW |
5423 | page_counter_init(&memcg->memory, NULL); |
5424 | page_counter_init(&memcg->memsw, NULL); | |
5425 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
5426 | /* |
5427 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5428 | * much sense so let cgroup subsystem know about this | |
5429 | * unfortunate state in our controller. | |
5430 | */ | |
d142e3e6 | 5431 | if (parent != root_mem_cgroup) |
073219e9 | 5432 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5433 | } |
0999821b | 5434 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 5435 | |
2f7dd7a4 JW |
5436 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
5437 | if (ret) | |
5438 | return ret; | |
5439 | ||
5440 | /* | |
5441 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
5442 | * orders reading memcg->initialized against its callers | |
5443 | * reading the memcg members. | |
5444 | */ | |
5445 | smp_store_release(&memcg->initialized, 1); | |
5446 | ||
5447 | return 0; | |
8cdea7c0 BS |
5448 | } |
5449 | ||
5f578161 MH |
5450 | /* |
5451 | * Announce all parents that a group from their hierarchy is gone. | |
5452 | */ | |
5453 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
5454 | { | |
5455 | struct mem_cgroup *parent = memcg; | |
5456 | ||
5457 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 5458 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
5459 | |
5460 | /* | |
5461 | * if the root memcg is not hierarchical we have to check it | |
5462 | * explicitely. | |
5463 | */ | |
5464 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 5465 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
5466 | } |
5467 | ||
eb95419b | 5468 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5469 | { |
eb95419b | 5470 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5471 | struct mem_cgroup_event *event, *tmp; |
4fb1a86f | 5472 | struct cgroup_subsys_state *iter; |
79bd9814 TH |
5473 | |
5474 | /* | |
5475 | * Unregister events and notify userspace. | |
5476 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5477 | * directory to avoid race between userspace and kernelspace. | |
5478 | */ | |
fba94807 TH |
5479 | spin_lock(&memcg->event_list_lock); |
5480 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5481 | list_del_init(&event->list); |
5482 | schedule_work(&event->remove); | |
5483 | } | |
fba94807 | 5484 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5485 | |
10d5ebf4 LZ |
5486 | kmem_cgroup_css_offline(memcg); |
5487 | ||
5f578161 | 5488 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
4fb1a86f FB |
5489 | |
5490 | /* | |
5491 | * This requires that offlining is serialized. Right now that is | |
5492 | * guaranteed because css_killed_work_fn() holds the cgroup_mutex. | |
5493 | */ | |
5494 | css_for_each_descendant_post(iter, css) | |
5495 | mem_cgroup_reparent_charges(mem_cgroup_from_css(iter)); | |
5496 | ||
776ed0f0 | 5497 | memcg_unregister_all_caches(memcg); |
33cb876e | 5498 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
5499 | } |
5500 | ||
eb95419b | 5501 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5502 | { |
eb95419b | 5503 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
96f1c58d JW |
5504 | /* |
5505 | * XXX: css_offline() would be where we should reparent all | |
5506 | * memory to prepare the cgroup for destruction. However, | |
3e32cb2e | 5507 | * memcg does not do css_tryget_online() and page_counter charging |
96f1c58d JW |
5508 | * under the same RCU lock region, which means that charging |
5509 | * could race with offlining. Offlining only happens to | |
5510 | * cgroups with no tasks in them but charges can show up | |
5511 | * without any tasks from the swapin path when the target | |
5512 | * memcg is looked up from the swapout record and not from the | |
5513 | * current task as it usually is. A race like this can leak | |
5514 | * charges and put pages with stale cgroup pointers into | |
5515 | * circulation: | |
5516 | * | |
5517 | * #0 #1 | |
5518 | * lookup_swap_cgroup_id() | |
5519 | * rcu_read_lock() | |
5520 | * mem_cgroup_lookup() | |
ec903c0c | 5521 | * css_tryget_online() |
96f1c58d | 5522 | * rcu_read_unlock() |
ec903c0c | 5523 | * disable css_tryget_online() |
96f1c58d JW |
5524 | * call_rcu() |
5525 | * offline_css() | |
5526 | * reparent_charges() | |
3e32cb2e | 5527 | * page_counter_try_charge() |
96f1c58d JW |
5528 | * css_put() |
5529 | * css_free() | |
5530 | * pc->mem_cgroup = dead memcg | |
5531 | * add page to lru | |
5532 | * | |
5533 | * The bulk of the charges are still moved in offline_css() to | |
5534 | * avoid pinning a lot of pages in case a long-term reference | |
5535 | * like a swapout record is deferring the css_free() to long | |
5536 | * after offlining. But this makes sure we catch any charges | |
5537 | * made after offlining: | |
5538 | */ | |
5539 | mem_cgroup_reparent_charges(memcg); | |
c268e994 | 5540 | |
10d5ebf4 | 5541 | memcg_destroy_kmem(memcg); |
465939a1 | 5542 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
5543 | } |
5544 | ||
1ced953b TH |
5545 | /** |
5546 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5547 | * @css: the target css | |
5548 | * | |
5549 | * Reset the states of the mem_cgroup associated with @css. This is | |
5550 | * invoked when the userland requests disabling on the default hierarchy | |
5551 | * but the memcg is pinned through dependency. The memcg should stop | |
5552 | * applying policies and should revert to the vanilla state as it may be | |
5553 | * made visible again. | |
5554 | * | |
5555 | * The current implementation only resets the essential configurations. | |
5556 | * This needs to be expanded to cover all the visible parts. | |
5557 | */ | |
5558 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5559 | { | |
5560 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5561 | ||
3e32cb2e JW |
5562 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
5563 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
5564 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
5565 | memcg->soft_limit = 0; | |
1ced953b TH |
5566 | } |
5567 | ||
02491447 | 5568 | #ifdef CONFIG_MMU |
7dc74be0 | 5569 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5570 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5571 | { |
05b84301 | 5572 | int ret; |
9476db97 JW |
5573 | |
5574 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 5575 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 5576 | if (!ret) { |
854ffa8d | 5577 | mc.precharge += count; |
854ffa8d DN |
5578 | return ret; |
5579 | } | |
692e7c45 | 5580 | if (ret == -EINTR) { |
00501b53 | 5581 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
5582 | return ret; |
5583 | } | |
9476db97 JW |
5584 | |
5585 | /* Try charges one by one with reclaim */ | |
854ffa8d | 5586 | while (count--) { |
00501b53 | 5587 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
5588 | /* |
5589 | * In case of failure, any residual charges against | |
5590 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
5591 | * later on. However, cancel any charges that are |
5592 | * bypassed to root right away or they'll be lost. | |
9476db97 | 5593 | */ |
692e7c45 | 5594 | if (ret == -EINTR) |
00501b53 | 5595 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 5596 | if (ret) |
38c5d72f | 5597 | return ret; |
854ffa8d | 5598 | mc.precharge++; |
9476db97 | 5599 | cond_resched(); |
854ffa8d | 5600 | } |
9476db97 | 5601 | return 0; |
4ffef5fe DN |
5602 | } |
5603 | ||
5604 | /** | |
8d32ff84 | 5605 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5606 | * @vma: the vma the pte to be checked belongs |
5607 | * @addr: the address corresponding to the pte to be checked | |
5608 | * @ptent: the pte to be checked | |
02491447 | 5609 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5610 | * |
5611 | * Returns | |
5612 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5613 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5614 | * move charge. if @target is not NULL, the page is stored in target->page | |
5615 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5616 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5617 | * target for charge migration. if @target is not NULL, the entry is stored | |
5618 | * in target->ent. | |
4ffef5fe DN |
5619 | * |
5620 | * Called with pte lock held. | |
5621 | */ | |
4ffef5fe DN |
5622 | union mc_target { |
5623 | struct page *page; | |
02491447 | 5624 | swp_entry_t ent; |
4ffef5fe DN |
5625 | }; |
5626 | ||
4ffef5fe | 5627 | enum mc_target_type { |
8d32ff84 | 5628 | MC_TARGET_NONE = 0, |
4ffef5fe | 5629 | MC_TARGET_PAGE, |
02491447 | 5630 | MC_TARGET_SWAP, |
4ffef5fe DN |
5631 | }; |
5632 | ||
90254a65 DN |
5633 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5634 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5635 | { |
90254a65 | 5636 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5637 | |
90254a65 DN |
5638 | if (!page || !page_mapped(page)) |
5639 | return NULL; | |
5640 | if (PageAnon(page)) { | |
5641 | /* we don't move shared anon */ | |
4b91355e | 5642 | if (!move_anon()) |
90254a65 | 5643 | return NULL; |
87946a72 DN |
5644 | } else if (!move_file()) |
5645 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5646 | return NULL; |
5647 | if (!get_page_unless_zero(page)) | |
5648 | return NULL; | |
5649 | ||
5650 | return page; | |
5651 | } | |
5652 | ||
4b91355e | 5653 | #ifdef CONFIG_SWAP |
90254a65 DN |
5654 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5655 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5656 | { | |
90254a65 DN |
5657 | struct page *page = NULL; |
5658 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5659 | ||
5660 | if (!move_anon() || non_swap_entry(ent)) | |
5661 | return NULL; | |
4b91355e KH |
5662 | /* |
5663 | * Because lookup_swap_cache() updates some statistics counter, | |
5664 | * we call find_get_page() with swapper_space directly. | |
5665 | */ | |
33806f06 | 5666 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
5667 | if (do_swap_account) |
5668 | entry->val = ent.val; | |
5669 | ||
5670 | return page; | |
5671 | } | |
4b91355e KH |
5672 | #else |
5673 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5674 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5675 | { | |
5676 | return NULL; | |
5677 | } | |
5678 | #endif | |
90254a65 | 5679 | |
87946a72 DN |
5680 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5681 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5682 | { | |
5683 | struct page *page = NULL; | |
87946a72 DN |
5684 | struct address_space *mapping; |
5685 | pgoff_t pgoff; | |
5686 | ||
5687 | if (!vma->vm_file) /* anonymous vma */ | |
5688 | return NULL; | |
5689 | if (!move_file()) | |
5690 | return NULL; | |
5691 | ||
87946a72 DN |
5692 | mapping = vma->vm_file->f_mapping; |
5693 | if (pte_none(ptent)) | |
5694 | pgoff = linear_page_index(vma, addr); | |
5695 | else /* pte_file(ptent) is true */ | |
5696 | pgoff = pte_to_pgoff(ptent); | |
5697 | ||
5698 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5699 | #ifdef CONFIG_SWAP |
5700 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5701 | if (shmem_mapping(mapping)) { |
5702 | page = find_get_entry(mapping, pgoff); | |
5703 | if (radix_tree_exceptional_entry(page)) { | |
5704 | swp_entry_t swp = radix_to_swp_entry(page); | |
5705 | if (do_swap_account) | |
5706 | *entry = swp; | |
5707 | page = find_get_page(swap_address_space(swp), swp.val); | |
5708 | } | |
5709 | } else | |
5710 | page = find_get_page(mapping, pgoff); | |
5711 | #else | |
5712 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5713 | #endif |
87946a72 DN |
5714 | return page; |
5715 | } | |
5716 | ||
8d32ff84 | 5717 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5718 | unsigned long addr, pte_t ptent, union mc_target *target) |
5719 | { | |
5720 | struct page *page = NULL; | |
5721 | struct page_cgroup *pc; | |
8d32ff84 | 5722 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5723 | swp_entry_t ent = { .val = 0 }; |
5724 | ||
5725 | if (pte_present(ptent)) | |
5726 | page = mc_handle_present_pte(vma, addr, ptent); | |
5727 | else if (is_swap_pte(ptent)) | |
5728 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5729 | else if (pte_none(ptent) || pte_file(ptent)) |
5730 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5731 | |
5732 | if (!page && !ent.val) | |
8d32ff84 | 5733 | return ret; |
02491447 DN |
5734 | if (page) { |
5735 | pc = lookup_page_cgroup(page); | |
5736 | /* | |
0a31bc97 JW |
5737 | * Do only loose check w/o serialization. |
5738 | * mem_cgroup_move_account() checks the pc is valid or | |
5739 | * not under LRU exclusion. | |
02491447 DN |
5740 | */ |
5741 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5742 | ret = MC_TARGET_PAGE; | |
5743 | if (target) | |
5744 | target->page = page; | |
5745 | } | |
5746 | if (!ret || !target) | |
5747 | put_page(page); | |
5748 | } | |
90254a65 DN |
5749 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5750 | if (ent.val && !ret && | |
34c00c31 | 5751 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5752 | ret = MC_TARGET_SWAP; |
5753 | if (target) | |
5754 | target->ent = ent; | |
4ffef5fe | 5755 | } |
4ffef5fe DN |
5756 | return ret; |
5757 | } | |
5758 | ||
12724850 NH |
5759 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5760 | /* | |
5761 | * We don't consider swapping or file mapped pages because THP does not | |
5762 | * support them for now. | |
5763 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5764 | */ | |
5765 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5766 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5767 | { | |
5768 | struct page *page = NULL; | |
5769 | struct page_cgroup *pc; | |
5770 | enum mc_target_type ret = MC_TARGET_NONE; | |
5771 | ||
5772 | page = pmd_page(pmd); | |
309381fe | 5773 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
5774 | if (!move_anon()) |
5775 | return ret; | |
5776 | pc = lookup_page_cgroup(page); | |
5777 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5778 | ret = MC_TARGET_PAGE; | |
5779 | if (target) { | |
5780 | get_page(page); | |
5781 | target->page = page; | |
5782 | } | |
5783 | } | |
5784 | return ret; | |
5785 | } | |
5786 | #else | |
5787 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5788 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5789 | { | |
5790 | return MC_TARGET_NONE; | |
5791 | } | |
5792 | #endif | |
5793 | ||
4ffef5fe DN |
5794 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5795 | unsigned long addr, unsigned long end, | |
5796 | struct mm_walk *walk) | |
5797 | { | |
5798 | struct vm_area_struct *vma = walk->private; | |
5799 | pte_t *pte; | |
5800 | spinlock_t *ptl; | |
5801 | ||
bf929152 | 5802 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
5803 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5804 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5805 | spin_unlock(ptl); |
1a5a9906 | 5806 | return 0; |
12724850 | 5807 | } |
03319327 | 5808 | |
45f83cef AA |
5809 | if (pmd_trans_unstable(pmd)) |
5810 | return 0; | |
4ffef5fe DN |
5811 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5812 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5813 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5814 | mc.precharge++; /* increment precharge temporarily */ |
5815 | pte_unmap_unlock(pte - 1, ptl); | |
5816 | cond_resched(); | |
5817 | ||
7dc74be0 DN |
5818 | return 0; |
5819 | } | |
5820 | ||
4ffef5fe DN |
5821 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5822 | { | |
5823 | unsigned long precharge; | |
5824 | struct vm_area_struct *vma; | |
5825 | ||
dfe076b0 | 5826 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5827 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5828 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5829 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5830 | .mm = mm, | |
5831 | .private = vma, | |
5832 | }; | |
5833 | if (is_vm_hugetlb_page(vma)) | |
5834 | continue; | |
4ffef5fe DN |
5835 | walk_page_range(vma->vm_start, vma->vm_end, |
5836 | &mem_cgroup_count_precharge_walk); | |
5837 | } | |
dfe076b0 | 5838 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5839 | |
5840 | precharge = mc.precharge; | |
5841 | mc.precharge = 0; | |
5842 | ||
5843 | return precharge; | |
5844 | } | |
5845 | ||
4ffef5fe DN |
5846 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5847 | { | |
dfe076b0 DN |
5848 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5849 | ||
5850 | VM_BUG_ON(mc.moving_task); | |
5851 | mc.moving_task = current; | |
5852 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5853 | } |
5854 | ||
dfe076b0 DN |
5855 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5856 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5857 | { |
2bd9bb20 KH |
5858 | struct mem_cgroup *from = mc.from; |
5859 | struct mem_cgroup *to = mc.to; | |
4050377b | 5860 | int i; |
2bd9bb20 | 5861 | |
4ffef5fe | 5862 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5863 | if (mc.precharge) { |
00501b53 | 5864 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5865 | mc.precharge = 0; |
5866 | } | |
5867 | /* | |
5868 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5869 | * we must uncharge here. | |
5870 | */ | |
5871 | if (mc.moved_charge) { | |
00501b53 | 5872 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5873 | mc.moved_charge = 0; |
4ffef5fe | 5874 | } |
483c30b5 DN |
5875 | /* we must fixup refcnts and charges */ |
5876 | if (mc.moved_swap) { | |
483c30b5 | 5877 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5878 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5879 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5880 | |
05b84301 | 5881 | /* |
3e32cb2e JW |
5882 | * we charged both to->memory and to->memsw, so we |
5883 | * should uncharge to->memory. | |
05b84301 | 5884 | */ |
ce00a967 | 5885 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5886 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5887 | ||
5888 | for (i = 0; i < mc.moved_swap; i++) | |
5889 | css_put(&mc.from->css); | |
5890 | ||
4050377b | 5891 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
5892 | mc.moved_swap = 0; |
5893 | } | |
dfe076b0 DN |
5894 | memcg_oom_recover(from); |
5895 | memcg_oom_recover(to); | |
5896 | wake_up_all(&mc.waitq); | |
5897 | } | |
5898 | ||
5899 | static void mem_cgroup_clear_mc(void) | |
5900 | { | |
5901 | struct mem_cgroup *from = mc.from; | |
5902 | ||
5903 | /* | |
5904 | * we must clear moving_task before waking up waiters at the end of | |
5905 | * task migration. | |
5906 | */ | |
5907 | mc.moving_task = NULL; | |
5908 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5909 | spin_lock(&mc.lock); |
4ffef5fe DN |
5910 | mc.from = NULL; |
5911 | mc.to = NULL; | |
2bd9bb20 | 5912 | spin_unlock(&mc.lock); |
32047e2a | 5913 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5914 | } |
5915 | ||
eb95419b | 5916 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5917 | struct cgroup_taskset *tset) |
7dc74be0 | 5918 | { |
2f7ee569 | 5919 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5920 | int ret = 0; |
eb95419b | 5921 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 5922 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 5923 | |
ee5e8472 GC |
5924 | /* |
5925 | * We are now commited to this value whatever it is. Changes in this | |
5926 | * tunable will only affect upcoming migrations, not the current one. | |
5927 | * So we need to save it, and keep it going. | |
5928 | */ | |
5929 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
5930 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
5931 | struct mm_struct *mm; |
5932 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5933 | ||
c0ff4b85 | 5934 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5935 | |
5936 | mm = get_task_mm(p); | |
5937 | if (!mm) | |
5938 | return 0; | |
7dc74be0 | 5939 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5940 | if (mm->owner == p) { |
5941 | VM_BUG_ON(mc.from); | |
5942 | VM_BUG_ON(mc.to); | |
5943 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5944 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5945 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5946 | mem_cgroup_start_move(from); |
2bd9bb20 | 5947 | spin_lock(&mc.lock); |
4ffef5fe | 5948 | mc.from = from; |
c0ff4b85 | 5949 | mc.to = memcg; |
ee5e8472 | 5950 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 5951 | spin_unlock(&mc.lock); |
dfe076b0 | 5952 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5953 | |
5954 | ret = mem_cgroup_precharge_mc(mm); | |
5955 | if (ret) | |
5956 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5957 | } |
5958 | mmput(mm); | |
7dc74be0 DN |
5959 | } |
5960 | return ret; | |
5961 | } | |
5962 | ||
eb95419b | 5963 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5964 | struct cgroup_taskset *tset) |
7dc74be0 | 5965 | { |
4ffef5fe | 5966 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5967 | } |
5968 | ||
4ffef5fe DN |
5969 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5970 | unsigned long addr, unsigned long end, | |
5971 | struct mm_walk *walk) | |
7dc74be0 | 5972 | { |
4ffef5fe DN |
5973 | int ret = 0; |
5974 | struct vm_area_struct *vma = walk->private; | |
5975 | pte_t *pte; | |
5976 | spinlock_t *ptl; | |
12724850 NH |
5977 | enum mc_target_type target_type; |
5978 | union mc_target target; | |
5979 | struct page *page; | |
5980 | struct page_cgroup *pc; | |
4ffef5fe | 5981 | |
12724850 NH |
5982 | /* |
5983 | * We don't take compound_lock() here but no race with splitting thp | |
5984 | * happens because: | |
5985 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5986 | * under splitting, which means there's no concurrent thp split, | |
5987 | * - if another thread runs into split_huge_page() just after we | |
5988 | * entered this if-block, the thread must wait for page table lock | |
5989 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5990 | * part of thp split is not executed yet. | |
5991 | */ | |
bf929152 | 5992 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5993 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5994 | spin_unlock(ptl); |
12724850 NH |
5995 | return 0; |
5996 | } | |
5997 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5998 | if (target_type == MC_TARGET_PAGE) { | |
5999 | page = target.page; | |
6000 | if (!isolate_lru_page(page)) { | |
6001 | pc = lookup_page_cgroup(page); | |
6002 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6003 | pc, mc.from, mc.to)) { |
12724850 NH |
6004 | mc.precharge -= HPAGE_PMD_NR; |
6005 | mc.moved_charge += HPAGE_PMD_NR; | |
6006 | } | |
6007 | putback_lru_page(page); | |
6008 | } | |
6009 | put_page(page); | |
6010 | } | |
bf929152 | 6011 | spin_unlock(ptl); |
1a5a9906 | 6012 | return 0; |
12724850 NH |
6013 | } |
6014 | ||
45f83cef AA |
6015 | if (pmd_trans_unstable(pmd)) |
6016 | return 0; | |
4ffef5fe DN |
6017 | retry: |
6018 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6019 | for (; addr != end; addr += PAGE_SIZE) { | |
6020 | pte_t ptent = *(pte++); | |
02491447 | 6021 | swp_entry_t ent; |
4ffef5fe DN |
6022 | |
6023 | if (!mc.precharge) | |
6024 | break; | |
6025 | ||
8d32ff84 | 6026 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6027 | case MC_TARGET_PAGE: |
6028 | page = target.page; | |
6029 | if (isolate_lru_page(page)) | |
6030 | goto put; | |
6031 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6032 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6033 | mc.from, mc.to)) { |
4ffef5fe | 6034 | mc.precharge--; |
854ffa8d DN |
6035 | /* we uncharge from mc.from later. */ |
6036 | mc.moved_charge++; | |
4ffef5fe DN |
6037 | } |
6038 | putback_lru_page(page); | |
8d32ff84 | 6039 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6040 | put_page(page); |
6041 | break; | |
02491447 DN |
6042 | case MC_TARGET_SWAP: |
6043 | ent = target.ent; | |
e91cbb42 | 6044 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6045 | mc.precharge--; |
483c30b5 DN |
6046 | /* we fixup refcnts and charges later. */ |
6047 | mc.moved_swap++; | |
6048 | } | |
02491447 | 6049 | break; |
4ffef5fe DN |
6050 | default: |
6051 | break; | |
6052 | } | |
6053 | } | |
6054 | pte_unmap_unlock(pte - 1, ptl); | |
6055 | cond_resched(); | |
6056 | ||
6057 | if (addr != end) { | |
6058 | /* | |
6059 | * We have consumed all precharges we got in can_attach(). | |
6060 | * We try charge one by one, but don't do any additional | |
6061 | * charges to mc.to if we have failed in charge once in attach() | |
6062 | * phase. | |
6063 | */ | |
854ffa8d | 6064 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6065 | if (!ret) |
6066 | goto retry; | |
6067 | } | |
6068 | ||
6069 | return ret; | |
6070 | } | |
6071 | ||
6072 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6073 | { | |
6074 | struct vm_area_struct *vma; | |
6075 | ||
6076 | lru_add_drain_all(); | |
dfe076b0 DN |
6077 | retry: |
6078 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6079 | /* | |
6080 | * Someone who are holding the mmap_sem might be waiting in | |
6081 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6082 | * and retry. Because we cancel precharges, we might not be able | |
6083 | * to move enough charges, but moving charge is a best-effort | |
6084 | * feature anyway, so it wouldn't be a big problem. | |
6085 | */ | |
6086 | __mem_cgroup_clear_mc(); | |
6087 | cond_resched(); | |
6088 | goto retry; | |
6089 | } | |
4ffef5fe DN |
6090 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6091 | int ret; | |
6092 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6093 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6094 | .mm = mm, | |
6095 | .private = vma, | |
6096 | }; | |
6097 | if (is_vm_hugetlb_page(vma)) | |
6098 | continue; | |
4ffef5fe DN |
6099 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6100 | &mem_cgroup_move_charge_walk); | |
6101 | if (ret) | |
6102 | /* | |
6103 | * means we have consumed all precharges and failed in | |
6104 | * doing additional charge. Just abandon here. | |
6105 | */ | |
6106 | break; | |
6107 | } | |
dfe076b0 | 6108 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6109 | } |
6110 | ||
eb95419b | 6111 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6112 | struct cgroup_taskset *tset) |
67e465a7 | 6113 | { |
2f7ee569 | 6114 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6115 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6116 | |
dfe076b0 | 6117 | if (mm) { |
a433658c KM |
6118 | if (mc.to) |
6119 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6120 | mmput(mm); |
6121 | } | |
a433658c KM |
6122 | if (mc.to) |
6123 | mem_cgroup_clear_mc(); | |
67e465a7 | 6124 | } |
5cfb80a7 | 6125 | #else /* !CONFIG_MMU */ |
eb95419b | 6126 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6127 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6128 | { |
6129 | return 0; | |
6130 | } | |
eb95419b | 6131 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6132 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6133 | { |
6134 | } | |
eb95419b | 6135 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6136 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6137 | { |
6138 | } | |
6139 | #endif | |
67e465a7 | 6140 | |
f00baae7 TH |
6141 | /* |
6142 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6143 | * to verify whether we're attached to the default hierarchy on each mount |
6144 | * attempt. | |
f00baae7 | 6145 | */ |
eb95419b | 6146 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6147 | { |
6148 | /* | |
aa6ec29b | 6149 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6150 | * guarantees that @root doesn't have any children, so turning it |
6151 | * on for the root memcg is enough. | |
6152 | */ | |
aa6ec29b | 6153 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 6154 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
6155 | } |
6156 | ||
073219e9 | 6157 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6158 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6159 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6160 | .css_offline = mem_cgroup_css_offline, |
6161 | .css_free = mem_cgroup_css_free, | |
1ced953b | 6162 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6163 | .can_attach = mem_cgroup_can_attach, |
6164 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6165 | .attach = mem_cgroup_move_task, |
f00baae7 | 6166 | .bind = mem_cgroup_bind, |
5577964e | 6167 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 6168 | .early_init = 0, |
8cdea7c0 | 6169 | }; |
c077719b | 6170 | |
c255a458 | 6171 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6172 | static int __init enable_swap_account(char *s) |
6173 | { | |
a2c8990a | 6174 | if (!strcmp(s, "1")) |
a42c390c | 6175 | really_do_swap_account = 1; |
a2c8990a | 6176 | else if (!strcmp(s, "0")) |
a42c390c MH |
6177 | really_do_swap_account = 0; |
6178 | return 1; | |
6179 | } | |
a2c8990a | 6180 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6181 | |
2d11085e MH |
6182 | static void __init memsw_file_init(void) |
6183 | { | |
2cf669a5 TH |
6184 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6185 | memsw_cgroup_files)); | |
6acc8b02 MH |
6186 | } |
6187 | ||
6188 | static void __init enable_swap_cgroup(void) | |
6189 | { | |
6190 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6191 | do_swap_account = 1; | |
6192 | memsw_file_init(); | |
6193 | } | |
2d11085e | 6194 | } |
6acc8b02 | 6195 | |
2d11085e | 6196 | #else |
6acc8b02 | 6197 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6198 | { |
6199 | } | |
c077719b | 6200 | #endif |
2d11085e | 6201 | |
0a31bc97 JW |
6202 | #ifdef CONFIG_MEMCG_SWAP |
6203 | /** | |
6204 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6205 | * @page: page whose memsw charge to transfer | |
6206 | * @entry: swap entry to move the charge to | |
6207 | * | |
6208 | * Transfer the memsw charge of @page to @entry. | |
6209 | */ | |
6210 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6211 | { | |
6212 | struct page_cgroup *pc; | |
6213 | unsigned short oldid; | |
6214 | ||
6215 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6216 | VM_BUG_ON_PAGE(page_count(page), page); | |
6217 | ||
6218 | if (!do_swap_account) | |
6219 | return; | |
6220 | ||
6221 | pc = lookup_page_cgroup(page); | |
6222 | ||
6223 | /* Readahead page, never charged */ | |
6224 | if (!PageCgroupUsed(pc)) | |
6225 | return; | |
6226 | ||
6227 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEMSW), page); | |
6228 | ||
6229 | oldid = swap_cgroup_record(entry, mem_cgroup_id(pc->mem_cgroup)); | |
6230 | VM_BUG_ON_PAGE(oldid, page); | |
6231 | ||
6232 | pc->flags &= ~PCG_MEMSW; | |
6233 | css_get(&pc->mem_cgroup->css); | |
6234 | mem_cgroup_swap_statistics(pc->mem_cgroup, true); | |
6235 | } | |
6236 | ||
6237 | /** | |
6238 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
6239 | * @entry: swap entry to uncharge | |
6240 | * | |
6241 | * Drop the memsw charge associated with @entry. | |
6242 | */ | |
6243 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
6244 | { | |
6245 | struct mem_cgroup *memcg; | |
6246 | unsigned short id; | |
6247 | ||
6248 | if (!do_swap_account) | |
6249 | return; | |
6250 | ||
6251 | id = swap_cgroup_record(entry, 0); | |
6252 | rcu_read_lock(); | |
6253 | memcg = mem_cgroup_lookup(id); | |
6254 | if (memcg) { | |
ce00a967 | 6255 | if (!mem_cgroup_is_root(memcg)) |
3e32cb2e | 6256 | page_counter_uncharge(&memcg->memsw, 1); |
0a31bc97 JW |
6257 | mem_cgroup_swap_statistics(memcg, false); |
6258 | css_put(&memcg->css); | |
6259 | } | |
6260 | rcu_read_unlock(); | |
6261 | } | |
6262 | #endif | |
6263 | ||
00501b53 JW |
6264 | /** |
6265 | * mem_cgroup_try_charge - try charging a page | |
6266 | * @page: page to charge | |
6267 | * @mm: mm context of the victim | |
6268 | * @gfp_mask: reclaim mode | |
6269 | * @memcgp: charged memcg return | |
6270 | * | |
6271 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6272 | * pages according to @gfp_mask if necessary. | |
6273 | * | |
6274 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
6275 | * Otherwise, an error code is returned. | |
6276 | * | |
6277 | * After page->mapping has been set up, the caller must finalize the | |
6278 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
6279 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
6280 | */ | |
6281 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
6282 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
6283 | { | |
6284 | struct mem_cgroup *memcg = NULL; | |
6285 | unsigned int nr_pages = 1; | |
6286 | int ret = 0; | |
6287 | ||
6288 | if (mem_cgroup_disabled()) | |
6289 | goto out; | |
6290 | ||
6291 | if (PageSwapCache(page)) { | |
6292 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
6293 | /* | |
6294 | * Every swap fault against a single page tries to charge the | |
6295 | * page, bail as early as possible. shmem_unuse() encounters | |
6296 | * already charged pages, too. The USED bit is protected by | |
6297 | * the page lock, which serializes swap cache removal, which | |
6298 | * in turn serializes uncharging. | |
6299 | */ | |
6300 | if (PageCgroupUsed(pc)) | |
6301 | goto out; | |
6302 | } | |
6303 | ||
6304 | if (PageTransHuge(page)) { | |
6305 | nr_pages <<= compound_order(page); | |
6306 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6307 | } | |
6308 | ||
6309 | if (do_swap_account && PageSwapCache(page)) | |
6310 | memcg = try_get_mem_cgroup_from_page(page); | |
6311 | if (!memcg) | |
6312 | memcg = get_mem_cgroup_from_mm(mm); | |
6313 | ||
6314 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6315 | ||
6316 | css_put(&memcg->css); | |
6317 | ||
6318 | if (ret == -EINTR) { | |
6319 | memcg = root_mem_cgroup; | |
6320 | ret = 0; | |
6321 | } | |
6322 | out: | |
6323 | *memcgp = memcg; | |
6324 | return ret; | |
6325 | } | |
6326 | ||
6327 | /** | |
6328 | * mem_cgroup_commit_charge - commit a page charge | |
6329 | * @page: page to charge | |
6330 | * @memcg: memcg to charge the page to | |
6331 | * @lrucare: page might be on LRU already | |
6332 | * | |
6333 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6334 | * after page->mapping has been set up. This must happen atomically | |
6335 | * as part of the page instantiation, i.e. under the page table lock | |
6336 | * for anonymous pages, under the page lock for page and swap cache. | |
6337 | * | |
6338 | * In addition, the page must not be on the LRU during the commit, to | |
6339 | * prevent racing with task migration. If it might be, use @lrucare. | |
6340 | * | |
6341 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6342 | */ | |
6343 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
6344 | bool lrucare) | |
6345 | { | |
6346 | unsigned int nr_pages = 1; | |
6347 | ||
6348 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6349 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6350 | ||
6351 | if (mem_cgroup_disabled()) | |
6352 | return; | |
6353 | /* | |
6354 | * Swap faults will attempt to charge the same page multiple | |
6355 | * times. But reuse_swap_page() might have removed the page | |
6356 | * from swapcache already, so we can't check PageSwapCache(). | |
6357 | */ | |
6358 | if (!memcg) | |
6359 | return; | |
6360 | ||
6abb5a86 JW |
6361 | commit_charge(page, memcg, lrucare); |
6362 | ||
00501b53 JW |
6363 | if (PageTransHuge(page)) { |
6364 | nr_pages <<= compound_order(page); | |
6365 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6366 | } | |
6367 | ||
6abb5a86 JW |
6368 | local_irq_disable(); |
6369 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
6370 | memcg_check_events(memcg, page); | |
6371 | local_irq_enable(); | |
00501b53 JW |
6372 | |
6373 | if (do_swap_account && PageSwapCache(page)) { | |
6374 | swp_entry_t entry = { .val = page_private(page) }; | |
6375 | /* | |
6376 | * The swap entry might not get freed for a long time, | |
6377 | * let's not wait for it. The page already received a | |
6378 | * memory+swap charge, drop the swap entry duplicate. | |
6379 | */ | |
6380 | mem_cgroup_uncharge_swap(entry); | |
6381 | } | |
6382 | } | |
6383 | ||
6384 | /** | |
6385 | * mem_cgroup_cancel_charge - cancel a page charge | |
6386 | * @page: page to charge | |
6387 | * @memcg: memcg to charge the page to | |
6388 | * | |
6389 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6390 | */ | |
6391 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
6392 | { | |
6393 | unsigned int nr_pages = 1; | |
6394 | ||
6395 | if (mem_cgroup_disabled()) | |
6396 | return; | |
6397 | /* | |
6398 | * Swap faults will attempt to charge the same page multiple | |
6399 | * times. But reuse_swap_page() might have removed the page | |
6400 | * from swapcache already, so we can't check PageSwapCache(). | |
6401 | */ | |
6402 | if (!memcg) | |
6403 | return; | |
6404 | ||
6405 | if (PageTransHuge(page)) { | |
6406 | nr_pages <<= compound_order(page); | |
6407 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6408 | } | |
6409 | ||
6410 | cancel_charge(memcg, nr_pages); | |
6411 | } | |
6412 | ||
747db954 JW |
6413 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
6414 | unsigned long nr_mem, unsigned long nr_memsw, | |
6415 | unsigned long nr_anon, unsigned long nr_file, | |
6416 | unsigned long nr_huge, struct page *dummy_page) | |
6417 | { | |
6418 | unsigned long flags; | |
6419 | ||
ce00a967 JW |
6420 | if (!mem_cgroup_is_root(memcg)) { |
6421 | if (nr_mem) | |
3e32cb2e | 6422 | page_counter_uncharge(&memcg->memory, nr_mem); |
ce00a967 | 6423 | if (nr_memsw) |
3e32cb2e | 6424 | page_counter_uncharge(&memcg->memsw, nr_memsw); |
ce00a967 JW |
6425 | memcg_oom_recover(memcg); |
6426 | } | |
747db954 JW |
6427 | |
6428 | local_irq_save(flags); | |
6429 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
6430 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
6431 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
6432 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
6433 | __this_cpu_add(memcg->stat->nr_page_events, nr_anon + nr_file); | |
6434 | memcg_check_events(memcg, dummy_page); | |
6435 | local_irq_restore(flags); | |
6436 | } | |
6437 | ||
6438 | static void uncharge_list(struct list_head *page_list) | |
6439 | { | |
6440 | struct mem_cgroup *memcg = NULL; | |
6441 | unsigned long nr_memsw = 0; | |
6442 | unsigned long nr_anon = 0; | |
6443 | unsigned long nr_file = 0; | |
6444 | unsigned long nr_huge = 0; | |
6445 | unsigned long pgpgout = 0; | |
6446 | unsigned long nr_mem = 0; | |
6447 | struct list_head *next; | |
6448 | struct page *page; | |
6449 | ||
6450 | next = page_list->next; | |
6451 | do { | |
6452 | unsigned int nr_pages = 1; | |
6453 | struct page_cgroup *pc; | |
6454 | ||
6455 | page = list_entry(next, struct page, lru); | |
6456 | next = page->lru.next; | |
6457 | ||
6458 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6459 | VM_BUG_ON_PAGE(page_count(page), page); | |
6460 | ||
6461 | pc = lookup_page_cgroup(page); | |
6462 | if (!PageCgroupUsed(pc)) | |
6463 | continue; | |
6464 | ||
6465 | /* | |
6466 | * Nobody should be changing or seriously looking at | |
6467 | * pc->mem_cgroup and pc->flags at this point, we have | |
6468 | * fully exclusive access to the page. | |
6469 | */ | |
6470 | ||
6471 | if (memcg != pc->mem_cgroup) { | |
6472 | if (memcg) { | |
6473 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6474 | nr_anon, nr_file, nr_huge, page); | |
6475 | pgpgout = nr_mem = nr_memsw = 0; | |
6476 | nr_anon = nr_file = nr_huge = 0; | |
6477 | } | |
6478 | memcg = pc->mem_cgroup; | |
6479 | } | |
6480 | ||
6481 | if (PageTransHuge(page)) { | |
6482 | nr_pages <<= compound_order(page); | |
6483 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6484 | nr_huge += nr_pages; | |
6485 | } | |
6486 | ||
6487 | if (PageAnon(page)) | |
6488 | nr_anon += nr_pages; | |
6489 | else | |
6490 | nr_file += nr_pages; | |
6491 | ||
6492 | if (pc->flags & PCG_MEM) | |
6493 | nr_mem += nr_pages; | |
6494 | if (pc->flags & PCG_MEMSW) | |
6495 | nr_memsw += nr_pages; | |
6496 | pc->flags = 0; | |
6497 | ||
6498 | pgpgout++; | |
6499 | } while (next != page_list); | |
6500 | ||
6501 | if (memcg) | |
6502 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6503 | nr_anon, nr_file, nr_huge, page); | |
6504 | } | |
6505 | ||
0a31bc97 JW |
6506 | /** |
6507 | * mem_cgroup_uncharge - uncharge a page | |
6508 | * @page: page to uncharge | |
6509 | * | |
6510 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6511 | * mem_cgroup_commit_charge(). | |
6512 | */ | |
6513 | void mem_cgroup_uncharge(struct page *page) | |
6514 | { | |
0a31bc97 | 6515 | struct page_cgroup *pc; |
0a31bc97 JW |
6516 | |
6517 | if (mem_cgroup_disabled()) | |
6518 | return; | |
6519 | ||
747db954 | 6520 | /* Don't touch page->lru of any random page, pre-check: */ |
0a31bc97 | 6521 | pc = lookup_page_cgroup(page); |
0a31bc97 JW |
6522 | if (!PageCgroupUsed(pc)) |
6523 | return; | |
6524 | ||
747db954 JW |
6525 | INIT_LIST_HEAD(&page->lru); |
6526 | uncharge_list(&page->lru); | |
6527 | } | |
0a31bc97 | 6528 | |
747db954 JW |
6529 | /** |
6530 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6531 | * @page_list: list of pages to uncharge | |
6532 | * | |
6533 | * Uncharge a list of pages previously charged with | |
6534 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6535 | */ | |
6536 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6537 | { | |
6538 | if (mem_cgroup_disabled()) | |
6539 | return; | |
0a31bc97 | 6540 | |
747db954 JW |
6541 | if (!list_empty(page_list)) |
6542 | uncharge_list(page_list); | |
0a31bc97 JW |
6543 | } |
6544 | ||
6545 | /** | |
6546 | * mem_cgroup_migrate - migrate a charge to another page | |
6547 | * @oldpage: currently charged page | |
6548 | * @newpage: page to transfer the charge to | |
6549 | * @lrucare: both pages might be on the LRU already | |
6550 | * | |
6551 | * Migrate the charge from @oldpage to @newpage. | |
6552 | * | |
6553 | * Both pages must be locked, @newpage->mapping must be set up. | |
6554 | */ | |
6555 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
6556 | bool lrucare) | |
6557 | { | |
0a31bc97 JW |
6558 | struct page_cgroup *pc; |
6559 | int isolated; | |
6560 | ||
6561 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6562 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
6563 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
6564 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
6565 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
6566 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6567 | newpage); | |
0a31bc97 JW |
6568 | |
6569 | if (mem_cgroup_disabled()) | |
6570 | return; | |
6571 | ||
6572 | /* Page cache replacement: new page already charged? */ | |
6573 | pc = lookup_page_cgroup(newpage); | |
6574 | if (PageCgroupUsed(pc)) | |
6575 | return; | |
6576 | ||
6577 | /* Re-entrant migration: old page already uncharged? */ | |
6578 | pc = lookup_page_cgroup(oldpage); | |
6579 | if (!PageCgroupUsed(pc)) | |
6580 | return; | |
6581 | ||
6582 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEM), oldpage); | |
6583 | VM_BUG_ON_PAGE(do_swap_account && !(pc->flags & PCG_MEMSW), oldpage); | |
6584 | ||
0a31bc97 JW |
6585 | if (lrucare) |
6586 | lock_page_lru(oldpage, &isolated); | |
6587 | ||
6588 | pc->flags = 0; | |
6589 | ||
6590 | if (lrucare) | |
6591 | unlock_page_lru(oldpage, isolated); | |
6592 | ||
6abb5a86 | 6593 | commit_charge(newpage, pc->mem_cgroup, lrucare); |
0a31bc97 JW |
6594 | } |
6595 | ||
2d11085e | 6596 | /* |
1081312f MH |
6597 | * subsys_initcall() for memory controller. |
6598 | * | |
6599 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6600 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6601 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6602 | * be initialized from here. | |
2d11085e MH |
6603 | */ |
6604 | static int __init mem_cgroup_init(void) | |
6605 | { | |
6606 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6607 | enable_swap_cgroup(); |
bb4cc1a8 | 6608 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6609 | memcg_stock_init(); |
2d11085e MH |
6610 | return 0; |
6611 | } | |
6612 | subsys_initcall(mem_cgroup_init); |