]>
Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
8cdea7c0 BS |
17 | * This program is free software; you can redistribute it and/or modify |
18 | * it under the terms of the GNU General Public License as published by | |
19 | * the Free Software Foundation; either version 2 of the License, or | |
20 | * (at your option) any later version. | |
21 | * | |
22 | * This program is distributed in the hope that it will be useful, | |
23 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
24 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
25 | * GNU General Public License for more details. | |
26 | */ | |
27 | ||
28 | #include <linux/res_counter.h> | |
29 | #include <linux/memcontrol.h> | |
30 | #include <linux/cgroup.h> | |
78fb7466 | 31 | #include <linux/mm.h> |
4ffef5fe | 32 | #include <linux/hugetlb.h> |
d13d1443 | 33 | #include <linux/pagemap.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
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 AM |
167 | struct rb_node tree_node; /* RB tree node */ |
168 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
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; | |
201 | u64 threshold; | |
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; | |
287 | /* | |
288 | * the counter to account for memory usage | |
289 | */ | |
290 | struct res_counter res; | |
59927fb9 | 291 | |
70ddf637 AV |
292 | /* vmpressure notifications */ |
293 | struct vmpressure vmpressure; | |
294 | ||
2f7dd7a4 JW |
295 | /* css_online() has been completed */ |
296 | int initialized; | |
297 | ||
465939a1 LZ |
298 | /* |
299 | * the counter to account for mem+swap usage. | |
300 | */ | |
301 | struct res_counter memsw; | |
59927fb9 | 302 | |
510fc4e1 GC |
303 | /* |
304 | * the counter to account for kernel memory usage. | |
305 | */ | |
306 | struct res_counter kmem; | |
18f59ea7 BS |
307 | /* |
308 | * Should the accounting and control be hierarchical, per subtree? | |
309 | */ | |
310 | bool use_hierarchy; | |
510fc4e1 | 311 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
312 | |
313 | bool oom_lock; | |
314 | atomic_t under_oom; | |
3812c8c8 | 315 | atomic_t oom_wakeups; |
79dfdacc | 316 | |
1f4c025b | 317 | int swappiness; |
3c11ecf4 KH |
318 | /* OOM-Killer disable */ |
319 | int oom_kill_disable; | |
a7885eb8 | 320 | |
2e72b634 KS |
321 | /* protect arrays of thresholds */ |
322 | struct mutex thresholds_lock; | |
323 | ||
324 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 325 | struct mem_cgroup_thresholds thresholds; |
907860ed | 326 | |
2e72b634 | 327 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 328 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 329 | |
9490ff27 KH |
330 | /* For oom notifier event fd */ |
331 | struct list_head oom_notify; | |
185efc0f | 332 | |
7dc74be0 DN |
333 | /* |
334 | * Should we move charges of a task when a task is moved into this | |
335 | * mem_cgroup ? And what type of charges should we move ? | |
336 | */ | |
f894ffa8 | 337 | unsigned long move_charge_at_immigrate; |
619d094b KH |
338 | /* |
339 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
340 | */ | |
341 | atomic_t moving_account; | |
312734c0 KH |
342 | /* taken only while moving_account > 0 */ |
343 | spinlock_t move_lock; | |
d52aa412 | 344 | /* |
c62b1a3b | 345 | * percpu counter. |
d52aa412 | 346 | */ |
3a7951b4 | 347 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
348 | /* |
349 | * used when a cpu is offlined or other synchronizations | |
350 | * See mem_cgroup_read_stat(). | |
351 | */ | |
352 | struct mem_cgroup_stat_cpu nocpu_base; | |
353 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 354 | |
5f578161 | 355 | atomic_t dead_count; |
4bd2c1ee | 356 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
2e685cad | 357 | struct cg_proto tcp_mem; |
d1a4c0b3 | 358 | #endif |
2633d7a0 | 359 | #if defined(CONFIG_MEMCG_KMEM) |
bd673145 VD |
360 | /* analogous to slab_common's slab_caches list, but per-memcg; |
361 | * protected by memcg_slab_mutex */ | |
2633d7a0 | 362 | struct list_head memcg_slab_caches; |
2633d7a0 GC |
363 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ |
364 | int kmemcg_id; | |
365 | #endif | |
45cf7ebd GC |
366 | |
367 | int last_scanned_node; | |
368 | #if MAX_NUMNODES > 1 | |
369 | nodemask_t scan_nodes; | |
370 | atomic_t numainfo_events; | |
371 | atomic_t numainfo_updating; | |
372 | #endif | |
70ddf637 | 373 | |
fba94807 TH |
374 | /* List of events which userspace want to receive */ |
375 | struct list_head event_list; | |
376 | spinlock_t event_list_lock; | |
377 | ||
54f72fe0 JW |
378 | struct mem_cgroup_per_node *nodeinfo[0]; |
379 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
380 | }; |
381 | ||
510fc4e1 GC |
382 | /* internal only representation about the status of kmem accounting. */ |
383 | enum { | |
6de64beb | 384 | KMEM_ACCOUNTED_ACTIVE, /* accounted by this cgroup itself */ |
7de37682 | 385 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
386 | }; |
387 | ||
510fc4e1 GC |
388 | #ifdef CONFIG_MEMCG_KMEM |
389 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
390 | { | |
391 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
392 | } | |
7de37682 GC |
393 | |
394 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
395 | { | |
396 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
397 | } | |
398 | ||
399 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) | |
400 | { | |
10d5ebf4 LZ |
401 | /* |
402 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
403 | * will call css_put() if it sees the memcg is dead. | |
404 | */ | |
405 | smp_wmb(); | |
7de37682 GC |
406 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
407 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
408 | } | |
409 | ||
410 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
411 | { | |
412 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
413 | &memcg->kmem_account_flags); | |
414 | } | |
510fc4e1 GC |
415 | #endif |
416 | ||
7dc74be0 DN |
417 | /* Stuffs for move charges at task migration. */ |
418 | /* | |
ee5e8472 GC |
419 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
420 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
421 | */ |
422 | enum move_type { | |
4ffef5fe | 423 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 424 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
425 | NR_MOVE_TYPE, |
426 | }; | |
427 | ||
4ffef5fe DN |
428 | /* "mc" and its members are protected by cgroup_mutex */ |
429 | static struct move_charge_struct { | |
b1dd693e | 430 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
431 | struct mem_cgroup *from; |
432 | struct mem_cgroup *to; | |
ee5e8472 | 433 | unsigned long immigrate_flags; |
4ffef5fe | 434 | unsigned long precharge; |
854ffa8d | 435 | unsigned long moved_charge; |
483c30b5 | 436 | unsigned long moved_swap; |
8033b97c DN |
437 | struct task_struct *moving_task; /* a task moving charges */ |
438 | wait_queue_head_t waitq; /* a waitq for other context */ | |
439 | } mc = { | |
2bd9bb20 | 440 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
441 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
442 | }; | |
4ffef5fe | 443 | |
90254a65 DN |
444 | static bool move_anon(void) |
445 | { | |
ee5e8472 | 446 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
447 | } |
448 | ||
87946a72 DN |
449 | static bool move_file(void) |
450 | { | |
ee5e8472 | 451 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
452 | } |
453 | ||
4e416953 BS |
454 | /* |
455 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
456 | * limit reclaim to prevent infinite loops, if they ever occur. | |
457 | */ | |
a0db00fc | 458 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 459 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 460 | |
217bc319 KH |
461 | enum charge_type { |
462 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 463 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 464 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 465 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
466 | NR_CHARGE_TYPE, |
467 | }; | |
468 | ||
8c7c6e34 | 469 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
470 | enum res_type { |
471 | _MEM, | |
472 | _MEMSWAP, | |
473 | _OOM_TYPE, | |
510fc4e1 | 474 | _KMEM, |
86ae53e1 GC |
475 | }; |
476 | ||
a0db00fc KS |
477 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
478 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 479 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
480 | /* Used for OOM nofiier */ |
481 | #define OOM_CONTROL (0) | |
8c7c6e34 | 482 | |
0999821b GC |
483 | /* |
484 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
485 | * As a consequence, any change that needs to protect against new child cgroups | |
486 | * appearing has to hold it as well. | |
487 | */ | |
488 | static DEFINE_MUTEX(memcg_create_mutex); | |
489 | ||
b2145145 WL |
490 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
491 | { | |
a7c6d554 | 492 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
493 | } |
494 | ||
70ddf637 AV |
495 | /* Some nice accessors for the vmpressure. */ |
496 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
497 | { | |
498 | if (!memcg) | |
499 | memcg = root_mem_cgroup; | |
500 | return &memcg->vmpressure; | |
501 | } | |
502 | ||
503 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
504 | { | |
505 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
506 | } | |
507 | ||
7ffc0edc MH |
508 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
509 | { | |
510 | return (memcg == root_mem_cgroup); | |
511 | } | |
512 | ||
4219b2da LZ |
513 | /* |
514 | * We restrict the id in the range of [1, 65535], so it can fit into | |
515 | * an unsigned short. | |
516 | */ | |
517 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
518 | ||
34c00c31 LZ |
519 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
520 | { | |
15a4c835 | 521 | return memcg->css.id; |
34c00c31 LZ |
522 | } |
523 | ||
524 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
525 | { | |
526 | struct cgroup_subsys_state *css; | |
527 | ||
7d699ddb | 528 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
529 | return mem_cgroup_from_css(css); |
530 | } | |
531 | ||
e1aab161 | 532 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 533 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 534 | |
e1aab161 GC |
535 | void sock_update_memcg(struct sock *sk) |
536 | { | |
376be5ff | 537 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 538 | struct mem_cgroup *memcg; |
3f134619 | 539 | struct cg_proto *cg_proto; |
e1aab161 GC |
540 | |
541 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
542 | ||
f3f511e1 GC |
543 | /* Socket cloning can throw us here with sk_cgrp already |
544 | * filled. It won't however, necessarily happen from | |
545 | * process context. So the test for root memcg given | |
546 | * the current task's memcg won't help us in this case. | |
547 | * | |
548 | * Respecting the original socket's memcg is a better | |
549 | * decision in this case. | |
550 | */ | |
551 | if (sk->sk_cgrp) { | |
552 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 553 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
554 | return; |
555 | } | |
556 | ||
e1aab161 GC |
557 | rcu_read_lock(); |
558 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 559 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 560 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
561 | memcg_proto_active(cg_proto) && |
562 | css_tryget_online(&memcg->css)) { | |
3f134619 | 563 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
564 | } |
565 | rcu_read_unlock(); | |
566 | } | |
567 | } | |
568 | EXPORT_SYMBOL(sock_update_memcg); | |
569 | ||
570 | void sock_release_memcg(struct sock *sk) | |
571 | { | |
376be5ff | 572 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
573 | struct mem_cgroup *memcg; |
574 | WARN_ON(!sk->sk_cgrp->memcg); | |
575 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 576 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
577 | } |
578 | } | |
d1a4c0b3 GC |
579 | |
580 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
581 | { | |
582 | if (!memcg || mem_cgroup_is_root(memcg)) | |
583 | return NULL; | |
584 | ||
2e685cad | 585 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
586 | } |
587 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 588 | |
3f134619 GC |
589 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
590 | { | |
2e685cad | 591 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
592 | return; |
593 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
594 | } | |
595 | #else | |
596 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
597 | { | |
598 | } | |
599 | #endif | |
600 | ||
a8964b9b | 601 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
602 | /* |
603 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
b8627835 LZ |
604 | * The main reason for not using cgroup id for this: |
605 | * this works better in sparse environments, where we have a lot of memcgs, | |
606 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
607 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
608 | * 200 entry array for that. | |
55007d84 GC |
609 | * |
610 | * The current size of the caches array is stored in | |
611 | * memcg_limited_groups_array_size. It will double each time we have to | |
612 | * increase it. | |
613 | */ | |
614 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
615 | int memcg_limited_groups_array_size; |
616 | ||
55007d84 GC |
617 | /* |
618 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
619 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
620 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
621 | * tunable, but that is strictly not necessary. | |
622 | * | |
b8627835 | 623 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
624 | * this constant directly from cgroup, but it is understandable that this is |
625 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 626 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
627 | * increase ours as well if it increases. |
628 | */ | |
629 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 630 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 631 | |
d7f25f8a GC |
632 | /* |
633 | * A lot of the calls to the cache allocation functions are expected to be | |
634 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
635 | * conditional to this static branch, we'll have to allow modules that does | |
636 | * kmem_cache_alloc and the such to see this symbol as well | |
637 | */ | |
a8964b9b | 638 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 639 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 640 | |
f3bb3043 VD |
641 | static void memcg_free_cache_id(int id); |
642 | ||
a8964b9b GC |
643 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
644 | { | |
55007d84 | 645 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 646 | static_key_slow_dec(&memcg_kmem_enabled_key); |
f3bb3043 | 647 | memcg_free_cache_id(memcg->kmemcg_id); |
55007d84 | 648 | } |
bea207c8 GC |
649 | /* |
650 | * This check can't live in kmem destruction function, | |
651 | * since the charges will outlive the cgroup | |
652 | */ | |
653 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
654 | } |
655 | #else | |
656 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
657 | { | |
658 | } | |
659 | #endif /* CONFIG_MEMCG_KMEM */ | |
660 | ||
661 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
662 | { | |
663 | disarm_sock_keys(memcg); | |
664 | disarm_kmem_keys(memcg); | |
665 | } | |
666 | ||
c0ff4b85 | 667 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 668 | |
f64c3f54 | 669 | static struct mem_cgroup_per_zone * |
e231875b | 670 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 671 | { |
e231875b JZ |
672 | int nid = zone_to_nid(zone); |
673 | int zid = zone_idx(zone); | |
674 | ||
54f72fe0 | 675 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
676 | } |
677 | ||
c0ff4b85 | 678 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 679 | { |
c0ff4b85 | 680 | return &memcg->css; |
d324236b WF |
681 | } |
682 | ||
f64c3f54 | 683 | static struct mem_cgroup_per_zone * |
e231875b | 684 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 685 | { |
97a6c37b JW |
686 | int nid = page_to_nid(page); |
687 | int zid = page_zonenum(page); | |
f64c3f54 | 688 | |
e231875b | 689 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
690 | } |
691 | ||
bb4cc1a8 AM |
692 | static struct mem_cgroup_tree_per_zone * |
693 | soft_limit_tree_node_zone(int nid, int zid) | |
694 | { | |
695 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
696 | } | |
697 | ||
698 | static struct mem_cgroup_tree_per_zone * | |
699 | soft_limit_tree_from_page(struct page *page) | |
700 | { | |
701 | int nid = page_to_nid(page); | |
702 | int zid = page_zonenum(page); | |
703 | ||
704 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
705 | } | |
706 | ||
cf2c8127 JW |
707 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
708 | struct mem_cgroup_tree_per_zone *mctz, | |
709 | unsigned long long new_usage_in_excess) | |
bb4cc1a8 AM |
710 | { |
711 | struct rb_node **p = &mctz->rb_root.rb_node; | |
712 | struct rb_node *parent = NULL; | |
713 | struct mem_cgroup_per_zone *mz_node; | |
714 | ||
715 | if (mz->on_tree) | |
716 | return; | |
717 | ||
718 | mz->usage_in_excess = new_usage_in_excess; | |
719 | if (!mz->usage_in_excess) | |
720 | return; | |
721 | while (*p) { | |
722 | parent = *p; | |
723 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
724 | tree_node); | |
725 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
726 | p = &(*p)->rb_left; | |
727 | /* | |
728 | * We can't avoid mem cgroups that are over their soft | |
729 | * limit by the same amount | |
730 | */ | |
731 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
732 | p = &(*p)->rb_right; | |
733 | } | |
734 | rb_link_node(&mz->tree_node, parent, p); | |
735 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
736 | mz->on_tree = true; | |
737 | } | |
738 | ||
cf2c8127 JW |
739 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
740 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
741 | { |
742 | if (!mz->on_tree) | |
743 | return; | |
744 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
745 | mz->on_tree = false; | |
746 | } | |
747 | ||
cf2c8127 JW |
748 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
749 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 750 | { |
0a31bc97 JW |
751 | unsigned long flags; |
752 | ||
753 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 754 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 755 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
756 | } |
757 | ||
758 | ||
759 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
760 | { | |
761 | unsigned long long excess; | |
762 | struct mem_cgroup_per_zone *mz; | |
763 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 764 | |
e231875b | 765 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
766 | /* |
767 | * Necessary to update all ancestors when hierarchy is used. | |
768 | * because their event counter is not touched. | |
769 | */ | |
770 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 771 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bb4cc1a8 AM |
772 | excess = res_counter_soft_limit_excess(&memcg->res); |
773 | /* | |
774 | * We have to update the tree if mz is on RB-tree or | |
775 | * mem is over its softlimit. | |
776 | */ | |
777 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
778 | unsigned long flags; |
779 | ||
780 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
781 | /* if on-tree, remove it */ |
782 | if (mz->on_tree) | |
cf2c8127 | 783 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
784 | /* |
785 | * Insert again. mz->usage_in_excess will be updated. | |
786 | * If excess is 0, no tree ops. | |
787 | */ | |
cf2c8127 | 788 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 789 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
790 | } |
791 | } | |
792 | } | |
793 | ||
794 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
795 | { | |
bb4cc1a8 | 796 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
797 | struct mem_cgroup_per_zone *mz; |
798 | int nid, zid; | |
bb4cc1a8 | 799 | |
e231875b JZ |
800 | for_each_node(nid) { |
801 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
802 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
803 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 804 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
805 | } |
806 | } | |
807 | } | |
808 | ||
809 | static struct mem_cgroup_per_zone * | |
810 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
811 | { | |
812 | struct rb_node *rightmost = NULL; | |
813 | struct mem_cgroup_per_zone *mz; | |
814 | ||
815 | retry: | |
816 | mz = NULL; | |
817 | rightmost = rb_last(&mctz->rb_root); | |
818 | if (!rightmost) | |
819 | goto done; /* Nothing to reclaim from */ | |
820 | ||
821 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
822 | /* | |
823 | * Remove the node now but someone else can add it back, | |
824 | * we will to add it back at the end of reclaim to its correct | |
825 | * position in the tree. | |
826 | */ | |
cf2c8127 | 827 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 | 828 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || |
ec903c0c | 829 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
830 | goto retry; |
831 | done: | |
832 | return mz; | |
833 | } | |
834 | ||
835 | static struct mem_cgroup_per_zone * | |
836 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
837 | { | |
838 | struct mem_cgroup_per_zone *mz; | |
839 | ||
0a31bc97 | 840 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 841 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 842 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
843 | return mz; |
844 | } | |
845 | ||
711d3d2c KH |
846 | /* |
847 | * Implementation Note: reading percpu statistics for memcg. | |
848 | * | |
849 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
850 | * synchronization to implement "quick" read. There are trade-off between | |
851 | * reading cost and precision of value. Then, we may have a chance to implement | |
852 | * a periodic synchronizion of counter in memcg's counter. | |
853 | * | |
854 | * But this _read() function is used for user interface now. The user accounts | |
855 | * memory usage by memory cgroup and he _always_ requires exact value because | |
856 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
857 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
858 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
859 | * | |
860 | * If there are kernel internal actions which can make use of some not-exact | |
861 | * value, and reading all cpu value can be performance bottleneck in some | |
862 | * common workload, threashold and synchonization as vmstat[] should be | |
863 | * implemented. | |
864 | */ | |
c0ff4b85 | 865 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 866 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 867 | { |
7a159cc9 | 868 | long val = 0; |
c62b1a3b | 869 | int cpu; |
c62b1a3b | 870 | |
711d3d2c KH |
871 | get_online_cpus(); |
872 | for_each_online_cpu(cpu) | |
c0ff4b85 | 873 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 874 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
875 | spin_lock(&memcg->pcp_counter_lock); |
876 | val += memcg->nocpu_base.count[idx]; | |
877 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
878 | #endif |
879 | put_online_cpus(); | |
c62b1a3b KH |
880 | return val; |
881 | } | |
882 | ||
c0ff4b85 | 883 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
884 | enum mem_cgroup_events_index idx) |
885 | { | |
886 | unsigned long val = 0; | |
887 | int cpu; | |
888 | ||
9c567512 | 889 | get_online_cpus(); |
e9f8974f | 890 | for_each_online_cpu(cpu) |
c0ff4b85 | 891 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 892 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
893 | spin_lock(&memcg->pcp_counter_lock); |
894 | val += memcg->nocpu_base.events[idx]; | |
895 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 896 | #endif |
9c567512 | 897 | put_online_cpus(); |
e9f8974f JW |
898 | return val; |
899 | } | |
900 | ||
c0ff4b85 | 901 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 902 | struct page *page, |
0a31bc97 | 903 | int nr_pages) |
d52aa412 | 904 | { |
b2402857 KH |
905 | /* |
906 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
907 | * counted as CACHE even if it's on ANON LRU. | |
908 | */ | |
0a31bc97 | 909 | if (PageAnon(page)) |
b2402857 | 910 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 911 | nr_pages); |
d52aa412 | 912 | else |
b2402857 | 913 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 914 | nr_pages); |
55e462b0 | 915 | |
b070e65c DR |
916 | if (PageTransHuge(page)) |
917 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
918 | nr_pages); | |
919 | ||
e401f176 KH |
920 | /* pagein of a big page is an event. So, ignore page size */ |
921 | if (nr_pages > 0) | |
c0ff4b85 | 922 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 923 | else { |
c0ff4b85 | 924 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
925 | nr_pages = -nr_pages; /* for event */ |
926 | } | |
e401f176 | 927 | |
13114716 | 928 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
929 | } |
930 | ||
e231875b | 931 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
932 | { |
933 | struct mem_cgroup_per_zone *mz; | |
934 | ||
935 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
936 | return mz->lru_size[lru]; | |
937 | } | |
938 | ||
e231875b JZ |
939 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
940 | int nid, | |
941 | unsigned int lru_mask) | |
bb2a0de9 | 942 | { |
e231875b | 943 | unsigned long nr = 0; |
889976db YH |
944 | int zid; |
945 | ||
e231875b | 946 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 947 | |
e231875b JZ |
948 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
949 | struct mem_cgroup_per_zone *mz; | |
950 | enum lru_list lru; | |
951 | ||
952 | for_each_lru(lru) { | |
953 | if (!(BIT(lru) & lru_mask)) | |
954 | continue; | |
955 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
956 | nr += mz->lru_size[lru]; | |
957 | } | |
958 | } | |
959 | return nr; | |
889976db | 960 | } |
bb2a0de9 | 961 | |
c0ff4b85 | 962 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 963 | unsigned int lru_mask) |
6d12e2d8 | 964 | { |
e231875b | 965 | unsigned long nr = 0; |
889976db | 966 | int nid; |
6d12e2d8 | 967 | |
31aaea4a | 968 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
969 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
970 | return nr; | |
d52aa412 KH |
971 | } |
972 | ||
f53d7ce3 JW |
973 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
974 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
975 | { |
976 | unsigned long val, next; | |
977 | ||
13114716 | 978 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 979 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 980 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
981 | if ((long)next - (long)val < 0) { |
982 | switch (target) { | |
983 | case MEM_CGROUP_TARGET_THRESH: | |
984 | next = val + THRESHOLDS_EVENTS_TARGET; | |
985 | break; | |
bb4cc1a8 AM |
986 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
987 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
988 | break; | |
f53d7ce3 JW |
989 | case MEM_CGROUP_TARGET_NUMAINFO: |
990 | next = val + NUMAINFO_EVENTS_TARGET; | |
991 | break; | |
992 | default: | |
993 | break; | |
994 | } | |
995 | __this_cpu_write(memcg->stat->targets[target], next); | |
996 | return true; | |
7a159cc9 | 997 | } |
f53d7ce3 | 998 | return false; |
d2265e6f KH |
999 | } |
1000 | ||
1001 | /* | |
1002 | * Check events in order. | |
1003 | * | |
1004 | */ | |
c0ff4b85 | 1005 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
1006 | { |
1007 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
1008 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1009 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1010 | bool do_softlimit; |
82b3f2a7 | 1011 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1012 | |
bb4cc1a8 AM |
1013 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1014 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1015 | #if MAX_NUMNODES > 1 |
1016 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1017 | MEM_CGROUP_TARGET_NUMAINFO); | |
1018 | #endif | |
c0ff4b85 | 1019 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1020 | if (unlikely(do_softlimit)) |
1021 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1022 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1023 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1024 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1025 | #endif |
0a31bc97 | 1026 | } |
d2265e6f KH |
1027 | } |
1028 | ||
cf475ad2 | 1029 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1030 | { |
31a78f23 BS |
1031 | /* |
1032 | * mm_update_next_owner() may clear mm->owner to NULL | |
1033 | * if it races with swapoff, page migration, etc. | |
1034 | * So this can be called with p == NULL. | |
1035 | */ | |
1036 | if (unlikely(!p)) | |
1037 | return NULL; | |
1038 | ||
073219e9 | 1039 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1040 | } |
1041 | ||
df381975 | 1042 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1043 | { |
c0ff4b85 | 1044 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 1045 | |
54595fe2 KH |
1046 | rcu_read_lock(); |
1047 | do { | |
6f6acb00 MH |
1048 | /* |
1049 | * Page cache insertions can happen withou an | |
1050 | * actual mm context, e.g. during disk probing | |
1051 | * on boot, loopback IO, acct() writes etc. | |
1052 | */ | |
1053 | if (unlikely(!mm)) | |
df381975 | 1054 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1055 | else { |
1056 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1057 | if (unlikely(!memcg)) | |
1058 | memcg = root_mem_cgroup; | |
1059 | } | |
ec903c0c | 1060 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1061 | rcu_read_unlock(); |
c0ff4b85 | 1062 | return memcg; |
54595fe2 KH |
1063 | } |
1064 | ||
16248d8f MH |
1065 | /* |
1066 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1067 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1068 | * | |
1069 | * helper function to be used by mem_cgroup_iter | |
1070 | */ | |
1071 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
694fbc0f | 1072 | struct mem_cgroup *last_visited) |
16248d8f | 1073 | { |
492eb21b | 1074 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 1075 | |
bd8815a6 | 1076 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 1077 | skip_node: |
492eb21b | 1078 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
1079 | |
1080 | /* | |
1081 | * Even if we found a group we have to make sure it is | |
1082 | * alive. css && !memcg means that the groups should be | |
1083 | * skipped and we should continue the tree walk. | |
1084 | * last_visited css is safe to use because it is | |
1085 | * protected by css_get and the tree walk is rcu safe. | |
0eef6156 MH |
1086 | * |
1087 | * We do not take a reference on the root of the tree walk | |
1088 | * because we might race with the root removal when it would | |
1089 | * be the only node in the iterated hierarchy and mem_cgroup_iter | |
1090 | * would end up in an endless loop because it expects that at | |
1091 | * least one valid node will be returned. Root cannot disappear | |
1092 | * because caller of the iterator should hold it already so | |
1093 | * skipping css reference should be safe. | |
16248d8f | 1094 | */ |
492eb21b | 1095 | if (next_css) { |
2f7dd7a4 JW |
1096 | struct mem_cgroup *memcg = mem_cgroup_from_css(next_css); |
1097 | ||
1098 | if (next_css == &root->css) | |
1099 | return memcg; | |
1100 | ||
1101 | if (css_tryget_online(next_css)) { | |
1102 | /* | |
1103 | * Make sure the memcg is initialized: | |
1104 | * mem_cgroup_css_online() orders the the | |
1105 | * initialization against setting the flag. | |
1106 | */ | |
1107 | if (smp_load_acquire(&memcg->initialized)) | |
1108 | return memcg; | |
1109 | css_put(next_css); | |
1110 | } | |
0eef6156 MH |
1111 | |
1112 | prev_css = next_css; | |
1113 | goto skip_node; | |
16248d8f MH |
1114 | } |
1115 | ||
1116 | return NULL; | |
1117 | } | |
1118 | ||
519ebea3 JW |
1119 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1120 | { | |
1121 | /* | |
1122 | * When a group in the hierarchy below root is destroyed, the | |
1123 | * hierarchy iterator can no longer be trusted since it might | |
1124 | * have pointed to the destroyed group. Invalidate it. | |
1125 | */ | |
1126 | atomic_inc(&root->dead_count); | |
1127 | } | |
1128 | ||
1129 | static struct mem_cgroup * | |
1130 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1131 | struct mem_cgroup *root, | |
1132 | int *sequence) | |
1133 | { | |
1134 | struct mem_cgroup *position = NULL; | |
1135 | /* | |
1136 | * A cgroup destruction happens in two stages: offlining and | |
1137 | * release. They are separated by a RCU grace period. | |
1138 | * | |
1139 | * If the iterator is valid, we may still race with an | |
1140 | * offlining. The RCU lock ensures the object won't be | |
1141 | * released, tryget will fail if we lost the race. | |
1142 | */ | |
1143 | *sequence = atomic_read(&root->dead_count); | |
1144 | if (iter->last_dead_count == *sequence) { | |
1145 | smp_rmb(); | |
1146 | position = iter->last_visited; | |
ecc736fc MH |
1147 | |
1148 | /* | |
1149 | * We cannot take a reference to root because we might race | |
1150 | * with root removal and returning NULL would end up in | |
1151 | * an endless loop on the iterator user level when root | |
1152 | * would be returned all the time. | |
1153 | */ | |
1154 | if (position && position != root && | |
ec903c0c | 1155 | !css_tryget_online(&position->css)) |
519ebea3 JW |
1156 | position = NULL; |
1157 | } | |
1158 | return position; | |
1159 | } | |
1160 | ||
1161 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1162 | struct mem_cgroup *last_visited, | |
1163 | struct mem_cgroup *new_position, | |
ecc736fc | 1164 | struct mem_cgroup *root, |
519ebea3 JW |
1165 | int sequence) |
1166 | { | |
ecc736fc MH |
1167 | /* root reference counting symmetric to mem_cgroup_iter_load */ |
1168 | if (last_visited && last_visited != root) | |
519ebea3 JW |
1169 | css_put(&last_visited->css); |
1170 | /* | |
1171 | * We store the sequence count from the time @last_visited was | |
1172 | * loaded successfully instead of rereading it here so that we | |
1173 | * don't lose destruction events in between. We could have | |
1174 | * raced with the destruction of @new_position after all. | |
1175 | */ | |
1176 | iter->last_visited = new_position; | |
1177 | smp_wmb(); | |
1178 | iter->last_dead_count = sequence; | |
1179 | } | |
1180 | ||
5660048c JW |
1181 | /** |
1182 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1183 | * @root: hierarchy root | |
1184 | * @prev: previously returned memcg, NULL on first invocation | |
1185 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1186 | * | |
1187 | * Returns references to children of the hierarchy below @root, or | |
1188 | * @root itself, or %NULL after a full round-trip. | |
1189 | * | |
1190 | * Caller must pass the return value in @prev on subsequent | |
1191 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1192 | * to cancel a hierarchy walk before the round-trip is complete. | |
1193 | * | |
1194 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1195 | * divide up the memcgs in the hierarchy among all concurrent | |
1196 | * reclaimers operating on the same zone and priority. | |
1197 | */ | |
694fbc0f | 1198 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1199 | struct mem_cgroup *prev, |
694fbc0f | 1200 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1201 | { |
9f3a0d09 | 1202 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1203 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1204 | |
694fbc0f AM |
1205 | if (mem_cgroup_disabled()) |
1206 | return NULL; | |
5660048c | 1207 | |
9f3a0d09 JW |
1208 | if (!root) |
1209 | root = root_mem_cgroup; | |
7d74b06f | 1210 | |
9f3a0d09 | 1211 | if (prev && !reclaim) |
542f85f9 | 1212 | last_visited = prev; |
14067bb3 | 1213 | |
9f3a0d09 JW |
1214 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1215 | if (prev) | |
c40046f3 | 1216 | goto out_css_put; |
694fbc0f | 1217 | return root; |
9f3a0d09 | 1218 | } |
14067bb3 | 1219 | |
542f85f9 | 1220 | rcu_read_lock(); |
9f3a0d09 | 1221 | while (!memcg) { |
527a5ec9 | 1222 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1223 | int uninitialized_var(seq); |
711d3d2c | 1224 | |
527a5ec9 | 1225 | if (reclaim) { |
527a5ec9 JW |
1226 | struct mem_cgroup_per_zone *mz; |
1227 | ||
e231875b | 1228 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); |
527a5ec9 | 1229 | iter = &mz->reclaim_iter[reclaim->priority]; |
542f85f9 | 1230 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1231 | iter->last_visited = NULL; |
542f85f9 MH |
1232 | goto out_unlock; |
1233 | } | |
5f578161 | 1234 | |
519ebea3 | 1235 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1236 | } |
7d74b06f | 1237 | |
694fbc0f | 1238 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1239 | |
527a5ec9 | 1240 | if (reclaim) { |
ecc736fc MH |
1241 | mem_cgroup_iter_update(iter, last_visited, memcg, root, |
1242 | seq); | |
542f85f9 | 1243 | |
19f39402 | 1244 | if (!memcg) |
527a5ec9 JW |
1245 | iter->generation++; |
1246 | else if (!prev && memcg) | |
1247 | reclaim->generation = iter->generation; | |
1248 | } | |
9f3a0d09 | 1249 | |
694fbc0f | 1250 | if (prev && !memcg) |
542f85f9 | 1251 | goto out_unlock; |
9f3a0d09 | 1252 | } |
542f85f9 MH |
1253 | out_unlock: |
1254 | rcu_read_unlock(); | |
c40046f3 MH |
1255 | out_css_put: |
1256 | if (prev && prev != root) | |
1257 | css_put(&prev->css); | |
1258 | ||
9f3a0d09 | 1259 | return memcg; |
14067bb3 | 1260 | } |
7d74b06f | 1261 | |
5660048c JW |
1262 | /** |
1263 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1264 | * @root: hierarchy root | |
1265 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1266 | */ | |
1267 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1268 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1269 | { |
1270 | if (!root) | |
1271 | root = root_mem_cgroup; | |
1272 | if (prev && prev != root) | |
1273 | css_put(&prev->css); | |
1274 | } | |
7d74b06f | 1275 | |
9f3a0d09 JW |
1276 | /* |
1277 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1278 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1279 | * be used for reference counting. | |
1280 | */ | |
1281 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1282 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1283 | iter != NULL; \ |
527a5ec9 | 1284 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1285 | |
9f3a0d09 | 1286 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1287 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1288 | iter != NULL; \ |
527a5ec9 | 1289 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1290 | |
68ae564b | 1291 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1292 | { |
c0ff4b85 | 1293 | struct mem_cgroup *memcg; |
456f998e | 1294 | |
456f998e | 1295 | rcu_read_lock(); |
c0ff4b85 R |
1296 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1297 | if (unlikely(!memcg)) | |
456f998e YH |
1298 | goto out; |
1299 | ||
1300 | switch (idx) { | |
456f998e | 1301 | case PGFAULT: |
0e574a93 JW |
1302 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1303 | break; | |
1304 | case PGMAJFAULT: | |
1305 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1306 | break; |
1307 | default: | |
1308 | BUG(); | |
1309 | } | |
1310 | out: | |
1311 | rcu_read_unlock(); | |
1312 | } | |
68ae564b | 1313 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1314 | |
925b7673 JW |
1315 | /** |
1316 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1317 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1318 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1319 | * |
1320 | * Returns the lru list vector holding pages for the given @zone and | |
1321 | * @mem. This can be the global zone lruvec, if the memory controller | |
1322 | * is disabled. | |
1323 | */ | |
1324 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1325 | struct mem_cgroup *memcg) | |
1326 | { | |
1327 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1328 | struct lruvec *lruvec; |
925b7673 | 1329 | |
bea8c150 HD |
1330 | if (mem_cgroup_disabled()) { |
1331 | lruvec = &zone->lruvec; | |
1332 | goto out; | |
1333 | } | |
925b7673 | 1334 | |
e231875b | 1335 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1336 | lruvec = &mz->lruvec; |
1337 | out: | |
1338 | /* | |
1339 | * Since a node can be onlined after the mem_cgroup was created, | |
1340 | * we have to be prepared to initialize lruvec->zone here; | |
1341 | * and if offlined then reonlined, we need to reinitialize it. | |
1342 | */ | |
1343 | if (unlikely(lruvec->zone != zone)) | |
1344 | lruvec->zone = zone; | |
1345 | return lruvec; | |
925b7673 JW |
1346 | } |
1347 | ||
925b7673 | 1348 | /** |
fa9add64 | 1349 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1350 | * @page: the page |
fa9add64 | 1351 | * @zone: zone of the page |
925b7673 | 1352 | */ |
fa9add64 | 1353 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1354 | { |
08e552c6 | 1355 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1356 | struct mem_cgroup *memcg; |
1357 | struct page_cgroup *pc; | |
bea8c150 | 1358 | struct lruvec *lruvec; |
6d12e2d8 | 1359 | |
bea8c150 HD |
1360 | if (mem_cgroup_disabled()) { |
1361 | lruvec = &zone->lruvec; | |
1362 | goto out; | |
1363 | } | |
925b7673 | 1364 | |
08e552c6 | 1365 | pc = lookup_page_cgroup(page); |
38c5d72f | 1366 | memcg = pc->mem_cgroup; |
7512102c HD |
1367 | |
1368 | /* | |
fa9add64 | 1369 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1370 | * an uncharged page off lru does nothing to secure |
1371 | * its former mem_cgroup from sudden removal. | |
1372 | * | |
1373 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1374 | * under page_cgroup lock: between them, they make all uses | |
1375 | * of pc->mem_cgroup safe. | |
1376 | */ | |
fa9add64 | 1377 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1378 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1379 | ||
e231875b | 1380 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1381 | lruvec = &mz->lruvec; |
1382 | out: | |
1383 | /* | |
1384 | * Since a node can be onlined after the mem_cgroup was created, | |
1385 | * we have to be prepared to initialize lruvec->zone here; | |
1386 | * and if offlined then reonlined, we need to reinitialize it. | |
1387 | */ | |
1388 | if (unlikely(lruvec->zone != zone)) | |
1389 | lruvec->zone = zone; | |
1390 | return lruvec; | |
08e552c6 | 1391 | } |
b69408e8 | 1392 | |
925b7673 | 1393 | /** |
fa9add64 HD |
1394 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1395 | * @lruvec: mem_cgroup per zone lru vector | |
1396 | * @lru: index of lru list the page is sitting on | |
1397 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1398 | * |
fa9add64 HD |
1399 | * This function must be called when a page is added to or removed from an |
1400 | * lru list. | |
3f58a829 | 1401 | */ |
fa9add64 HD |
1402 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1403 | int nr_pages) | |
3f58a829 MK |
1404 | { |
1405 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1406 | unsigned long *lru_size; |
3f58a829 MK |
1407 | |
1408 | if (mem_cgroup_disabled()) | |
1409 | return; | |
1410 | ||
fa9add64 HD |
1411 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1412 | lru_size = mz->lru_size + lru; | |
1413 | *lru_size += nr_pages; | |
1414 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1415 | } |
544122e5 | 1416 | |
3e92041d | 1417 | /* |
c0ff4b85 | 1418 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1419 | * hierarchy subtree |
1420 | */ | |
c3ac9a8a JW |
1421 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1422 | struct mem_cgroup *memcg) | |
3e92041d | 1423 | { |
91c63734 JW |
1424 | if (root_memcg == memcg) |
1425 | return true; | |
3a981f48 | 1426 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1427 | return false; |
b47f77b5 | 1428 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1429 | } |
1430 | ||
1431 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1432 | struct mem_cgroup *memcg) | |
1433 | { | |
1434 | bool ret; | |
1435 | ||
91c63734 | 1436 | rcu_read_lock(); |
c3ac9a8a | 1437 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1438 | rcu_read_unlock(); |
1439 | return ret; | |
3e92041d MH |
1440 | } |
1441 | ||
ffbdccf5 DR |
1442 | bool task_in_mem_cgroup(struct task_struct *task, |
1443 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1444 | { |
0b7f569e | 1445 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1446 | struct task_struct *p; |
ffbdccf5 | 1447 | bool ret; |
4c4a2214 | 1448 | |
158e0a2d | 1449 | p = find_lock_task_mm(task); |
de077d22 | 1450 | if (p) { |
df381975 | 1451 | curr = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1452 | task_unlock(p); |
1453 | } else { | |
1454 | /* | |
1455 | * All threads may have already detached their mm's, but the oom | |
1456 | * killer still needs to detect if they have already been oom | |
1457 | * killed to prevent needlessly killing additional tasks. | |
1458 | */ | |
ffbdccf5 | 1459 | rcu_read_lock(); |
de077d22 DR |
1460 | curr = mem_cgroup_from_task(task); |
1461 | if (curr) | |
1462 | css_get(&curr->css); | |
ffbdccf5 | 1463 | rcu_read_unlock(); |
de077d22 | 1464 | } |
d31f56db | 1465 | /* |
c0ff4b85 | 1466 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1467 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1468 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1469 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1470 | */ |
c0ff4b85 | 1471 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1472 | css_put(&curr->css); |
4c4a2214 DR |
1473 | return ret; |
1474 | } | |
1475 | ||
c56d5c7d | 1476 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1477 | { |
9b272977 | 1478 | unsigned long inactive_ratio; |
14797e23 | 1479 | unsigned long inactive; |
9b272977 | 1480 | unsigned long active; |
c772be93 | 1481 | unsigned long gb; |
14797e23 | 1482 | |
4d7dcca2 HD |
1483 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1484 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1485 | |
c772be93 KM |
1486 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1487 | if (gb) | |
1488 | inactive_ratio = int_sqrt(10 * gb); | |
1489 | else | |
1490 | inactive_ratio = 1; | |
1491 | ||
9b272977 | 1492 | return inactive * inactive_ratio < active; |
14797e23 KM |
1493 | } |
1494 | ||
6d61ef40 BS |
1495 | #define mem_cgroup_from_res_counter(counter, member) \ |
1496 | container_of(counter, struct mem_cgroup, member) | |
1497 | ||
19942822 | 1498 | /** |
9d11ea9f | 1499 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1500 | * @memcg: the memory cgroup |
19942822 | 1501 | * |
9d11ea9f | 1502 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1503 | * pages. |
19942822 | 1504 | */ |
c0ff4b85 | 1505 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1506 | { |
9d11ea9f JW |
1507 | unsigned long long margin; |
1508 | ||
c0ff4b85 | 1509 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1510 | if (do_swap_account) |
c0ff4b85 | 1511 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1512 | return margin >> PAGE_SHIFT; |
19942822 JW |
1513 | } |
1514 | ||
1f4c025b | 1515 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1516 | { |
a7885eb8 | 1517 | /* root ? */ |
14208b0e | 1518 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1519 | return vm_swappiness; |
1520 | ||
bf1ff263 | 1521 | return memcg->swappiness; |
a7885eb8 KM |
1522 | } |
1523 | ||
619d094b KH |
1524 | /* |
1525 | * memcg->moving_account is used for checking possibility that some thread is | |
1526 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1527 | * a memcg, other threads should check memcg->moving_account under | |
1528 | * rcu_read_lock(), like this: | |
1529 | * | |
1530 | * CPU-A CPU-B | |
1531 | * rcu_read_lock() | |
1532 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1533 | * take heavy locks. | |
1534 | * synchronize_rcu() update something. | |
1535 | * rcu_read_unlock() | |
1536 | * start move here. | |
1537 | */ | |
4331f7d3 KH |
1538 | |
1539 | /* for quick checking without looking up memcg */ | |
1540 | atomic_t memcg_moving __read_mostly; | |
1541 | ||
c0ff4b85 | 1542 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1543 | { |
4331f7d3 | 1544 | atomic_inc(&memcg_moving); |
619d094b | 1545 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1546 | synchronize_rcu(); |
1547 | } | |
1548 | ||
c0ff4b85 | 1549 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1550 | { |
619d094b KH |
1551 | /* |
1552 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1553 | * We check NULL in callee rather than caller. | |
1554 | */ | |
4331f7d3 KH |
1555 | if (memcg) { |
1556 | atomic_dec(&memcg_moving); | |
619d094b | 1557 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1558 | } |
32047e2a | 1559 | } |
619d094b | 1560 | |
32047e2a | 1561 | /* |
bdcbb659 | 1562 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1563 | * |
bdcbb659 QH |
1564 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1565 | * moving cgroups. This is for waiting at high-memory pressure | |
1566 | * caused by "move". | |
32047e2a | 1567 | */ |
c0ff4b85 | 1568 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1569 | { |
2bd9bb20 KH |
1570 | struct mem_cgroup *from; |
1571 | struct mem_cgroup *to; | |
4b534334 | 1572 | bool ret = false; |
2bd9bb20 KH |
1573 | /* |
1574 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1575 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1576 | */ | |
1577 | spin_lock(&mc.lock); | |
1578 | from = mc.from; | |
1579 | to = mc.to; | |
1580 | if (!from) | |
1581 | goto unlock; | |
3e92041d | 1582 | |
c0ff4b85 R |
1583 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1584 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1585 | unlock: |
1586 | spin_unlock(&mc.lock); | |
4b534334 KH |
1587 | return ret; |
1588 | } | |
1589 | ||
c0ff4b85 | 1590 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1591 | { |
1592 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1593 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1594 | DEFINE_WAIT(wait); |
1595 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1596 | /* moving charge context might have finished. */ | |
1597 | if (mc.moving_task) | |
1598 | schedule(); | |
1599 | finish_wait(&mc.waitq, &wait); | |
1600 | return true; | |
1601 | } | |
1602 | } | |
1603 | return false; | |
1604 | } | |
1605 | ||
312734c0 KH |
1606 | /* |
1607 | * Take this lock when | |
1608 | * - a code tries to modify page's memcg while it's USED. | |
1609 | * - a code tries to modify page state accounting in a memcg. | |
312734c0 KH |
1610 | */ |
1611 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1612 | unsigned long *flags) | |
1613 | { | |
1614 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1615 | } | |
1616 | ||
1617 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1618 | unsigned long *flags) | |
1619 | { | |
1620 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1621 | } | |
1622 | ||
58cf188e | 1623 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1624 | /** |
58cf188e | 1625 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1626 | * @memcg: The memory cgroup that went over limit |
1627 | * @p: Task that is going to be killed | |
1628 | * | |
1629 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1630 | * enabled | |
1631 | */ | |
1632 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1633 | { | |
e61734c5 | 1634 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1635 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1636 | struct mem_cgroup *iter; |
1637 | unsigned int i; | |
e222432b | 1638 | |
58cf188e | 1639 | if (!p) |
e222432b BS |
1640 | return; |
1641 | ||
08088cb9 | 1642 | mutex_lock(&oom_info_lock); |
e222432b BS |
1643 | rcu_read_lock(); |
1644 | ||
e61734c5 TH |
1645 | pr_info("Task in "); |
1646 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1647 | pr_info(" killed as a result of limit of "); | |
1648 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1649 | pr_info("\n"); | |
e222432b | 1650 | |
e222432b BS |
1651 | rcu_read_unlock(); |
1652 | ||
d045197f | 1653 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1654 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1655 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1656 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1657 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1658 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1659 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1660 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1661 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1662 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1663 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1664 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1665 | |
1666 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1667 | pr_info("Memory cgroup stats for "); |
1668 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1669 | pr_cont(":"); |
1670 | ||
1671 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1672 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1673 | continue; | |
1674 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1675 | K(mem_cgroup_read_stat(iter, i))); | |
1676 | } | |
1677 | ||
1678 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1679 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1680 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1681 | ||
1682 | pr_cont("\n"); | |
1683 | } | |
08088cb9 | 1684 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1685 | } |
1686 | ||
81d39c20 KH |
1687 | /* |
1688 | * This function returns the number of memcg under hierarchy tree. Returns | |
1689 | * 1(self count) if no children. | |
1690 | */ | |
c0ff4b85 | 1691 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1692 | { |
1693 | int num = 0; | |
7d74b06f KH |
1694 | struct mem_cgroup *iter; |
1695 | ||
c0ff4b85 | 1696 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1697 | num++; |
81d39c20 KH |
1698 | return num; |
1699 | } | |
1700 | ||
a63d83f4 DR |
1701 | /* |
1702 | * Return the memory (and swap, if configured) limit for a memcg. | |
1703 | */ | |
9cbb78bb | 1704 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1705 | { |
1706 | u64 limit; | |
a63d83f4 | 1707 | |
f3e8eb70 | 1708 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1709 | |
a63d83f4 | 1710 | /* |
9a5a8f19 | 1711 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1712 | */ |
9a5a8f19 MH |
1713 | if (mem_cgroup_swappiness(memcg)) { |
1714 | u64 memsw; | |
1715 | ||
1716 | limit += total_swap_pages << PAGE_SHIFT; | |
1717 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1718 | ||
1719 | /* | |
1720 | * If memsw is finite and limits the amount of swap space | |
1721 | * available to this memcg, return that limit. | |
1722 | */ | |
1723 | limit = min(limit, memsw); | |
1724 | } | |
1725 | ||
1726 | return limit; | |
a63d83f4 DR |
1727 | } |
1728 | ||
19965460 DR |
1729 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1730 | int order) | |
9cbb78bb DR |
1731 | { |
1732 | struct mem_cgroup *iter; | |
1733 | unsigned long chosen_points = 0; | |
1734 | unsigned long totalpages; | |
1735 | unsigned int points = 0; | |
1736 | struct task_struct *chosen = NULL; | |
1737 | ||
876aafbf | 1738 | /* |
465adcf1 DR |
1739 | * If current has a pending SIGKILL or is exiting, then automatically |
1740 | * select it. The goal is to allow it to allocate so that it may | |
1741 | * quickly exit and free its memory. | |
876aafbf | 1742 | */ |
465adcf1 | 1743 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1744 | set_thread_flag(TIF_MEMDIE); |
1745 | return; | |
1746 | } | |
1747 | ||
1748 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1749 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1750 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1751 | struct css_task_iter it; |
9cbb78bb DR |
1752 | struct task_struct *task; |
1753 | ||
72ec7029 TH |
1754 | css_task_iter_start(&iter->css, &it); |
1755 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1756 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1757 | false)) { | |
1758 | case OOM_SCAN_SELECT: | |
1759 | if (chosen) | |
1760 | put_task_struct(chosen); | |
1761 | chosen = task; | |
1762 | chosen_points = ULONG_MAX; | |
1763 | get_task_struct(chosen); | |
1764 | /* fall through */ | |
1765 | case OOM_SCAN_CONTINUE: | |
1766 | continue; | |
1767 | case OOM_SCAN_ABORT: | |
72ec7029 | 1768 | css_task_iter_end(&it); |
9cbb78bb DR |
1769 | mem_cgroup_iter_break(memcg, iter); |
1770 | if (chosen) | |
1771 | put_task_struct(chosen); | |
1772 | return; | |
1773 | case OOM_SCAN_OK: | |
1774 | break; | |
1775 | }; | |
1776 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1777 | if (!points || points < chosen_points) |
1778 | continue; | |
1779 | /* Prefer thread group leaders for display purposes */ | |
1780 | if (points == chosen_points && | |
1781 | thread_group_leader(chosen)) | |
1782 | continue; | |
1783 | ||
1784 | if (chosen) | |
1785 | put_task_struct(chosen); | |
1786 | chosen = task; | |
1787 | chosen_points = points; | |
1788 | get_task_struct(chosen); | |
9cbb78bb | 1789 | } |
72ec7029 | 1790 | css_task_iter_end(&it); |
9cbb78bb DR |
1791 | } |
1792 | ||
1793 | if (!chosen) | |
1794 | return; | |
1795 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1796 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1797 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1798 | } |
1799 | ||
4d0c066d KH |
1800 | /** |
1801 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1802 | * @memcg: the target memcg |
4d0c066d KH |
1803 | * @nid: the node ID to be checked. |
1804 | * @noswap : specify true here if the user wants flle only information. | |
1805 | * | |
1806 | * This function returns whether the specified memcg contains any | |
1807 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1808 | * pages in the node. | |
1809 | */ | |
c0ff4b85 | 1810 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1811 | int nid, bool noswap) |
1812 | { | |
c0ff4b85 | 1813 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1814 | return true; |
1815 | if (noswap || !total_swap_pages) | |
1816 | return false; | |
c0ff4b85 | 1817 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1818 | return true; |
1819 | return false; | |
1820 | ||
1821 | } | |
bb4cc1a8 | 1822 | #if MAX_NUMNODES > 1 |
889976db YH |
1823 | |
1824 | /* | |
1825 | * Always updating the nodemask is not very good - even if we have an empty | |
1826 | * list or the wrong list here, we can start from some node and traverse all | |
1827 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1828 | * | |
1829 | */ | |
c0ff4b85 | 1830 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1831 | { |
1832 | int nid; | |
453a9bf3 KH |
1833 | /* |
1834 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1835 | * pagein/pageout changes since the last update. | |
1836 | */ | |
c0ff4b85 | 1837 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1838 | return; |
c0ff4b85 | 1839 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1840 | return; |
1841 | ||
889976db | 1842 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1843 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1844 | |
31aaea4a | 1845 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1846 | |
c0ff4b85 R |
1847 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1848 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1849 | } |
453a9bf3 | 1850 | |
c0ff4b85 R |
1851 | atomic_set(&memcg->numainfo_events, 0); |
1852 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1853 | } |
1854 | ||
1855 | /* | |
1856 | * Selecting a node where we start reclaim from. Because what we need is just | |
1857 | * reducing usage counter, start from anywhere is O,K. Considering | |
1858 | * memory reclaim from current node, there are pros. and cons. | |
1859 | * | |
1860 | * Freeing memory from current node means freeing memory from a node which | |
1861 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1862 | * hit limits, it will see a contention on a node. But freeing from remote | |
1863 | * node means more costs for memory reclaim because of memory latency. | |
1864 | * | |
1865 | * Now, we use round-robin. Better algorithm is welcomed. | |
1866 | */ | |
c0ff4b85 | 1867 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1868 | { |
1869 | int node; | |
1870 | ||
c0ff4b85 R |
1871 | mem_cgroup_may_update_nodemask(memcg); |
1872 | node = memcg->last_scanned_node; | |
889976db | 1873 | |
c0ff4b85 | 1874 | node = next_node(node, memcg->scan_nodes); |
889976db | 1875 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1876 | node = first_node(memcg->scan_nodes); |
889976db YH |
1877 | /* |
1878 | * We call this when we hit limit, not when pages are added to LRU. | |
1879 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1880 | * memcg is too small and all pages are not on LRU. In that case, | |
1881 | * we use curret node. | |
1882 | */ | |
1883 | if (unlikely(node == MAX_NUMNODES)) | |
1884 | node = numa_node_id(); | |
1885 | ||
c0ff4b85 | 1886 | memcg->last_scanned_node = node; |
889976db YH |
1887 | return node; |
1888 | } | |
1889 | ||
bb4cc1a8 AM |
1890 | /* |
1891 | * Check all nodes whether it contains reclaimable pages or not. | |
1892 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1893 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1894 | * enough new information. We need to do double check. | |
1895 | */ | |
1896 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1897 | { | |
1898 | int nid; | |
1899 | ||
1900 | /* | |
1901 | * quick check...making use of scan_node. | |
1902 | * We can skip unused nodes. | |
1903 | */ | |
1904 | if (!nodes_empty(memcg->scan_nodes)) { | |
1905 | for (nid = first_node(memcg->scan_nodes); | |
1906 | nid < MAX_NUMNODES; | |
1907 | nid = next_node(nid, memcg->scan_nodes)) { | |
1908 | ||
1909 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1910 | return true; | |
1911 | } | |
1912 | } | |
1913 | /* | |
1914 | * Check rest of nodes. | |
1915 | */ | |
1916 | for_each_node_state(nid, N_MEMORY) { | |
1917 | if (node_isset(nid, memcg->scan_nodes)) | |
1918 | continue; | |
1919 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1920 | return true; | |
1921 | } | |
1922 | return false; | |
1923 | } | |
1924 | ||
889976db | 1925 | #else |
c0ff4b85 | 1926 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1927 | { |
1928 | return 0; | |
1929 | } | |
4d0c066d | 1930 | |
bb4cc1a8 AM |
1931 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
1932 | { | |
1933 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
1934 | } | |
889976db YH |
1935 | #endif |
1936 | ||
0608f43d AM |
1937 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1938 | struct zone *zone, | |
1939 | gfp_t gfp_mask, | |
1940 | unsigned long *total_scanned) | |
1941 | { | |
1942 | struct mem_cgroup *victim = NULL; | |
1943 | int total = 0; | |
1944 | int loop = 0; | |
1945 | unsigned long excess; | |
1946 | unsigned long nr_scanned; | |
1947 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1948 | .zone = zone, | |
1949 | .priority = 0, | |
1950 | }; | |
1951 | ||
1952 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; | |
1953 | ||
1954 | while (1) { | |
1955 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1956 | if (!victim) { | |
1957 | loop++; | |
1958 | if (loop >= 2) { | |
1959 | /* | |
1960 | * If we have not been able to reclaim | |
1961 | * anything, it might because there are | |
1962 | * no reclaimable pages under this hierarchy | |
1963 | */ | |
1964 | if (!total) | |
1965 | break; | |
1966 | /* | |
1967 | * We want to do more targeted reclaim. | |
1968 | * excess >> 2 is not to excessive so as to | |
1969 | * reclaim too much, nor too less that we keep | |
1970 | * coming back to reclaim from this cgroup | |
1971 | */ | |
1972 | if (total >= (excess >> 2) || | |
1973 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1974 | break; | |
1975 | } | |
1976 | continue; | |
1977 | } | |
1978 | if (!mem_cgroup_reclaimable(victim, false)) | |
1979 | continue; | |
1980 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
1981 | zone, &nr_scanned); | |
1982 | *total_scanned += nr_scanned; | |
1983 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
1984 | break; | |
6d61ef40 | 1985 | } |
0608f43d AM |
1986 | mem_cgroup_iter_break(root_memcg, victim); |
1987 | return total; | |
6d61ef40 BS |
1988 | } |
1989 | ||
0056f4e6 JW |
1990 | #ifdef CONFIG_LOCKDEP |
1991 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1992 | .name = "memcg_oom_lock", | |
1993 | }; | |
1994 | #endif | |
1995 | ||
fb2a6fc5 JW |
1996 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1997 | ||
867578cb KH |
1998 | /* |
1999 | * Check OOM-Killer is already running under our hierarchy. | |
2000 | * If someone is running, return false. | |
2001 | */ | |
fb2a6fc5 | 2002 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2003 | { |
79dfdacc | 2004 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2005 | |
fb2a6fc5 JW |
2006 | spin_lock(&memcg_oom_lock); |
2007 | ||
9f3a0d09 | 2008 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2009 | if (iter->oom_lock) { |
79dfdacc MH |
2010 | /* |
2011 | * this subtree of our hierarchy is already locked | |
2012 | * so we cannot give a lock. | |
2013 | */ | |
79dfdacc | 2014 | failed = iter; |
9f3a0d09 JW |
2015 | mem_cgroup_iter_break(memcg, iter); |
2016 | break; | |
23751be0 JW |
2017 | } else |
2018 | iter->oom_lock = true; | |
7d74b06f | 2019 | } |
867578cb | 2020 | |
fb2a6fc5 JW |
2021 | if (failed) { |
2022 | /* | |
2023 | * OK, we failed to lock the whole subtree so we have | |
2024 | * to clean up what we set up to the failing subtree | |
2025 | */ | |
2026 | for_each_mem_cgroup_tree(iter, memcg) { | |
2027 | if (iter == failed) { | |
2028 | mem_cgroup_iter_break(memcg, iter); | |
2029 | break; | |
2030 | } | |
2031 | iter->oom_lock = false; | |
79dfdacc | 2032 | } |
0056f4e6 JW |
2033 | } else |
2034 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2035 | |
2036 | spin_unlock(&memcg_oom_lock); | |
2037 | ||
2038 | return !failed; | |
a636b327 | 2039 | } |
0b7f569e | 2040 | |
fb2a6fc5 | 2041 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2042 | { |
7d74b06f KH |
2043 | struct mem_cgroup *iter; |
2044 | ||
fb2a6fc5 | 2045 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 2046 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 2047 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2048 | iter->oom_lock = false; |
fb2a6fc5 | 2049 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2050 | } |
2051 | ||
c0ff4b85 | 2052 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2053 | { |
2054 | struct mem_cgroup *iter; | |
2055 | ||
c0ff4b85 | 2056 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2057 | atomic_inc(&iter->under_oom); |
2058 | } | |
2059 | ||
c0ff4b85 | 2060 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2061 | { |
2062 | struct mem_cgroup *iter; | |
2063 | ||
867578cb KH |
2064 | /* |
2065 | * When a new child is created while the hierarchy is under oom, | |
2066 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2067 | * atomic_add_unless() here. | |
2068 | */ | |
c0ff4b85 | 2069 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2070 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2071 | } |
2072 | ||
867578cb KH |
2073 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2074 | ||
dc98df5a | 2075 | struct oom_wait_info { |
d79154bb | 2076 | struct mem_cgroup *memcg; |
dc98df5a KH |
2077 | wait_queue_t wait; |
2078 | }; | |
2079 | ||
2080 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2081 | unsigned mode, int sync, void *arg) | |
2082 | { | |
d79154bb HD |
2083 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2084 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2085 | struct oom_wait_info *oom_wait_info; |
2086 | ||
2087 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2088 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2089 | |
dc98df5a | 2090 | /* |
d79154bb | 2091 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2092 | * Then we can use css_is_ancestor without taking care of RCU. |
2093 | */ | |
c0ff4b85 R |
2094 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2095 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2096 | return 0; |
dc98df5a KH |
2097 | return autoremove_wake_function(wait, mode, sync, arg); |
2098 | } | |
2099 | ||
c0ff4b85 | 2100 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2101 | { |
3812c8c8 | 2102 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2103 | /* for filtering, pass "memcg" as argument. */ |
2104 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2105 | } |
2106 | ||
c0ff4b85 | 2107 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2108 | { |
c0ff4b85 R |
2109 | if (memcg && atomic_read(&memcg->under_oom)) |
2110 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2111 | } |
2112 | ||
3812c8c8 | 2113 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2114 | { |
3812c8c8 JW |
2115 | if (!current->memcg_oom.may_oom) |
2116 | return; | |
867578cb | 2117 | /* |
49426420 JW |
2118 | * We are in the middle of the charge context here, so we |
2119 | * don't want to block when potentially sitting on a callstack | |
2120 | * that holds all kinds of filesystem and mm locks. | |
2121 | * | |
2122 | * Also, the caller may handle a failed allocation gracefully | |
2123 | * (like optional page cache readahead) and so an OOM killer | |
2124 | * invocation might not even be necessary. | |
2125 | * | |
2126 | * That's why we don't do anything here except remember the | |
2127 | * OOM context and then deal with it at the end of the page | |
2128 | * fault when the stack is unwound, the locks are released, | |
2129 | * and when we know whether the fault was overall successful. | |
867578cb | 2130 | */ |
49426420 JW |
2131 | css_get(&memcg->css); |
2132 | current->memcg_oom.memcg = memcg; | |
2133 | current->memcg_oom.gfp_mask = mask; | |
2134 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2135 | } |
2136 | ||
2137 | /** | |
2138 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2139 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2140 | * |
49426420 JW |
2141 | * This has to be called at the end of a page fault if the memcg OOM |
2142 | * handler was enabled. | |
3812c8c8 | 2143 | * |
49426420 | 2144 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2145 | * sleep on a waitqueue until the userspace task resolves the |
2146 | * situation. Sleeping directly in the charge context with all kinds | |
2147 | * of locks held is not a good idea, instead we remember an OOM state | |
2148 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2149 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2150 | * |
2151 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2152 | * completed, %false otherwise. |
3812c8c8 | 2153 | */ |
49426420 | 2154 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2155 | { |
49426420 | 2156 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2157 | struct oom_wait_info owait; |
49426420 | 2158 | bool locked; |
3812c8c8 JW |
2159 | |
2160 | /* OOM is global, do not handle */ | |
3812c8c8 | 2161 | if (!memcg) |
49426420 | 2162 | return false; |
3812c8c8 | 2163 | |
49426420 JW |
2164 | if (!handle) |
2165 | goto cleanup; | |
3812c8c8 JW |
2166 | |
2167 | owait.memcg = memcg; | |
2168 | owait.wait.flags = 0; | |
2169 | owait.wait.func = memcg_oom_wake_function; | |
2170 | owait.wait.private = current; | |
2171 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2172 | |
3812c8c8 | 2173 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2174 | mem_cgroup_mark_under_oom(memcg); |
2175 | ||
2176 | locked = mem_cgroup_oom_trylock(memcg); | |
2177 | ||
2178 | if (locked) | |
2179 | mem_cgroup_oom_notify(memcg); | |
2180 | ||
2181 | if (locked && !memcg->oom_kill_disable) { | |
2182 | mem_cgroup_unmark_under_oom(memcg); | |
2183 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2184 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2185 | current->memcg_oom.order); | |
2186 | } else { | |
3812c8c8 | 2187 | schedule(); |
49426420 JW |
2188 | mem_cgroup_unmark_under_oom(memcg); |
2189 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2190 | } | |
2191 | ||
2192 | if (locked) { | |
fb2a6fc5 JW |
2193 | mem_cgroup_oom_unlock(memcg); |
2194 | /* | |
2195 | * There is no guarantee that an OOM-lock contender | |
2196 | * sees the wakeups triggered by the OOM kill | |
2197 | * uncharges. Wake any sleepers explicitely. | |
2198 | */ | |
2199 | memcg_oom_recover(memcg); | |
2200 | } | |
49426420 JW |
2201 | cleanup: |
2202 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2203 | css_put(&memcg->css); |
867578cb | 2204 | return true; |
0b7f569e KH |
2205 | } |
2206 | ||
d69b042f | 2207 | /* |
b5ffc856 | 2208 | * Used to update mapped file or writeback or other statistics. |
32047e2a KH |
2209 | * |
2210 | * Notes: Race condition | |
2211 | * | |
0a31bc97 JW |
2212 | * Charging occurs during page instantiation, while the page is |
2213 | * unmapped and locked in page migration, or while the page table is | |
2214 | * locked in THP migration. No race is possible. | |
32047e2a | 2215 | * |
0a31bc97 | 2216 | * Uncharge happens to pages with zero references, no race possible. |
32047e2a | 2217 | * |
0a31bc97 JW |
2218 | * Charge moving between groups is protected by checking mm->moving |
2219 | * account and taking the move_lock in the slowpath. | |
d69b042f | 2220 | */ |
26174efd | 2221 | |
89c06bd5 KH |
2222 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2223 | bool *locked, unsigned long *flags) | |
2224 | { | |
2225 | struct mem_cgroup *memcg; | |
2226 | struct page_cgroup *pc; | |
2227 | ||
2228 | pc = lookup_page_cgroup(page); | |
2229 | again: | |
2230 | memcg = pc->mem_cgroup; | |
2231 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2232 | return; | |
2233 | /* | |
2234 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2235 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2236 | * rcu_read_lock(), any calls to move_account will be delayed until |
bdcbb659 | 2237 | * rcu_read_unlock(). |
89c06bd5 | 2238 | */ |
bdcbb659 QH |
2239 | VM_BUG_ON(!rcu_read_lock_held()); |
2240 | if (atomic_read(&memcg->moving_account) <= 0) | |
89c06bd5 KH |
2241 | return; |
2242 | ||
2243 | move_lock_mem_cgroup(memcg, flags); | |
2244 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2245 | move_unlock_mem_cgroup(memcg, flags); | |
2246 | goto again; | |
2247 | } | |
2248 | *locked = true; | |
2249 | } | |
2250 | ||
2251 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2252 | { | |
2253 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2254 | ||
2255 | /* | |
2256 | * It's guaranteed that pc->mem_cgroup never changes while | |
2257 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2258 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2259 | */ |
2260 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2261 | } | |
2262 | ||
2a7106f2 | 2263 | void mem_cgroup_update_page_stat(struct page *page, |
68b4876d | 2264 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2265 | { |
c0ff4b85 | 2266 | struct mem_cgroup *memcg; |
32047e2a | 2267 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2268 | unsigned long uninitialized_var(flags); |
d69b042f | 2269 | |
cfa44946 | 2270 | if (mem_cgroup_disabled()) |
d69b042f | 2271 | return; |
89c06bd5 | 2272 | |
658b72c5 | 2273 | VM_BUG_ON(!rcu_read_lock_held()); |
c0ff4b85 R |
2274 | memcg = pc->mem_cgroup; |
2275 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2276 | return; |
26174efd | 2277 | |
c0ff4b85 | 2278 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2279 | } |
26174efd | 2280 | |
cdec2e42 KH |
2281 | /* |
2282 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2283 | * TODO: maybe necessary to use big numbers in big irons. | |
2284 | */ | |
7ec99d62 | 2285 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2286 | struct memcg_stock_pcp { |
2287 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2288 | unsigned int nr_pages; |
cdec2e42 | 2289 | struct work_struct work; |
26fe6168 | 2290 | unsigned long flags; |
a0db00fc | 2291 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2292 | }; |
2293 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2294 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2295 | |
a0956d54 SS |
2296 | /** |
2297 | * consume_stock: Try to consume stocked charge on this cpu. | |
2298 | * @memcg: memcg to consume from. | |
2299 | * @nr_pages: how many pages to charge. | |
2300 | * | |
2301 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2302 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2303 | * service an allocation will refill the stock. | |
2304 | * | |
2305 | * returns true if successful, false otherwise. | |
cdec2e42 | 2306 | */ |
a0956d54 | 2307 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2308 | { |
2309 | struct memcg_stock_pcp *stock; | |
2310 | bool ret = true; | |
2311 | ||
a0956d54 SS |
2312 | if (nr_pages > CHARGE_BATCH) |
2313 | return false; | |
2314 | ||
cdec2e42 | 2315 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2316 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2317 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2318 | else /* need to call res_counter_charge */ |
2319 | ret = false; | |
2320 | put_cpu_var(memcg_stock); | |
2321 | return ret; | |
2322 | } | |
2323 | ||
2324 | /* | |
2325 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2326 | */ | |
2327 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2328 | { | |
2329 | struct mem_cgroup *old = stock->cached; | |
2330 | ||
11c9ea4e JW |
2331 | if (stock->nr_pages) { |
2332 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2333 | ||
2334 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2335 | if (do_swap_account) |
11c9ea4e JW |
2336 | res_counter_uncharge(&old->memsw, bytes); |
2337 | stock->nr_pages = 0; | |
cdec2e42 KH |
2338 | } |
2339 | stock->cached = NULL; | |
cdec2e42 KH |
2340 | } |
2341 | ||
2342 | /* | |
2343 | * This must be called under preempt disabled or must be called by | |
2344 | * a thread which is pinned to local cpu. | |
2345 | */ | |
2346 | static void drain_local_stock(struct work_struct *dummy) | |
2347 | { | |
7c8e0181 | 2348 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2349 | drain_stock(stock); |
26fe6168 | 2350 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2351 | } |
2352 | ||
e4777496 MH |
2353 | static void __init memcg_stock_init(void) |
2354 | { | |
2355 | int cpu; | |
2356 | ||
2357 | for_each_possible_cpu(cpu) { | |
2358 | struct memcg_stock_pcp *stock = | |
2359 | &per_cpu(memcg_stock, cpu); | |
2360 | INIT_WORK(&stock->work, drain_local_stock); | |
2361 | } | |
2362 | } | |
2363 | ||
cdec2e42 KH |
2364 | /* |
2365 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2366 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2367 | */ |
c0ff4b85 | 2368 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2369 | { |
2370 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2371 | ||
c0ff4b85 | 2372 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2373 | drain_stock(stock); |
c0ff4b85 | 2374 | stock->cached = memcg; |
cdec2e42 | 2375 | } |
11c9ea4e | 2376 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2377 | put_cpu_var(memcg_stock); |
2378 | } | |
2379 | ||
2380 | /* | |
c0ff4b85 | 2381 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2382 | * of the hierarchy under it. sync flag says whether we should block |
2383 | * until the work is done. | |
cdec2e42 | 2384 | */ |
c0ff4b85 | 2385 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2386 | { |
26fe6168 | 2387 | int cpu, curcpu; |
d38144b7 | 2388 | |
cdec2e42 | 2389 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2390 | get_online_cpus(); |
5af12d0e | 2391 | curcpu = get_cpu(); |
cdec2e42 KH |
2392 | for_each_online_cpu(cpu) { |
2393 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2394 | struct mem_cgroup *memcg; |
26fe6168 | 2395 | |
c0ff4b85 R |
2396 | memcg = stock->cached; |
2397 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2398 | continue; |
c0ff4b85 | 2399 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2400 | continue; |
d1a05b69 MH |
2401 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2402 | if (cpu == curcpu) | |
2403 | drain_local_stock(&stock->work); | |
2404 | else | |
2405 | schedule_work_on(cpu, &stock->work); | |
2406 | } | |
cdec2e42 | 2407 | } |
5af12d0e | 2408 | put_cpu(); |
d38144b7 MH |
2409 | |
2410 | if (!sync) | |
2411 | goto out; | |
2412 | ||
2413 | for_each_online_cpu(cpu) { | |
2414 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2415 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2416 | flush_work(&stock->work); |
2417 | } | |
2418 | out: | |
f894ffa8 | 2419 | put_online_cpus(); |
d38144b7 MH |
2420 | } |
2421 | ||
2422 | /* | |
2423 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2424 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2425 | * expects some charges will be back to res_counter later but cannot wait for | |
2426 | * it. | |
2427 | */ | |
c0ff4b85 | 2428 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2429 | { |
9f50fad6 MH |
2430 | /* |
2431 | * If someone calls draining, avoid adding more kworker runs. | |
2432 | */ | |
2433 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2434 | return; | |
c0ff4b85 | 2435 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2436 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2437 | } |
2438 | ||
2439 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2440 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2441 | { |
2442 | /* called when force_empty is called */ | |
9f50fad6 | 2443 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2444 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2445 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2446 | } |
2447 | ||
711d3d2c KH |
2448 | /* |
2449 | * This function drains percpu counter value from DEAD cpu and | |
2450 | * move it to local cpu. Note that this function can be preempted. | |
2451 | */ | |
c0ff4b85 | 2452 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2453 | { |
2454 | int i; | |
2455 | ||
c0ff4b85 | 2456 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2457 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2458 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2459 | |
c0ff4b85 R |
2460 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2461 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2462 | } |
e9f8974f | 2463 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2464 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2465 | |
c0ff4b85 R |
2466 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2467 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2468 | } |
c0ff4b85 | 2469 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2470 | } |
2471 | ||
0db0628d | 2472 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2473 | unsigned long action, |
2474 | void *hcpu) | |
2475 | { | |
2476 | int cpu = (unsigned long)hcpu; | |
2477 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2478 | struct mem_cgroup *iter; |
cdec2e42 | 2479 | |
619d094b | 2480 | if (action == CPU_ONLINE) |
1489ebad | 2481 | return NOTIFY_OK; |
1489ebad | 2482 | |
d833049b | 2483 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2484 | return NOTIFY_OK; |
711d3d2c | 2485 | |
9f3a0d09 | 2486 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2487 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2488 | ||
cdec2e42 KH |
2489 | stock = &per_cpu(memcg_stock, cpu); |
2490 | drain_stock(stock); | |
2491 | return NOTIFY_OK; | |
2492 | } | |
2493 | ||
00501b53 JW |
2494 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2495 | unsigned int nr_pages) | |
8a9f3ccd | 2496 | { |
7ec99d62 | 2497 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2498 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 JW |
2499 | struct mem_cgroup *mem_over_limit; |
2500 | struct res_counter *fail_res; | |
2501 | unsigned long nr_reclaimed; | |
6539cc05 | 2502 | unsigned long long size; |
b70a2a21 JW |
2503 | bool may_swap = true; |
2504 | bool drained = false; | |
05b84301 | 2505 | int ret = 0; |
a636b327 | 2506 | |
ce00a967 JW |
2507 | if (mem_cgroup_is_root(memcg)) |
2508 | goto done; | |
6539cc05 | 2509 | retry: |
b6b6cc72 MH |
2510 | if (consume_stock(memcg, nr_pages)) |
2511 | goto done; | |
8a9f3ccd | 2512 | |
6539cc05 | 2513 | size = batch * PAGE_SIZE; |
3fbe7244 JW |
2514 | if (!do_swap_account || |
2515 | !res_counter_charge(&memcg->memsw, size, &fail_res)) { | |
2516 | if (!res_counter_charge(&memcg->res, size, &fail_res)) | |
6539cc05 | 2517 | goto done_restock; |
3fbe7244 JW |
2518 | if (do_swap_account) |
2519 | res_counter_uncharge(&memcg->memsw, size); | |
2520 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
2521 | } else { | |
6539cc05 | 2522 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
b70a2a21 | 2523 | may_swap = false; |
3fbe7244 | 2524 | } |
7a81b88c | 2525 | |
6539cc05 JW |
2526 | if (batch > nr_pages) { |
2527 | batch = nr_pages; | |
2528 | goto retry; | |
2529 | } | |
6d61ef40 | 2530 | |
06b078fc JW |
2531 | /* |
2532 | * Unlike in global OOM situations, memcg is not in a physical | |
2533 | * memory shortage. Allow dying and OOM-killed tasks to | |
2534 | * bypass the last charges so that they can exit quickly and | |
2535 | * free their memory. | |
2536 | */ | |
2537 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2538 | fatal_signal_pending(current) || | |
2539 | current->flags & PF_EXITING)) | |
2540 | goto bypass; | |
2541 | ||
2542 | if (unlikely(task_in_memcg_oom(current))) | |
2543 | goto nomem; | |
2544 | ||
6539cc05 JW |
2545 | if (!(gfp_mask & __GFP_WAIT)) |
2546 | goto nomem; | |
4b534334 | 2547 | |
b70a2a21 JW |
2548 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2549 | gfp_mask, may_swap); | |
6539cc05 | 2550 | |
61e02c74 | 2551 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2552 | goto retry; |
28c34c29 | 2553 | |
b70a2a21 JW |
2554 | if (!drained) { |
2555 | drain_all_stock_async(mem_over_limit); | |
2556 | drained = true; | |
2557 | goto retry; | |
2558 | } | |
2559 | ||
28c34c29 JW |
2560 | if (gfp_mask & __GFP_NORETRY) |
2561 | goto nomem; | |
6539cc05 JW |
2562 | /* |
2563 | * Even though the limit is exceeded at this point, reclaim | |
2564 | * may have been able to free some pages. Retry the charge | |
2565 | * before killing the task. | |
2566 | * | |
2567 | * Only for regular pages, though: huge pages are rather | |
2568 | * unlikely to succeed so close to the limit, and we fall back | |
2569 | * to regular pages anyway in case of failure. | |
2570 | */ | |
61e02c74 | 2571 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2572 | goto retry; |
2573 | /* | |
2574 | * At task move, charge accounts can be doubly counted. So, it's | |
2575 | * better to wait until the end of task_move if something is going on. | |
2576 | */ | |
2577 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2578 | goto retry; | |
2579 | ||
9b130619 JW |
2580 | if (nr_retries--) |
2581 | goto retry; | |
2582 | ||
06b078fc JW |
2583 | if (gfp_mask & __GFP_NOFAIL) |
2584 | goto bypass; | |
2585 | ||
6539cc05 JW |
2586 | if (fatal_signal_pending(current)) |
2587 | goto bypass; | |
2588 | ||
61e02c74 | 2589 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2590 | nomem: |
6d1fdc48 | 2591 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2592 | return -ENOMEM; |
867578cb | 2593 | bypass: |
ce00a967 | 2594 | return -EINTR; |
6539cc05 JW |
2595 | |
2596 | done_restock: | |
2597 | if (batch > nr_pages) | |
2598 | refill_stock(memcg, batch - nr_pages); | |
2599 | done: | |
05b84301 | 2600 | return ret; |
7a81b88c | 2601 | } |
8a9f3ccd | 2602 | |
00501b53 | 2603 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2604 | { |
05b84301 | 2605 | unsigned long bytes = nr_pages * PAGE_SIZE; |
e7018b8d | 2606 | |
ce00a967 JW |
2607 | if (mem_cgroup_is_root(memcg)) |
2608 | return; | |
2609 | ||
05b84301 JW |
2610 | res_counter_uncharge(&memcg->res, bytes); |
2611 | if (do_swap_account) | |
2612 | res_counter_uncharge(&memcg->memsw, bytes); | |
854ffa8d DN |
2613 | } |
2614 | ||
d01dd17f KH |
2615 | /* |
2616 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2617 | * This is useful when moving usage to parent cgroup. | |
2618 | */ | |
2619 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2620 | unsigned int nr_pages) | |
2621 | { | |
2622 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2623 | ||
ce00a967 JW |
2624 | if (mem_cgroup_is_root(memcg)) |
2625 | return; | |
2626 | ||
d01dd17f KH |
2627 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); |
2628 | if (do_swap_account) | |
2629 | res_counter_uncharge_until(&memcg->memsw, | |
2630 | memcg->memsw.parent, bytes); | |
2631 | } | |
2632 | ||
a3b2d692 KH |
2633 | /* |
2634 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2635 | * rcu_read_lock(). The caller is responsible for calling |
2636 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2637 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2638 | */ |
2639 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2640 | { | |
a3b2d692 KH |
2641 | /* ID 0 is unused ID */ |
2642 | if (!id) | |
2643 | return NULL; | |
34c00c31 | 2644 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2645 | } |
2646 | ||
0a31bc97 JW |
2647 | /* |
2648 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2649 | * @page: the page | |
2650 | * | |
2651 | * Look up, get a css reference, and return the memcg that owns @page. | |
2652 | * | |
2653 | * The page must be locked to prevent racing with swap-in and page | |
2654 | * cache charges. If coming from an unlocked page table, the caller | |
2655 | * must ensure the page is on the LRU or this can race with charging. | |
2656 | */ | |
e42d9d5d | 2657 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2658 | { |
c0ff4b85 | 2659 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2660 | struct page_cgroup *pc; |
a3b2d692 | 2661 | unsigned short id; |
b5a84319 KH |
2662 | swp_entry_t ent; |
2663 | ||
309381fe | 2664 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2665 | |
3c776e64 | 2666 | pc = lookup_page_cgroup(page); |
a3b2d692 | 2667 | if (PageCgroupUsed(pc)) { |
c0ff4b85 | 2668 | memcg = pc->mem_cgroup; |
ec903c0c | 2669 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2670 | memcg = NULL; |
e42d9d5d | 2671 | } else if (PageSwapCache(page)) { |
3c776e64 | 2672 | ent.val = page_private(page); |
9fb4b7cc | 2673 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2674 | rcu_read_lock(); |
c0ff4b85 | 2675 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2676 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2677 | memcg = NULL; |
a3b2d692 | 2678 | rcu_read_unlock(); |
3c776e64 | 2679 | } |
c0ff4b85 | 2680 | return memcg; |
b5a84319 KH |
2681 | } |
2682 | ||
0a31bc97 JW |
2683 | static void lock_page_lru(struct page *page, int *isolated) |
2684 | { | |
2685 | struct zone *zone = page_zone(page); | |
2686 | ||
2687 | spin_lock_irq(&zone->lru_lock); | |
2688 | if (PageLRU(page)) { | |
2689 | struct lruvec *lruvec; | |
2690 | ||
2691 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2692 | ClearPageLRU(page); | |
2693 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2694 | *isolated = 1; | |
2695 | } else | |
2696 | *isolated = 0; | |
2697 | } | |
2698 | ||
2699 | static void unlock_page_lru(struct page *page, int isolated) | |
2700 | { | |
2701 | struct zone *zone = page_zone(page); | |
2702 | ||
2703 | if (isolated) { | |
2704 | struct lruvec *lruvec; | |
2705 | ||
2706 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2707 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2708 | SetPageLRU(page); | |
2709 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2710 | } | |
2711 | spin_unlock_irq(&zone->lru_lock); | |
2712 | } | |
2713 | ||
00501b53 | 2714 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2715 | bool lrucare) |
7a81b88c | 2716 | { |
ce587e65 | 2717 | struct page_cgroup *pc = lookup_page_cgroup(page); |
0a31bc97 | 2718 | int isolated; |
9ce70c02 | 2719 | |
309381fe | 2720 | VM_BUG_ON_PAGE(PageCgroupUsed(pc), page); |
ca3e0214 KH |
2721 | /* |
2722 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2723 | * accessed by any other context at this point. | |
2724 | */ | |
9ce70c02 HD |
2725 | |
2726 | /* | |
2727 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2728 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2729 | */ | |
0a31bc97 JW |
2730 | if (lrucare) |
2731 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2732 | |
0a31bc97 JW |
2733 | /* |
2734 | * Nobody should be changing or seriously looking at | |
2735 | * pc->mem_cgroup and pc->flags at this point: | |
2736 | * | |
2737 | * - the page is uncharged | |
2738 | * | |
2739 | * - the page is off-LRU | |
2740 | * | |
2741 | * - an anonymous fault has exclusive page access, except for | |
2742 | * a locked page table | |
2743 | * | |
2744 | * - a page cache insertion, a swapin fault, or a migration | |
2745 | * have the page locked | |
2746 | */ | |
c0ff4b85 | 2747 | pc->mem_cgroup = memcg; |
0a31bc97 | 2748 | pc->flags = PCG_USED | PCG_MEM | (do_swap_account ? PCG_MEMSW : 0); |
9ce70c02 | 2749 | |
0a31bc97 JW |
2750 | if (lrucare) |
2751 | unlock_page_lru(page, isolated); | |
7a81b88c | 2752 | } |
66e1707b | 2753 | |
7cf27982 GC |
2754 | static DEFINE_MUTEX(set_limit_mutex); |
2755 | ||
7ae1e1d0 | 2756 | #ifdef CONFIG_MEMCG_KMEM |
bd673145 VD |
2757 | /* |
2758 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or | |
2759 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. | |
2760 | */ | |
2761 | static DEFINE_MUTEX(memcg_slab_mutex); | |
2762 | ||
d6441637 VD |
2763 | static DEFINE_MUTEX(activate_kmem_mutex); |
2764 | ||
1f458cbf GC |
2765 | /* |
2766 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2767 | * in the memcg_cache_params struct. | |
2768 | */ | |
2769 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2770 | { | |
2771 | struct kmem_cache *cachep; | |
2772 | ||
2773 | VM_BUG_ON(p->is_root_cache); | |
2774 | cachep = p->root_cache; | |
7a67d7ab | 2775 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2776 | } |
2777 | ||
749c5415 | 2778 | #ifdef CONFIG_SLABINFO |
2da8ca82 | 2779 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
749c5415 | 2780 | { |
2da8ca82 | 2781 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
749c5415 GC |
2782 | struct memcg_cache_params *params; |
2783 | ||
cf2b8fbf | 2784 | if (!memcg_kmem_is_active(memcg)) |
749c5415 GC |
2785 | return -EIO; |
2786 | ||
2787 | print_slabinfo_header(m); | |
2788 | ||
bd673145 | 2789 | mutex_lock(&memcg_slab_mutex); |
749c5415 GC |
2790 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) |
2791 | cache_show(memcg_params_to_cache(params), m); | |
bd673145 | 2792 | mutex_unlock(&memcg_slab_mutex); |
749c5415 GC |
2793 | |
2794 | return 0; | |
2795 | } | |
2796 | #endif | |
2797 | ||
c67a8a68 | 2798 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
7ae1e1d0 GC |
2799 | { |
2800 | struct res_counter *fail_res; | |
7ae1e1d0 | 2801 | int ret = 0; |
7ae1e1d0 GC |
2802 | |
2803 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2804 | if (ret) | |
2805 | return ret; | |
2806 | ||
00501b53 | 2807 | ret = try_charge(memcg, gfp, size >> PAGE_SHIFT); |
7ae1e1d0 GC |
2808 | if (ret == -EINTR) { |
2809 | /* | |
00501b53 JW |
2810 | * try_charge() chose to bypass to root due to OOM kill or |
2811 | * fatal signal. Since our only options are to either fail | |
2812 | * the allocation or charge it to this cgroup, do it as a | |
2813 | * temporary condition. But we can't fail. From a kmem/slab | |
2814 | * perspective, the cache has already been selected, by | |
2815 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2816 | * our minds. |
2817 | * | |
2818 | * This condition will only trigger if the task entered | |
00501b53 JW |
2819 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2820 | * during try_charge() above. Tasks that were already dying | |
2821 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2822 | * directed to the root cgroup in memcontrol.h |
2823 | */ | |
2824 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
2825 | if (do_swap_account) | |
2826 | res_counter_charge_nofail(&memcg->memsw, size, | |
2827 | &fail_res); | |
2828 | ret = 0; | |
2829 | } else if (ret) | |
2830 | res_counter_uncharge(&memcg->kmem, size); | |
2831 | ||
2832 | return ret; | |
2833 | } | |
2834 | ||
c67a8a68 | 2835 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) |
7ae1e1d0 | 2836 | { |
7ae1e1d0 GC |
2837 | res_counter_uncharge(&memcg->res, size); |
2838 | if (do_swap_account) | |
2839 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
2840 | |
2841 | /* Not down to 0 */ | |
2842 | if (res_counter_uncharge(&memcg->kmem, size)) | |
2843 | return; | |
2844 | ||
10d5ebf4 LZ |
2845 | /* |
2846 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
2847 | * this last uncharge is racing with the offlining code or it is | |
2848 | * outliving the memcg existence. | |
2849 | * | |
2850 | * The memory barrier imposed by test&clear is paired with the | |
2851 | * explicit one in memcg_kmem_mark_dead(). | |
2852 | */ | |
7de37682 | 2853 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 2854 | css_put(&memcg->css); |
7ae1e1d0 GC |
2855 | } |
2856 | ||
2633d7a0 GC |
2857 | /* |
2858 | * helper for acessing a memcg's index. It will be used as an index in the | |
2859 | * child cache array in kmem_cache, and also to derive its name. This function | |
2860 | * will return -1 when this is not a kmem-limited memcg. | |
2861 | */ | |
2862 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2863 | { | |
2864 | return memcg ? memcg->kmemcg_id : -1; | |
2865 | } | |
2866 | ||
f3bb3043 | 2867 | static int memcg_alloc_cache_id(void) |
55007d84 | 2868 | { |
f3bb3043 VD |
2869 | int id, size; |
2870 | int err; | |
2871 | ||
2872 | id = ida_simple_get(&kmem_limited_groups, | |
2873 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2874 | if (id < 0) | |
2875 | return id; | |
55007d84 | 2876 | |
f3bb3043 VD |
2877 | if (id < memcg_limited_groups_array_size) |
2878 | return id; | |
2879 | ||
2880 | /* | |
2881 | * There's no space for the new id in memcg_caches arrays, | |
2882 | * so we have to grow them. | |
2883 | */ | |
2884 | ||
2885 | size = 2 * (id + 1); | |
55007d84 GC |
2886 | if (size < MEMCG_CACHES_MIN_SIZE) |
2887 | size = MEMCG_CACHES_MIN_SIZE; | |
2888 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2889 | size = MEMCG_CACHES_MAX_SIZE; | |
2890 | ||
f3bb3043 VD |
2891 | mutex_lock(&memcg_slab_mutex); |
2892 | err = memcg_update_all_caches(size); | |
2893 | mutex_unlock(&memcg_slab_mutex); | |
2894 | ||
2895 | if (err) { | |
2896 | ida_simple_remove(&kmem_limited_groups, id); | |
2897 | return err; | |
2898 | } | |
2899 | return id; | |
2900 | } | |
2901 | ||
2902 | static void memcg_free_cache_id(int id) | |
2903 | { | |
2904 | ida_simple_remove(&kmem_limited_groups, id); | |
55007d84 GC |
2905 | } |
2906 | ||
2907 | /* | |
2908 | * We should update the current array size iff all caches updates succeed. This | |
2909 | * can only be done from the slab side. The slab mutex needs to be held when | |
2910 | * calling this. | |
2911 | */ | |
2912 | void memcg_update_array_size(int num) | |
2913 | { | |
f3bb3043 | 2914 | memcg_limited_groups_array_size = num; |
55007d84 GC |
2915 | } |
2916 | ||
776ed0f0 VD |
2917 | static void memcg_register_cache(struct mem_cgroup *memcg, |
2918 | struct kmem_cache *root_cache) | |
2633d7a0 | 2919 | { |
93f39eea VD |
2920 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by |
2921 | memcg_slab_mutex */ | |
bd673145 | 2922 | struct kmem_cache *cachep; |
d7f25f8a GC |
2923 | int id; |
2924 | ||
bd673145 VD |
2925 | lockdep_assert_held(&memcg_slab_mutex); |
2926 | ||
2927 | id = memcg_cache_id(memcg); | |
2928 | ||
2929 | /* | |
2930 | * Since per-memcg caches are created asynchronously on first | |
2931 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
2932 | * create the same cache, but only one of them may succeed. | |
2933 | */ | |
2934 | if (cache_from_memcg_idx(root_cache, id)) | |
1aa13254 VD |
2935 | return; |
2936 | ||
073ee1c6 | 2937 | cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1); |
776ed0f0 | 2938 | cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf); |
2edefe11 | 2939 | /* |
bd673145 VD |
2940 | * If we could not create a memcg cache, do not complain, because |
2941 | * that's not critical at all as we can always proceed with the root | |
2942 | * cache. | |
2edefe11 | 2943 | */ |
bd673145 VD |
2944 | if (!cachep) |
2945 | return; | |
2edefe11 | 2946 | |
33a690c4 | 2947 | css_get(&memcg->css); |
bd673145 | 2948 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
1aa13254 | 2949 | |
d7f25f8a | 2950 | /* |
959c8963 VD |
2951 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
2952 | * barrier here to ensure nobody will see the kmem_cache partially | |
2953 | * initialized. | |
d7f25f8a | 2954 | */ |
959c8963 VD |
2955 | smp_wmb(); |
2956 | ||
bd673145 VD |
2957 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
2958 | root_cache->memcg_params->memcg_caches[id] = cachep; | |
1aa13254 | 2959 | } |
d7f25f8a | 2960 | |
776ed0f0 | 2961 | static void memcg_unregister_cache(struct kmem_cache *cachep) |
1aa13254 | 2962 | { |
bd673145 | 2963 | struct kmem_cache *root_cache; |
1aa13254 VD |
2964 | struct mem_cgroup *memcg; |
2965 | int id; | |
2966 | ||
bd673145 | 2967 | lockdep_assert_held(&memcg_slab_mutex); |
d7f25f8a | 2968 | |
bd673145 | 2969 | BUG_ON(is_root_cache(cachep)); |
2edefe11 | 2970 | |
bd673145 VD |
2971 | root_cache = cachep->memcg_params->root_cache; |
2972 | memcg = cachep->memcg_params->memcg; | |
96403da2 | 2973 | id = memcg_cache_id(memcg); |
d7f25f8a | 2974 | |
bd673145 VD |
2975 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
2976 | root_cache->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a | 2977 | |
bd673145 VD |
2978 | list_del(&cachep->memcg_params->list); |
2979 | ||
2980 | kmem_cache_destroy(cachep); | |
33a690c4 VD |
2981 | |
2982 | /* drop the reference taken in memcg_register_cache */ | |
2983 | css_put(&memcg->css); | |
2633d7a0 GC |
2984 | } |
2985 | ||
0e9d92f2 GC |
2986 | /* |
2987 | * During the creation a new cache, we need to disable our accounting mechanism | |
2988 | * altogether. This is true even if we are not creating, but rather just | |
2989 | * enqueing new caches to be created. | |
2990 | * | |
2991 | * This is because that process will trigger allocations; some visible, like | |
2992 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
2993 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
2994 | * objects during debug. | |
2995 | * | |
2996 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
2997 | * to it. This may not be a bounded recursion: since the first cache creation | |
2998 | * failed to complete (waiting on the allocation), we'll just try to create the | |
2999 | * cache again, failing at the same point. | |
3000 | * | |
3001 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3002 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3003 | * inside the following two functions. | |
3004 | */ | |
3005 | static inline void memcg_stop_kmem_account(void) | |
3006 | { | |
3007 | VM_BUG_ON(!current->mm); | |
3008 | current->memcg_kmem_skip_account++; | |
3009 | } | |
3010 | ||
3011 | static inline void memcg_resume_kmem_account(void) | |
3012 | { | |
3013 | VM_BUG_ON(!current->mm); | |
3014 | current->memcg_kmem_skip_account--; | |
3015 | } | |
3016 | ||
776ed0f0 | 3017 | int __memcg_cleanup_cache_params(struct kmem_cache *s) |
7cf27982 GC |
3018 | { |
3019 | struct kmem_cache *c; | |
b8529907 | 3020 | int i, failed = 0; |
7cf27982 | 3021 | |
bd673145 | 3022 | mutex_lock(&memcg_slab_mutex); |
7a67d7ab QH |
3023 | for_each_memcg_cache_index(i) { |
3024 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
3025 | if (!c) |
3026 | continue; | |
3027 | ||
776ed0f0 | 3028 | memcg_unregister_cache(c); |
b8529907 VD |
3029 | |
3030 | if (cache_from_memcg_idx(s, i)) | |
3031 | failed++; | |
7cf27982 | 3032 | } |
bd673145 | 3033 | mutex_unlock(&memcg_slab_mutex); |
b8529907 | 3034 | return failed; |
7cf27982 GC |
3035 | } |
3036 | ||
776ed0f0 | 3037 | static void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3038 | { |
3039 | struct kmem_cache *cachep; | |
bd673145 | 3040 | struct memcg_cache_params *params, *tmp; |
1f458cbf GC |
3041 | |
3042 | if (!memcg_kmem_is_active(memcg)) | |
3043 | return; | |
3044 | ||
bd673145 VD |
3045 | mutex_lock(&memcg_slab_mutex); |
3046 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { | |
1f458cbf | 3047 | cachep = memcg_params_to_cache(params); |
bd673145 VD |
3048 | kmem_cache_shrink(cachep); |
3049 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
776ed0f0 | 3050 | memcg_unregister_cache(cachep); |
1f458cbf | 3051 | } |
bd673145 | 3052 | mutex_unlock(&memcg_slab_mutex); |
1f458cbf GC |
3053 | } |
3054 | ||
776ed0f0 | 3055 | struct memcg_register_cache_work { |
5722d094 VD |
3056 | struct mem_cgroup *memcg; |
3057 | struct kmem_cache *cachep; | |
3058 | struct work_struct work; | |
3059 | }; | |
3060 | ||
776ed0f0 | 3061 | static void memcg_register_cache_func(struct work_struct *w) |
d7f25f8a | 3062 | { |
776ed0f0 VD |
3063 | struct memcg_register_cache_work *cw = |
3064 | container_of(w, struct memcg_register_cache_work, work); | |
5722d094 VD |
3065 | struct mem_cgroup *memcg = cw->memcg; |
3066 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 3067 | |
bd673145 | 3068 | mutex_lock(&memcg_slab_mutex); |
776ed0f0 | 3069 | memcg_register_cache(memcg, cachep); |
bd673145 VD |
3070 | mutex_unlock(&memcg_slab_mutex); |
3071 | ||
5722d094 | 3072 | css_put(&memcg->css); |
d7f25f8a GC |
3073 | kfree(cw); |
3074 | } | |
3075 | ||
3076 | /* | |
3077 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3078 | */ |
776ed0f0 VD |
3079 | static void __memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3080 | struct kmem_cache *cachep) | |
d7f25f8a | 3081 | { |
776ed0f0 | 3082 | struct memcg_register_cache_work *cw; |
d7f25f8a | 3083 | |
776ed0f0 | 3084 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
ca0dde97 LZ |
3085 | if (cw == NULL) { |
3086 | css_put(&memcg->css); | |
d7f25f8a GC |
3087 | return; |
3088 | } | |
3089 | ||
3090 | cw->memcg = memcg; | |
3091 | cw->cachep = cachep; | |
3092 | ||
776ed0f0 | 3093 | INIT_WORK(&cw->work, memcg_register_cache_func); |
d7f25f8a GC |
3094 | schedule_work(&cw->work); |
3095 | } | |
3096 | ||
776ed0f0 VD |
3097 | static void memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3098 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
3099 | { |
3100 | /* | |
3101 | * We need to stop accounting when we kmalloc, because if the | |
3102 | * corresponding kmalloc cache is not yet created, the first allocation | |
776ed0f0 | 3103 | * in __memcg_schedule_register_cache will recurse. |
0e9d92f2 GC |
3104 | * |
3105 | * However, it is better to enclose the whole function. Depending on | |
3106 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3107 | * trigger an allocation. This too, will make us recurse. Because at | |
3108 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3109 | * the safest choice is to do it like this, wrapping the whole function. | |
3110 | */ | |
3111 | memcg_stop_kmem_account(); | |
776ed0f0 | 3112 | __memcg_schedule_register_cache(memcg, cachep); |
0e9d92f2 GC |
3113 | memcg_resume_kmem_account(); |
3114 | } | |
c67a8a68 VD |
3115 | |
3116 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) | |
3117 | { | |
3118 | int res; | |
3119 | ||
3120 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, | |
3121 | PAGE_SIZE << order); | |
3122 | if (!res) | |
3123 | atomic_add(1 << order, &cachep->memcg_params->nr_pages); | |
3124 | return res; | |
3125 | } | |
3126 | ||
3127 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) | |
3128 | { | |
3129 | memcg_uncharge_kmem(cachep->memcg_params->memcg, PAGE_SIZE << order); | |
3130 | atomic_sub(1 << order, &cachep->memcg_params->nr_pages); | |
3131 | } | |
3132 | ||
d7f25f8a GC |
3133 | /* |
3134 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3135 | * We try to use the current memcg's version of the cache. | |
3136 | * | |
3137 | * If the cache does not exist yet, if we are the first user of it, | |
3138 | * we either create it immediately, if possible, or create it asynchronously | |
3139 | * in a workqueue. | |
3140 | * In the latter case, we will let the current allocation go through with | |
3141 | * the original cache. | |
3142 | * | |
3143 | * Can't be called in interrupt context or from kernel threads. | |
3144 | * This function needs to be called with rcu_read_lock() held. | |
3145 | */ | |
3146 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3147 | gfp_t gfp) | |
3148 | { | |
3149 | struct mem_cgroup *memcg; | |
959c8963 | 3150 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
3151 | |
3152 | VM_BUG_ON(!cachep->memcg_params); | |
3153 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3154 | ||
0e9d92f2 GC |
3155 | if (!current->mm || current->memcg_kmem_skip_account) |
3156 | return cachep; | |
3157 | ||
d7f25f8a GC |
3158 | rcu_read_lock(); |
3159 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a | 3160 | |
cf2b8fbf | 3161 | if (!memcg_kmem_is_active(memcg)) |
ca0dde97 | 3162 | goto out; |
d7f25f8a | 3163 | |
959c8963 VD |
3164 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
3165 | if (likely(memcg_cachep)) { | |
3166 | cachep = memcg_cachep; | |
ca0dde97 | 3167 | goto out; |
d7f25f8a GC |
3168 | } |
3169 | ||
ca0dde97 | 3170 | /* The corresponding put will be done in the workqueue. */ |
ec903c0c | 3171 | if (!css_tryget_online(&memcg->css)) |
ca0dde97 LZ |
3172 | goto out; |
3173 | rcu_read_unlock(); | |
3174 | ||
3175 | /* | |
3176 | * If we are in a safe context (can wait, and not in interrupt | |
3177 | * context), we could be be predictable and return right away. | |
3178 | * This would guarantee that the allocation being performed | |
3179 | * already belongs in the new cache. | |
3180 | * | |
3181 | * However, there are some clashes that can arrive from locking. | |
3182 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
3183 | * memcg_create_kmem_cache, this means no further allocation |
3184 | * could happen with the slab_mutex held. So it's better to | |
3185 | * defer everything. | |
ca0dde97 | 3186 | */ |
776ed0f0 | 3187 | memcg_schedule_register_cache(memcg, cachep); |
ca0dde97 LZ |
3188 | return cachep; |
3189 | out: | |
3190 | rcu_read_unlock(); | |
3191 | return cachep; | |
d7f25f8a | 3192 | } |
d7f25f8a | 3193 | |
7ae1e1d0 GC |
3194 | /* |
3195 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3196 | * possible for the given order. But the page is not allocated yet, so we'll | |
3197 | * need a further commit step to do the final arrangements. | |
3198 | * | |
3199 | * It is possible for the task to switch cgroups in this mean time, so at | |
3200 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3201 | * the handle argument to return to the caller which cgroup we should commit | |
3202 | * against. We could also return the memcg directly and avoid the pointer | |
3203 | * passing, but a boolean return value gives better semantics considering | |
3204 | * the compiled-out case as well. | |
3205 | * | |
3206 | * Returning true means the allocation is possible. | |
3207 | */ | |
3208 | bool | |
3209 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3210 | { | |
3211 | struct mem_cgroup *memcg; | |
3212 | int ret; | |
3213 | ||
3214 | *_memcg = NULL; | |
6d42c232 GC |
3215 | |
3216 | /* | |
3217 | * Disabling accounting is only relevant for some specific memcg | |
3218 | * internal allocations. Therefore we would initially not have such | |
52383431 VD |
3219 | * check here, since direct calls to the page allocator that are |
3220 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen | |
3221 | * outside memcg core. We are mostly concerned with cache allocations, | |
3222 | * and by having this test at memcg_kmem_get_cache, we are already able | |
3223 | * to relay the allocation to the root cache and bypass the memcg cache | |
3224 | * altogether. | |
6d42c232 GC |
3225 | * |
3226 | * There is one exception, though: the SLUB allocator does not create | |
3227 | * large order caches, but rather service large kmallocs directly from | |
3228 | * the page allocator. Therefore, the following sequence when backed by | |
3229 | * the SLUB allocator: | |
3230 | * | |
f894ffa8 AM |
3231 | * memcg_stop_kmem_account(); |
3232 | * kmalloc(<large_number>) | |
3233 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3234 | * |
3235 | * would effectively ignore the fact that we should skip accounting, | |
3236 | * since it will drive us directly to this function without passing | |
3237 | * through the cache selector memcg_kmem_get_cache. Such large | |
3238 | * allocations are extremely rare but can happen, for instance, for the | |
3239 | * cache arrays. We bring this test here. | |
3240 | */ | |
3241 | if (!current->mm || current->memcg_kmem_skip_account) | |
3242 | return true; | |
3243 | ||
df381975 | 3244 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 3245 | |
cf2b8fbf | 3246 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
3247 | css_put(&memcg->css); |
3248 | return true; | |
3249 | } | |
3250 | ||
7ae1e1d0 GC |
3251 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3252 | if (!ret) | |
3253 | *_memcg = memcg; | |
7ae1e1d0 GC |
3254 | |
3255 | css_put(&memcg->css); | |
3256 | return (ret == 0); | |
3257 | } | |
3258 | ||
3259 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3260 | int order) | |
3261 | { | |
3262 | struct page_cgroup *pc; | |
3263 | ||
3264 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3265 | ||
3266 | /* The page allocation failed. Revert */ | |
3267 | if (!page) { | |
3268 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3269 | return; |
3270 | } | |
a840cda6 JW |
3271 | /* |
3272 | * The page is freshly allocated and not visible to any | |
3273 | * outside callers yet. Set up pc non-atomically. | |
3274 | */ | |
7ae1e1d0 | 3275 | pc = lookup_page_cgroup(page); |
7ae1e1d0 | 3276 | pc->mem_cgroup = memcg; |
a840cda6 | 3277 | pc->flags = PCG_USED; |
7ae1e1d0 GC |
3278 | } |
3279 | ||
3280 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3281 | { | |
3282 | struct mem_cgroup *memcg = NULL; | |
3283 | struct page_cgroup *pc; | |
3284 | ||
3285 | ||
3286 | pc = lookup_page_cgroup(page); | |
7ae1e1d0 GC |
3287 | if (!PageCgroupUsed(pc)) |
3288 | return; | |
3289 | ||
a840cda6 JW |
3290 | memcg = pc->mem_cgroup; |
3291 | pc->flags = 0; | |
7ae1e1d0 GC |
3292 | |
3293 | /* | |
3294 | * We trust that only if there is a memcg associated with the page, it | |
3295 | * is a valid allocation | |
3296 | */ | |
3297 | if (!memcg) | |
3298 | return; | |
3299 | ||
309381fe | 3300 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
7ae1e1d0 | 3301 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); |
7ae1e1d0 | 3302 | } |
1f458cbf | 3303 | #else |
776ed0f0 | 3304 | static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3305 | { |
3306 | } | |
7ae1e1d0 GC |
3307 | #endif /* CONFIG_MEMCG_KMEM */ |
3308 | ||
ca3e0214 KH |
3309 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3310 | ||
ca3e0214 KH |
3311 | /* |
3312 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3313 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3314 | * charge/uncharge will be never happen and move_account() is done under | |
3315 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3316 | */ |
e94c8a9c | 3317 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3318 | { |
3319 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3320 | struct page_cgroup *pc; |
b070e65c | 3321 | struct mem_cgroup *memcg; |
e94c8a9c | 3322 | int i; |
ca3e0214 | 3323 | |
3d37c4a9 KH |
3324 | if (mem_cgroup_disabled()) |
3325 | return; | |
b070e65c DR |
3326 | |
3327 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3328 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3329 | pc = head_pc + i; | |
b070e65c | 3330 | pc->mem_cgroup = memcg; |
0a31bc97 | 3331 | pc->flags = head_pc->flags; |
e94c8a9c | 3332 | } |
b070e65c DR |
3333 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3334 | HPAGE_PMD_NR); | |
ca3e0214 | 3335 | } |
12d27107 | 3336 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3337 | |
f817ed48 | 3338 | /** |
de3638d9 | 3339 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3340 | * @page: the page |
7ec99d62 | 3341 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3342 | * @pc: page_cgroup of the page. |
3343 | * @from: mem_cgroup which the page is moved from. | |
3344 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3345 | * | |
3346 | * The caller must confirm following. | |
08e552c6 | 3347 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3348 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3349 | * |
2f3479b1 KH |
3350 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3351 | * from old cgroup. | |
f817ed48 | 3352 | */ |
7ec99d62 JW |
3353 | static int mem_cgroup_move_account(struct page *page, |
3354 | unsigned int nr_pages, | |
3355 | struct page_cgroup *pc, | |
3356 | struct mem_cgroup *from, | |
2f3479b1 | 3357 | struct mem_cgroup *to) |
f817ed48 | 3358 | { |
de3638d9 JW |
3359 | unsigned long flags; |
3360 | int ret; | |
987eba66 | 3361 | |
f817ed48 | 3362 | VM_BUG_ON(from == to); |
309381fe | 3363 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3364 | /* |
3365 | * The page is isolated from LRU. So, collapse function | |
3366 | * will not handle this page. But page splitting can happen. | |
3367 | * Do this check under compound_page_lock(). The caller should | |
3368 | * hold it. | |
3369 | */ | |
3370 | ret = -EBUSY; | |
7ec99d62 | 3371 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3372 | goto out; |
3373 | ||
0a31bc97 JW |
3374 | /* |
3375 | * Prevent mem_cgroup_migrate() from looking at pc->mem_cgroup | |
3376 | * of its source page while we change it: page migration takes | |
3377 | * both pages off the LRU, but page cache replacement doesn't. | |
3378 | */ | |
3379 | if (!trylock_page(page)) | |
3380 | goto out; | |
de3638d9 JW |
3381 | |
3382 | ret = -EINVAL; | |
3383 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
0a31bc97 | 3384 | goto out_unlock; |
de3638d9 | 3385 | |
312734c0 | 3386 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3387 | |
0a31bc97 | 3388 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
3389 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
3390 | nr_pages); | |
3391 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3392 | nr_pages); | |
3393 | } | |
3ea67d06 | 3394 | |
59d1d256 JW |
3395 | if (PageWriteback(page)) { |
3396 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3397 | nr_pages); | |
3398 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3399 | nr_pages); | |
3400 | } | |
3ea67d06 | 3401 | |
0a31bc97 JW |
3402 | /* |
3403 | * It is safe to change pc->mem_cgroup here because the page | |
3404 | * is referenced, charged, and isolated - we can't race with | |
3405 | * uncharging, charging, migration, or LRU putback. | |
3406 | */ | |
d69b042f | 3407 | |
854ffa8d | 3408 | /* caller should have done css_get */ |
08e552c6 | 3409 | pc->mem_cgroup = to; |
312734c0 | 3410 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 | 3411 | ret = 0; |
0a31bc97 JW |
3412 | |
3413 | local_irq_disable(); | |
3414 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 3415 | memcg_check_events(to, page); |
0a31bc97 | 3416 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 3417 | memcg_check_events(from, page); |
0a31bc97 JW |
3418 | local_irq_enable(); |
3419 | out_unlock: | |
3420 | unlock_page(page); | |
de3638d9 | 3421 | out: |
f817ed48 KH |
3422 | return ret; |
3423 | } | |
3424 | ||
2ef37d3f MH |
3425 | /** |
3426 | * mem_cgroup_move_parent - moves page to the parent group | |
3427 | * @page: the page to move | |
3428 | * @pc: page_cgroup of the page | |
3429 | * @child: page's cgroup | |
3430 | * | |
3431 | * move charges to its parent or the root cgroup if the group has no | |
3432 | * parent (aka use_hierarchy==0). | |
3433 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3434 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3435 | * it signals a race with a page removal/uncharge or migration. In the | |
3436 | * first case the page is on the way out and it will vanish from the LRU | |
3437 | * on the next attempt and the call should be retried later. | |
3438 | * Isolation from the LRU fails only if page has been isolated from | |
3439 | * the LRU since we looked at it and that usually means either global | |
3440 | * reclaim or migration going on. The page will either get back to the | |
3441 | * LRU or vanish. | |
3442 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3443 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3444 | * disappear in the next attempt. | |
f817ed48 | 3445 | */ |
5564e88b JW |
3446 | static int mem_cgroup_move_parent(struct page *page, |
3447 | struct page_cgroup *pc, | |
6068bf01 | 3448 | struct mem_cgroup *child) |
f817ed48 | 3449 | { |
f817ed48 | 3450 | struct mem_cgroup *parent; |
7ec99d62 | 3451 | unsigned int nr_pages; |
4be4489f | 3452 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3453 | int ret; |
3454 | ||
d8423011 | 3455 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3456 | |
57f9fd7d DN |
3457 | ret = -EBUSY; |
3458 | if (!get_page_unless_zero(page)) | |
3459 | goto out; | |
3460 | if (isolate_lru_page(page)) | |
3461 | goto put; | |
52dbb905 | 3462 | |
7ec99d62 | 3463 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3464 | |
cc926f78 KH |
3465 | parent = parent_mem_cgroup(child); |
3466 | /* | |
3467 | * If no parent, move charges to root cgroup. | |
3468 | */ | |
3469 | if (!parent) | |
3470 | parent = root_mem_cgroup; | |
f817ed48 | 3471 | |
2ef37d3f | 3472 | if (nr_pages > 1) { |
309381fe | 3473 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
987eba66 | 3474 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3475 | } |
987eba66 | 3476 | |
cc926f78 | 3477 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3478 | pc, child, parent); |
cc926f78 KH |
3479 | if (!ret) |
3480 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3481 | |
7ec99d62 | 3482 | if (nr_pages > 1) |
987eba66 | 3483 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3484 | putback_lru_page(page); |
57f9fd7d | 3485 | put: |
40d58138 | 3486 | put_page(page); |
57f9fd7d | 3487 | out: |
f817ed48 KH |
3488 | return ret; |
3489 | } | |
3490 | ||
c255a458 | 3491 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
3492 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
3493 | bool charge) | |
d13d1443 | 3494 | { |
0a31bc97 JW |
3495 | int val = (charge) ? 1 : -1; |
3496 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 3497 | } |
02491447 DN |
3498 | |
3499 | /** | |
3500 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3501 | * @entry: swap entry to be moved | |
3502 | * @from: mem_cgroup which the entry is moved from | |
3503 | * @to: mem_cgroup which the entry is moved to | |
3504 | * | |
3505 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3506 | * as the mem_cgroup's id of @from. | |
3507 | * | |
3508 | * Returns 0 on success, -EINVAL on failure. | |
3509 | * | |
3510 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3511 | * both res and memsw, and called css_get(). | |
3512 | */ | |
3513 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3514 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3515 | { |
3516 | unsigned short old_id, new_id; | |
3517 | ||
34c00c31 LZ |
3518 | old_id = mem_cgroup_id(from); |
3519 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3520 | |
3521 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3522 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3523 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3524 | /* |
483c30b5 DN |
3525 | * This function is only called from task migration context now. |
3526 | * It postpones res_counter and refcount handling till the end | |
3527 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
3528 | * improvement. But we cannot postpone css_get(to) because if |
3529 | * the process that has been moved to @to does swap-in, the | |
3530 | * refcount of @to might be decreased to 0. | |
3531 | * | |
3532 | * We are in attach() phase, so the cgroup is guaranteed to be | |
3533 | * alive, so we can just call css_get(). | |
02491447 | 3534 | */ |
4050377b | 3535 | css_get(&to->css); |
02491447 DN |
3536 | return 0; |
3537 | } | |
3538 | return -EINVAL; | |
3539 | } | |
3540 | #else | |
3541 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3542 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3543 | { |
3544 | return -EINVAL; | |
3545 | } | |
8c7c6e34 | 3546 | #endif |
d13d1443 | 3547 | |
f212ad7c DN |
3548 | #ifdef CONFIG_DEBUG_VM |
3549 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3550 | { | |
3551 | struct page_cgroup *pc; | |
3552 | ||
3553 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3554 | /* |
3555 | * Can be NULL while feeding pages into the page allocator for | |
3556 | * the first time, i.e. during boot or memory hotplug; | |
3557 | * or when mem_cgroup_disabled(). | |
3558 | */ | |
f212ad7c DN |
3559 | if (likely(pc) && PageCgroupUsed(pc)) |
3560 | return pc; | |
3561 | return NULL; | |
3562 | } | |
3563 | ||
3564 | bool mem_cgroup_bad_page_check(struct page *page) | |
3565 | { | |
3566 | if (mem_cgroup_disabled()) | |
3567 | return false; | |
3568 | ||
3569 | return lookup_page_cgroup_used(page) != NULL; | |
3570 | } | |
3571 | ||
3572 | void mem_cgroup_print_bad_page(struct page *page) | |
3573 | { | |
3574 | struct page_cgroup *pc; | |
3575 | ||
3576 | pc = lookup_page_cgroup_used(page); | |
3577 | if (pc) { | |
d045197f AM |
3578 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
3579 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
3580 | } |
3581 | } | |
3582 | #endif | |
3583 | ||
d38d2a75 | 3584 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3585 | unsigned long long val) |
628f4235 | 3586 | { |
81d39c20 | 3587 | int retry_count; |
628f4235 | 3588 | int ret = 0; |
81d39c20 KH |
3589 | int children = mem_cgroup_count_children(memcg); |
3590 | u64 curusage, oldusage; | |
3c11ecf4 | 3591 | int enlarge; |
81d39c20 KH |
3592 | |
3593 | /* | |
3594 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3595 | * is depends on callers. We set our retry-count to be function | |
3596 | * of # of children which we should visit in this loop. | |
3597 | */ | |
3598 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3599 | ||
3600 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3601 | |
3c11ecf4 | 3602 | enlarge = 0; |
8c7c6e34 | 3603 | while (retry_count) { |
628f4235 KH |
3604 | if (signal_pending(current)) { |
3605 | ret = -EINTR; | |
3606 | break; | |
3607 | } | |
8c7c6e34 KH |
3608 | /* |
3609 | * Rather than hide all in some function, I do this in | |
3610 | * open coded manner. You see what this really does. | |
aaad153e | 3611 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3612 | */ |
3613 | mutex_lock(&set_limit_mutex); | |
3fbe7244 | 3614 | if (res_counter_read_u64(&memcg->memsw, RES_LIMIT) < val) { |
8c7c6e34 KH |
3615 | ret = -EINVAL; |
3616 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3617 | break; |
3618 | } | |
3c11ecf4 | 3619 | |
3fbe7244 | 3620 | if (res_counter_read_u64(&memcg->res, RES_LIMIT) < val) |
3c11ecf4 KH |
3621 | enlarge = 1; |
3622 | ||
8c7c6e34 KH |
3623 | ret = res_counter_set_limit(&memcg->res, val); |
3624 | mutex_unlock(&set_limit_mutex); | |
3625 | ||
3626 | if (!ret) | |
3627 | break; | |
3628 | ||
b70a2a21 JW |
3629 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
3630 | ||
81d39c20 KH |
3631 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3632 | /* Usage is reduced ? */ | |
f894ffa8 | 3633 | if (curusage >= oldusage) |
81d39c20 KH |
3634 | retry_count--; |
3635 | else | |
3636 | oldusage = curusage; | |
8c7c6e34 | 3637 | } |
3c11ecf4 KH |
3638 | if (!ret && enlarge) |
3639 | memcg_oom_recover(memcg); | |
14797e23 | 3640 | |
8c7c6e34 KH |
3641 | return ret; |
3642 | } | |
3643 | ||
338c8431 LZ |
3644 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3645 | unsigned long long val) | |
8c7c6e34 | 3646 | { |
81d39c20 | 3647 | int retry_count; |
3fbe7244 | 3648 | u64 oldusage, curusage; |
81d39c20 KH |
3649 | int children = mem_cgroup_count_children(memcg); |
3650 | int ret = -EBUSY; | |
3c11ecf4 | 3651 | int enlarge = 0; |
8c7c6e34 | 3652 | |
81d39c20 | 3653 | /* see mem_cgroup_resize_res_limit */ |
f894ffa8 | 3654 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
81d39c20 | 3655 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
8c7c6e34 KH |
3656 | while (retry_count) { |
3657 | if (signal_pending(current)) { | |
3658 | ret = -EINTR; | |
3659 | break; | |
3660 | } | |
3661 | /* | |
3662 | * Rather than hide all in some function, I do this in | |
3663 | * open coded manner. You see what this really does. | |
aaad153e | 3664 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3665 | */ |
3666 | mutex_lock(&set_limit_mutex); | |
3fbe7244 | 3667 | if (res_counter_read_u64(&memcg->res, RES_LIMIT) > val) { |
8c7c6e34 KH |
3668 | ret = -EINVAL; |
3669 | mutex_unlock(&set_limit_mutex); | |
3670 | break; | |
3671 | } | |
3fbe7244 | 3672 | if (res_counter_read_u64(&memcg->memsw, RES_LIMIT) < val) |
3c11ecf4 | 3673 | enlarge = 1; |
8c7c6e34 KH |
3674 | ret = res_counter_set_limit(&memcg->memsw, val); |
3675 | mutex_unlock(&set_limit_mutex); | |
3676 | ||
3677 | if (!ret) | |
3678 | break; | |
3679 | ||
b70a2a21 JW |
3680 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3681 | ||
8c7c6e34 | 3682 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3683 | /* Usage is reduced ? */ |
8c7c6e34 | 3684 | if (curusage >= oldusage) |
628f4235 | 3685 | retry_count--; |
81d39c20 KH |
3686 | else |
3687 | oldusage = curusage; | |
628f4235 | 3688 | } |
3c11ecf4 KH |
3689 | if (!ret && enlarge) |
3690 | memcg_oom_recover(memcg); | |
628f4235 KH |
3691 | return ret; |
3692 | } | |
3693 | ||
0608f43d AM |
3694 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3695 | gfp_t gfp_mask, | |
3696 | unsigned long *total_scanned) | |
3697 | { | |
3698 | unsigned long nr_reclaimed = 0; | |
3699 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3700 | unsigned long reclaimed; | |
3701 | int loop = 0; | |
3702 | struct mem_cgroup_tree_per_zone *mctz; | |
3703 | unsigned long long excess; | |
3704 | unsigned long nr_scanned; | |
3705 | ||
3706 | if (order > 0) | |
3707 | return 0; | |
3708 | ||
3709 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3710 | /* | |
3711 | * This loop can run a while, specially if mem_cgroup's continuously | |
3712 | * keep exceeding their soft limit and putting the system under | |
3713 | * pressure | |
3714 | */ | |
3715 | do { | |
3716 | if (next_mz) | |
3717 | mz = next_mz; | |
3718 | else | |
3719 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3720 | if (!mz) | |
3721 | break; | |
3722 | ||
3723 | nr_scanned = 0; | |
3724 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3725 | gfp_mask, &nr_scanned); | |
3726 | nr_reclaimed += reclaimed; | |
3727 | *total_scanned += nr_scanned; | |
0a31bc97 | 3728 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3729 | |
3730 | /* | |
3731 | * If we failed to reclaim anything from this memory cgroup | |
3732 | * it is time to move on to the next cgroup | |
3733 | */ | |
3734 | next_mz = NULL; | |
3735 | if (!reclaimed) { | |
3736 | do { | |
3737 | /* | |
3738 | * Loop until we find yet another one. | |
3739 | * | |
3740 | * By the time we get the soft_limit lock | |
3741 | * again, someone might have aded the | |
3742 | * group back on the RB tree. Iterate to | |
3743 | * make sure we get a different mem. | |
3744 | * mem_cgroup_largest_soft_limit_node returns | |
3745 | * NULL if no other cgroup is present on | |
3746 | * the tree | |
3747 | */ | |
3748 | next_mz = | |
3749 | __mem_cgroup_largest_soft_limit_node(mctz); | |
3750 | if (next_mz == mz) | |
3751 | css_put(&next_mz->memcg->css); | |
3752 | else /* next_mz == NULL or other memcg */ | |
3753 | break; | |
3754 | } while (1); | |
3755 | } | |
cf2c8127 | 3756 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3757 | excess = res_counter_soft_limit_excess(&mz->memcg->res); |
3758 | /* | |
3759 | * One school of thought says that we should not add | |
3760 | * back the node to the tree if reclaim returns 0. | |
3761 | * But our reclaim could return 0, simply because due | |
3762 | * to priority we are exposing a smaller subset of | |
3763 | * memory to reclaim from. Consider this as a longer | |
3764 | * term TODO. | |
3765 | */ | |
3766 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3767 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3768 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3769 | css_put(&mz->memcg->css); |
3770 | loop++; | |
3771 | /* | |
3772 | * Could not reclaim anything and there are no more | |
3773 | * mem cgroups to try or we seem to be looping without | |
3774 | * reclaiming anything. | |
3775 | */ | |
3776 | if (!nr_reclaimed && | |
3777 | (next_mz == NULL || | |
3778 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3779 | break; | |
3780 | } while (!nr_reclaimed); | |
3781 | if (next_mz) | |
3782 | css_put(&next_mz->memcg->css); | |
3783 | return nr_reclaimed; | |
3784 | } | |
3785 | ||
2ef37d3f MH |
3786 | /** |
3787 | * mem_cgroup_force_empty_list - clears LRU of a group | |
3788 | * @memcg: group to clear | |
3789 | * @node: NUMA node | |
3790 | * @zid: zone id | |
3791 | * @lru: lru to to clear | |
3792 | * | |
3c935d18 | 3793 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
3794 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
3795 | * group. | |
cc847582 | 3796 | */ |
2ef37d3f | 3797 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3798 | int node, int zid, enum lru_list lru) |
cc847582 | 3799 | { |
bea8c150 | 3800 | struct lruvec *lruvec; |
2ef37d3f | 3801 | unsigned long flags; |
072c56c1 | 3802 | struct list_head *list; |
925b7673 JW |
3803 | struct page *busy; |
3804 | struct zone *zone; | |
072c56c1 | 3805 | |
08e552c6 | 3806 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
3807 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
3808 | list = &lruvec->lists[lru]; | |
cc847582 | 3809 | |
f817ed48 | 3810 | busy = NULL; |
2ef37d3f | 3811 | do { |
925b7673 | 3812 | struct page_cgroup *pc; |
5564e88b JW |
3813 | struct page *page; |
3814 | ||
08e552c6 | 3815 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3816 | if (list_empty(list)) { |
08e552c6 | 3817 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3818 | break; |
f817ed48 | 3819 | } |
925b7673 JW |
3820 | page = list_entry(list->prev, struct page, lru); |
3821 | if (busy == page) { | |
3822 | list_move(&page->lru, list); | |
648bcc77 | 3823 | busy = NULL; |
08e552c6 | 3824 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3825 | continue; |
3826 | } | |
08e552c6 | 3827 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3828 | |
925b7673 | 3829 | pc = lookup_page_cgroup(page); |
5564e88b | 3830 | |
3c935d18 | 3831 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 3832 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 3833 | busy = page; |
f817ed48 KH |
3834 | } else |
3835 | busy = NULL; | |
2a7a0e0f | 3836 | cond_resched(); |
2ef37d3f | 3837 | } while (!list_empty(list)); |
cc847582 KH |
3838 | } |
3839 | ||
3840 | /* | |
c26251f9 MH |
3841 | * make mem_cgroup's charge to be 0 if there is no task by moving |
3842 | * all the charges and pages to the parent. | |
cc847582 | 3843 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
3844 | * |
3845 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 3846 | */ |
ab5196c2 | 3847 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 3848 | { |
c26251f9 | 3849 | int node, zid; |
bea207c8 | 3850 | u64 usage; |
f817ed48 | 3851 | |
fce66477 | 3852 | do { |
52d4b9ac KH |
3853 | /* This is for making all *used* pages to be on LRU. */ |
3854 | lru_add_drain_all(); | |
c0ff4b85 | 3855 | drain_all_stock_sync(memcg); |
c0ff4b85 | 3856 | mem_cgroup_start_move(memcg); |
31aaea4a | 3857 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 3858 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3859 | enum lru_list lru; |
3860 | for_each_lru(lru) { | |
2ef37d3f | 3861 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3862 | node, zid, lru); |
f817ed48 | 3863 | } |
1ecaab2b | 3864 | } |
f817ed48 | 3865 | } |
c0ff4b85 R |
3866 | mem_cgroup_end_move(memcg); |
3867 | memcg_oom_recover(memcg); | |
52d4b9ac | 3868 | cond_resched(); |
f817ed48 | 3869 | |
2ef37d3f | 3870 | /* |
bea207c8 GC |
3871 | * Kernel memory may not necessarily be trackable to a specific |
3872 | * process. So they are not migrated, and therefore we can't | |
3873 | * expect their value to drop to 0 here. | |
3874 | * Having res filled up with kmem only is enough. | |
3875 | * | |
2ef37d3f MH |
3876 | * This is a safety check because mem_cgroup_force_empty_list |
3877 | * could have raced with mem_cgroup_replace_page_cache callers | |
3878 | * so the lru seemed empty but the page could have been added | |
3879 | * right after the check. RES_USAGE should be safe as we always | |
3880 | * charge before adding to the LRU. | |
3881 | */ | |
bea207c8 GC |
3882 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
3883 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
3884 | } while (usage > 0); | |
c26251f9 MH |
3885 | } |
3886 | ||
ea280e7b TH |
3887 | /* |
3888 | * Test whether @memcg has children, dead or alive. Note that this | |
3889 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3890 | * returns %true if there are child csses according to the cgroup | |
3891 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3892 | */ | |
b5f99b53 GC |
3893 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3894 | { | |
ea280e7b TH |
3895 | bool ret; |
3896 | ||
696ac172 | 3897 | /* |
ea280e7b TH |
3898 | * The lock does not prevent addition or deletion of children, but |
3899 | * it prevents a new child from being initialized based on this | |
3900 | * parent in css_online(), so it's enough to decide whether | |
3901 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3902 | */ |
ea280e7b TH |
3903 | lockdep_assert_held(&memcg_create_mutex); |
3904 | ||
3905 | rcu_read_lock(); | |
3906 | ret = css_next_child(NULL, &memcg->css); | |
3907 | rcu_read_unlock(); | |
3908 | return ret; | |
b5f99b53 GC |
3909 | } |
3910 | ||
c26251f9 MH |
3911 | /* |
3912 | * Reclaims as many pages from the given memcg as possible and moves | |
3913 | * the rest to the parent. | |
3914 | * | |
3915 | * Caller is responsible for holding css reference for memcg. | |
3916 | */ | |
3917 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3918 | { | |
3919 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3920 | |
c1e862c1 KH |
3921 | /* we call try-to-free pages for make this cgroup empty */ |
3922 | lru_add_drain_all(); | |
f817ed48 | 3923 | /* try to free all pages in this cgroup */ |
569530fb | 3924 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 3925 | int progress; |
c1e862c1 | 3926 | |
c26251f9 MH |
3927 | if (signal_pending(current)) |
3928 | return -EINTR; | |
3929 | ||
b70a2a21 JW |
3930 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3931 | GFP_KERNEL, true); | |
c1e862c1 | 3932 | if (!progress) { |
f817ed48 | 3933 | nr_retries--; |
c1e862c1 | 3934 | /* maybe some writeback is necessary */ |
8aa7e847 | 3935 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3936 | } |
f817ed48 KH |
3937 | |
3938 | } | |
ab5196c2 MH |
3939 | |
3940 | return 0; | |
cc847582 KH |
3941 | } |
3942 | ||
6770c64e TH |
3943 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3944 | char *buf, size_t nbytes, | |
3945 | loff_t off) | |
c1e862c1 | 3946 | { |
6770c64e | 3947 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3948 | |
d8423011 MH |
3949 | if (mem_cgroup_is_root(memcg)) |
3950 | return -EINVAL; | |
6770c64e | 3951 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3952 | } |
3953 | ||
182446d0 TH |
3954 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3955 | struct cftype *cft) | |
18f59ea7 | 3956 | { |
182446d0 | 3957 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3958 | } |
3959 | ||
182446d0 TH |
3960 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3961 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3962 | { |
3963 | int retval = 0; | |
182446d0 | 3964 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3965 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3966 | |
0999821b | 3967 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3968 | |
3969 | if (memcg->use_hierarchy == val) | |
3970 | goto out; | |
3971 | ||
18f59ea7 | 3972 | /* |
af901ca1 | 3973 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3974 | * in the child subtrees. If it is unset, then the change can |
3975 | * occur, provided the current cgroup has no children. | |
3976 | * | |
3977 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3978 | * set if there are no children. | |
3979 | */ | |
c0ff4b85 | 3980 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3981 | (val == 1 || val == 0)) { |
ea280e7b | 3982 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3983 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3984 | else |
3985 | retval = -EBUSY; | |
3986 | } else | |
3987 | retval = -EINVAL; | |
567fb435 GC |
3988 | |
3989 | out: | |
0999821b | 3990 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3991 | |
3992 | return retval; | |
3993 | } | |
3994 | ||
ce00a967 JW |
3995 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
3996 | enum mem_cgroup_stat_index idx) | |
3997 | { | |
3998 | struct mem_cgroup *iter; | |
3999 | long val = 0; | |
4000 | ||
4001 | /* Per-cpu values can be negative, use a signed accumulator */ | |
4002 | for_each_mem_cgroup_tree(iter, memcg) | |
4003 | val += mem_cgroup_read_stat(iter, idx); | |
4004 | ||
4005 | if (val < 0) /* race ? */ | |
4006 | val = 0; | |
4007 | return val; | |
4008 | } | |
4009 | ||
4010 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
4011 | { | |
4012 | u64 val; | |
4013 | ||
4014 | if (!mem_cgroup_is_root(memcg)) { | |
4015 | if (!swap) | |
4016 | return res_counter_read_u64(&memcg->res, RES_USAGE); | |
4017 | else | |
4018 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
4019 | } | |
4020 | ||
4021 | /* | |
4022 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
4023 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
4024 | */ | |
4025 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
4026 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
4027 | ||
4028 | if (swap) | |
4029 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
4030 | ||
4031 | return val << PAGE_SHIFT; | |
4032 | } | |
4033 | ||
4034 | ||
791badbd | 4035 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 4036 | struct cftype *cft) |
8cdea7c0 | 4037 | { |
182446d0 | 4038 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
05b84301 JW |
4039 | enum res_type type = MEMFILE_TYPE(cft->private); |
4040 | int name = MEMFILE_ATTR(cft->private); | |
af36f906 | 4041 | |
8c7c6e34 KH |
4042 | switch (type) { |
4043 | case _MEM: | |
ce00a967 JW |
4044 | if (name == RES_USAGE) |
4045 | return mem_cgroup_usage(memcg, false); | |
05b84301 | 4046 | return res_counter_read_u64(&memcg->res, name); |
8c7c6e34 | 4047 | case _MEMSWAP: |
ce00a967 JW |
4048 | if (name == RES_USAGE) |
4049 | return mem_cgroup_usage(memcg, true); | |
05b84301 | 4050 | return res_counter_read_u64(&memcg->memsw, name); |
510fc4e1 | 4051 | case _KMEM: |
05b84301 | 4052 | return res_counter_read_u64(&memcg->kmem, name); |
510fc4e1 | 4053 | break; |
8c7c6e34 KH |
4054 | default: |
4055 | BUG(); | |
8c7c6e34 | 4056 | } |
8cdea7c0 | 4057 | } |
510fc4e1 | 4058 | |
510fc4e1 | 4059 | #ifdef CONFIG_MEMCG_KMEM |
d6441637 VD |
4060 | /* should be called with activate_kmem_mutex held */ |
4061 | static int __memcg_activate_kmem(struct mem_cgroup *memcg, | |
4062 | unsigned long long limit) | |
4063 | { | |
4064 | int err = 0; | |
4065 | int memcg_id; | |
4066 | ||
4067 | if (memcg_kmem_is_active(memcg)) | |
4068 | return 0; | |
4069 | ||
4070 | /* | |
4071 | * We are going to allocate memory for data shared by all memory | |
4072 | * cgroups so let's stop accounting here. | |
4073 | */ | |
4074 | memcg_stop_kmem_account(); | |
4075 | ||
510fc4e1 GC |
4076 | /* |
4077 | * For simplicity, we won't allow this to be disabled. It also can't | |
4078 | * be changed if the cgroup has children already, or if tasks had | |
4079 | * already joined. | |
4080 | * | |
4081 | * If tasks join before we set the limit, a person looking at | |
4082 | * kmem.usage_in_bytes will have no way to determine when it took | |
4083 | * place, which makes the value quite meaningless. | |
4084 | * | |
4085 | * After it first became limited, changes in the value of the limit are | |
4086 | * of course permitted. | |
510fc4e1 | 4087 | */ |
0999821b | 4088 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
4089 | if (cgroup_has_tasks(memcg->css.cgroup) || |
4090 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
4091 | err = -EBUSY; |
4092 | mutex_unlock(&memcg_create_mutex); | |
4093 | if (err) | |
4094 | goto out; | |
510fc4e1 | 4095 | |
f3bb3043 | 4096 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
4097 | if (memcg_id < 0) { |
4098 | err = memcg_id; | |
4099 | goto out; | |
4100 | } | |
4101 | ||
d6441637 VD |
4102 | memcg->kmemcg_id = memcg_id; |
4103 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
d6441637 VD |
4104 | |
4105 | /* | |
4106 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
4107 | * active bit set yet, so this should succeed. | |
4108 | */ | |
4109 | err = res_counter_set_limit(&memcg->kmem, limit); | |
4110 | VM_BUG_ON(err); | |
4111 | ||
4112 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
4113 | /* | |
4114 | * Setting the active bit after enabling static branching will | |
4115 | * guarantee no one starts accounting before all call sites are | |
4116 | * patched. | |
4117 | */ | |
4118 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 4119 | out: |
d6441637 VD |
4120 | memcg_resume_kmem_account(); |
4121 | return err; | |
d6441637 VD |
4122 | } |
4123 | ||
4124 | static int memcg_activate_kmem(struct mem_cgroup *memcg, | |
4125 | unsigned long long limit) | |
4126 | { | |
4127 | int ret; | |
4128 | ||
4129 | mutex_lock(&activate_kmem_mutex); | |
4130 | ret = __memcg_activate_kmem(memcg, limit); | |
4131 | mutex_unlock(&activate_kmem_mutex); | |
4132 | return ret; | |
4133 | } | |
4134 | ||
4135 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
4136 | unsigned long long val) | |
4137 | { | |
4138 | int ret; | |
4139 | ||
4140 | if (!memcg_kmem_is_active(memcg)) | |
4141 | ret = memcg_activate_kmem(memcg, val); | |
4142 | else | |
4143 | ret = res_counter_set_limit(&memcg->kmem, val); | |
510fc4e1 GC |
4144 | return ret; |
4145 | } | |
4146 | ||
55007d84 | 4147 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 4148 | { |
55007d84 | 4149 | int ret = 0; |
510fc4e1 | 4150 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 4151 | |
d6441637 VD |
4152 | if (!parent) |
4153 | return 0; | |
55007d84 | 4154 | |
d6441637 | 4155 | mutex_lock(&activate_kmem_mutex); |
55007d84 | 4156 | /* |
d6441637 VD |
4157 | * If the parent cgroup is not kmem-active now, it cannot be activated |
4158 | * after this point, because it has at least one child already. | |
55007d84 | 4159 | */ |
d6441637 VD |
4160 | if (memcg_kmem_is_active(parent)) |
4161 | ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX); | |
4162 | mutex_unlock(&activate_kmem_mutex); | |
55007d84 | 4163 | return ret; |
510fc4e1 | 4164 | } |
d6441637 VD |
4165 | #else |
4166 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
4167 | unsigned long long val) | |
4168 | { | |
4169 | return -EINVAL; | |
4170 | } | |
6d043990 | 4171 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 4172 | |
628f4235 KH |
4173 | /* |
4174 | * The user of this function is... | |
4175 | * RES_LIMIT. | |
4176 | */ | |
451af504 TH |
4177 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
4178 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 4179 | { |
451af504 | 4180 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
86ae53e1 GC |
4181 | enum res_type type; |
4182 | int name; | |
628f4235 KH |
4183 | unsigned long long val; |
4184 | int ret; | |
4185 | ||
451af504 TH |
4186 | buf = strstrip(buf); |
4187 | type = MEMFILE_TYPE(of_cft(of)->private); | |
4188 | name = MEMFILE_ATTR(of_cft(of)->private); | |
af36f906 | 4189 | |
8c7c6e34 | 4190 | switch (name) { |
628f4235 | 4191 | case RES_LIMIT: |
4b3bde4c BS |
4192 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4193 | ret = -EINVAL; | |
4194 | break; | |
4195 | } | |
628f4235 | 4196 | /* This function does all necessary parse...reuse it */ |
451af504 | 4197 | ret = res_counter_memparse_write_strategy(buf, &val); |
8c7c6e34 KH |
4198 | if (ret) |
4199 | break; | |
4200 | if (type == _MEM) | |
628f4235 | 4201 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 4202 | else if (type == _MEMSWAP) |
8c7c6e34 | 4203 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 4204 | else if (type == _KMEM) |
d6441637 | 4205 | ret = memcg_update_kmem_limit(memcg, val); |
510fc4e1 GC |
4206 | else |
4207 | return -EINVAL; | |
628f4235 | 4208 | break; |
296c81d8 | 4209 | case RES_SOFT_LIMIT: |
451af504 | 4210 | ret = res_counter_memparse_write_strategy(buf, &val); |
296c81d8 BS |
4211 | if (ret) |
4212 | break; | |
4213 | /* | |
4214 | * For memsw, soft limits are hard to implement in terms | |
4215 | * of semantics, for now, we support soft limits for | |
4216 | * control without swap | |
4217 | */ | |
4218 | if (type == _MEM) | |
4219 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
4220 | else | |
4221 | ret = -EINVAL; | |
4222 | break; | |
628f4235 KH |
4223 | default: |
4224 | ret = -EINVAL; /* should be BUG() ? */ | |
4225 | break; | |
4226 | } | |
451af504 | 4227 | return ret ?: nbytes; |
8cdea7c0 BS |
4228 | } |
4229 | ||
fee7b548 KH |
4230 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
4231 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
4232 | { | |
fee7b548 KH |
4233 | unsigned long long min_limit, min_memsw_limit, tmp; |
4234 | ||
4235 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4236 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
4237 | if (!memcg->use_hierarchy) |
4238 | goto out; | |
4239 | ||
5c9d535b TH |
4240 | while (memcg->css.parent) { |
4241 | memcg = mem_cgroup_from_css(memcg->css.parent); | |
fee7b548 KH |
4242 | if (!memcg->use_hierarchy) |
4243 | break; | |
4244 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4245 | min_limit = min(min_limit, tmp); | |
4246 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4247 | min_memsw_limit = min(min_memsw_limit, tmp); | |
4248 | } | |
4249 | out: | |
4250 | *mem_limit = min_limit; | |
4251 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
4252 | } |
4253 | ||
6770c64e TH |
4254 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
4255 | size_t nbytes, loff_t off) | |
c84872e1 | 4256 | { |
6770c64e | 4257 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
86ae53e1 GC |
4258 | int name; |
4259 | enum res_type type; | |
c84872e1 | 4260 | |
6770c64e TH |
4261 | type = MEMFILE_TYPE(of_cft(of)->private); |
4262 | name = MEMFILE_ATTR(of_cft(of)->private); | |
af36f906 | 4263 | |
8c7c6e34 | 4264 | switch (name) { |
29f2a4da | 4265 | case RES_MAX_USAGE: |
8c7c6e34 | 4266 | if (type == _MEM) |
c0ff4b85 | 4267 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 4268 | else if (type == _MEMSWAP) |
c0ff4b85 | 4269 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
4270 | else if (type == _KMEM) |
4271 | res_counter_reset_max(&memcg->kmem); | |
4272 | else | |
4273 | return -EINVAL; | |
29f2a4da PE |
4274 | break; |
4275 | case RES_FAILCNT: | |
8c7c6e34 | 4276 | if (type == _MEM) |
c0ff4b85 | 4277 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 4278 | else if (type == _MEMSWAP) |
c0ff4b85 | 4279 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
4280 | else if (type == _KMEM) |
4281 | res_counter_reset_failcnt(&memcg->kmem); | |
4282 | else | |
4283 | return -EINVAL; | |
29f2a4da PE |
4284 | break; |
4285 | } | |
f64c3f54 | 4286 | |
6770c64e | 4287 | return nbytes; |
c84872e1 PE |
4288 | } |
4289 | ||
182446d0 | 4290 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4291 | struct cftype *cft) |
4292 | { | |
182446d0 | 4293 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
4294 | } |
4295 | ||
02491447 | 4296 | #ifdef CONFIG_MMU |
182446d0 | 4297 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4298 | struct cftype *cft, u64 val) |
4299 | { | |
182446d0 | 4300 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
4301 | |
4302 | if (val >= (1 << NR_MOVE_TYPE)) | |
4303 | return -EINVAL; | |
ee5e8472 | 4304 | |
7dc74be0 | 4305 | /* |
ee5e8472 GC |
4306 | * No kind of locking is needed in here, because ->can_attach() will |
4307 | * check this value once in the beginning of the process, and then carry | |
4308 | * on with stale data. This means that changes to this value will only | |
4309 | * affect task migrations starting after the change. | |
7dc74be0 | 4310 | */ |
c0ff4b85 | 4311 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4312 | return 0; |
4313 | } | |
02491447 | 4314 | #else |
182446d0 | 4315 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
4316 | struct cftype *cft, u64 val) |
4317 | { | |
4318 | return -ENOSYS; | |
4319 | } | |
4320 | #endif | |
7dc74be0 | 4321 | |
406eb0c9 | 4322 | #ifdef CONFIG_NUMA |
2da8ca82 | 4323 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 4324 | { |
25485de6 GT |
4325 | struct numa_stat { |
4326 | const char *name; | |
4327 | unsigned int lru_mask; | |
4328 | }; | |
4329 | ||
4330 | static const struct numa_stat stats[] = { | |
4331 | { "total", LRU_ALL }, | |
4332 | { "file", LRU_ALL_FILE }, | |
4333 | { "anon", LRU_ALL_ANON }, | |
4334 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4335 | }; | |
4336 | const struct numa_stat *stat; | |
406eb0c9 | 4337 | int nid; |
25485de6 | 4338 | unsigned long nr; |
2da8ca82 | 4339 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 4340 | |
25485de6 GT |
4341 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
4342 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
4343 | seq_printf(m, "%s=%lu", stat->name, nr); | |
4344 | for_each_node_state(nid, N_MEMORY) { | |
4345 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4346 | stat->lru_mask); | |
4347 | seq_printf(m, " N%d=%lu", nid, nr); | |
4348 | } | |
4349 | seq_putc(m, '\n'); | |
406eb0c9 | 4350 | } |
406eb0c9 | 4351 | |
071aee13 YH |
4352 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
4353 | struct mem_cgroup *iter; | |
4354 | ||
4355 | nr = 0; | |
4356 | for_each_mem_cgroup_tree(iter, memcg) | |
4357 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
4358 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
4359 | for_each_node_state(nid, N_MEMORY) { | |
4360 | nr = 0; | |
4361 | for_each_mem_cgroup_tree(iter, memcg) | |
4362 | nr += mem_cgroup_node_nr_lru_pages( | |
4363 | iter, nid, stat->lru_mask); | |
4364 | seq_printf(m, " N%d=%lu", nid, nr); | |
4365 | } | |
4366 | seq_putc(m, '\n'); | |
406eb0c9 | 4367 | } |
406eb0c9 | 4368 | |
406eb0c9 YH |
4369 | return 0; |
4370 | } | |
4371 | #endif /* CONFIG_NUMA */ | |
4372 | ||
af7c4b0e JW |
4373 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
4374 | { | |
4375 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4376 | } | |
4377 | ||
2da8ca82 | 4378 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4379 | { |
2da8ca82 | 4380 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
af7c4b0e JW |
4381 | struct mem_cgroup *mi; |
4382 | unsigned int i; | |
406eb0c9 | 4383 | |
af7c4b0e | 4384 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4385 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4386 | continue; |
af7c4b0e JW |
4387 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4388 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4389 | } |
7b854121 | 4390 | |
af7c4b0e JW |
4391 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4392 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4393 | mem_cgroup_read_events(memcg, i)); | |
4394 | ||
4395 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4396 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4397 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4398 | ||
14067bb3 | 4399 | /* Hierarchical information */ |
fee7b548 KH |
4400 | { |
4401 | unsigned long long limit, memsw_limit; | |
d79154bb | 4402 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 4403 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 4404 | if (do_swap_account) |
78ccf5b5 JW |
4405 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4406 | memsw_limit); | |
fee7b548 | 4407 | } |
7f016ee8 | 4408 | |
af7c4b0e JW |
4409 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4410 | long long val = 0; | |
4411 | ||
bff6bb83 | 4412 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4413 | continue; |
af7c4b0e JW |
4414 | for_each_mem_cgroup_tree(mi, memcg) |
4415 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4416 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4417 | } | |
4418 | ||
4419 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4420 | unsigned long long val = 0; | |
4421 | ||
4422 | for_each_mem_cgroup_tree(mi, memcg) | |
4423 | val += mem_cgroup_read_events(mi, i); | |
4424 | seq_printf(m, "total_%s %llu\n", | |
4425 | mem_cgroup_events_names[i], val); | |
4426 | } | |
4427 | ||
4428 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4429 | unsigned long long val = 0; | |
4430 | ||
4431 | for_each_mem_cgroup_tree(mi, memcg) | |
4432 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4433 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4434 | } |
14067bb3 | 4435 | |
7f016ee8 | 4436 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4437 | { |
4438 | int nid, zid; | |
4439 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4440 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4441 | unsigned long recent_rotated[2] = {0, 0}; |
4442 | unsigned long recent_scanned[2] = {0, 0}; | |
4443 | ||
4444 | for_each_online_node(nid) | |
4445 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 4446 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 4447 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4448 | |
89abfab1 HD |
4449 | recent_rotated[0] += rstat->recent_rotated[0]; |
4450 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4451 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4452 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4453 | } |
78ccf5b5 JW |
4454 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4455 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4456 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4457 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4458 | } |
4459 | #endif | |
4460 | ||
d2ceb9b7 KH |
4461 | return 0; |
4462 | } | |
4463 | ||
182446d0 TH |
4464 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4465 | struct cftype *cft) | |
a7885eb8 | 4466 | { |
182446d0 | 4467 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4468 | |
1f4c025b | 4469 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4470 | } |
4471 | ||
182446d0 TH |
4472 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4473 | struct cftype *cft, u64 val) | |
a7885eb8 | 4474 | { |
182446d0 | 4475 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4476 | |
3dae7fec | 4477 | if (val > 100) |
a7885eb8 KM |
4478 | return -EINVAL; |
4479 | ||
14208b0e | 4480 | if (css->parent) |
3dae7fec JW |
4481 | memcg->swappiness = val; |
4482 | else | |
4483 | vm_swappiness = val; | |
068b38c1 | 4484 | |
a7885eb8 KM |
4485 | return 0; |
4486 | } | |
4487 | ||
2e72b634 KS |
4488 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4489 | { | |
4490 | struct mem_cgroup_threshold_ary *t; | |
4491 | u64 usage; | |
4492 | int i; | |
4493 | ||
4494 | rcu_read_lock(); | |
4495 | if (!swap) | |
2c488db2 | 4496 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4497 | else |
2c488db2 | 4498 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4499 | |
4500 | if (!t) | |
4501 | goto unlock; | |
4502 | ||
ce00a967 | 4503 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4504 | |
4505 | /* | |
748dad36 | 4506 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4507 | * If it's not true, a threshold was crossed after last |
4508 | * call of __mem_cgroup_threshold(). | |
4509 | */ | |
5407a562 | 4510 | i = t->current_threshold; |
2e72b634 KS |
4511 | |
4512 | /* | |
4513 | * Iterate backward over array of thresholds starting from | |
4514 | * current_threshold and check if a threshold is crossed. | |
4515 | * If none of thresholds below usage is crossed, we read | |
4516 | * only one element of the array here. | |
4517 | */ | |
4518 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4519 | eventfd_signal(t->entries[i].eventfd, 1); | |
4520 | ||
4521 | /* i = current_threshold + 1 */ | |
4522 | i++; | |
4523 | ||
4524 | /* | |
4525 | * Iterate forward over array of thresholds starting from | |
4526 | * current_threshold+1 and check if a threshold is crossed. | |
4527 | * If none of thresholds above usage is crossed, we read | |
4528 | * only one element of the array here. | |
4529 | */ | |
4530 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4531 | eventfd_signal(t->entries[i].eventfd, 1); | |
4532 | ||
4533 | /* Update current_threshold */ | |
5407a562 | 4534 | t->current_threshold = i - 1; |
2e72b634 KS |
4535 | unlock: |
4536 | rcu_read_unlock(); | |
4537 | } | |
4538 | ||
4539 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4540 | { | |
ad4ca5f4 KS |
4541 | while (memcg) { |
4542 | __mem_cgroup_threshold(memcg, false); | |
4543 | if (do_swap_account) | |
4544 | __mem_cgroup_threshold(memcg, true); | |
4545 | ||
4546 | memcg = parent_mem_cgroup(memcg); | |
4547 | } | |
2e72b634 KS |
4548 | } |
4549 | ||
4550 | static int compare_thresholds(const void *a, const void *b) | |
4551 | { | |
4552 | const struct mem_cgroup_threshold *_a = a; | |
4553 | const struct mem_cgroup_threshold *_b = b; | |
4554 | ||
2bff24a3 GT |
4555 | if (_a->threshold > _b->threshold) |
4556 | return 1; | |
4557 | ||
4558 | if (_a->threshold < _b->threshold) | |
4559 | return -1; | |
4560 | ||
4561 | return 0; | |
2e72b634 KS |
4562 | } |
4563 | ||
c0ff4b85 | 4564 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4565 | { |
4566 | struct mem_cgroup_eventfd_list *ev; | |
4567 | ||
2bcf2e92 MH |
4568 | spin_lock(&memcg_oom_lock); |
4569 | ||
c0ff4b85 | 4570 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4571 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4572 | |
4573 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4574 | return 0; |
4575 | } | |
4576 | ||
c0ff4b85 | 4577 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4578 | { |
7d74b06f KH |
4579 | struct mem_cgroup *iter; |
4580 | ||
c0ff4b85 | 4581 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4582 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4583 | } |
4584 | ||
59b6f873 | 4585 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4586 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4587 | { |
2c488db2 KS |
4588 | struct mem_cgroup_thresholds *thresholds; |
4589 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 4590 | u64 threshold, usage; |
2c488db2 | 4591 | int i, size, ret; |
2e72b634 KS |
4592 | |
4593 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4594 | if (ret) | |
4595 | return ret; | |
4596 | ||
4597 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4598 | |
05b84301 | 4599 | if (type == _MEM) { |
2c488db2 | 4600 | thresholds = &memcg->thresholds; |
ce00a967 | 4601 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4602 | } else if (type == _MEMSWAP) { |
2c488db2 | 4603 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4604 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4605 | } else |
2e72b634 KS |
4606 | BUG(); |
4607 | ||
2e72b634 | 4608 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4609 | if (thresholds->primary) |
2e72b634 KS |
4610 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4611 | ||
2c488db2 | 4612 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4613 | |
4614 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4615 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4616 | GFP_KERNEL); |
2c488db2 | 4617 | if (!new) { |
2e72b634 KS |
4618 | ret = -ENOMEM; |
4619 | goto unlock; | |
4620 | } | |
2c488db2 | 4621 | new->size = size; |
2e72b634 KS |
4622 | |
4623 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4624 | if (thresholds->primary) { |
4625 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4626 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4627 | } |
4628 | ||
2e72b634 | 4629 | /* Add new threshold */ |
2c488db2 KS |
4630 | new->entries[size - 1].eventfd = eventfd; |
4631 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4632 | |
4633 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4634 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4635 | compare_thresholds, NULL); |
4636 | ||
4637 | /* Find current threshold */ | |
2c488db2 | 4638 | new->current_threshold = -1; |
2e72b634 | 4639 | for (i = 0; i < size; i++) { |
748dad36 | 4640 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4641 | /* |
2c488db2 KS |
4642 | * new->current_threshold will not be used until |
4643 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4644 | * it here. |
4645 | */ | |
2c488db2 | 4646 | ++new->current_threshold; |
748dad36 SZ |
4647 | } else |
4648 | break; | |
2e72b634 KS |
4649 | } |
4650 | ||
2c488db2 KS |
4651 | /* Free old spare buffer and save old primary buffer as spare */ |
4652 | kfree(thresholds->spare); | |
4653 | thresholds->spare = thresholds->primary; | |
4654 | ||
4655 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4656 | |
907860ed | 4657 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4658 | synchronize_rcu(); |
4659 | ||
2e72b634 KS |
4660 | unlock: |
4661 | mutex_unlock(&memcg->thresholds_lock); | |
4662 | ||
4663 | return ret; | |
4664 | } | |
4665 | ||
59b6f873 | 4666 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4667 | struct eventfd_ctx *eventfd, const char *args) |
4668 | { | |
59b6f873 | 4669 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4670 | } |
4671 | ||
59b6f873 | 4672 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4673 | struct eventfd_ctx *eventfd, const char *args) |
4674 | { | |
59b6f873 | 4675 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4676 | } |
4677 | ||
59b6f873 | 4678 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4679 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4680 | { |
2c488db2 KS |
4681 | struct mem_cgroup_thresholds *thresholds; |
4682 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 4683 | u64 usage; |
2c488db2 | 4684 | int i, j, size; |
2e72b634 KS |
4685 | |
4686 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4687 | |
4688 | if (type == _MEM) { | |
2c488db2 | 4689 | thresholds = &memcg->thresholds; |
ce00a967 | 4690 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4691 | } else if (type == _MEMSWAP) { |
2c488db2 | 4692 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4693 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4694 | } else |
2e72b634 KS |
4695 | BUG(); |
4696 | ||
371528ca AV |
4697 | if (!thresholds->primary) |
4698 | goto unlock; | |
4699 | ||
2e72b634 KS |
4700 | /* Check if a threshold crossed before removing */ |
4701 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4702 | ||
4703 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4704 | size = 0; |
4705 | for (i = 0; i < thresholds->primary->size; i++) { | |
4706 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4707 | size++; |
4708 | } | |
4709 | ||
2c488db2 | 4710 | new = thresholds->spare; |
907860ed | 4711 | |
2e72b634 KS |
4712 | /* Set thresholds array to NULL if we don't have thresholds */ |
4713 | if (!size) { | |
2c488db2 KS |
4714 | kfree(new); |
4715 | new = NULL; | |
907860ed | 4716 | goto swap_buffers; |
2e72b634 KS |
4717 | } |
4718 | ||
2c488db2 | 4719 | new->size = size; |
2e72b634 KS |
4720 | |
4721 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4722 | new->current_threshold = -1; |
4723 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4724 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4725 | continue; |
4726 | ||
2c488db2 | 4727 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4728 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4729 | /* |
2c488db2 | 4730 | * new->current_threshold will not be used |
2e72b634 KS |
4731 | * until rcu_assign_pointer(), so it's safe to increment |
4732 | * it here. | |
4733 | */ | |
2c488db2 | 4734 | ++new->current_threshold; |
2e72b634 KS |
4735 | } |
4736 | j++; | |
4737 | } | |
4738 | ||
907860ed | 4739 | swap_buffers: |
2c488db2 KS |
4740 | /* Swap primary and spare array */ |
4741 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4742 | /* If all events are unregistered, free the spare array */ |
4743 | if (!new) { | |
4744 | kfree(thresholds->spare); | |
4745 | thresholds->spare = NULL; | |
4746 | } | |
4747 | ||
2c488db2 | 4748 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4749 | |
907860ed | 4750 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4751 | synchronize_rcu(); |
371528ca | 4752 | unlock: |
2e72b634 | 4753 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4754 | } |
c1e862c1 | 4755 | |
59b6f873 | 4756 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4757 | struct eventfd_ctx *eventfd) |
4758 | { | |
59b6f873 | 4759 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4760 | } |
4761 | ||
59b6f873 | 4762 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4763 | struct eventfd_ctx *eventfd) |
4764 | { | |
59b6f873 | 4765 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4766 | } |
4767 | ||
59b6f873 | 4768 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4769 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4770 | { |
9490ff27 | 4771 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4772 | |
9490ff27 KH |
4773 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4774 | if (!event) | |
4775 | return -ENOMEM; | |
4776 | ||
1af8efe9 | 4777 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4778 | |
4779 | event->eventfd = eventfd; | |
4780 | list_add(&event->list, &memcg->oom_notify); | |
4781 | ||
4782 | /* already in OOM ? */ | |
79dfdacc | 4783 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4784 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4785 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4786 | |
4787 | return 0; | |
4788 | } | |
4789 | ||
59b6f873 | 4790 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4791 | struct eventfd_ctx *eventfd) |
9490ff27 | 4792 | { |
9490ff27 | 4793 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4794 | |
1af8efe9 | 4795 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4796 | |
c0ff4b85 | 4797 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4798 | if (ev->eventfd == eventfd) { |
4799 | list_del(&ev->list); | |
4800 | kfree(ev); | |
4801 | } | |
4802 | } | |
4803 | ||
1af8efe9 | 4804 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4805 | } |
4806 | ||
2da8ca82 | 4807 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4808 | { |
2da8ca82 | 4809 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4810 | |
791badbd TH |
4811 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4812 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4813 | return 0; |
4814 | } | |
4815 | ||
182446d0 | 4816 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4817 | struct cftype *cft, u64 val) |
4818 | { | |
182446d0 | 4819 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4820 | |
4821 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4822 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4823 | return -EINVAL; |
4824 | ||
c0ff4b85 | 4825 | memcg->oom_kill_disable = val; |
4d845ebf | 4826 | if (!val) |
c0ff4b85 | 4827 | memcg_oom_recover(memcg); |
3dae7fec | 4828 | |
3c11ecf4 KH |
4829 | return 0; |
4830 | } | |
4831 | ||
c255a458 | 4832 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4833 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4834 | { |
55007d84 GC |
4835 | int ret; |
4836 | ||
2633d7a0 | 4837 | memcg->kmemcg_id = -1; |
55007d84 GC |
4838 | ret = memcg_propagate_kmem(memcg); |
4839 | if (ret) | |
4840 | return ret; | |
2633d7a0 | 4841 | |
1d62e436 | 4842 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4843 | } |
e5671dfa | 4844 | |
10d5ebf4 | 4845 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4846 | { |
1d62e436 | 4847 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
4848 | } |
4849 | ||
4850 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
4851 | { | |
4852 | if (!memcg_kmem_is_active(memcg)) | |
4853 | return; | |
4854 | ||
4855 | /* | |
4856 | * kmem charges can outlive the cgroup. In the case of slab | |
4857 | * pages, for instance, a page contain objects from various | |
4858 | * processes. As we prevent from taking a reference for every | |
4859 | * such allocation we have to be careful when doing uncharge | |
4860 | * (see memcg_uncharge_kmem) and here during offlining. | |
4861 | * | |
4862 | * The idea is that that only the _last_ uncharge which sees | |
4863 | * the dead memcg will drop the last reference. An additional | |
4864 | * reference is taken here before the group is marked dead | |
4865 | * which is then paired with css_put during uncharge resp. here. | |
4866 | * | |
4867 | * Although this might sound strange as this path is called from | |
ec903c0c TH |
4868 | * css_offline() when the referencemight have dropped down to 0 and |
4869 | * shouldn't be incremented anymore (css_tryget_online() would | |
4870 | * fail) we do not have other options because of the kmem | |
4871 | * allocations lifetime. | |
10d5ebf4 LZ |
4872 | */ |
4873 | css_get(&memcg->css); | |
7de37682 GC |
4874 | |
4875 | memcg_kmem_mark_dead(memcg); | |
4876 | ||
4877 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
4878 | return; | |
4879 | ||
7de37682 | 4880 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 4881 | css_put(&memcg->css); |
d1a4c0b3 | 4882 | } |
e5671dfa | 4883 | #else |
cbe128e3 | 4884 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4885 | { |
4886 | return 0; | |
4887 | } | |
d1a4c0b3 | 4888 | |
10d5ebf4 LZ |
4889 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4890 | { | |
4891 | } | |
4892 | ||
4893 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
4894 | { |
4895 | } | |
e5671dfa GC |
4896 | #endif |
4897 | ||
3bc942f3 TH |
4898 | /* |
4899 | * DO NOT USE IN NEW FILES. | |
4900 | * | |
4901 | * "cgroup.event_control" implementation. | |
4902 | * | |
4903 | * This is way over-engineered. It tries to support fully configurable | |
4904 | * events for each user. Such level of flexibility is completely | |
4905 | * unnecessary especially in the light of the planned unified hierarchy. | |
4906 | * | |
4907 | * Please deprecate this and replace with something simpler if at all | |
4908 | * possible. | |
4909 | */ | |
4910 | ||
79bd9814 TH |
4911 | /* |
4912 | * Unregister event and free resources. | |
4913 | * | |
4914 | * Gets called from workqueue. | |
4915 | */ | |
3bc942f3 | 4916 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4917 | { |
3bc942f3 TH |
4918 | struct mem_cgroup_event *event = |
4919 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4920 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4921 | |
4922 | remove_wait_queue(event->wqh, &event->wait); | |
4923 | ||
59b6f873 | 4924 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4925 | |
4926 | /* Notify userspace the event is going away. */ | |
4927 | eventfd_signal(event->eventfd, 1); | |
4928 | ||
4929 | eventfd_ctx_put(event->eventfd); | |
4930 | kfree(event); | |
59b6f873 | 4931 | css_put(&memcg->css); |
79bd9814 TH |
4932 | } |
4933 | ||
4934 | /* | |
4935 | * Gets called on POLLHUP on eventfd when user closes it. | |
4936 | * | |
4937 | * Called with wqh->lock held and interrupts disabled. | |
4938 | */ | |
3bc942f3 TH |
4939 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4940 | int sync, void *key) | |
79bd9814 | 4941 | { |
3bc942f3 TH |
4942 | struct mem_cgroup_event *event = |
4943 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4944 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4945 | unsigned long flags = (unsigned long)key; |
4946 | ||
4947 | if (flags & POLLHUP) { | |
4948 | /* | |
4949 | * If the event has been detached at cgroup removal, we | |
4950 | * can simply return knowing the other side will cleanup | |
4951 | * for us. | |
4952 | * | |
4953 | * We can't race against event freeing since the other | |
4954 | * side will require wqh->lock via remove_wait_queue(), | |
4955 | * which we hold. | |
4956 | */ | |
fba94807 | 4957 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4958 | if (!list_empty(&event->list)) { |
4959 | list_del_init(&event->list); | |
4960 | /* | |
4961 | * We are in atomic context, but cgroup_event_remove() | |
4962 | * may sleep, so we have to call it in workqueue. | |
4963 | */ | |
4964 | schedule_work(&event->remove); | |
4965 | } | |
fba94807 | 4966 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4967 | } |
4968 | ||
4969 | return 0; | |
4970 | } | |
4971 | ||
3bc942f3 | 4972 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4973 | wait_queue_head_t *wqh, poll_table *pt) |
4974 | { | |
3bc942f3 TH |
4975 | struct mem_cgroup_event *event = |
4976 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4977 | |
4978 | event->wqh = wqh; | |
4979 | add_wait_queue(wqh, &event->wait); | |
4980 | } | |
4981 | ||
4982 | /* | |
3bc942f3 TH |
4983 | * DO NOT USE IN NEW FILES. |
4984 | * | |
79bd9814 TH |
4985 | * Parse input and register new cgroup event handler. |
4986 | * | |
4987 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4988 | * Interpretation of args is defined by control file implementation. | |
4989 | */ | |
451af504 TH |
4990 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4991 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4992 | { |
451af504 | 4993 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4994 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4995 | struct mem_cgroup_event *event; |
79bd9814 TH |
4996 | struct cgroup_subsys_state *cfile_css; |
4997 | unsigned int efd, cfd; | |
4998 | struct fd efile; | |
4999 | struct fd cfile; | |
fba94807 | 5000 | const char *name; |
79bd9814 TH |
5001 | char *endp; |
5002 | int ret; | |
5003 | ||
451af504 TH |
5004 | buf = strstrip(buf); |
5005 | ||
5006 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
5007 | if (*endp != ' ') |
5008 | return -EINVAL; | |
451af504 | 5009 | buf = endp + 1; |
79bd9814 | 5010 | |
451af504 | 5011 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
5012 | if ((*endp != ' ') && (*endp != '\0')) |
5013 | return -EINVAL; | |
451af504 | 5014 | buf = endp + 1; |
79bd9814 TH |
5015 | |
5016 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
5017 | if (!event) | |
5018 | return -ENOMEM; | |
5019 | ||
59b6f873 | 5020 | event->memcg = memcg; |
79bd9814 | 5021 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
5022 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
5023 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
5024 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
5025 | |
5026 | efile = fdget(efd); | |
5027 | if (!efile.file) { | |
5028 | ret = -EBADF; | |
5029 | goto out_kfree; | |
5030 | } | |
5031 | ||
5032 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
5033 | if (IS_ERR(event->eventfd)) { | |
5034 | ret = PTR_ERR(event->eventfd); | |
5035 | goto out_put_efile; | |
5036 | } | |
5037 | ||
5038 | cfile = fdget(cfd); | |
5039 | if (!cfile.file) { | |
5040 | ret = -EBADF; | |
5041 | goto out_put_eventfd; | |
5042 | } | |
5043 | ||
5044 | /* the process need read permission on control file */ | |
5045 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
5046 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
5047 | if (ret < 0) | |
5048 | goto out_put_cfile; | |
5049 | ||
fba94807 TH |
5050 | /* |
5051 | * Determine the event callbacks and set them in @event. This used | |
5052 | * to be done via struct cftype but cgroup core no longer knows | |
5053 | * about these events. The following is crude but the whole thing | |
5054 | * is for compatibility anyway. | |
3bc942f3 TH |
5055 | * |
5056 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
5057 | */ |
5058 | name = cfile.file->f_dentry->d_name.name; | |
5059 | ||
5060 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
5061 | event->register_event = mem_cgroup_usage_register_event; | |
5062 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
5063 | } else if (!strcmp(name, "memory.oom_control")) { | |
5064 | event->register_event = mem_cgroup_oom_register_event; | |
5065 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
5066 | } else if (!strcmp(name, "memory.pressure_level")) { | |
5067 | event->register_event = vmpressure_register_event; | |
5068 | event->unregister_event = vmpressure_unregister_event; | |
5069 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
5070 | event->register_event = memsw_cgroup_usage_register_event; |
5071 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
5072 | } else { |
5073 | ret = -EINVAL; | |
5074 | goto out_put_cfile; | |
5075 | } | |
5076 | ||
79bd9814 | 5077 | /* |
b5557c4c TH |
5078 | * Verify @cfile should belong to @css. Also, remaining events are |
5079 | * automatically removed on cgroup destruction but the removal is | |
5080 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 5081 | */ |
ec903c0c TH |
5082 | cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent, |
5083 | &memory_cgrp_subsys); | |
79bd9814 | 5084 | ret = -EINVAL; |
5a17f543 | 5085 | if (IS_ERR(cfile_css)) |
79bd9814 | 5086 | goto out_put_cfile; |
5a17f543 TH |
5087 | if (cfile_css != css) { |
5088 | css_put(cfile_css); | |
79bd9814 | 5089 | goto out_put_cfile; |
5a17f543 | 5090 | } |
79bd9814 | 5091 | |
451af504 | 5092 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
5093 | if (ret) |
5094 | goto out_put_css; | |
5095 | ||
5096 | efile.file->f_op->poll(efile.file, &event->pt); | |
5097 | ||
fba94807 TH |
5098 | spin_lock(&memcg->event_list_lock); |
5099 | list_add(&event->list, &memcg->event_list); | |
5100 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
5101 | |
5102 | fdput(cfile); | |
5103 | fdput(efile); | |
5104 | ||
451af504 | 5105 | return nbytes; |
79bd9814 TH |
5106 | |
5107 | out_put_css: | |
b5557c4c | 5108 | css_put(css); |
79bd9814 TH |
5109 | out_put_cfile: |
5110 | fdput(cfile); | |
5111 | out_put_eventfd: | |
5112 | eventfd_ctx_put(event->eventfd); | |
5113 | out_put_efile: | |
5114 | fdput(efile); | |
5115 | out_kfree: | |
5116 | kfree(event); | |
5117 | ||
5118 | return ret; | |
5119 | } | |
5120 | ||
8cdea7c0 BS |
5121 | static struct cftype mem_cgroup_files[] = { |
5122 | { | |
0eea1030 | 5123 | .name = "usage_in_bytes", |
8c7c6e34 | 5124 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 5125 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5126 | }, |
c84872e1 PE |
5127 | { |
5128 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5129 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5130 | .write = mem_cgroup_reset, |
791badbd | 5131 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5132 | }, |
8cdea7c0 | 5133 | { |
0eea1030 | 5134 | .name = "limit_in_bytes", |
8c7c6e34 | 5135 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5136 | .write = mem_cgroup_write, |
791badbd | 5137 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5138 | }, |
296c81d8 BS |
5139 | { |
5140 | .name = "soft_limit_in_bytes", | |
5141 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5142 | .write = mem_cgroup_write, |
791badbd | 5143 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5144 | }, |
8cdea7c0 BS |
5145 | { |
5146 | .name = "failcnt", | |
8c7c6e34 | 5147 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5148 | .write = mem_cgroup_reset, |
791badbd | 5149 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5150 | }, |
d2ceb9b7 KH |
5151 | { |
5152 | .name = "stat", | |
2da8ca82 | 5153 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5154 | }, |
c1e862c1 KH |
5155 | { |
5156 | .name = "force_empty", | |
6770c64e | 5157 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5158 | }, |
18f59ea7 BS |
5159 | { |
5160 | .name = "use_hierarchy", | |
5161 | .write_u64 = mem_cgroup_hierarchy_write, | |
5162 | .read_u64 = mem_cgroup_hierarchy_read, | |
5163 | }, | |
79bd9814 | 5164 | { |
3bc942f3 | 5165 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5166 | .write = memcg_write_event_control, |
79bd9814 TH |
5167 | .flags = CFTYPE_NO_PREFIX, |
5168 | .mode = S_IWUGO, | |
5169 | }, | |
a7885eb8 KM |
5170 | { |
5171 | .name = "swappiness", | |
5172 | .read_u64 = mem_cgroup_swappiness_read, | |
5173 | .write_u64 = mem_cgroup_swappiness_write, | |
5174 | }, | |
7dc74be0 DN |
5175 | { |
5176 | .name = "move_charge_at_immigrate", | |
5177 | .read_u64 = mem_cgroup_move_charge_read, | |
5178 | .write_u64 = mem_cgroup_move_charge_write, | |
5179 | }, | |
9490ff27 KH |
5180 | { |
5181 | .name = "oom_control", | |
2da8ca82 | 5182 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5183 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5184 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5185 | }, | |
70ddf637 AV |
5186 | { |
5187 | .name = "pressure_level", | |
70ddf637 | 5188 | }, |
406eb0c9 YH |
5189 | #ifdef CONFIG_NUMA |
5190 | { | |
5191 | .name = "numa_stat", | |
2da8ca82 | 5192 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5193 | }, |
5194 | #endif | |
510fc4e1 GC |
5195 | #ifdef CONFIG_MEMCG_KMEM |
5196 | { | |
5197 | .name = "kmem.limit_in_bytes", | |
5198 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5199 | .write = mem_cgroup_write, |
791badbd | 5200 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5201 | }, |
5202 | { | |
5203 | .name = "kmem.usage_in_bytes", | |
5204 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5205 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5206 | }, |
5207 | { | |
5208 | .name = "kmem.failcnt", | |
5209 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5210 | .write = mem_cgroup_reset, |
791badbd | 5211 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5212 | }, |
5213 | { | |
5214 | .name = "kmem.max_usage_in_bytes", | |
5215 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5216 | .write = mem_cgroup_reset, |
791badbd | 5217 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5218 | }, |
749c5415 GC |
5219 | #ifdef CONFIG_SLABINFO |
5220 | { | |
5221 | .name = "kmem.slabinfo", | |
2da8ca82 | 5222 | .seq_show = mem_cgroup_slabinfo_read, |
749c5415 GC |
5223 | }, |
5224 | #endif | |
8c7c6e34 | 5225 | #endif |
6bc10349 | 5226 | { }, /* terminate */ |
af36f906 | 5227 | }; |
8c7c6e34 | 5228 | |
2d11085e MH |
5229 | #ifdef CONFIG_MEMCG_SWAP |
5230 | static struct cftype memsw_cgroup_files[] = { | |
5231 | { | |
5232 | .name = "memsw.usage_in_bytes", | |
5233 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 5234 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5235 | }, |
5236 | { | |
5237 | .name = "memsw.max_usage_in_bytes", | |
5238 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 5239 | .write = mem_cgroup_reset, |
791badbd | 5240 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5241 | }, |
5242 | { | |
5243 | .name = "memsw.limit_in_bytes", | |
5244 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 5245 | .write = mem_cgroup_write, |
791badbd | 5246 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5247 | }, |
5248 | { | |
5249 | .name = "memsw.failcnt", | |
5250 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 5251 | .write = mem_cgroup_reset, |
791badbd | 5252 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5253 | }, |
5254 | { }, /* terminate */ | |
5255 | }; | |
5256 | #endif | |
c0ff4b85 | 5257 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5258 | { |
5259 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 5260 | struct mem_cgroup_per_zone *mz; |
41e3355d | 5261 | int zone, tmp = node; |
1ecaab2b KH |
5262 | /* |
5263 | * This routine is called against possible nodes. | |
5264 | * But it's BUG to call kmalloc() against offline node. | |
5265 | * | |
5266 | * TODO: this routine can waste much memory for nodes which will | |
5267 | * never be onlined. It's better to use memory hotplug callback | |
5268 | * function. | |
5269 | */ | |
41e3355d KH |
5270 | if (!node_state(node, N_NORMAL_MEMORY)) |
5271 | tmp = -1; | |
17295c88 | 5272 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5273 | if (!pn) |
5274 | return 1; | |
1ecaab2b | 5275 | |
1ecaab2b KH |
5276 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
5277 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 5278 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
5279 | mz->usage_in_excess = 0; |
5280 | mz->on_tree = false; | |
d79154bb | 5281 | mz->memcg = memcg; |
1ecaab2b | 5282 | } |
54f72fe0 | 5283 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5284 | return 0; |
5285 | } | |
5286 | ||
c0ff4b85 | 5287 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5288 | { |
54f72fe0 | 5289 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
5290 | } |
5291 | ||
33327948 KH |
5292 | static struct mem_cgroup *mem_cgroup_alloc(void) |
5293 | { | |
d79154bb | 5294 | struct mem_cgroup *memcg; |
8ff69e2c | 5295 | size_t size; |
33327948 | 5296 | |
8ff69e2c VD |
5297 | size = sizeof(struct mem_cgroup); |
5298 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 5299 | |
8ff69e2c | 5300 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 5301 | if (!memcg) |
e7bbcdf3 DC |
5302 | return NULL; |
5303 | ||
d79154bb HD |
5304 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
5305 | if (!memcg->stat) | |
d2e61b8d | 5306 | goto out_free; |
d79154bb HD |
5307 | spin_lock_init(&memcg->pcp_counter_lock); |
5308 | return memcg; | |
d2e61b8d DC |
5309 | |
5310 | out_free: | |
8ff69e2c | 5311 | kfree(memcg); |
d2e61b8d | 5312 | return NULL; |
33327948 KH |
5313 | } |
5314 | ||
59927fb9 | 5315 | /* |
c8b2a36f GC |
5316 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
5317 | * (scanning all at force_empty is too costly...) | |
5318 | * | |
5319 | * Instead of clearing all references at force_empty, we remember | |
5320 | * the number of reference from swap_cgroup and free mem_cgroup when | |
5321 | * it goes down to 0. | |
5322 | * | |
5323 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 5324 | */ |
c8b2a36f GC |
5325 | |
5326 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 5327 | { |
c8b2a36f | 5328 | int node; |
59927fb9 | 5329 | |
bb4cc1a8 | 5330 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
5331 | |
5332 | for_each_node(node) | |
5333 | free_mem_cgroup_per_zone_info(memcg, node); | |
5334 | ||
5335 | free_percpu(memcg->stat); | |
5336 | ||
3f134619 GC |
5337 | /* |
5338 | * We need to make sure that (at least for now), the jump label | |
5339 | * destruction code runs outside of the cgroup lock. This is because | |
5340 | * get_online_cpus(), which is called from the static_branch update, | |
5341 | * can't be called inside the cgroup_lock. cpusets are the ones | |
5342 | * enforcing this dependency, so if they ever change, we might as well. | |
5343 | * | |
5344 | * schedule_work() will guarantee this happens. Be careful if you need | |
5345 | * to move this code around, and make sure it is outside | |
5346 | * the cgroup_lock. | |
5347 | */ | |
a8964b9b | 5348 | disarm_static_keys(memcg); |
8ff69e2c | 5349 | kfree(memcg); |
59927fb9 | 5350 | } |
3afe36b1 | 5351 | |
7bcc1bb1 DN |
5352 | /* |
5353 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
5354 | */ | |
e1aab161 | 5355 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 5356 | { |
c0ff4b85 | 5357 | if (!memcg->res.parent) |
7bcc1bb1 | 5358 | return NULL; |
c0ff4b85 | 5359 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 5360 | } |
e1aab161 | 5361 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 5362 | |
bb4cc1a8 AM |
5363 | static void __init mem_cgroup_soft_limit_tree_init(void) |
5364 | { | |
5365 | struct mem_cgroup_tree_per_node *rtpn; | |
5366 | struct mem_cgroup_tree_per_zone *rtpz; | |
5367 | int tmp, node, zone; | |
5368 | ||
5369 | for_each_node(node) { | |
5370 | tmp = node; | |
5371 | if (!node_state(node, N_NORMAL_MEMORY)) | |
5372 | tmp = -1; | |
5373 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
5374 | BUG_ON(!rtpn); | |
5375 | ||
5376 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5377 | ||
5378 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5379 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5380 | rtpz->rb_root = RB_ROOT; | |
5381 | spin_lock_init(&rtpz->lock); | |
5382 | } | |
5383 | } | |
5384 | } | |
5385 | ||
0eb253e2 | 5386 | static struct cgroup_subsys_state * __ref |
eb95419b | 5387 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 5388 | { |
d142e3e6 | 5389 | struct mem_cgroup *memcg; |
04046e1a | 5390 | long error = -ENOMEM; |
6d12e2d8 | 5391 | int node; |
8cdea7c0 | 5392 | |
c0ff4b85 R |
5393 | memcg = mem_cgroup_alloc(); |
5394 | if (!memcg) | |
04046e1a | 5395 | return ERR_PTR(error); |
78fb7466 | 5396 | |
3ed28fa1 | 5397 | for_each_node(node) |
c0ff4b85 | 5398 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5399 | goto free_out; |
f64c3f54 | 5400 | |
c077719b | 5401 | /* root ? */ |
eb95419b | 5402 | if (parent_css == NULL) { |
a41c58a6 | 5403 | root_mem_cgroup = memcg; |
d142e3e6 GC |
5404 | res_counter_init(&memcg->res, NULL); |
5405 | res_counter_init(&memcg->memsw, NULL); | |
5406 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 5407 | } |
28dbc4b6 | 5408 | |
d142e3e6 GC |
5409 | memcg->last_scanned_node = MAX_NUMNODES; |
5410 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
5411 | memcg->move_charge_at_immigrate = 0; |
5412 | mutex_init(&memcg->thresholds_lock); | |
5413 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5414 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5415 | INIT_LIST_HEAD(&memcg->event_list); |
5416 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
5417 | |
5418 | return &memcg->css; | |
5419 | ||
5420 | free_out: | |
5421 | __mem_cgroup_free(memcg); | |
5422 | return ERR_PTR(error); | |
5423 | } | |
5424 | ||
5425 | static int | |
eb95419b | 5426 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 5427 | { |
eb95419b | 5428 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 5429 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 5430 | int ret; |
d142e3e6 | 5431 | |
15a4c835 | 5432 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
5433 | return -ENOSPC; |
5434 | ||
63876986 | 5435 | if (!parent) |
d142e3e6 GC |
5436 | return 0; |
5437 | ||
0999821b | 5438 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
5439 | |
5440 | memcg->use_hierarchy = parent->use_hierarchy; | |
5441 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5442 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5443 | ||
5444 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
5445 | res_counter_init(&memcg->res, &parent->res); |
5446 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 5447 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 5448 | |
7bcc1bb1 | 5449 | /* |
8d76a979 LZ |
5450 | * No need to take a reference to the parent because cgroup |
5451 | * core guarantees its existence. | |
7bcc1bb1 | 5452 | */ |
18f59ea7 | 5453 | } else { |
ce00a967 JW |
5454 | res_counter_init(&memcg->res, NULL); |
5455 | res_counter_init(&memcg->memsw, NULL); | |
5456 | res_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
5457 | /* |
5458 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5459 | * much sense so let cgroup subsystem know about this | |
5460 | * unfortunate state in our controller. | |
5461 | */ | |
d142e3e6 | 5462 | if (parent != root_mem_cgroup) |
073219e9 | 5463 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5464 | } |
0999821b | 5465 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 5466 | |
2f7dd7a4 JW |
5467 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
5468 | if (ret) | |
5469 | return ret; | |
5470 | ||
5471 | /* | |
5472 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
5473 | * orders reading memcg->initialized against its callers | |
5474 | * reading the memcg members. | |
5475 | */ | |
5476 | smp_store_release(&memcg->initialized, 1); | |
5477 | ||
5478 | return 0; | |
8cdea7c0 BS |
5479 | } |
5480 | ||
5f578161 MH |
5481 | /* |
5482 | * Announce all parents that a group from their hierarchy is gone. | |
5483 | */ | |
5484 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
5485 | { | |
5486 | struct mem_cgroup *parent = memcg; | |
5487 | ||
5488 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 5489 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
5490 | |
5491 | /* | |
5492 | * if the root memcg is not hierarchical we have to check it | |
5493 | * explicitely. | |
5494 | */ | |
5495 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 5496 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
5497 | } |
5498 | ||
eb95419b | 5499 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5500 | { |
eb95419b | 5501 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5502 | struct mem_cgroup_event *event, *tmp; |
4fb1a86f | 5503 | struct cgroup_subsys_state *iter; |
79bd9814 TH |
5504 | |
5505 | /* | |
5506 | * Unregister events and notify userspace. | |
5507 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5508 | * directory to avoid race between userspace and kernelspace. | |
5509 | */ | |
fba94807 TH |
5510 | spin_lock(&memcg->event_list_lock); |
5511 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5512 | list_del_init(&event->list); |
5513 | schedule_work(&event->remove); | |
5514 | } | |
fba94807 | 5515 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5516 | |
10d5ebf4 LZ |
5517 | kmem_cgroup_css_offline(memcg); |
5518 | ||
5f578161 | 5519 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
4fb1a86f FB |
5520 | |
5521 | /* | |
5522 | * This requires that offlining is serialized. Right now that is | |
5523 | * guaranteed because css_killed_work_fn() holds the cgroup_mutex. | |
5524 | */ | |
5525 | css_for_each_descendant_post(iter, css) | |
5526 | mem_cgroup_reparent_charges(mem_cgroup_from_css(iter)); | |
5527 | ||
776ed0f0 | 5528 | memcg_unregister_all_caches(memcg); |
33cb876e | 5529 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
5530 | } |
5531 | ||
eb95419b | 5532 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5533 | { |
eb95419b | 5534 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
96f1c58d JW |
5535 | /* |
5536 | * XXX: css_offline() would be where we should reparent all | |
5537 | * memory to prepare the cgroup for destruction. However, | |
ec903c0c | 5538 | * memcg does not do css_tryget_online() and res_counter charging |
96f1c58d JW |
5539 | * under the same RCU lock region, which means that charging |
5540 | * could race with offlining. Offlining only happens to | |
5541 | * cgroups with no tasks in them but charges can show up | |
5542 | * without any tasks from the swapin path when the target | |
5543 | * memcg is looked up from the swapout record and not from the | |
5544 | * current task as it usually is. A race like this can leak | |
5545 | * charges and put pages with stale cgroup pointers into | |
5546 | * circulation: | |
5547 | * | |
5548 | * #0 #1 | |
5549 | * lookup_swap_cgroup_id() | |
5550 | * rcu_read_lock() | |
5551 | * mem_cgroup_lookup() | |
ec903c0c | 5552 | * css_tryget_online() |
96f1c58d | 5553 | * rcu_read_unlock() |
ec903c0c | 5554 | * disable css_tryget_online() |
96f1c58d JW |
5555 | * call_rcu() |
5556 | * offline_css() | |
5557 | * reparent_charges() | |
5558 | * res_counter_charge() | |
5559 | * css_put() | |
5560 | * css_free() | |
5561 | * pc->mem_cgroup = dead memcg | |
5562 | * add page to lru | |
5563 | * | |
5564 | * The bulk of the charges are still moved in offline_css() to | |
5565 | * avoid pinning a lot of pages in case a long-term reference | |
5566 | * like a swapout record is deferring the css_free() to long | |
5567 | * after offlining. But this makes sure we catch any charges | |
5568 | * made after offlining: | |
5569 | */ | |
5570 | mem_cgroup_reparent_charges(memcg); | |
c268e994 | 5571 | |
10d5ebf4 | 5572 | memcg_destroy_kmem(memcg); |
465939a1 | 5573 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
5574 | } |
5575 | ||
1ced953b TH |
5576 | /** |
5577 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5578 | * @css: the target css | |
5579 | * | |
5580 | * Reset the states of the mem_cgroup associated with @css. This is | |
5581 | * invoked when the userland requests disabling on the default hierarchy | |
5582 | * but the memcg is pinned through dependency. The memcg should stop | |
5583 | * applying policies and should revert to the vanilla state as it may be | |
5584 | * made visible again. | |
5585 | * | |
5586 | * The current implementation only resets the essential configurations. | |
5587 | * This needs to be expanded to cover all the visible parts. | |
5588 | */ | |
5589 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5590 | { | |
5591 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5592 | ||
5593 | mem_cgroup_resize_limit(memcg, ULLONG_MAX); | |
5594 | mem_cgroup_resize_memsw_limit(memcg, ULLONG_MAX); | |
5595 | memcg_update_kmem_limit(memcg, ULLONG_MAX); | |
5596 | res_counter_set_soft_limit(&memcg->res, ULLONG_MAX); | |
5597 | } | |
5598 | ||
02491447 | 5599 | #ifdef CONFIG_MMU |
7dc74be0 | 5600 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5601 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5602 | { |
05b84301 | 5603 | int ret; |
9476db97 JW |
5604 | |
5605 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 5606 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 5607 | if (!ret) { |
854ffa8d | 5608 | mc.precharge += count; |
854ffa8d DN |
5609 | return ret; |
5610 | } | |
692e7c45 | 5611 | if (ret == -EINTR) { |
00501b53 | 5612 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
5613 | return ret; |
5614 | } | |
9476db97 JW |
5615 | |
5616 | /* Try charges one by one with reclaim */ | |
854ffa8d | 5617 | while (count--) { |
00501b53 | 5618 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
5619 | /* |
5620 | * In case of failure, any residual charges against | |
5621 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
5622 | * later on. However, cancel any charges that are |
5623 | * bypassed to root right away or they'll be lost. | |
9476db97 | 5624 | */ |
692e7c45 | 5625 | if (ret == -EINTR) |
00501b53 | 5626 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 5627 | if (ret) |
38c5d72f | 5628 | return ret; |
854ffa8d | 5629 | mc.precharge++; |
9476db97 | 5630 | cond_resched(); |
854ffa8d | 5631 | } |
9476db97 | 5632 | return 0; |
4ffef5fe DN |
5633 | } |
5634 | ||
5635 | /** | |
8d32ff84 | 5636 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5637 | * @vma: the vma the pte to be checked belongs |
5638 | * @addr: the address corresponding to the pte to be checked | |
5639 | * @ptent: the pte to be checked | |
02491447 | 5640 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5641 | * |
5642 | * Returns | |
5643 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5644 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5645 | * move charge. if @target is not NULL, the page is stored in target->page | |
5646 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5647 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5648 | * target for charge migration. if @target is not NULL, the entry is stored | |
5649 | * in target->ent. | |
4ffef5fe DN |
5650 | * |
5651 | * Called with pte lock held. | |
5652 | */ | |
4ffef5fe DN |
5653 | union mc_target { |
5654 | struct page *page; | |
02491447 | 5655 | swp_entry_t ent; |
4ffef5fe DN |
5656 | }; |
5657 | ||
4ffef5fe | 5658 | enum mc_target_type { |
8d32ff84 | 5659 | MC_TARGET_NONE = 0, |
4ffef5fe | 5660 | MC_TARGET_PAGE, |
02491447 | 5661 | MC_TARGET_SWAP, |
4ffef5fe DN |
5662 | }; |
5663 | ||
90254a65 DN |
5664 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5665 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5666 | { |
90254a65 | 5667 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5668 | |
90254a65 DN |
5669 | if (!page || !page_mapped(page)) |
5670 | return NULL; | |
5671 | if (PageAnon(page)) { | |
5672 | /* we don't move shared anon */ | |
4b91355e | 5673 | if (!move_anon()) |
90254a65 | 5674 | return NULL; |
87946a72 DN |
5675 | } else if (!move_file()) |
5676 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5677 | return NULL; |
5678 | if (!get_page_unless_zero(page)) | |
5679 | return NULL; | |
5680 | ||
5681 | return page; | |
5682 | } | |
5683 | ||
4b91355e | 5684 | #ifdef CONFIG_SWAP |
90254a65 DN |
5685 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5686 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5687 | { | |
90254a65 DN |
5688 | struct page *page = NULL; |
5689 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5690 | ||
5691 | if (!move_anon() || non_swap_entry(ent)) | |
5692 | return NULL; | |
4b91355e KH |
5693 | /* |
5694 | * Because lookup_swap_cache() updates some statistics counter, | |
5695 | * we call find_get_page() with swapper_space directly. | |
5696 | */ | |
33806f06 | 5697 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
5698 | if (do_swap_account) |
5699 | entry->val = ent.val; | |
5700 | ||
5701 | return page; | |
5702 | } | |
4b91355e KH |
5703 | #else |
5704 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5705 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5706 | { | |
5707 | return NULL; | |
5708 | } | |
5709 | #endif | |
90254a65 | 5710 | |
87946a72 DN |
5711 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5712 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5713 | { | |
5714 | struct page *page = NULL; | |
87946a72 DN |
5715 | struct address_space *mapping; |
5716 | pgoff_t pgoff; | |
5717 | ||
5718 | if (!vma->vm_file) /* anonymous vma */ | |
5719 | return NULL; | |
5720 | if (!move_file()) | |
5721 | return NULL; | |
5722 | ||
87946a72 DN |
5723 | mapping = vma->vm_file->f_mapping; |
5724 | if (pte_none(ptent)) | |
5725 | pgoff = linear_page_index(vma, addr); | |
5726 | else /* pte_file(ptent) is true */ | |
5727 | pgoff = pte_to_pgoff(ptent); | |
5728 | ||
5729 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5730 | #ifdef CONFIG_SWAP |
5731 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5732 | if (shmem_mapping(mapping)) { |
5733 | page = find_get_entry(mapping, pgoff); | |
5734 | if (radix_tree_exceptional_entry(page)) { | |
5735 | swp_entry_t swp = radix_to_swp_entry(page); | |
5736 | if (do_swap_account) | |
5737 | *entry = swp; | |
5738 | page = find_get_page(swap_address_space(swp), swp.val); | |
5739 | } | |
5740 | } else | |
5741 | page = find_get_page(mapping, pgoff); | |
5742 | #else | |
5743 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5744 | #endif |
87946a72 DN |
5745 | return page; |
5746 | } | |
5747 | ||
8d32ff84 | 5748 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5749 | unsigned long addr, pte_t ptent, union mc_target *target) |
5750 | { | |
5751 | struct page *page = NULL; | |
5752 | struct page_cgroup *pc; | |
8d32ff84 | 5753 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5754 | swp_entry_t ent = { .val = 0 }; |
5755 | ||
5756 | if (pte_present(ptent)) | |
5757 | page = mc_handle_present_pte(vma, addr, ptent); | |
5758 | else if (is_swap_pte(ptent)) | |
5759 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5760 | else if (pte_none(ptent) || pte_file(ptent)) |
5761 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5762 | |
5763 | if (!page && !ent.val) | |
8d32ff84 | 5764 | return ret; |
02491447 DN |
5765 | if (page) { |
5766 | pc = lookup_page_cgroup(page); | |
5767 | /* | |
0a31bc97 JW |
5768 | * Do only loose check w/o serialization. |
5769 | * mem_cgroup_move_account() checks the pc is valid or | |
5770 | * not under LRU exclusion. | |
02491447 DN |
5771 | */ |
5772 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5773 | ret = MC_TARGET_PAGE; | |
5774 | if (target) | |
5775 | target->page = page; | |
5776 | } | |
5777 | if (!ret || !target) | |
5778 | put_page(page); | |
5779 | } | |
90254a65 DN |
5780 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5781 | if (ent.val && !ret && | |
34c00c31 | 5782 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5783 | ret = MC_TARGET_SWAP; |
5784 | if (target) | |
5785 | target->ent = ent; | |
4ffef5fe | 5786 | } |
4ffef5fe DN |
5787 | return ret; |
5788 | } | |
5789 | ||
12724850 NH |
5790 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5791 | /* | |
5792 | * We don't consider swapping or file mapped pages because THP does not | |
5793 | * support them for now. | |
5794 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5795 | */ | |
5796 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5797 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5798 | { | |
5799 | struct page *page = NULL; | |
5800 | struct page_cgroup *pc; | |
5801 | enum mc_target_type ret = MC_TARGET_NONE; | |
5802 | ||
5803 | page = pmd_page(pmd); | |
309381fe | 5804 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
5805 | if (!move_anon()) |
5806 | return ret; | |
5807 | pc = lookup_page_cgroup(page); | |
5808 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5809 | ret = MC_TARGET_PAGE; | |
5810 | if (target) { | |
5811 | get_page(page); | |
5812 | target->page = page; | |
5813 | } | |
5814 | } | |
5815 | return ret; | |
5816 | } | |
5817 | #else | |
5818 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5819 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5820 | { | |
5821 | return MC_TARGET_NONE; | |
5822 | } | |
5823 | #endif | |
5824 | ||
4ffef5fe DN |
5825 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5826 | unsigned long addr, unsigned long end, | |
5827 | struct mm_walk *walk) | |
5828 | { | |
5829 | struct vm_area_struct *vma = walk->private; | |
5830 | pte_t *pte; | |
5831 | spinlock_t *ptl; | |
5832 | ||
bf929152 | 5833 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
5834 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5835 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5836 | spin_unlock(ptl); |
1a5a9906 | 5837 | return 0; |
12724850 | 5838 | } |
03319327 | 5839 | |
45f83cef AA |
5840 | if (pmd_trans_unstable(pmd)) |
5841 | return 0; | |
4ffef5fe DN |
5842 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5843 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5844 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5845 | mc.precharge++; /* increment precharge temporarily */ |
5846 | pte_unmap_unlock(pte - 1, ptl); | |
5847 | cond_resched(); | |
5848 | ||
7dc74be0 DN |
5849 | return 0; |
5850 | } | |
5851 | ||
4ffef5fe DN |
5852 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5853 | { | |
5854 | unsigned long precharge; | |
5855 | struct vm_area_struct *vma; | |
5856 | ||
dfe076b0 | 5857 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5858 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5859 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5860 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5861 | .mm = mm, | |
5862 | .private = vma, | |
5863 | }; | |
5864 | if (is_vm_hugetlb_page(vma)) | |
5865 | continue; | |
4ffef5fe DN |
5866 | walk_page_range(vma->vm_start, vma->vm_end, |
5867 | &mem_cgroup_count_precharge_walk); | |
5868 | } | |
dfe076b0 | 5869 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5870 | |
5871 | precharge = mc.precharge; | |
5872 | mc.precharge = 0; | |
5873 | ||
5874 | return precharge; | |
5875 | } | |
5876 | ||
4ffef5fe DN |
5877 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5878 | { | |
dfe076b0 DN |
5879 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5880 | ||
5881 | VM_BUG_ON(mc.moving_task); | |
5882 | mc.moving_task = current; | |
5883 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5884 | } |
5885 | ||
dfe076b0 DN |
5886 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5887 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5888 | { |
2bd9bb20 KH |
5889 | struct mem_cgroup *from = mc.from; |
5890 | struct mem_cgroup *to = mc.to; | |
4050377b | 5891 | int i; |
2bd9bb20 | 5892 | |
4ffef5fe | 5893 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5894 | if (mc.precharge) { |
00501b53 | 5895 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5896 | mc.precharge = 0; |
5897 | } | |
5898 | /* | |
5899 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5900 | * we must uncharge here. | |
5901 | */ | |
5902 | if (mc.moved_charge) { | |
00501b53 | 5903 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5904 | mc.moved_charge = 0; |
4ffef5fe | 5905 | } |
483c30b5 DN |
5906 | /* we must fixup refcnts and charges */ |
5907 | if (mc.moved_swap) { | |
483c30b5 | 5908 | /* uncharge swap account from the old cgroup */ |
ce00a967 JW |
5909 | if (!mem_cgroup_is_root(mc.from)) |
5910 | res_counter_uncharge(&mc.from->memsw, | |
5911 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
5912 | |
5913 | for (i = 0; i < mc.moved_swap; i++) | |
5914 | css_put(&mc.from->css); | |
483c30b5 | 5915 | |
05b84301 JW |
5916 | /* |
5917 | * we charged both to->res and to->memsw, so we should | |
5918 | * uncharge to->res. | |
5919 | */ | |
ce00a967 JW |
5920 | if (!mem_cgroup_is_root(mc.to)) |
5921 | res_counter_uncharge(&mc.to->res, | |
5922 | PAGE_SIZE * mc.moved_swap); | |
4050377b | 5923 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
5924 | mc.moved_swap = 0; |
5925 | } | |
dfe076b0 DN |
5926 | memcg_oom_recover(from); |
5927 | memcg_oom_recover(to); | |
5928 | wake_up_all(&mc.waitq); | |
5929 | } | |
5930 | ||
5931 | static void mem_cgroup_clear_mc(void) | |
5932 | { | |
5933 | struct mem_cgroup *from = mc.from; | |
5934 | ||
5935 | /* | |
5936 | * we must clear moving_task before waking up waiters at the end of | |
5937 | * task migration. | |
5938 | */ | |
5939 | mc.moving_task = NULL; | |
5940 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5941 | spin_lock(&mc.lock); |
4ffef5fe DN |
5942 | mc.from = NULL; |
5943 | mc.to = NULL; | |
2bd9bb20 | 5944 | spin_unlock(&mc.lock); |
32047e2a | 5945 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5946 | } |
5947 | ||
eb95419b | 5948 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5949 | struct cgroup_taskset *tset) |
7dc74be0 | 5950 | { |
2f7ee569 | 5951 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5952 | int ret = 0; |
eb95419b | 5953 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 5954 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 5955 | |
ee5e8472 GC |
5956 | /* |
5957 | * We are now commited to this value whatever it is. Changes in this | |
5958 | * tunable will only affect upcoming migrations, not the current one. | |
5959 | * So we need to save it, and keep it going. | |
5960 | */ | |
5961 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
5962 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
5963 | struct mm_struct *mm; |
5964 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5965 | ||
c0ff4b85 | 5966 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5967 | |
5968 | mm = get_task_mm(p); | |
5969 | if (!mm) | |
5970 | return 0; | |
7dc74be0 | 5971 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5972 | if (mm->owner == p) { |
5973 | VM_BUG_ON(mc.from); | |
5974 | VM_BUG_ON(mc.to); | |
5975 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5976 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5977 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5978 | mem_cgroup_start_move(from); |
2bd9bb20 | 5979 | spin_lock(&mc.lock); |
4ffef5fe | 5980 | mc.from = from; |
c0ff4b85 | 5981 | mc.to = memcg; |
ee5e8472 | 5982 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 5983 | spin_unlock(&mc.lock); |
dfe076b0 | 5984 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5985 | |
5986 | ret = mem_cgroup_precharge_mc(mm); | |
5987 | if (ret) | |
5988 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5989 | } |
5990 | mmput(mm); | |
7dc74be0 DN |
5991 | } |
5992 | return ret; | |
5993 | } | |
5994 | ||
eb95419b | 5995 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5996 | struct cgroup_taskset *tset) |
7dc74be0 | 5997 | { |
4ffef5fe | 5998 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5999 | } |
6000 | ||
4ffef5fe DN |
6001 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6002 | unsigned long addr, unsigned long end, | |
6003 | struct mm_walk *walk) | |
7dc74be0 | 6004 | { |
4ffef5fe DN |
6005 | int ret = 0; |
6006 | struct vm_area_struct *vma = walk->private; | |
6007 | pte_t *pte; | |
6008 | spinlock_t *ptl; | |
12724850 NH |
6009 | enum mc_target_type target_type; |
6010 | union mc_target target; | |
6011 | struct page *page; | |
6012 | struct page_cgroup *pc; | |
4ffef5fe | 6013 | |
12724850 NH |
6014 | /* |
6015 | * We don't take compound_lock() here but no race with splitting thp | |
6016 | * happens because: | |
6017 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
6018 | * under splitting, which means there's no concurrent thp split, | |
6019 | * - if another thread runs into split_huge_page() just after we | |
6020 | * entered this if-block, the thread must wait for page table lock | |
6021 | * to be unlocked in __split_huge_page_splitting(), where the main | |
6022 | * part of thp split is not executed yet. | |
6023 | */ | |
bf929152 | 6024 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 6025 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6026 | spin_unlock(ptl); |
12724850 NH |
6027 | return 0; |
6028 | } | |
6029 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6030 | if (target_type == MC_TARGET_PAGE) { | |
6031 | page = target.page; | |
6032 | if (!isolate_lru_page(page)) { | |
6033 | pc = lookup_page_cgroup(page); | |
6034 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6035 | pc, mc.from, mc.to)) { |
12724850 NH |
6036 | mc.precharge -= HPAGE_PMD_NR; |
6037 | mc.moved_charge += HPAGE_PMD_NR; | |
6038 | } | |
6039 | putback_lru_page(page); | |
6040 | } | |
6041 | put_page(page); | |
6042 | } | |
bf929152 | 6043 | spin_unlock(ptl); |
1a5a9906 | 6044 | return 0; |
12724850 NH |
6045 | } |
6046 | ||
45f83cef AA |
6047 | if (pmd_trans_unstable(pmd)) |
6048 | return 0; | |
4ffef5fe DN |
6049 | retry: |
6050 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6051 | for (; addr != end; addr += PAGE_SIZE) { | |
6052 | pte_t ptent = *(pte++); | |
02491447 | 6053 | swp_entry_t ent; |
4ffef5fe DN |
6054 | |
6055 | if (!mc.precharge) | |
6056 | break; | |
6057 | ||
8d32ff84 | 6058 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6059 | case MC_TARGET_PAGE: |
6060 | page = target.page; | |
6061 | if (isolate_lru_page(page)) | |
6062 | goto put; | |
6063 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6064 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6065 | mc.from, mc.to)) { |
4ffef5fe | 6066 | mc.precharge--; |
854ffa8d DN |
6067 | /* we uncharge from mc.from later. */ |
6068 | mc.moved_charge++; | |
4ffef5fe DN |
6069 | } |
6070 | putback_lru_page(page); | |
8d32ff84 | 6071 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6072 | put_page(page); |
6073 | break; | |
02491447 DN |
6074 | case MC_TARGET_SWAP: |
6075 | ent = target.ent; | |
e91cbb42 | 6076 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6077 | mc.precharge--; |
483c30b5 DN |
6078 | /* we fixup refcnts and charges later. */ |
6079 | mc.moved_swap++; | |
6080 | } | |
02491447 | 6081 | break; |
4ffef5fe DN |
6082 | default: |
6083 | break; | |
6084 | } | |
6085 | } | |
6086 | pte_unmap_unlock(pte - 1, ptl); | |
6087 | cond_resched(); | |
6088 | ||
6089 | if (addr != end) { | |
6090 | /* | |
6091 | * We have consumed all precharges we got in can_attach(). | |
6092 | * We try charge one by one, but don't do any additional | |
6093 | * charges to mc.to if we have failed in charge once in attach() | |
6094 | * phase. | |
6095 | */ | |
854ffa8d | 6096 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6097 | if (!ret) |
6098 | goto retry; | |
6099 | } | |
6100 | ||
6101 | return ret; | |
6102 | } | |
6103 | ||
6104 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6105 | { | |
6106 | struct vm_area_struct *vma; | |
6107 | ||
6108 | lru_add_drain_all(); | |
dfe076b0 DN |
6109 | retry: |
6110 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6111 | /* | |
6112 | * Someone who are holding the mmap_sem might be waiting in | |
6113 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6114 | * and retry. Because we cancel precharges, we might not be able | |
6115 | * to move enough charges, but moving charge is a best-effort | |
6116 | * feature anyway, so it wouldn't be a big problem. | |
6117 | */ | |
6118 | __mem_cgroup_clear_mc(); | |
6119 | cond_resched(); | |
6120 | goto retry; | |
6121 | } | |
4ffef5fe DN |
6122 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6123 | int ret; | |
6124 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6125 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6126 | .mm = mm, | |
6127 | .private = vma, | |
6128 | }; | |
6129 | if (is_vm_hugetlb_page(vma)) | |
6130 | continue; | |
4ffef5fe DN |
6131 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6132 | &mem_cgroup_move_charge_walk); | |
6133 | if (ret) | |
6134 | /* | |
6135 | * means we have consumed all precharges and failed in | |
6136 | * doing additional charge. Just abandon here. | |
6137 | */ | |
6138 | break; | |
6139 | } | |
dfe076b0 | 6140 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6141 | } |
6142 | ||
eb95419b | 6143 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6144 | struct cgroup_taskset *tset) |
67e465a7 | 6145 | { |
2f7ee569 | 6146 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6147 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6148 | |
dfe076b0 | 6149 | if (mm) { |
a433658c KM |
6150 | if (mc.to) |
6151 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6152 | mmput(mm); |
6153 | } | |
a433658c KM |
6154 | if (mc.to) |
6155 | mem_cgroup_clear_mc(); | |
67e465a7 | 6156 | } |
5cfb80a7 | 6157 | #else /* !CONFIG_MMU */ |
eb95419b | 6158 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6159 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6160 | { |
6161 | return 0; | |
6162 | } | |
eb95419b | 6163 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6164 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6165 | { |
6166 | } | |
eb95419b | 6167 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6168 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6169 | { |
6170 | } | |
6171 | #endif | |
67e465a7 | 6172 | |
f00baae7 TH |
6173 | /* |
6174 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6175 | * to verify whether we're attached to the default hierarchy on each mount |
6176 | * attempt. | |
f00baae7 | 6177 | */ |
eb95419b | 6178 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6179 | { |
6180 | /* | |
aa6ec29b | 6181 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6182 | * guarantees that @root doesn't have any children, so turning it |
6183 | * on for the root memcg is enough. | |
6184 | */ | |
aa6ec29b | 6185 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 6186 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
6187 | } |
6188 | ||
073219e9 | 6189 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6190 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6191 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6192 | .css_offline = mem_cgroup_css_offline, |
6193 | .css_free = mem_cgroup_css_free, | |
1ced953b | 6194 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6195 | .can_attach = mem_cgroup_can_attach, |
6196 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6197 | .attach = mem_cgroup_move_task, |
f00baae7 | 6198 | .bind = mem_cgroup_bind, |
5577964e | 6199 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 6200 | .early_init = 0, |
8cdea7c0 | 6201 | }; |
c077719b | 6202 | |
c255a458 | 6203 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6204 | static int __init enable_swap_account(char *s) |
6205 | { | |
a2c8990a | 6206 | if (!strcmp(s, "1")) |
a42c390c | 6207 | really_do_swap_account = 1; |
a2c8990a | 6208 | else if (!strcmp(s, "0")) |
a42c390c MH |
6209 | really_do_swap_account = 0; |
6210 | return 1; | |
6211 | } | |
a2c8990a | 6212 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6213 | |
2d11085e MH |
6214 | static void __init memsw_file_init(void) |
6215 | { | |
2cf669a5 TH |
6216 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6217 | memsw_cgroup_files)); | |
6acc8b02 MH |
6218 | } |
6219 | ||
6220 | static void __init enable_swap_cgroup(void) | |
6221 | { | |
6222 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6223 | do_swap_account = 1; | |
6224 | memsw_file_init(); | |
6225 | } | |
2d11085e | 6226 | } |
6acc8b02 | 6227 | |
2d11085e | 6228 | #else |
6acc8b02 | 6229 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6230 | { |
6231 | } | |
c077719b | 6232 | #endif |
2d11085e | 6233 | |
0a31bc97 JW |
6234 | #ifdef CONFIG_MEMCG_SWAP |
6235 | /** | |
6236 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6237 | * @page: page whose memsw charge to transfer | |
6238 | * @entry: swap entry to move the charge to | |
6239 | * | |
6240 | * Transfer the memsw charge of @page to @entry. | |
6241 | */ | |
6242 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6243 | { | |
6244 | struct page_cgroup *pc; | |
6245 | unsigned short oldid; | |
6246 | ||
6247 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6248 | VM_BUG_ON_PAGE(page_count(page), page); | |
6249 | ||
6250 | if (!do_swap_account) | |
6251 | return; | |
6252 | ||
6253 | pc = lookup_page_cgroup(page); | |
6254 | ||
6255 | /* Readahead page, never charged */ | |
6256 | if (!PageCgroupUsed(pc)) | |
6257 | return; | |
6258 | ||
6259 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEMSW), page); | |
6260 | ||
6261 | oldid = swap_cgroup_record(entry, mem_cgroup_id(pc->mem_cgroup)); | |
6262 | VM_BUG_ON_PAGE(oldid, page); | |
6263 | ||
6264 | pc->flags &= ~PCG_MEMSW; | |
6265 | css_get(&pc->mem_cgroup->css); | |
6266 | mem_cgroup_swap_statistics(pc->mem_cgroup, true); | |
6267 | } | |
6268 | ||
6269 | /** | |
6270 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
6271 | * @entry: swap entry to uncharge | |
6272 | * | |
6273 | * Drop the memsw charge associated with @entry. | |
6274 | */ | |
6275 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
6276 | { | |
6277 | struct mem_cgroup *memcg; | |
6278 | unsigned short id; | |
6279 | ||
6280 | if (!do_swap_account) | |
6281 | return; | |
6282 | ||
6283 | id = swap_cgroup_record(entry, 0); | |
6284 | rcu_read_lock(); | |
6285 | memcg = mem_cgroup_lookup(id); | |
6286 | if (memcg) { | |
ce00a967 JW |
6287 | if (!mem_cgroup_is_root(memcg)) |
6288 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
0a31bc97 JW |
6289 | mem_cgroup_swap_statistics(memcg, false); |
6290 | css_put(&memcg->css); | |
6291 | } | |
6292 | rcu_read_unlock(); | |
6293 | } | |
6294 | #endif | |
6295 | ||
00501b53 JW |
6296 | /** |
6297 | * mem_cgroup_try_charge - try charging a page | |
6298 | * @page: page to charge | |
6299 | * @mm: mm context of the victim | |
6300 | * @gfp_mask: reclaim mode | |
6301 | * @memcgp: charged memcg return | |
6302 | * | |
6303 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6304 | * pages according to @gfp_mask if necessary. | |
6305 | * | |
6306 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
6307 | * Otherwise, an error code is returned. | |
6308 | * | |
6309 | * After page->mapping has been set up, the caller must finalize the | |
6310 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
6311 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
6312 | */ | |
6313 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
6314 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
6315 | { | |
6316 | struct mem_cgroup *memcg = NULL; | |
6317 | unsigned int nr_pages = 1; | |
6318 | int ret = 0; | |
6319 | ||
6320 | if (mem_cgroup_disabled()) | |
6321 | goto out; | |
6322 | ||
6323 | if (PageSwapCache(page)) { | |
6324 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
6325 | /* | |
6326 | * Every swap fault against a single page tries to charge the | |
6327 | * page, bail as early as possible. shmem_unuse() encounters | |
6328 | * already charged pages, too. The USED bit is protected by | |
6329 | * the page lock, which serializes swap cache removal, which | |
6330 | * in turn serializes uncharging. | |
6331 | */ | |
6332 | if (PageCgroupUsed(pc)) | |
6333 | goto out; | |
6334 | } | |
6335 | ||
6336 | if (PageTransHuge(page)) { | |
6337 | nr_pages <<= compound_order(page); | |
6338 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6339 | } | |
6340 | ||
6341 | if (do_swap_account && PageSwapCache(page)) | |
6342 | memcg = try_get_mem_cgroup_from_page(page); | |
6343 | if (!memcg) | |
6344 | memcg = get_mem_cgroup_from_mm(mm); | |
6345 | ||
6346 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6347 | ||
6348 | css_put(&memcg->css); | |
6349 | ||
6350 | if (ret == -EINTR) { | |
6351 | memcg = root_mem_cgroup; | |
6352 | ret = 0; | |
6353 | } | |
6354 | out: | |
6355 | *memcgp = memcg; | |
6356 | return ret; | |
6357 | } | |
6358 | ||
6359 | /** | |
6360 | * mem_cgroup_commit_charge - commit a page charge | |
6361 | * @page: page to charge | |
6362 | * @memcg: memcg to charge the page to | |
6363 | * @lrucare: page might be on LRU already | |
6364 | * | |
6365 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6366 | * after page->mapping has been set up. This must happen atomically | |
6367 | * as part of the page instantiation, i.e. under the page table lock | |
6368 | * for anonymous pages, under the page lock for page and swap cache. | |
6369 | * | |
6370 | * In addition, the page must not be on the LRU during the commit, to | |
6371 | * prevent racing with task migration. If it might be, use @lrucare. | |
6372 | * | |
6373 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6374 | */ | |
6375 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
6376 | bool lrucare) | |
6377 | { | |
6378 | unsigned int nr_pages = 1; | |
6379 | ||
6380 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6381 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6382 | ||
6383 | if (mem_cgroup_disabled()) | |
6384 | return; | |
6385 | /* | |
6386 | * Swap faults will attempt to charge the same page multiple | |
6387 | * times. But reuse_swap_page() might have removed the page | |
6388 | * from swapcache already, so we can't check PageSwapCache(). | |
6389 | */ | |
6390 | if (!memcg) | |
6391 | return; | |
6392 | ||
6abb5a86 JW |
6393 | commit_charge(page, memcg, lrucare); |
6394 | ||
00501b53 JW |
6395 | if (PageTransHuge(page)) { |
6396 | nr_pages <<= compound_order(page); | |
6397 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6398 | } | |
6399 | ||
6abb5a86 JW |
6400 | local_irq_disable(); |
6401 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
6402 | memcg_check_events(memcg, page); | |
6403 | local_irq_enable(); | |
00501b53 JW |
6404 | |
6405 | if (do_swap_account && PageSwapCache(page)) { | |
6406 | swp_entry_t entry = { .val = page_private(page) }; | |
6407 | /* | |
6408 | * The swap entry might not get freed for a long time, | |
6409 | * let's not wait for it. The page already received a | |
6410 | * memory+swap charge, drop the swap entry duplicate. | |
6411 | */ | |
6412 | mem_cgroup_uncharge_swap(entry); | |
6413 | } | |
6414 | } | |
6415 | ||
6416 | /** | |
6417 | * mem_cgroup_cancel_charge - cancel a page charge | |
6418 | * @page: page to charge | |
6419 | * @memcg: memcg to charge the page to | |
6420 | * | |
6421 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6422 | */ | |
6423 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
6424 | { | |
6425 | unsigned int nr_pages = 1; | |
6426 | ||
6427 | if (mem_cgroup_disabled()) | |
6428 | return; | |
6429 | /* | |
6430 | * Swap faults will attempt to charge the same page multiple | |
6431 | * times. But reuse_swap_page() might have removed the page | |
6432 | * from swapcache already, so we can't check PageSwapCache(). | |
6433 | */ | |
6434 | if (!memcg) | |
6435 | return; | |
6436 | ||
6437 | if (PageTransHuge(page)) { | |
6438 | nr_pages <<= compound_order(page); | |
6439 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6440 | } | |
6441 | ||
6442 | cancel_charge(memcg, nr_pages); | |
6443 | } | |
6444 | ||
747db954 JW |
6445 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
6446 | unsigned long nr_mem, unsigned long nr_memsw, | |
6447 | unsigned long nr_anon, unsigned long nr_file, | |
6448 | unsigned long nr_huge, struct page *dummy_page) | |
6449 | { | |
6450 | unsigned long flags; | |
6451 | ||
ce00a967 JW |
6452 | if (!mem_cgroup_is_root(memcg)) { |
6453 | if (nr_mem) | |
6454 | res_counter_uncharge(&memcg->res, | |
6455 | nr_mem * PAGE_SIZE); | |
6456 | if (nr_memsw) | |
6457 | res_counter_uncharge(&memcg->memsw, | |
6458 | nr_memsw * PAGE_SIZE); | |
6459 | memcg_oom_recover(memcg); | |
6460 | } | |
747db954 JW |
6461 | |
6462 | local_irq_save(flags); | |
6463 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
6464 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
6465 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
6466 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
6467 | __this_cpu_add(memcg->stat->nr_page_events, nr_anon + nr_file); | |
6468 | memcg_check_events(memcg, dummy_page); | |
6469 | local_irq_restore(flags); | |
6470 | } | |
6471 | ||
6472 | static void uncharge_list(struct list_head *page_list) | |
6473 | { | |
6474 | struct mem_cgroup *memcg = NULL; | |
6475 | unsigned long nr_memsw = 0; | |
6476 | unsigned long nr_anon = 0; | |
6477 | unsigned long nr_file = 0; | |
6478 | unsigned long nr_huge = 0; | |
6479 | unsigned long pgpgout = 0; | |
6480 | unsigned long nr_mem = 0; | |
6481 | struct list_head *next; | |
6482 | struct page *page; | |
6483 | ||
6484 | next = page_list->next; | |
6485 | do { | |
6486 | unsigned int nr_pages = 1; | |
6487 | struct page_cgroup *pc; | |
6488 | ||
6489 | page = list_entry(next, struct page, lru); | |
6490 | next = page->lru.next; | |
6491 | ||
6492 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6493 | VM_BUG_ON_PAGE(page_count(page), page); | |
6494 | ||
6495 | pc = lookup_page_cgroup(page); | |
6496 | if (!PageCgroupUsed(pc)) | |
6497 | continue; | |
6498 | ||
6499 | /* | |
6500 | * Nobody should be changing or seriously looking at | |
6501 | * pc->mem_cgroup and pc->flags at this point, we have | |
6502 | * fully exclusive access to the page. | |
6503 | */ | |
6504 | ||
6505 | if (memcg != pc->mem_cgroup) { | |
6506 | if (memcg) { | |
6507 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6508 | nr_anon, nr_file, nr_huge, page); | |
6509 | pgpgout = nr_mem = nr_memsw = 0; | |
6510 | nr_anon = nr_file = nr_huge = 0; | |
6511 | } | |
6512 | memcg = pc->mem_cgroup; | |
6513 | } | |
6514 | ||
6515 | if (PageTransHuge(page)) { | |
6516 | nr_pages <<= compound_order(page); | |
6517 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6518 | nr_huge += nr_pages; | |
6519 | } | |
6520 | ||
6521 | if (PageAnon(page)) | |
6522 | nr_anon += nr_pages; | |
6523 | else | |
6524 | nr_file += nr_pages; | |
6525 | ||
6526 | if (pc->flags & PCG_MEM) | |
6527 | nr_mem += nr_pages; | |
6528 | if (pc->flags & PCG_MEMSW) | |
6529 | nr_memsw += nr_pages; | |
6530 | pc->flags = 0; | |
6531 | ||
6532 | pgpgout++; | |
6533 | } while (next != page_list); | |
6534 | ||
6535 | if (memcg) | |
6536 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6537 | nr_anon, nr_file, nr_huge, page); | |
6538 | } | |
6539 | ||
0a31bc97 JW |
6540 | /** |
6541 | * mem_cgroup_uncharge - uncharge a page | |
6542 | * @page: page to uncharge | |
6543 | * | |
6544 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6545 | * mem_cgroup_commit_charge(). | |
6546 | */ | |
6547 | void mem_cgroup_uncharge(struct page *page) | |
6548 | { | |
0a31bc97 | 6549 | struct page_cgroup *pc; |
0a31bc97 JW |
6550 | |
6551 | if (mem_cgroup_disabled()) | |
6552 | return; | |
6553 | ||
747db954 | 6554 | /* Don't touch page->lru of any random page, pre-check: */ |
0a31bc97 | 6555 | pc = lookup_page_cgroup(page); |
0a31bc97 JW |
6556 | if (!PageCgroupUsed(pc)) |
6557 | return; | |
6558 | ||
747db954 JW |
6559 | INIT_LIST_HEAD(&page->lru); |
6560 | uncharge_list(&page->lru); | |
6561 | } | |
0a31bc97 | 6562 | |
747db954 JW |
6563 | /** |
6564 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6565 | * @page_list: list of pages to uncharge | |
6566 | * | |
6567 | * Uncharge a list of pages previously charged with | |
6568 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6569 | */ | |
6570 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6571 | { | |
6572 | if (mem_cgroup_disabled()) | |
6573 | return; | |
0a31bc97 | 6574 | |
747db954 JW |
6575 | if (!list_empty(page_list)) |
6576 | uncharge_list(page_list); | |
0a31bc97 JW |
6577 | } |
6578 | ||
6579 | /** | |
6580 | * mem_cgroup_migrate - migrate a charge to another page | |
6581 | * @oldpage: currently charged page | |
6582 | * @newpage: page to transfer the charge to | |
6583 | * @lrucare: both pages might be on the LRU already | |
6584 | * | |
6585 | * Migrate the charge from @oldpage to @newpage. | |
6586 | * | |
6587 | * Both pages must be locked, @newpage->mapping must be set up. | |
6588 | */ | |
6589 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
6590 | bool lrucare) | |
6591 | { | |
0a31bc97 JW |
6592 | struct page_cgroup *pc; |
6593 | int isolated; | |
6594 | ||
6595 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6596 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
6597 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
6598 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
6599 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
6600 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6601 | newpage); | |
0a31bc97 JW |
6602 | |
6603 | if (mem_cgroup_disabled()) | |
6604 | return; | |
6605 | ||
6606 | /* Page cache replacement: new page already charged? */ | |
6607 | pc = lookup_page_cgroup(newpage); | |
6608 | if (PageCgroupUsed(pc)) | |
6609 | return; | |
6610 | ||
6611 | /* Re-entrant migration: old page already uncharged? */ | |
6612 | pc = lookup_page_cgroup(oldpage); | |
6613 | if (!PageCgroupUsed(pc)) | |
6614 | return; | |
6615 | ||
6616 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEM), oldpage); | |
6617 | VM_BUG_ON_PAGE(do_swap_account && !(pc->flags & PCG_MEMSW), oldpage); | |
6618 | ||
0a31bc97 JW |
6619 | if (lrucare) |
6620 | lock_page_lru(oldpage, &isolated); | |
6621 | ||
6622 | pc->flags = 0; | |
6623 | ||
6624 | if (lrucare) | |
6625 | unlock_page_lru(oldpage, isolated); | |
6626 | ||
6abb5a86 | 6627 | commit_charge(newpage, pc->mem_cgroup, lrucare); |
0a31bc97 JW |
6628 | } |
6629 | ||
2d11085e | 6630 | /* |
1081312f MH |
6631 | * subsys_initcall() for memory controller. |
6632 | * | |
6633 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6634 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6635 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6636 | * be initialized from here. | |
2d11085e MH |
6637 | */ |
6638 | static int __init mem_cgroup_init(void) | |
6639 | { | |
6640 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6641 | enable_swap_cgroup(); |
bb4cc1a8 | 6642 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6643 | memcg_stock_init(); |
2d11085e MH |
6644 | return 0; |
6645 | } | |
6646 | subsys_initcall(mem_cgroup_init); |