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 | ||
465939a1 LZ |
295 | /* |
296 | * the counter to account for mem+swap usage. | |
297 | */ | |
298 | struct res_counter memsw; | |
59927fb9 | 299 | |
510fc4e1 GC |
300 | /* |
301 | * the counter to account for kernel memory usage. | |
302 | */ | |
303 | struct res_counter kmem; | |
18f59ea7 BS |
304 | /* |
305 | * Should the accounting and control be hierarchical, per subtree? | |
306 | */ | |
307 | bool use_hierarchy; | |
510fc4e1 | 308 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
309 | |
310 | bool oom_lock; | |
311 | atomic_t under_oom; | |
3812c8c8 | 312 | atomic_t oom_wakeups; |
79dfdacc | 313 | |
1f4c025b | 314 | int swappiness; |
3c11ecf4 KH |
315 | /* OOM-Killer disable */ |
316 | int oom_kill_disable; | |
a7885eb8 | 317 | |
22a668d7 KH |
318 | /* set when res.limit == memsw.limit */ |
319 | bool memsw_is_minimum; | |
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 | |
75822b44 BS |
483 | /* |
484 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
485 | */ | |
486 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
487 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
488 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
489 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
490 | ||
0999821b GC |
491 | /* |
492 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
493 | * As a consequence, any change that needs to protect against new child cgroups | |
494 | * appearing has to hold it as well. | |
495 | */ | |
496 | static DEFINE_MUTEX(memcg_create_mutex); | |
497 | ||
b2145145 WL |
498 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
499 | { | |
a7c6d554 | 500 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
501 | } |
502 | ||
70ddf637 AV |
503 | /* Some nice accessors for the vmpressure. */ |
504 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
505 | { | |
506 | if (!memcg) | |
507 | memcg = root_mem_cgroup; | |
508 | return &memcg->vmpressure; | |
509 | } | |
510 | ||
511 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
512 | { | |
513 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
514 | } | |
515 | ||
7ffc0edc MH |
516 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
517 | { | |
518 | return (memcg == root_mem_cgroup); | |
519 | } | |
520 | ||
4219b2da LZ |
521 | /* |
522 | * We restrict the id in the range of [1, 65535], so it can fit into | |
523 | * an unsigned short. | |
524 | */ | |
525 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
526 | ||
34c00c31 LZ |
527 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
528 | { | |
15a4c835 | 529 | return memcg->css.id; |
34c00c31 LZ |
530 | } |
531 | ||
532 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
533 | { | |
534 | struct cgroup_subsys_state *css; | |
535 | ||
7d699ddb | 536 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
537 | return mem_cgroup_from_css(css); |
538 | } | |
539 | ||
e1aab161 | 540 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 541 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 542 | |
e1aab161 GC |
543 | void sock_update_memcg(struct sock *sk) |
544 | { | |
376be5ff | 545 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 546 | struct mem_cgroup *memcg; |
3f134619 | 547 | struct cg_proto *cg_proto; |
e1aab161 GC |
548 | |
549 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
550 | ||
f3f511e1 GC |
551 | /* Socket cloning can throw us here with sk_cgrp already |
552 | * filled. It won't however, necessarily happen from | |
553 | * process context. So the test for root memcg given | |
554 | * the current task's memcg won't help us in this case. | |
555 | * | |
556 | * Respecting the original socket's memcg is a better | |
557 | * decision in this case. | |
558 | */ | |
559 | if (sk->sk_cgrp) { | |
560 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 561 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
562 | return; |
563 | } | |
564 | ||
e1aab161 GC |
565 | rcu_read_lock(); |
566 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 567 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 568 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
569 | memcg_proto_active(cg_proto) && |
570 | css_tryget_online(&memcg->css)) { | |
3f134619 | 571 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
572 | } |
573 | rcu_read_unlock(); | |
574 | } | |
575 | } | |
576 | EXPORT_SYMBOL(sock_update_memcg); | |
577 | ||
578 | void sock_release_memcg(struct sock *sk) | |
579 | { | |
376be5ff | 580 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
581 | struct mem_cgroup *memcg; |
582 | WARN_ON(!sk->sk_cgrp->memcg); | |
583 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 584 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
585 | } |
586 | } | |
d1a4c0b3 GC |
587 | |
588 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
589 | { | |
590 | if (!memcg || mem_cgroup_is_root(memcg)) | |
591 | return NULL; | |
592 | ||
2e685cad | 593 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
594 | } |
595 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 596 | |
3f134619 GC |
597 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
598 | { | |
2e685cad | 599 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
600 | return; |
601 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
602 | } | |
603 | #else | |
604 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
605 | { | |
606 | } | |
607 | #endif | |
608 | ||
a8964b9b | 609 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
610 | /* |
611 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
b8627835 LZ |
612 | * The main reason for not using cgroup id for this: |
613 | * this works better in sparse environments, where we have a lot of memcgs, | |
614 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
615 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
616 | * 200 entry array for that. | |
55007d84 GC |
617 | * |
618 | * The current size of the caches array is stored in | |
619 | * memcg_limited_groups_array_size. It will double each time we have to | |
620 | * increase it. | |
621 | */ | |
622 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
623 | int memcg_limited_groups_array_size; |
624 | ||
55007d84 GC |
625 | /* |
626 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
627 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
628 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
629 | * tunable, but that is strictly not necessary. | |
630 | * | |
b8627835 | 631 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
632 | * this constant directly from cgroup, but it is understandable that this is |
633 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 634 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
635 | * increase ours as well if it increases. |
636 | */ | |
637 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 638 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 639 | |
d7f25f8a GC |
640 | /* |
641 | * A lot of the calls to the cache allocation functions are expected to be | |
642 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
643 | * conditional to this static branch, we'll have to allow modules that does | |
644 | * kmem_cache_alloc and the such to see this symbol as well | |
645 | */ | |
a8964b9b | 646 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 647 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
648 | |
649 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
650 | { | |
55007d84 | 651 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 652 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
653 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
654 | } | |
bea207c8 GC |
655 | /* |
656 | * This check can't live in kmem destruction function, | |
657 | * since the charges will outlive the cgroup | |
658 | */ | |
659 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
660 | } |
661 | #else | |
662 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
663 | { | |
664 | } | |
665 | #endif /* CONFIG_MEMCG_KMEM */ | |
666 | ||
667 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
668 | { | |
669 | disarm_sock_keys(memcg); | |
670 | disarm_kmem_keys(memcg); | |
671 | } | |
672 | ||
c0ff4b85 | 673 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 674 | |
f64c3f54 | 675 | static struct mem_cgroup_per_zone * |
e231875b | 676 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 677 | { |
e231875b JZ |
678 | int nid = zone_to_nid(zone); |
679 | int zid = zone_idx(zone); | |
680 | ||
54f72fe0 | 681 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
682 | } |
683 | ||
c0ff4b85 | 684 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 685 | { |
c0ff4b85 | 686 | return &memcg->css; |
d324236b WF |
687 | } |
688 | ||
f64c3f54 | 689 | static struct mem_cgroup_per_zone * |
e231875b | 690 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 691 | { |
97a6c37b JW |
692 | int nid = page_to_nid(page); |
693 | int zid = page_zonenum(page); | |
f64c3f54 | 694 | |
e231875b | 695 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
696 | } |
697 | ||
bb4cc1a8 AM |
698 | static struct mem_cgroup_tree_per_zone * |
699 | soft_limit_tree_node_zone(int nid, int zid) | |
700 | { | |
701 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
702 | } | |
703 | ||
704 | static struct mem_cgroup_tree_per_zone * | |
705 | soft_limit_tree_from_page(struct page *page) | |
706 | { | |
707 | int nid = page_to_nid(page); | |
708 | int zid = page_zonenum(page); | |
709 | ||
710 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
711 | } | |
712 | ||
cf2c8127 JW |
713 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
714 | struct mem_cgroup_tree_per_zone *mctz, | |
715 | unsigned long long new_usage_in_excess) | |
bb4cc1a8 AM |
716 | { |
717 | struct rb_node **p = &mctz->rb_root.rb_node; | |
718 | struct rb_node *parent = NULL; | |
719 | struct mem_cgroup_per_zone *mz_node; | |
720 | ||
721 | if (mz->on_tree) | |
722 | return; | |
723 | ||
724 | mz->usage_in_excess = new_usage_in_excess; | |
725 | if (!mz->usage_in_excess) | |
726 | return; | |
727 | while (*p) { | |
728 | parent = *p; | |
729 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
730 | tree_node); | |
731 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
732 | p = &(*p)->rb_left; | |
733 | /* | |
734 | * We can't avoid mem cgroups that are over their soft | |
735 | * limit by the same amount | |
736 | */ | |
737 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
738 | p = &(*p)->rb_right; | |
739 | } | |
740 | rb_link_node(&mz->tree_node, parent, p); | |
741 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
742 | mz->on_tree = true; | |
743 | } | |
744 | ||
cf2c8127 JW |
745 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
746 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
747 | { |
748 | if (!mz->on_tree) | |
749 | return; | |
750 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
751 | mz->on_tree = false; | |
752 | } | |
753 | ||
cf2c8127 JW |
754 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
755 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 756 | { |
0a31bc97 JW |
757 | unsigned long flags; |
758 | ||
759 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 760 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 761 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
762 | } |
763 | ||
764 | ||
765 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
766 | { | |
767 | unsigned long long excess; | |
768 | struct mem_cgroup_per_zone *mz; | |
769 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 770 | |
e231875b | 771 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
772 | /* |
773 | * Necessary to update all ancestors when hierarchy is used. | |
774 | * because their event counter is not touched. | |
775 | */ | |
776 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 777 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bb4cc1a8 AM |
778 | excess = res_counter_soft_limit_excess(&memcg->res); |
779 | /* | |
780 | * We have to update the tree if mz is on RB-tree or | |
781 | * mem is over its softlimit. | |
782 | */ | |
783 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
784 | unsigned long flags; |
785 | ||
786 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
787 | /* if on-tree, remove it */ |
788 | if (mz->on_tree) | |
cf2c8127 | 789 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
790 | /* |
791 | * Insert again. mz->usage_in_excess will be updated. | |
792 | * If excess is 0, no tree ops. | |
793 | */ | |
cf2c8127 | 794 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 795 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
796 | } |
797 | } | |
798 | } | |
799 | ||
800 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
801 | { | |
bb4cc1a8 | 802 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
803 | struct mem_cgroup_per_zone *mz; |
804 | int nid, zid; | |
bb4cc1a8 | 805 | |
e231875b JZ |
806 | for_each_node(nid) { |
807 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
808 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
809 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 810 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
811 | } |
812 | } | |
813 | } | |
814 | ||
815 | static struct mem_cgroup_per_zone * | |
816 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
817 | { | |
818 | struct rb_node *rightmost = NULL; | |
819 | struct mem_cgroup_per_zone *mz; | |
820 | ||
821 | retry: | |
822 | mz = NULL; | |
823 | rightmost = rb_last(&mctz->rb_root); | |
824 | if (!rightmost) | |
825 | goto done; /* Nothing to reclaim from */ | |
826 | ||
827 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
828 | /* | |
829 | * Remove the node now but someone else can add it back, | |
830 | * we will to add it back at the end of reclaim to its correct | |
831 | * position in the tree. | |
832 | */ | |
cf2c8127 | 833 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 | 834 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || |
ec903c0c | 835 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
836 | goto retry; |
837 | done: | |
838 | return mz; | |
839 | } | |
840 | ||
841 | static struct mem_cgroup_per_zone * | |
842 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
843 | { | |
844 | struct mem_cgroup_per_zone *mz; | |
845 | ||
0a31bc97 | 846 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 847 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 848 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
849 | return mz; |
850 | } | |
851 | ||
711d3d2c KH |
852 | /* |
853 | * Implementation Note: reading percpu statistics for memcg. | |
854 | * | |
855 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
856 | * synchronization to implement "quick" read. There are trade-off between | |
857 | * reading cost and precision of value. Then, we may have a chance to implement | |
858 | * a periodic synchronizion of counter in memcg's counter. | |
859 | * | |
860 | * But this _read() function is used for user interface now. The user accounts | |
861 | * memory usage by memory cgroup and he _always_ requires exact value because | |
862 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
863 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
864 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
865 | * | |
866 | * If there are kernel internal actions which can make use of some not-exact | |
867 | * value, and reading all cpu value can be performance bottleneck in some | |
868 | * common workload, threashold and synchonization as vmstat[] should be | |
869 | * implemented. | |
870 | */ | |
c0ff4b85 | 871 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 872 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 873 | { |
7a159cc9 | 874 | long val = 0; |
c62b1a3b | 875 | int cpu; |
c62b1a3b | 876 | |
711d3d2c KH |
877 | get_online_cpus(); |
878 | for_each_online_cpu(cpu) | |
c0ff4b85 | 879 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 880 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
881 | spin_lock(&memcg->pcp_counter_lock); |
882 | val += memcg->nocpu_base.count[idx]; | |
883 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
884 | #endif |
885 | put_online_cpus(); | |
c62b1a3b KH |
886 | return val; |
887 | } | |
888 | ||
c0ff4b85 | 889 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
890 | enum mem_cgroup_events_index idx) |
891 | { | |
892 | unsigned long val = 0; | |
893 | int cpu; | |
894 | ||
9c567512 | 895 | get_online_cpus(); |
e9f8974f | 896 | for_each_online_cpu(cpu) |
c0ff4b85 | 897 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 898 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
899 | spin_lock(&memcg->pcp_counter_lock); |
900 | val += memcg->nocpu_base.events[idx]; | |
901 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 902 | #endif |
9c567512 | 903 | put_online_cpus(); |
e9f8974f JW |
904 | return val; |
905 | } | |
906 | ||
c0ff4b85 | 907 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 908 | struct page *page, |
0a31bc97 | 909 | int nr_pages) |
d52aa412 | 910 | { |
b2402857 KH |
911 | /* |
912 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
913 | * counted as CACHE even if it's on ANON LRU. | |
914 | */ | |
0a31bc97 | 915 | if (PageAnon(page)) |
b2402857 | 916 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 917 | nr_pages); |
d52aa412 | 918 | else |
b2402857 | 919 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 920 | nr_pages); |
55e462b0 | 921 | |
b070e65c DR |
922 | if (PageTransHuge(page)) |
923 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
924 | nr_pages); | |
925 | ||
e401f176 KH |
926 | /* pagein of a big page is an event. So, ignore page size */ |
927 | if (nr_pages > 0) | |
c0ff4b85 | 928 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 929 | else { |
c0ff4b85 | 930 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
931 | nr_pages = -nr_pages; /* for event */ |
932 | } | |
e401f176 | 933 | |
13114716 | 934 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
935 | } |
936 | ||
e231875b | 937 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
938 | { |
939 | struct mem_cgroup_per_zone *mz; | |
940 | ||
941 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
942 | return mz->lru_size[lru]; | |
943 | } | |
944 | ||
e231875b JZ |
945 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
946 | int nid, | |
947 | unsigned int lru_mask) | |
bb2a0de9 | 948 | { |
e231875b | 949 | unsigned long nr = 0; |
889976db YH |
950 | int zid; |
951 | ||
e231875b | 952 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 953 | |
e231875b JZ |
954 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
955 | struct mem_cgroup_per_zone *mz; | |
956 | enum lru_list lru; | |
957 | ||
958 | for_each_lru(lru) { | |
959 | if (!(BIT(lru) & lru_mask)) | |
960 | continue; | |
961 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
962 | nr += mz->lru_size[lru]; | |
963 | } | |
964 | } | |
965 | return nr; | |
889976db | 966 | } |
bb2a0de9 | 967 | |
c0ff4b85 | 968 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 969 | unsigned int lru_mask) |
6d12e2d8 | 970 | { |
e231875b | 971 | unsigned long nr = 0; |
889976db | 972 | int nid; |
6d12e2d8 | 973 | |
31aaea4a | 974 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
975 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
976 | return nr; | |
d52aa412 KH |
977 | } |
978 | ||
f53d7ce3 JW |
979 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
980 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
981 | { |
982 | unsigned long val, next; | |
983 | ||
13114716 | 984 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 985 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 986 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
987 | if ((long)next - (long)val < 0) { |
988 | switch (target) { | |
989 | case MEM_CGROUP_TARGET_THRESH: | |
990 | next = val + THRESHOLDS_EVENTS_TARGET; | |
991 | break; | |
bb4cc1a8 AM |
992 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
993 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
994 | break; | |
f53d7ce3 JW |
995 | case MEM_CGROUP_TARGET_NUMAINFO: |
996 | next = val + NUMAINFO_EVENTS_TARGET; | |
997 | break; | |
998 | default: | |
999 | break; | |
1000 | } | |
1001 | __this_cpu_write(memcg->stat->targets[target], next); | |
1002 | return true; | |
7a159cc9 | 1003 | } |
f53d7ce3 | 1004 | return false; |
d2265e6f KH |
1005 | } |
1006 | ||
1007 | /* | |
1008 | * Check events in order. | |
1009 | * | |
1010 | */ | |
c0ff4b85 | 1011 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
1012 | { |
1013 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
1014 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1015 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1016 | bool do_softlimit; |
82b3f2a7 | 1017 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1018 | |
bb4cc1a8 AM |
1019 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1020 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1021 | #if MAX_NUMNODES > 1 |
1022 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1023 | MEM_CGROUP_TARGET_NUMAINFO); | |
1024 | #endif | |
c0ff4b85 | 1025 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1026 | if (unlikely(do_softlimit)) |
1027 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1028 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1029 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1030 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1031 | #endif |
0a31bc97 | 1032 | } |
d2265e6f KH |
1033 | } |
1034 | ||
cf475ad2 | 1035 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1036 | { |
31a78f23 BS |
1037 | /* |
1038 | * mm_update_next_owner() may clear mm->owner to NULL | |
1039 | * if it races with swapoff, page migration, etc. | |
1040 | * So this can be called with p == NULL. | |
1041 | */ | |
1042 | if (unlikely(!p)) | |
1043 | return NULL; | |
1044 | ||
073219e9 | 1045 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1046 | } |
1047 | ||
df381975 | 1048 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1049 | { |
c0ff4b85 | 1050 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 1051 | |
54595fe2 KH |
1052 | rcu_read_lock(); |
1053 | do { | |
6f6acb00 MH |
1054 | /* |
1055 | * Page cache insertions can happen withou an | |
1056 | * actual mm context, e.g. during disk probing | |
1057 | * on boot, loopback IO, acct() writes etc. | |
1058 | */ | |
1059 | if (unlikely(!mm)) | |
df381975 | 1060 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1061 | else { |
1062 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1063 | if (unlikely(!memcg)) | |
1064 | memcg = root_mem_cgroup; | |
1065 | } | |
ec903c0c | 1066 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1067 | rcu_read_unlock(); |
c0ff4b85 | 1068 | return memcg; |
54595fe2 KH |
1069 | } |
1070 | ||
16248d8f MH |
1071 | /* |
1072 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1073 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1074 | * | |
1075 | * helper function to be used by mem_cgroup_iter | |
1076 | */ | |
1077 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
694fbc0f | 1078 | struct mem_cgroup *last_visited) |
16248d8f | 1079 | { |
492eb21b | 1080 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 1081 | |
bd8815a6 | 1082 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 1083 | skip_node: |
492eb21b | 1084 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
1085 | |
1086 | /* | |
1087 | * Even if we found a group we have to make sure it is | |
1088 | * alive. css && !memcg means that the groups should be | |
1089 | * skipped and we should continue the tree walk. | |
1090 | * last_visited css is safe to use because it is | |
1091 | * protected by css_get and the tree walk is rcu safe. | |
0eef6156 MH |
1092 | * |
1093 | * We do not take a reference on the root of the tree walk | |
1094 | * because we might race with the root removal when it would | |
1095 | * be the only node in the iterated hierarchy and mem_cgroup_iter | |
1096 | * would end up in an endless loop because it expects that at | |
1097 | * least one valid node will be returned. Root cannot disappear | |
1098 | * because caller of the iterator should hold it already so | |
1099 | * skipping css reference should be safe. | |
16248d8f | 1100 | */ |
492eb21b | 1101 | if (next_css) { |
ce48225f | 1102 | if ((next_css == &root->css) || |
ec903c0c TH |
1103 | ((next_css->flags & CSS_ONLINE) && |
1104 | css_tryget_online(next_css))) | |
d8ad3055 | 1105 | return mem_cgroup_from_css(next_css); |
0eef6156 MH |
1106 | |
1107 | prev_css = next_css; | |
1108 | goto skip_node; | |
16248d8f MH |
1109 | } |
1110 | ||
1111 | return NULL; | |
1112 | } | |
1113 | ||
519ebea3 JW |
1114 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1115 | { | |
1116 | /* | |
1117 | * When a group in the hierarchy below root is destroyed, the | |
1118 | * hierarchy iterator can no longer be trusted since it might | |
1119 | * have pointed to the destroyed group. Invalidate it. | |
1120 | */ | |
1121 | atomic_inc(&root->dead_count); | |
1122 | } | |
1123 | ||
1124 | static struct mem_cgroup * | |
1125 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1126 | struct mem_cgroup *root, | |
1127 | int *sequence) | |
1128 | { | |
1129 | struct mem_cgroup *position = NULL; | |
1130 | /* | |
1131 | * A cgroup destruction happens in two stages: offlining and | |
1132 | * release. They are separated by a RCU grace period. | |
1133 | * | |
1134 | * If the iterator is valid, we may still race with an | |
1135 | * offlining. The RCU lock ensures the object won't be | |
1136 | * released, tryget will fail if we lost the race. | |
1137 | */ | |
1138 | *sequence = atomic_read(&root->dead_count); | |
1139 | if (iter->last_dead_count == *sequence) { | |
1140 | smp_rmb(); | |
1141 | position = iter->last_visited; | |
ecc736fc MH |
1142 | |
1143 | /* | |
1144 | * We cannot take a reference to root because we might race | |
1145 | * with root removal and returning NULL would end up in | |
1146 | * an endless loop on the iterator user level when root | |
1147 | * would be returned all the time. | |
1148 | */ | |
1149 | if (position && position != root && | |
ec903c0c | 1150 | !css_tryget_online(&position->css)) |
519ebea3 JW |
1151 | position = NULL; |
1152 | } | |
1153 | return position; | |
1154 | } | |
1155 | ||
1156 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1157 | struct mem_cgroup *last_visited, | |
1158 | struct mem_cgroup *new_position, | |
ecc736fc | 1159 | struct mem_cgroup *root, |
519ebea3 JW |
1160 | int sequence) |
1161 | { | |
ecc736fc MH |
1162 | /* root reference counting symmetric to mem_cgroup_iter_load */ |
1163 | if (last_visited && last_visited != root) | |
519ebea3 JW |
1164 | css_put(&last_visited->css); |
1165 | /* | |
1166 | * We store the sequence count from the time @last_visited was | |
1167 | * loaded successfully instead of rereading it here so that we | |
1168 | * don't lose destruction events in between. We could have | |
1169 | * raced with the destruction of @new_position after all. | |
1170 | */ | |
1171 | iter->last_visited = new_position; | |
1172 | smp_wmb(); | |
1173 | iter->last_dead_count = sequence; | |
1174 | } | |
1175 | ||
5660048c JW |
1176 | /** |
1177 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1178 | * @root: hierarchy root | |
1179 | * @prev: previously returned memcg, NULL on first invocation | |
1180 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1181 | * | |
1182 | * Returns references to children of the hierarchy below @root, or | |
1183 | * @root itself, or %NULL after a full round-trip. | |
1184 | * | |
1185 | * Caller must pass the return value in @prev on subsequent | |
1186 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1187 | * to cancel a hierarchy walk before the round-trip is complete. | |
1188 | * | |
1189 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1190 | * divide up the memcgs in the hierarchy among all concurrent | |
1191 | * reclaimers operating on the same zone and priority. | |
1192 | */ | |
694fbc0f | 1193 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1194 | struct mem_cgroup *prev, |
694fbc0f | 1195 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1196 | { |
9f3a0d09 | 1197 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1198 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1199 | |
694fbc0f AM |
1200 | if (mem_cgroup_disabled()) |
1201 | return NULL; | |
5660048c | 1202 | |
9f3a0d09 JW |
1203 | if (!root) |
1204 | root = root_mem_cgroup; | |
7d74b06f | 1205 | |
9f3a0d09 | 1206 | if (prev && !reclaim) |
542f85f9 | 1207 | last_visited = prev; |
14067bb3 | 1208 | |
9f3a0d09 JW |
1209 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1210 | if (prev) | |
c40046f3 | 1211 | goto out_css_put; |
694fbc0f | 1212 | return root; |
9f3a0d09 | 1213 | } |
14067bb3 | 1214 | |
542f85f9 | 1215 | rcu_read_lock(); |
9f3a0d09 | 1216 | while (!memcg) { |
527a5ec9 | 1217 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1218 | int uninitialized_var(seq); |
711d3d2c | 1219 | |
527a5ec9 | 1220 | if (reclaim) { |
527a5ec9 JW |
1221 | struct mem_cgroup_per_zone *mz; |
1222 | ||
e231875b | 1223 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); |
527a5ec9 | 1224 | iter = &mz->reclaim_iter[reclaim->priority]; |
542f85f9 | 1225 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1226 | iter->last_visited = NULL; |
542f85f9 MH |
1227 | goto out_unlock; |
1228 | } | |
5f578161 | 1229 | |
519ebea3 | 1230 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1231 | } |
7d74b06f | 1232 | |
694fbc0f | 1233 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1234 | |
527a5ec9 | 1235 | if (reclaim) { |
ecc736fc MH |
1236 | mem_cgroup_iter_update(iter, last_visited, memcg, root, |
1237 | seq); | |
542f85f9 | 1238 | |
19f39402 | 1239 | if (!memcg) |
527a5ec9 JW |
1240 | iter->generation++; |
1241 | else if (!prev && memcg) | |
1242 | reclaim->generation = iter->generation; | |
1243 | } | |
9f3a0d09 | 1244 | |
694fbc0f | 1245 | if (prev && !memcg) |
542f85f9 | 1246 | goto out_unlock; |
9f3a0d09 | 1247 | } |
542f85f9 MH |
1248 | out_unlock: |
1249 | rcu_read_unlock(); | |
c40046f3 MH |
1250 | out_css_put: |
1251 | if (prev && prev != root) | |
1252 | css_put(&prev->css); | |
1253 | ||
9f3a0d09 | 1254 | return memcg; |
14067bb3 | 1255 | } |
7d74b06f | 1256 | |
5660048c JW |
1257 | /** |
1258 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1259 | * @root: hierarchy root | |
1260 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1261 | */ | |
1262 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1263 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1264 | { |
1265 | if (!root) | |
1266 | root = root_mem_cgroup; | |
1267 | if (prev && prev != root) | |
1268 | css_put(&prev->css); | |
1269 | } | |
7d74b06f | 1270 | |
9f3a0d09 JW |
1271 | /* |
1272 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1273 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1274 | * be used for reference counting. | |
1275 | */ | |
1276 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1277 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1278 | iter != NULL; \ |
527a5ec9 | 1279 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1280 | |
9f3a0d09 | 1281 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1282 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1283 | iter != NULL; \ |
527a5ec9 | 1284 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1285 | |
68ae564b | 1286 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1287 | { |
c0ff4b85 | 1288 | struct mem_cgroup *memcg; |
456f998e | 1289 | |
456f998e | 1290 | rcu_read_lock(); |
c0ff4b85 R |
1291 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1292 | if (unlikely(!memcg)) | |
456f998e YH |
1293 | goto out; |
1294 | ||
1295 | switch (idx) { | |
456f998e | 1296 | case PGFAULT: |
0e574a93 JW |
1297 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1298 | break; | |
1299 | case PGMAJFAULT: | |
1300 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1301 | break; |
1302 | default: | |
1303 | BUG(); | |
1304 | } | |
1305 | out: | |
1306 | rcu_read_unlock(); | |
1307 | } | |
68ae564b | 1308 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1309 | |
925b7673 JW |
1310 | /** |
1311 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1312 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1313 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1314 | * |
1315 | * Returns the lru list vector holding pages for the given @zone and | |
1316 | * @mem. This can be the global zone lruvec, if the memory controller | |
1317 | * is disabled. | |
1318 | */ | |
1319 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1320 | struct mem_cgroup *memcg) | |
1321 | { | |
1322 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1323 | struct lruvec *lruvec; |
925b7673 | 1324 | |
bea8c150 HD |
1325 | if (mem_cgroup_disabled()) { |
1326 | lruvec = &zone->lruvec; | |
1327 | goto out; | |
1328 | } | |
925b7673 | 1329 | |
e231875b | 1330 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1331 | lruvec = &mz->lruvec; |
1332 | out: | |
1333 | /* | |
1334 | * Since a node can be onlined after the mem_cgroup was created, | |
1335 | * we have to be prepared to initialize lruvec->zone here; | |
1336 | * and if offlined then reonlined, we need to reinitialize it. | |
1337 | */ | |
1338 | if (unlikely(lruvec->zone != zone)) | |
1339 | lruvec->zone = zone; | |
1340 | return lruvec; | |
925b7673 JW |
1341 | } |
1342 | ||
925b7673 | 1343 | /** |
fa9add64 | 1344 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1345 | * @page: the page |
fa9add64 | 1346 | * @zone: zone of the page |
925b7673 | 1347 | */ |
fa9add64 | 1348 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1349 | { |
08e552c6 | 1350 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1351 | struct mem_cgroup *memcg; |
1352 | struct page_cgroup *pc; | |
bea8c150 | 1353 | struct lruvec *lruvec; |
6d12e2d8 | 1354 | |
bea8c150 HD |
1355 | if (mem_cgroup_disabled()) { |
1356 | lruvec = &zone->lruvec; | |
1357 | goto out; | |
1358 | } | |
925b7673 | 1359 | |
08e552c6 | 1360 | pc = lookup_page_cgroup(page); |
38c5d72f | 1361 | memcg = pc->mem_cgroup; |
7512102c HD |
1362 | |
1363 | /* | |
fa9add64 | 1364 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1365 | * an uncharged page off lru does nothing to secure |
1366 | * its former mem_cgroup from sudden removal. | |
1367 | * | |
1368 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1369 | * under page_cgroup lock: between them, they make all uses | |
1370 | * of pc->mem_cgroup safe. | |
1371 | */ | |
fa9add64 | 1372 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1373 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1374 | ||
e231875b | 1375 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1376 | lruvec = &mz->lruvec; |
1377 | out: | |
1378 | /* | |
1379 | * Since a node can be onlined after the mem_cgroup was created, | |
1380 | * we have to be prepared to initialize lruvec->zone here; | |
1381 | * and if offlined then reonlined, we need to reinitialize it. | |
1382 | */ | |
1383 | if (unlikely(lruvec->zone != zone)) | |
1384 | lruvec->zone = zone; | |
1385 | return lruvec; | |
08e552c6 | 1386 | } |
b69408e8 | 1387 | |
925b7673 | 1388 | /** |
fa9add64 HD |
1389 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1390 | * @lruvec: mem_cgroup per zone lru vector | |
1391 | * @lru: index of lru list the page is sitting on | |
1392 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1393 | * |
fa9add64 HD |
1394 | * This function must be called when a page is added to or removed from an |
1395 | * lru list. | |
3f58a829 | 1396 | */ |
fa9add64 HD |
1397 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1398 | int nr_pages) | |
3f58a829 MK |
1399 | { |
1400 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1401 | unsigned long *lru_size; |
3f58a829 MK |
1402 | |
1403 | if (mem_cgroup_disabled()) | |
1404 | return; | |
1405 | ||
fa9add64 HD |
1406 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1407 | lru_size = mz->lru_size + lru; | |
1408 | *lru_size += nr_pages; | |
1409 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1410 | } |
544122e5 | 1411 | |
3e92041d | 1412 | /* |
c0ff4b85 | 1413 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1414 | * hierarchy subtree |
1415 | */ | |
c3ac9a8a JW |
1416 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1417 | struct mem_cgroup *memcg) | |
3e92041d | 1418 | { |
91c63734 JW |
1419 | if (root_memcg == memcg) |
1420 | return true; | |
3a981f48 | 1421 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1422 | return false; |
b47f77b5 | 1423 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1424 | } |
1425 | ||
1426 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1427 | struct mem_cgroup *memcg) | |
1428 | { | |
1429 | bool ret; | |
1430 | ||
91c63734 | 1431 | rcu_read_lock(); |
c3ac9a8a | 1432 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1433 | rcu_read_unlock(); |
1434 | return ret; | |
3e92041d MH |
1435 | } |
1436 | ||
ffbdccf5 DR |
1437 | bool task_in_mem_cgroup(struct task_struct *task, |
1438 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1439 | { |
0b7f569e | 1440 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1441 | struct task_struct *p; |
ffbdccf5 | 1442 | bool ret; |
4c4a2214 | 1443 | |
158e0a2d | 1444 | p = find_lock_task_mm(task); |
de077d22 | 1445 | if (p) { |
df381975 | 1446 | curr = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1447 | task_unlock(p); |
1448 | } else { | |
1449 | /* | |
1450 | * All threads may have already detached their mm's, but the oom | |
1451 | * killer still needs to detect if they have already been oom | |
1452 | * killed to prevent needlessly killing additional tasks. | |
1453 | */ | |
ffbdccf5 | 1454 | rcu_read_lock(); |
de077d22 DR |
1455 | curr = mem_cgroup_from_task(task); |
1456 | if (curr) | |
1457 | css_get(&curr->css); | |
ffbdccf5 | 1458 | rcu_read_unlock(); |
de077d22 | 1459 | } |
d31f56db | 1460 | /* |
c0ff4b85 | 1461 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1462 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1463 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1464 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1465 | */ |
c0ff4b85 | 1466 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1467 | css_put(&curr->css); |
4c4a2214 DR |
1468 | return ret; |
1469 | } | |
1470 | ||
c56d5c7d | 1471 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1472 | { |
9b272977 | 1473 | unsigned long inactive_ratio; |
14797e23 | 1474 | unsigned long inactive; |
9b272977 | 1475 | unsigned long active; |
c772be93 | 1476 | unsigned long gb; |
14797e23 | 1477 | |
4d7dcca2 HD |
1478 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1479 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1480 | |
c772be93 KM |
1481 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1482 | if (gb) | |
1483 | inactive_ratio = int_sqrt(10 * gb); | |
1484 | else | |
1485 | inactive_ratio = 1; | |
1486 | ||
9b272977 | 1487 | return inactive * inactive_ratio < active; |
14797e23 KM |
1488 | } |
1489 | ||
6d61ef40 BS |
1490 | #define mem_cgroup_from_res_counter(counter, member) \ |
1491 | container_of(counter, struct mem_cgroup, member) | |
1492 | ||
19942822 | 1493 | /** |
9d11ea9f | 1494 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1495 | * @memcg: the memory cgroup |
19942822 | 1496 | * |
9d11ea9f | 1497 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1498 | * pages. |
19942822 | 1499 | */ |
c0ff4b85 | 1500 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1501 | { |
9d11ea9f JW |
1502 | unsigned long long margin; |
1503 | ||
c0ff4b85 | 1504 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1505 | if (do_swap_account) |
c0ff4b85 | 1506 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1507 | return margin >> PAGE_SHIFT; |
19942822 JW |
1508 | } |
1509 | ||
1f4c025b | 1510 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1511 | { |
a7885eb8 | 1512 | /* root ? */ |
14208b0e | 1513 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1514 | return vm_swappiness; |
1515 | ||
bf1ff263 | 1516 | return memcg->swappiness; |
a7885eb8 KM |
1517 | } |
1518 | ||
619d094b KH |
1519 | /* |
1520 | * memcg->moving_account is used for checking possibility that some thread is | |
1521 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1522 | * a memcg, other threads should check memcg->moving_account under | |
1523 | * rcu_read_lock(), like this: | |
1524 | * | |
1525 | * CPU-A CPU-B | |
1526 | * rcu_read_lock() | |
1527 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1528 | * take heavy locks. | |
1529 | * synchronize_rcu() update something. | |
1530 | * rcu_read_unlock() | |
1531 | * start move here. | |
1532 | */ | |
4331f7d3 KH |
1533 | |
1534 | /* for quick checking without looking up memcg */ | |
1535 | atomic_t memcg_moving __read_mostly; | |
1536 | ||
c0ff4b85 | 1537 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1538 | { |
4331f7d3 | 1539 | atomic_inc(&memcg_moving); |
619d094b | 1540 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1541 | synchronize_rcu(); |
1542 | } | |
1543 | ||
c0ff4b85 | 1544 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1545 | { |
619d094b KH |
1546 | /* |
1547 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1548 | * We check NULL in callee rather than caller. | |
1549 | */ | |
4331f7d3 KH |
1550 | if (memcg) { |
1551 | atomic_dec(&memcg_moving); | |
619d094b | 1552 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1553 | } |
32047e2a | 1554 | } |
619d094b | 1555 | |
32047e2a | 1556 | /* |
bdcbb659 | 1557 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1558 | * |
bdcbb659 QH |
1559 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1560 | * moving cgroups. This is for waiting at high-memory pressure | |
1561 | * caused by "move". | |
32047e2a | 1562 | */ |
c0ff4b85 | 1563 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1564 | { |
2bd9bb20 KH |
1565 | struct mem_cgroup *from; |
1566 | struct mem_cgroup *to; | |
4b534334 | 1567 | bool ret = false; |
2bd9bb20 KH |
1568 | /* |
1569 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1570 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1571 | */ | |
1572 | spin_lock(&mc.lock); | |
1573 | from = mc.from; | |
1574 | to = mc.to; | |
1575 | if (!from) | |
1576 | goto unlock; | |
3e92041d | 1577 | |
c0ff4b85 R |
1578 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1579 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1580 | unlock: |
1581 | spin_unlock(&mc.lock); | |
4b534334 KH |
1582 | return ret; |
1583 | } | |
1584 | ||
c0ff4b85 | 1585 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1586 | { |
1587 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1588 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1589 | DEFINE_WAIT(wait); |
1590 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1591 | /* moving charge context might have finished. */ | |
1592 | if (mc.moving_task) | |
1593 | schedule(); | |
1594 | finish_wait(&mc.waitq, &wait); | |
1595 | return true; | |
1596 | } | |
1597 | } | |
1598 | return false; | |
1599 | } | |
1600 | ||
312734c0 KH |
1601 | /* |
1602 | * Take this lock when | |
1603 | * - a code tries to modify page's memcg while it's USED. | |
1604 | * - a code tries to modify page state accounting in a memcg. | |
312734c0 KH |
1605 | */ |
1606 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1607 | unsigned long *flags) | |
1608 | { | |
1609 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1610 | } | |
1611 | ||
1612 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1613 | unsigned long *flags) | |
1614 | { | |
1615 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1616 | } | |
1617 | ||
58cf188e | 1618 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1619 | /** |
58cf188e | 1620 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1621 | * @memcg: The memory cgroup that went over limit |
1622 | * @p: Task that is going to be killed | |
1623 | * | |
1624 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1625 | * enabled | |
1626 | */ | |
1627 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1628 | { | |
e61734c5 | 1629 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1630 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1631 | struct mem_cgroup *iter; |
1632 | unsigned int i; | |
e222432b | 1633 | |
58cf188e | 1634 | if (!p) |
e222432b BS |
1635 | return; |
1636 | ||
08088cb9 | 1637 | mutex_lock(&oom_info_lock); |
e222432b BS |
1638 | rcu_read_lock(); |
1639 | ||
e61734c5 TH |
1640 | pr_info("Task in "); |
1641 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1642 | pr_info(" killed as a result of limit of "); | |
1643 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1644 | pr_info("\n"); | |
e222432b | 1645 | |
e222432b BS |
1646 | rcu_read_unlock(); |
1647 | ||
d045197f | 1648 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1649 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1650 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1651 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1652 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1653 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1654 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1655 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1656 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1657 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1658 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1659 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1660 | |
1661 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1662 | pr_info("Memory cgroup stats for "); |
1663 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1664 | pr_cont(":"); |
1665 | ||
1666 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1667 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1668 | continue; | |
1669 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1670 | K(mem_cgroup_read_stat(iter, i))); | |
1671 | } | |
1672 | ||
1673 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1674 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1675 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1676 | ||
1677 | pr_cont("\n"); | |
1678 | } | |
08088cb9 | 1679 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1680 | } |
1681 | ||
81d39c20 KH |
1682 | /* |
1683 | * This function returns the number of memcg under hierarchy tree. Returns | |
1684 | * 1(self count) if no children. | |
1685 | */ | |
c0ff4b85 | 1686 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1687 | { |
1688 | int num = 0; | |
7d74b06f KH |
1689 | struct mem_cgroup *iter; |
1690 | ||
c0ff4b85 | 1691 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1692 | num++; |
81d39c20 KH |
1693 | return num; |
1694 | } | |
1695 | ||
a63d83f4 DR |
1696 | /* |
1697 | * Return the memory (and swap, if configured) limit for a memcg. | |
1698 | */ | |
9cbb78bb | 1699 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1700 | { |
1701 | u64 limit; | |
a63d83f4 | 1702 | |
f3e8eb70 | 1703 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1704 | |
a63d83f4 | 1705 | /* |
9a5a8f19 | 1706 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1707 | */ |
9a5a8f19 MH |
1708 | if (mem_cgroup_swappiness(memcg)) { |
1709 | u64 memsw; | |
1710 | ||
1711 | limit += total_swap_pages << PAGE_SHIFT; | |
1712 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1713 | ||
1714 | /* | |
1715 | * If memsw is finite and limits the amount of swap space | |
1716 | * available to this memcg, return that limit. | |
1717 | */ | |
1718 | limit = min(limit, memsw); | |
1719 | } | |
1720 | ||
1721 | return limit; | |
a63d83f4 DR |
1722 | } |
1723 | ||
19965460 DR |
1724 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1725 | int order) | |
9cbb78bb DR |
1726 | { |
1727 | struct mem_cgroup *iter; | |
1728 | unsigned long chosen_points = 0; | |
1729 | unsigned long totalpages; | |
1730 | unsigned int points = 0; | |
1731 | struct task_struct *chosen = NULL; | |
1732 | ||
876aafbf | 1733 | /* |
465adcf1 DR |
1734 | * If current has a pending SIGKILL or is exiting, then automatically |
1735 | * select it. The goal is to allow it to allocate so that it may | |
1736 | * quickly exit and free its memory. | |
876aafbf | 1737 | */ |
465adcf1 | 1738 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1739 | set_thread_flag(TIF_MEMDIE); |
1740 | return; | |
1741 | } | |
1742 | ||
1743 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1744 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1745 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1746 | struct css_task_iter it; |
9cbb78bb DR |
1747 | struct task_struct *task; |
1748 | ||
72ec7029 TH |
1749 | css_task_iter_start(&iter->css, &it); |
1750 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1751 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1752 | false)) { | |
1753 | case OOM_SCAN_SELECT: | |
1754 | if (chosen) | |
1755 | put_task_struct(chosen); | |
1756 | chosen = task; | |
1757 | chosen_points = ULONG_MAX; | |
1758 | get_task_struct(chosen); | |
1759 | /* fall through */ | |
1760 | case OOM_SCAN_CONTINUE: | |
1761 | continue; | |
1762 | case OOM_SCAN_ABORT: | |
72ec7029 | 1763 | css_task_iter_end(&it); |
9cbb78bb DR |
1764 | mem_cgroup_iter_break(memcg, iter); |
1765 | if (chosen) | |
1766 | put_task_struct(chosen); | |
1767 | return; | |
1768 | case OOM_SCAN_OK: | |
1769 | break; | |
1770 | }; | |
1771 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1772 | if (!points || points < chosen_points) |
1773 | continue; | |
1774 | /* Prefer thread group leaders for display purposes */ | |
1775 | if (points == chosen_points && | |
1776 | thread_group_leader(chosen)) | |
1777 | continue; | |
1778 | ||
1779 | if (chosen) | |
1780 | put_task_struct(chosen); | |
1781 | chosen = task; | |
1782 | chosen_points = points; | |
1783 | get_task_struct(chosen); | |
9cbb78bb | 1784 | } |
72ec7029 | 1785 | css_task_iter_end(&it); |
9cbb78bb DR |
1786 | } |
1787 | ||
1788 | if (!chosen) | |
1789 | return; | |
1790 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1791 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1792 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1793 | } |
1794 | ||
5660048c JW |
1795 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1796 | gfp_t gfp_mask, | |
1797 | unsigned long flags) | |
1798 | { | |
1799 | unsigned long total = 0; | |
1800 | bool noswap = false; | |
1801 | int loop; | |
1802 | ||
1803 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1804 | noswap = true; | |
1805 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1806 | noswap = true; | |
1807 | ||
1808 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1809 | if (loop) | |
1810 | drain_all_stock_async(memcg); | |
1811 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1812 | /* | |
1813 | * Allow limit shrinkers, which are triggered directly | |
1814 | * by userspace, to catch signals and stop reclaim | |
1815 | * after minimal progress, regardless of the margin. | |
1816 | */ | |
1817 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1818 | break; | |
1819 | if (mem_cgroup_margin(memcg)) | |
1820 | break; | |
1821 | /* | |
1822 | * If nothing was reclaimed after two attempts, there | |
1823 | * may be no reclaimable pages in this hierarchy. | |
1824 | */ | |
1825 | if (loop && !total) | |
1826 | break; | |
1827 | } | |
1828 | return total; | |
1829 | } | |
1830 | ||
4d0c066d KH |
1831 | /** |
1832 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1833 | * @memcg: the target memcg |
4d0c066d KH |
1834 | * @nid: the node ID to be checked. |
1835 | * @noswap : specify true here if the user wants flle only information. | |
1836 | * | |
1837 | * This function returns whether the specified memcg contains any | |
1838 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1839 | * pages in the node. | |
1840 | */ | |
c0ff4b85 | 1841 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1842 | int nid, bool noswap) |
1843 | { | |
c0ff4b85 | 1844 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1845 | return true; |
1846 | if (noswap || !total_swap_pages) | |
1847 | return false; | |
c0ff4b85 | 1848 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1849 | return true; |
1850 | return false; | |
1851 | ||
1852 | } | |
bb4cc1a8 | 1853 | #if MAX_NUMNODES > 1 |
889976db YH |
1854 | |
1855 | /* | |
1856 | * Always updating the nodemask is not very good - even if we have an empty | |
1857 | * list or the wrong list here, we can start from some node and traverse all | |
1858 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1859 | * | |
1860 | */ | |
c0ff4b85 | 1861 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1862 | { |
1863 | int nid; | |
453a9bf3 KH |
1864 | /* |
1865 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1866 | * pagein/pageout changes since the last update. | |
1867 | */ | |
c0ff4b85 | 1868 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1869 | return; |
c0ff4b85 | 1870 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1871 | return; |
1872 | ||
889976db | 1873 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1874 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1875 | |
31aaea4a | 1876 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1877 | |
c0ff4b85 R |
1878 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1879 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1880 | } |
453a9bf3 | 1881 | |
c0ff4b85 R |
1882 | atomic_set(&memcg->numainfo_events, 0); |
1883 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1884 | } |
1885 | ||
1886 | /* | |
1887 | * Selecting a node where we start reclaim from. Because what we need is just | |
1888 | * reducing usage counter, start from anywhere is O,K. Considering | |
1889 | * memory reclaim from current node, there are pros. and cons. | |
1890 | * | |
1891 | * Freeing memory from current node means freeing memory from a node which | |
1892 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1893 | * hit limits, it will see a contention on a node. But freeing from remote | |
1894 | * node means more costs for memory reclaim because of memory latency. | |
1895 | * | |
1896 | * Now, we use round-robin. Better algorithm is welcomed. | |
1897 | */ | |
c0ff4b85 | 1898 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1899 | { |
1900 | int node; | |
1901 | ||
c0ff4b85 R |
1902 | mem_cgroup_may_update_nodemask(memcg); |
1903 | node = memcg->last_scanned_node; | |
889976db | 1904 | |
c0ff4b85 | 1905 | node = next_node(node, memcg->scan_nodes); |
889976db | 1906 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1907 | node = first_node(memcg->scan_nodes); |
889976db YH |
1908 | /* |
1909 | * We call this when we hit limit, not when pages are added to LRU. | |
1910 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1911 | * memcg is too small and all pages are not on LRU. In that case, | |
1912 | * we use curret node. | |
1913 | */ | |
1914 | if (unlikely(node == MAX_NUMNODES)) | |
1915 | node = numa_node_id(); | |
1916 | ||
c0ff4b85 | 1917 | memcg->last_scanned_node = node; |
889976db YH |
1918 | return node; |
1919 | } | |
1920 | ||
bb4cc1a8 AM |
1921 | /* |
1922 | * Check all nodes whether it contains reclaimable pages or not. | |
1923 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1924 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1925 | * enough new information. We need to do double check. | |
1926 | */ | |
1927 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1928 | { | |
1929 | int nid; | |
1930 | ||
1931 | /* | |
1932 | * quick check...making use of scan_node. | |
1933 | * We can skip unused nodes. | |
1934 | */ | |
1935 | if (!nodes_empty(memcg->scan_nodes)) { | |
1936 | for (nid = first_node(memcg->scan_nodes); | |
1937 | nid < MAX_NUMNODES; | |
1938 | nid = next_node(nid, memcg->scan_nodes)) { | |
1939 | ||
1940 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1941 | return true; | |
1942 | } | |
1943 | } | |
1944 | /* | |
1945 | * Check rest of nodes. | |
1946 | */ | |
1947 | for_each_node_state(nid, N_MEMORY) { | |
1948 | if (node_isset(nid, memcg->scan_nodes)) | |
1949 | continue; | |
1950 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1951 | return true; | |
1952 | } | |
1953 | return false; | |
1954 | } | |
1955 | ||
889976db | 1956 | #else |
c0ff4b85 | 1957 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1958 | { |
1959 | return 0; | |
1960 | } | |
4d0c066d | 1961 | |
bb4cc1a8 AM |
1962 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
1963 | { | |
1964 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
1965 | } | |
889976db YH |
1966 | #endif |
1967 | ||
0608f43d AM |
1968 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1969 | struct zone *zone, | |
1970 | gfp_t gfp_mask, | |
1971 | unsigned long *total_scanned) | |
1972 | { | |
1973 | struct mem_cgroup *victim = NULL; | |
1974 | int total = 0; | |
1975 | int loop = 0; | |
1976 | unsigned long excess; | |
1977 | unsigned long nr_scanned; | |
1978 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1979 | .zone = zone, | |
1980 | .priority = 0, | |
1981 | }; | |
1982 | ||
1983 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; | |
1984 | ||
1985 | while (1) { | |
1986 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1987 | if (!victim) { | |
1988 | loop++; | |
1989 | if (loop >= 2) { | |
1990 | /* | |
1991 | * If we have not been able to reclaim | |
1992 | * anything, it might because there are | |
1993 | * no reclaimable pages under this hierarchy | |
1994 | */ | |
1995 | if (!total) | |
1996 | break; | |
1997 | /* | |
1998 | * We want to do more targeted reclaim. | |
1999 | * excess >> 2 is not to excessive so as to | |
2000 | * reclaim too much, nor too less that we keep | |
2001 | * coming back to reclaim from this cgroup | |
2002 | */ | |
2003 | if (total >= (excess >> 2) || | |
2004 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
2005 | break; | |
2006 | } | |
2007 | continue; | |
2008 | } | |
2009 | if (!mem_cgroup_reclaimable(victim, false)) | |
2010 | continue; | |
2011 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
2012 | zone, &nr_scanned); | |
2013 | *total_scanned += nr_scanned; | |
2014 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
2015 | break; | |
6d61ef40 | 2016 | } |
0608f43d AM |
2017 | mem_cgroup_iter_break(root_memcg, victim); |
2018 | return total; | |
6d61ef40 BS |
2019 | } |
2020 | ||
0056f4e6 JW |
2021 | #ifdef CONFIG_LOCKDEP |
2022 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
2023 | .name = "memcg_oom_lock", | |
2024 | }; | |
2025 | #endif | |
2026 | ||
fb2a6fc5 JW |
2027 | static DEFINE_SPINLOCK(memcg_oom_lock); |
2028 | ||
867578cb KH |
2029 | /* |
2030 | * Check OOM-Killer is already running under our hierarchy. | |
2031 | * If someone is running, return false. | |
2032 | */ | |
fb2a6fc5 | 2033 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2034 | { |
79dfdacc | 2035 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2036 | |
fb2a6fc5 JW |
2037 | spin_lock(&memcg_oom_lock); |
2038 | ||
9f3a0d09 | 2039 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2040 | if (iter->oom_lock) { |
79dfdacc MH |
2041 | /* |
2042 | * this subtree of our hierarchy is already locked | |
2043 | * so we cannot give a lock. | |
2044 | */ | |
79dfdacc | 2045 | failed = iter; |
9f3a0d09 JW |
2046 | mem_cgroup_iter_break(memcg, iter); |
2047 | break; | |
23751be0 JW |
2048 | } else |
2049 | iter->oom_lock = true; | |
7d74b06f | 2050 | } |
867578cb | 2051 | |
fb2a6fc5 JW |
2052 | if (failed) { |
2053 | /* | |
2054 | * OK, we failed to lock the whole subtree so we have | |
2055 | * to clean up what we set up to the failing subtree | |
2056 | */ | |
2057 | for_each_mem_cgroup_tree(iter, memcg) { | |
2058 | if (iter == failed) { | |
2059 | mem_cgroup_iter_break(memcg, iter); | |
2060 | break; | |
2061 | } | |
2062 | iter->oom_lock = false; | |
79dfdacc | 2063 | } |
0056f4e6 JW |
2064 | } else |
2065 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2066 | |
2067 | spin_unlock(&memcg_oom_lock); | |
2068 | ||
2069 | return !failed; | |
a636b327 | 2070 | } |
0b7f569e | 2071 | |
fb2a6fc5 | 2072 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2073 | { |
7d74b06f KH |
2074 | struct mem_cgroup *iter; |
2075 | ||
fb2a6fc5 | 2076 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 2077 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 2078 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2079 | iter->oom_lock = false; |
fb2a6fc5 | 2080 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2081 | } |
2082 | ||
c0ff4b85 | 2083 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2084 | { |
2085 | struct mem_cgroup *iter; | |
2086 | ||
c0ff4b85 | 2087 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2088 | atomic_inc(&iter->under_oom); |
2089 | } | |
2090 | ||
c0ff4b85 | 2091 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2092 | { |
2093 | struct mem_cgroup *iter; | |
2094 | ||
867578cb KH |
2095 | /* |
2096 | * When a new child is created while the hierarchy is under oom, | |
2097 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2098 | * atomic_add_unless() here. | |
2099 | */ | |
c0ff4b85 | 2100 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2101 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2102 | } |
2103 | ||
867578cb KH |
2104 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2105 | ||
dc98df5a | 2106 | struct oom_wait_info { |
d79154bb | 2107 | struct mem_cgroup *memcg; |
dc98df5a KH |
2108 | wait_queue_t wait; |
2109 | }; | |
2110 | ||
2111 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2112 | unsigned mode, int sync, void *arg) | |
2113 | { | |
d79154bb HD |
2114 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2115 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2116 | struct oom_wait_info *oom_wait_info; |
2117 | ||
2118 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2119 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2120 | |
dc98df5a | 2121 | /* |
d79154bb | 2122 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2123 | * Then we can use css_is_ancestor without taking care of RCU. |
2124 | */ | |
c0ff4b85 R |
2125 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2126 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2127 | return 0; |
dc98df5a KH |
2128 | return autoremove_wake_function(wait, mode, sync, arg); |
2129 | } | |
2130 | ||
c0ff4b85 | 2131 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2132 | { |
3812c8c8 | 2133 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2134 | /* for filtering, pass "memcg" as argument. */ |
2135 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2136 | } |
2137 | ||
c0ff4b85 | 2138 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2139 | { |
c0ff4b85 R |
2140 | if (memcg && atomic_read(&memcg->under_oom)) |
2141 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2142 | } |
2143 | ||
3812c8c8 | 2144 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2145 | { |
3812c8c8 JW |
2146 | if (!current->memcg_oom.may_oom) |
2147 | return; | |
867578cb | 2148 | /* |
49426420 JW |
2149 | * We are in the middle of the charge context here, so we |
2150 | * don't want to block when potentially sitting on a callstack | |
2151 | * that holds all kinds of filesystem and mm locks. | |
2152 | * | |
2153 | * Also, the caller may handle a failed allocation gracefully | |
2154 | * (like optional page cache readahead) and so an OOM killer | |
2155 | * invocation might not even be necessary. | |
2156 | * | |
2157 | * That's why we don't do anything here except remember the | |
2158 | * OOM context and then deal with it at the end of the page | |
2159 | * fault when the stack is unwound, the locks are released, | |
2160 | * and when we know whether the fault was overall successful. | |
867578cb | 2161 | */ |
49426420 JW |
2162 | css_get(&memcg->css); |
2163 | current->memcg_oom.memcg = memcg; | |
2164 | current->memcg_oom.gfp_mask = mask; | |
2165 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2166 | } |
2167 | ||
2168 | /** | |
2169 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2170 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2171 | * |
49426420 JW |
2172 | * This has to be called at the end of a page fault if the memcg OOM |
2173 | * handler was enabled. | |
3812c8c8 | 2174 | * |
49426420 | 2175 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2176 | * sleep on a waitqueue until the userspace task resolves the |
2177 | * situation. Sleeping directly in the charge context with all kinds | |
2178 | * of locks held is not a good idea, instead we remember an OOM state | |
2179 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2180 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2181 | * |
2182 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2183 | * completed, %false otherwise. |
3812c8c8 | 2184 | */ |
49426420 | 2185 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2186 | { |
49426420 | 2187 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2188 | struct oom_wait_info owait; |
49426420 | 2189 | bool locked; |
3812c8c8 JW |
2190 | |
2191 | /* OOM is global, do not handle */ | |
3812c8c8 | 2192 | if (!memcg) |
49426420 | 2193 | return false; |
3812c8c8 | 2194 | |
49426420 JW |
2195 | if (!handle) |
2196 | goto cleanup; | |
3812c8c8 JW |
2197 | |
2198 | owait.memcg = memcg; | |
2199 | owait.wait.flags = 0; | |
2200 | owait.wait.func = memcg_oom_wake_function; | |
2201 | owait.wait.private = current; | |
2202 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2203 | |
3812c8c8 | 2204 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2205 | mem_cgroup_mark_under_oom(memcg); |
2206 | ||
2207 | locked = mem_cgroup_oom_trylock(memcg); | |
2208 | ||
2209 | if (locked) | |
2210 | mem_cgroup_oom_notify(memcg); | |
2211 | ||
2212 | if (locked && !memcg->oom_kill_disable) { | |
2213 | mem_cgroup_unmark_under_oom(memcg); | |
2214 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2215 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2216 | current->memcg_oom.order); | |
2217 | } else { | |
3812c8c8 | 2218 | schedule(); |
49426420 JW |
2219 | mem_cgroup_unmark_under_oom(memcg); |
2220 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2221 | } | |
2222 | ||
2223 | if (locked) { | |
fb2a6fc5 JW |
2224 | mem_cgroup_oom_unlock(memcg); |
2225 | /* | |
2226 | * There is no guarantee that an OOM-lock contender | |
2227 | * sees the wakeups triggered by the OOM kill | |
2228 | * uncharges. Wake any sleepers explicitely. | |
2229 | */ | |
2230 | memcg_oom_recover(memcg); | |
2231 | } | |
49426420 JW |
2232 | cleanup: |
2233 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2234 | css_put(&memcg->css); |
867578cb | 2235 | return true; |
0b7f569e KH |
2236 | } |
2237 | ||
d69b042f | 2238 | /* |
b5ffc856 | 2239 | * Used to update mapped file or writeback or other statistics. |
32047e2a KH |
2240 | * |
2241 | * Notes: Race condition | |
2242 | * | |
0a31bc97 JW |
2243 | * Charging occurs during page instantiation, while the page is |
2244 | * unmapped and locked in page migration, or while the page table is | |
2245 | * locked in THP migration. No race is possible. | |
32047e2a | 2246 | * |
0a31bc97 | 2247 | * Uncharge happens to pages with zero references, no race possible. |
32047e2a | 2248 | * |
0a31bc97 JW |
2249 | * Charge moving between groups is protected by checking mm->moving |
2250 | * account and taking the move_lock in the slowpath. | |
d69b042f | 2251 | */ |
26174efd | 2252 | |
89c06bd5 KH |
2253 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2254 | bool *locked, unsigned long *flags) | |
2255 | { | |
2256 | struct mem_cgroup *memcg; | |
2257 | struct page_cgroup *pc; | |
2258 | ||
2259 | pc = lookup_page_cgroup(page); | |
2260 | again: | |
2261 | memcg = pc->mem_cgroup; | |
2262 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2263 | return; | |
2264 | /* | |
2265 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2266 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2267 | * rcu_read_lock(), any calls to move_account will be delayed until |
bdcbb659 | 2268 | * rcu_read_unlock(). |
89c06bd5 | 2269 | */ |
bdcbb659 QH |
2270 | VM_BUG_ON(!rcu_read_lock_held()); |
2271 | if (atomic_read(&memcg->moving_account) <= 0) | |
89c06bd5 KH |
2272 | return; |
2273 | ||
2274 | move_lock_mem_cgroup(memcg, flags); | |
2275 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2276 | move_unlock_mem_cgroup(memcg, flags); | |
2277 | goto again; | |
2278 | } | |
2279 | *locked = true; | |
2280 | } | |
2281 | ||
2282 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2283 | { | |
2284 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2285 | ||
2286 | /* | |
2287 | * It's guaranteed that pc->mem_cgroup never changes while | |
2288 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2289 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2290 | */ |
2291 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2292 | } | |
2293 | ||
2a7106f2 | 2294 | void mem_cgroup_update_page_stat(struct page *page, |
68b4876d | 2295 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2296 | { |
c0ff4b85 | 2297 | struct mem_cgroup *memcg; |
32047e2a | 2298 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2299 | unsigned long uninitialized_var(flags); |
d69b042f | 2300 | |
cfa44946 | 2301 | if (mem_cgroup_disabled()) |
d69b042f | 2302 | return; |
89c06bd5 | 2303 | |
658b72c5 | 2304 | VM_BUG_ON(!rcu_read_lock_held()); |
c0ff4b85 R |
2305 | memcg = pc->mem_cgroup; |
2306 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2307 | return; |
26174efd | 2308 | |
c0ff4b85 | 2309 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2310 | } |
26174efd | 2311 | |
cdec2e42 KH |
2312 | /* |
2313 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2314 | * TODO: maybe necessary to use big numbers in big irons. | |
2315 | */ | |
7ec99d62 | 2316 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2317 | struct memcg_stock_pcp { |
2318 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2319 | unsigned int nr_pages; |
cdec2e42 | 2320 | struct work_struct work; |
26fe6168 | 2321 | unsigned long flags; |
a0db00fc | 2322 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2323 | }; |
2324 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2325 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2326 | |
a0956d54 SS |
2327 | /** |
2328 | * consume_stock: Try to consume stocked charge on this cpu. | |
2329 | * @memcg: memcg to consume from. | |
2330 | * @nr_pages: how many pages to charge. | |
2331 | * | |
2332 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2333 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2334 | * service an allocation will refill the stock. | |
2335 | * | |
2336 | * returns true if successful, false otherwise. | |
cdec2e42 | 2337 | */ |
a0956d54 | 2338 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2339 | { |
2340 | struct memcg_stock_pcp *stock; | |
2341 | bool ret = true; | |
2342 | ||
a0956d54 SS |
2343 | if (nr_pages > CHARGE_BATCH) |
2344 | return false; | |
2345 | ||
cdec2e42 | 2346 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2347 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2348 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2349 | else /* need to call res_counter_charge */ |
2350 | ret = false; | |
2351 | put_cpu_var(memcg_stock); | |
2352 | return ret; | |
2353 | } | |
2354 | ||
2355 | /* | |
2356 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2357 | */ | |
2358 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2359 | { | |
2360 | struct mem_cgroup *old = stock->cached; | |
2361 | ||
11c9ea4e JW |
2362 | if (stock->nr_pages) { |
2363 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2364 | ||
2365 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2366 | if (do_swap_account) |
11c9ea4e JW |
2367 | res_counter_uncharge(&old->memsw, bytes); |
2368 | stock->nr_pages = 0; | |
cdec2e42 KH |
2369 | } |
2370 | stock->cached = NULL; | |
cdec2e42 KH |
2371 | } |
2372 | ||
2373 | /* | |
2374 | * This must be called under preempt disabled or must be called by | |
2375 | * a thread which is pinned to local cpu. | |
2376 | */ | |
2377 | static void drain_local_stock(struct work_struct *dummy) | |
2378 | { | |
7c8e0181 | 2379 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2380 | drain_stock(stock); |
26fe6168 | 2381 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2382 | } |
2383 | ||
e4777496 MH |
2384 | static void __init memcg_stock_init(void) |
2385 | { | |
2386 | int cpu; | |
2387 | ||
2388 | for_each_possible_cpu(cpu) { | |
2389 | struct memcg_stock_pcp *stock = | |
2390 | &per_cpu(memcg_stock, cpu); | |
2391 | INIT_WORK(&stock->work, drain_local_stock); | |
2392 | } | |
2393 | } | |
2394 | ||
cdec2e42 KH |
2395 | /* |
2396 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2397 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2398 | */ |
c0ff4b85 | 2399 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2400 | { |
2401 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2402 | ||
c0ff4b85 | 2403 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2404 | drain_stock(stock); |
c0ff4b85 | 2405 | stock->cached = memcg; |
cdec2e42 | 2406 | } |
11c9ea4e | 2407 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2408 | put_cpu_var(memcg_stock); |
2409 | } | |
2410 | ||
2411 | /* | |
c0ff4b85 | 2412 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2413 | * of the hierarchy under it. sync flag says whether we should block |
2414 | * until the work is done. | |
cdec2e42 | 2415 | */ |
c0ff4b85 | 2416 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2417 | { |
26fe6168 | 2418 | int cpu, curcpu; |
d38144b7 | 2419 | |
cdec2e42 | 2420 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2421 | get_online_cpus(); |
5af12d0e | 2422 | curcpu = get_cpu(); |
cdec2e42 KH |
2423 | for_each_online_cpu(cpu) { |
2424 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2425 | struct mem_cgroup *memcg; |
26fe6168 | 2426 | |
c0ff4b85 R |
2427 | memcg = stock->cached; |
2428 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2429 | continue; |
c0ff4b85 | 2430 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2431 | continue; |
d1a05b69 MH |
2432 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2433 | if (cpu == curcpu) | |
2434 | drain_local_stock(&stock->work); | |
2435 | else | |
2436 | schedule_work_on(cpu, &stock->work); | |
2437 | } | |
cdec2e42 | 2438 | } |
5af12d0e | 2439 | put_cpu(); |
d38144b7 MH |
2440 | |
2441 | if (!sync) | |
2442 | goto out; | |
2443 | ||
2444 | for_each_online_cpu(cpu) { | |
2445 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2446 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2447 | flush_work(&stock->work); |
2448 | } | |
2449 | out: | |
f894ffa8 | 2450 | put_online_cpus(); |
d38144b7 MH |
2451 | } |
2452 | ||
2453 | /* | |
2454 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2455 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2456 | * expects some charges will be back to res_counter later but cannot wait for | |
2457 | * it. | |
2458 | */ | |
c0ff4b85 | 2459 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2460 | { |
9f50fad6 MH |
2461 | /* |
2462 | * If someone calls draining, avoid adding more kworker runs. | |
2463 | */ | |
2464 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2465 | return; | |
c0ff4b85 | 2466 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2467 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2468 | } |
2469 | ||
2470 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2471 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2472 | { |
2473 | /* called when force_empty is called */ | |
9f50fad6 | 2474 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2475 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2476 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2477 | } |
2478 | ||
711d3d2c KH |
2479 | /* |
2480 | * This function drains percpu counter value from DEAD cpu and | |
2481 | * move it to local cpu. Note that this function can be preempted. | |
2482 | */ | |
c0ff4b85 | 2483 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2484 | { |
2485 | int i; | |
2486 | ||
c0ff4b85 | 2487 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2488 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2489 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2490 | |
c0ff4b85 R |
2491 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2492 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2493 | } |
e9f8974f | 2494 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2495 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2496 | |
c0ff4b85 R |
2497 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2498 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2499 | } |
c0ff4b85 | 2500 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2501 | } |
2502 | ||
0db0628d | 2503 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2504 | unsigned long action, |
2505 | void *hcpu) | |
2506 | { | |
2507 | int cpu = (unsigned long)hcpu; | |
2508 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2509 | struct mem_cgroup *iter; |
cdec2e42 | 2510 | |
619d094b | 2511 | if (action == CPU_ONLINE) |
1489ebad | 2512 | return NOTIFY_OK; |
1489ebad | 2513 | |
d833049b | 2514 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2515 | return NOTIFY_OK; |
711d3d2c | 2516 | |
9f3a0d09 | 2517 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2518 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2519 | ||
cdec2e42 KH |
2520 | stock = &per_cpu(memcg_stock, cpu); |
2521 | drain_stock(stock); | |
2522 | return NOTIFY_OK; | |
2523 | } | |
2524 | ||
00501b53 JW |
2525 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2526 | unsigned int nr_pages) | |
8a9f3ccd | 2527 | { |
7ec99d62 | 2528 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2529 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 JW |
2530 | struct mem_cgroup *mem_over_limit; |
2531 | struct res_counter *fail_res; | |
2532 | unsigned long nr_reclaimed; | |
2533 | unsigned long flags = 0; | |
2534 | unsigned long long size; | |
05b84301 | 2535 | int ret = 0; |
a636b327 | 2536 | |
ce00a967 JW |
2537 | if (mem_cgroup_is_root(memcg)) |
2538 | goto done; | |
6539cc05 | 2539 | retry: |
b6b6cc72 MH |
2540 | if (consume_stock(memcg, nr_pages)) |
2541 | goto done; | |
8a9f3ccd | 2542 | |
6539cc05 JW |
2543 | size = batch * PAGE_SIZE; |
2544 | if (!res_counter_charge(&memcg->res, size, &fail_res)) { | |
2545 | if (!do_swap_account) | |
2546 | goto done_restock; | |
2547 | if (!res_counter_charge(&memcg->memsw, size, &fail_res)) | |
2548 | goto done_restock; | |
2549 | res_counter_uncharge(&memcg->res, size); | |
2550 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); | |
2551 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2552 | } else | |
2553 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
7a81b88c | 2554 | |
6539cc05 JW |
2555 | if (batch > nr_pages) { |
2556 | batch = nr_pages; | |
2557 | goto retry; | |
2558 | } | |
6d61ef40 | 2559 | |
06b078fc JW |
2560 | /* |
2561 | * Unlike in global OOM situations, memcg is not in a physical | |
2562 | * memory shortage. Allow dying and OOM-killed tasks to | |
2563 | * bypass the last charges so that they can exit quickly and | |
2564 | * free their memory. | |
2565 | */ | |
2566 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2567 | fatal_signal_pending(current) || | |
2568 | current->flags & PF_EXITING)) | |
2569 | goto bypass; | |
2570 | ||
2571 | if (unlikely(task_in_memcg_oom(current))) | |
2572 | goto nomem; | |
2573 | ||
6539cc05 JW |
2574 | if (!(gfp_mask & __GFP_WAIT)) |
2575 | goto nomem; | |
4b534334 | 2576 | |
6539cc05 JW |
2577 | nr_reclaimed = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
2578 | ||
61e02c74 | 2579 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2580 | goto retry; |
28c34c29 JW |
2581 | |
2582 | if (gfp_mask & __GFP_NORETRY) | |
2583 | goto nomem; | |
6539cc05 JW |
2584 | /* |
2585 | * Even though the limit is exceeded at this point, reclaim | |
2586 | * may have been able to free some pages. Retry the charge | |
2587 | * before killing the task. | |
2588 | * | |
2589 | * Only for regular pages, though: huge pages are rather | |
2590 | * unlikely to succeed so close to the limit, and we fall back | |
2591 | * to regular pages anyway in case of failure. | |
2592 | */ | |
61e02c74 | 2593 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2594 | goto retry; |
2595 | /* | |
2596 | * At task move, charge accounts can be doubly counted. So, it's | |
2597 | * better to wait until the end of task_move if something is going on. | |
2598 | */ | |
2599 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2600 | goto retry; | |
2601 | ||
9b130619 JW |
2602 | if (nr_retries--) |
2603 | goto retry; | |
2604 | ||
06b078fc JW |
2605 | if (gfp_mask & __GFP_NOFAIL) |
2606 | goto bypass; | |
2607 | ||
6539cc05 JW |
2608 | if (fatal_signal_pending(current)) |
2609 | goto bypass; | |
2610 | ||
61e02c74 | 2611 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2612 | nomem: |
6d1fdc48 | 2613 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2614 | return -ENOMEM; |
867578cb | 2615 | bypass: |
ce00a967 | 2616 | return -EINTR; |
6539cc05 JW |
2617 | |
2618 | done_restock: | |
2619 | if (batch > nr_pages) | |
2620 | refill_stock(memcg, batch - nr_pages); | |
2621 | done: | |
05b84301 | 2622 | return ret; |
7a81b88c | 2623 | } |
8a9f3ccd | 2624 | |
00501b53 | 2625 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2626 | { |
05b84301 | 2627 | unsigned long bytes = nr_pages * PAGE_SIZE; |
e7018b8d | 2628 | |
ce00a967 JW |
2629 | if (mem_cgroup_is_root(memcg)) |
2630 | return; | |
2631 | ||
05b84301 JW |
2632 | res_counter_uncharge(&memcg->res, bytes); |
2633 | if (do_swap_account) | |
2634 | res_counter_uncharge(&memcg->memsw, bytes); | |
854ffa8d DN |
2635 | } |
2636 | ||
d01dd17f KH |
2637 | /* |
2638 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2639 | * This is useful when moving usage to parent cgroup. | |
2640 | */ | |
2641 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2642 | unsigned int nr_pages) | |
2643 | { | |
2644 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2645 | ||
ce00a967 JW |
2646 | if (mem_cgroup_is_root(memcg)) |
2647 | return; | |
2648 | ||
d01dd17f KH |
2649 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); |
2650 | if (do_swap_account) | |
2651 | res_counter_uncharge_until(&memcg->memsw, | |
2652 | memcg->memsw.parent, bytes); | |
2653 | } | |
2654 | ||
a3b2d692 KH |
2655 | /* |
2656 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2657 | * rcu_read_lock(). The caller is responsible for calling |
2658 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2659 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2660 | */ |
2661 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2662 | { | |
a3b2d692 KH |
2663 | /* ID 0 is unused ID */ |
2664 | if (!id) | |
2665 | return NULL; | |
34c00c31 | 2666 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2667 | } |
2668 | ||
0a31bc97 JW |
2669 | /* |
2670 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2671 | * @page: the page | |
2672 | * | |
2673 | * Look up, get a css reference, and return the memcg that owns @page. | |
2674 | * | |
2675 | * The page must be locked to prevent racing with swap-in and page | |
2676 | * cache charges. If coming from an unlocked page table, the caller | |
2677 | * must ensure the page is on the LRU or this can race with charging. | |
2678 | */ | |
e42d9d5d | 2679 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2680 | { |
c0ff4b85 | 2681 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2682 | struct page_cgroup *pc; |
a3b2d692 | 2683 | unsigned short id; |
b5a84319 KH |
2684 | swp_entry_t ent; |
2685 | ||
309381fe | 2686 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2687 | |
3c776e64 | 2688 | pc = lookup_page_cgroup(page); |
a3b2d692 | 2689 | if (PageCgroupUsed(pc)) { |
c0ff4b85 | 2690 | memcg = pc->mem_cgroup; |
ec903c0c | 2691 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2692 | memcg = NULL; |
e42d9d5d | 2693 | } else if (PageSwapCache(page)) { |
3c776e64 | 2694 | ent.val = page_private(page); |
9fb4b7cc | 2695 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2696 | rcu_read_lock(); |
c0ff4b85 | 2697 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2698 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2699 | memcg = NULL; |
a3b2d692 | 2700 | rcu_read_unlock(); |
3c776e64 | 2701 | } |
c0ff4b85 | 2702 | return memcg; |
b5a84319 KH |
2703 | } |
2704 | ||
0a31bc97 JW |
2705 | static void lock_page_lru(struct page *page, int *isolated) |
2706 | { | |
2707 | struct zone *zone = page_zone(page); | |
2708 | ||
2709 | spin_lock_irq(&zone->lru_lock); | |
2710 | if (PageLRU(page)) { | |
2711 | struct lruvec *lruvec; | |
2712 | ||
2713 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2714 | ClearPageLRU(page); | |
2715 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2716 | *isolated = 1; | |
2717 | } else | |
2718 | *isolated = 0; | |
2719 | } | |
2720 | ||
2721 | static void unlock_page_lru(struct page *page, int isolated) | |
2722 | { | |
2723 | struct zone *zone = page_zone(page); | |
2724 | ||
2725 | if (isolated) { | |
2726 | struct lruvec *lruvec; | |
2727 | ||
2728 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2729 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2730 | SetPageLRU(page); | |
2731 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2732 | } | |
2733 | spin_unlock_irq(&zone->lru_lock); | |
2734 | } | |
2735 | ||
00501b53 | 2736 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2737 | bool lrucare) |
7a81b88c | 2738 | { |
ce587e65 | 2739 | struct page_cgroup *pc = lookup_page_cgroup(page); |
0a31bc97 | 2740 | int isolated; |
9ce70c02 | 2741 | |
309381fe | 2742 | VM_BUG_ON_PAGE(PageCgroupUsed(pc), page); |
ca3e0214 KH |
2743 | /* |
2744 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2745 | * accessed by any other context at this point. | |
2746 | */ | |
9ce70c02 HD |
2747 | |
2748 | /* | |
2749 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2750 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2751 | */ | |
0a31bc97 JW |
2752 | if (lrucare) |
2753 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2754 | |
0a31bc97 JW |
2755 | /* |
2756 | * Nobody should be changing or seriously looking at | |
2757 | * pc->mem_cgroup and pc->flags at this point: | |
2758 | * | |
2759 | * - the page is uncharged | |
2760 | * | |
2761 | * - the page is off-LRU | |
2762 | * | |
2763 | * - an anonymous fault has exclusive page access, except for | |
2764 | * a locked page table | |
2765 | * | |
2766 | * - a page cache insertion, a swapin fault, or a migration | |
2767 | * have the page locked | |
2768 | */ | |
c0ff4b85 | 2769 | pc->mem_cgroup = memcg; |
0a31bc97 | 2770 | pc->flags = PCG_USED | PCG_MEM | (do_swap_account ? PCG_MEMSW : 0); |
9ce70c02 | 2771 | |
0a31bc97 JW |
2772 | if (lrucare) |
2773 | unlock_page_lru(page, isolated); | |
7a81b88c | 2774 | } |
66e1707b | 2775 | |
7cf27982 GC |
2776 | static DEFINE_MUTEX(set_limit_mutex); |
2777 | ||
7ae1e1d0 | 2778 | #ifdef CONFIG_MEMCG_KMEM |
bd673145 VD |
2779 | /* |
2780 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or | |
2781 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. | |
2782 | */ | |
2783 | static DEFINE_MUTEX(memcg_slab_mutex); | |
2784 | ||
d6441637 VD |
2785 | static DEFINE_MUTEX(activate_kmem_mutex); |
2786 | ||
7ae1e1d0 GC |
2787 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) |
2788 | { | |
2789 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
6de64beb | 2790 | memcg_kmem_is_active(memcg); |
7ae1e1d0 GC |
2791 | } |
2792 | ||
1f458cbf GC |
2793 | /* |
2794 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2795 | * in the memcg_cache_params struct. | |
2796 | */ | |
2797 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2798 | { | |
2799 | struct kmem_cache *cachep; | |
2800 | ||
2801 | VM_BUG_ON(p->is_root_cache); | |
2802 | cachep = p->root_cache; | |
7a67d7ab | 2803 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2804 | } |
2805 | ||
749c5415 | 2806 | #ifdef CONFIG_SLABINFO |
2da8ca82 | 2807 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
749c5415 | 2808 | { |
2da8ca82 | 2809 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
749c5415 GC |
2810 | struct memcg_cache_params *params; |
2811 | ||
2812 | if (!memcg_can_account_kmem(memcg)) | |
2813 | return -EIO; | |
2814 | ||
2815 | print_slabinfo_header(m); | |
2816 | ||
bd673145 | 2817 | mutex_lock(&memcg_slab_mutex); |
749c5415 GC |
2818 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) |
2819 | cache_show(memcg_params_to_cache(params), m); | |
bd673145 | 2820 | mutex_unlock(&memcg_slab_mutex); |
749c5415 GC |
2821 | |
2822 | return 0; | |
2823 | } | |
2824 | #endif | |
2825 | ||
c67a8a68 | 2826 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
7ae1e1d0 GC |
2827 | { |
2828 | struct res_counter *fail_res; | |
7ae1e1d0 | 2829 | int ret = 0; |
7ae1e1d0 GC |
2830 | |
2831 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2832 | if (ret) | |
2833 | return ret; | |
2834 | ||
00501b53 | 2835 | ret = try_charge(memcg, gfp, size >> PAGE_SHIFT); |
7ae1e1d0 GC |
2836 | if (ret == -EINTR) { |
2837 | /* | |
00501b53 JW |
2838 | * try_charge() chose to bypass to root due to OOM kill or |
2839 | * fatal signal. Since our only options are to either fail | |
2840 | * the allocation or charge it to this cgroup, do it as a | |
2841 | * temporary condition. But we can't fail. From a kmem/slab | |
2842 | * perspective, the cache has already been selected, by | |
2843 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2844 | * our minds. |
2845 | * | |
2846 | * This condition will only trigger if the task entered | |
00501b53 JW |
2847 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2848 | * during try_charge() above. Tasks that were already dying | |
2849 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2850 | * directed to the root cgroup in memcontrol.h |
2851 | */ | |
2852 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
2853 | if (do_swap_account) | |
2854 | res_counter_charge_nofail(&memcg->memsw, size, | |
2855 | &fail_res); | |
2856 | ret = 0; | |
2857 | } else if (ret) | |
2858 | res_counter_uncharge(&memcg->kmem, size); | |
2859 | ||
2860 | return ret; | |
2861 | } | |
2862 | ||
c67a8a68 | 2863 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) |
7ae1e1d0 | 2864 | { |
7ae1e1d0 GC |
2865 | res_counter_uncharge(&memcg->res, size); |
2866 | if (do_swap_account) | |
2867 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
2868 | |
2869 | /* Not down to 0 */ | |
2870 | if (res_counter_uncharge(&memcg->kmem, size)) | |
2871 | return; | |
2872 | ||
10d5ebf4 LZ |
2873 | /* |
2874 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
2875 | * this last uncharge is racing with the offlining code or it is | |
2876 | * outliving the memcg existence. | |
2877 | * | |
2878 | * The memory barrier imposed by test&clear is paired with the | |
2879 | * explicit one in memcg_kmem_mark_dead(). | |
2880 | */ | |
7de37682 | 2881 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 2882 | css_put(&memcg->css); |
7ae1e1d0 GC |
2883 | } |
2884 | ||
2633d7a0 GC |
2885 | /* |
2886 | * helper for acessing a memcg's index. It will be used as an index in the | |
2887 | * child cache array in kmem_cache, and also to derive its name. This function | |
2888 | * will return -1 when this is not a kmem-limited memcg. | |
2889 | */ | |
2890 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2891 | { | |
2892 | return memcg ? memcg->kmemcg_id : -1; | |
2893 | } | |
2894 | ||
55007d84 GC |
2895 | static size_t memcg_caches_array_size(int num_groups) |
2896 | { | |
2897 | ssize_t size; | |
2898 | if (num_groups <= 0) | |
2899 | return 0; | |
2900 | ||
2901 | size = 2 * num_groups; | |
2902 | if (size < MEMCG_CACHES_MIN_SIZE) | |
2903 | size = MEMCG_CACHES_MIN_SIZE; | |
2904 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2905 | size = MEMCG_CACHES_MAX_SIZE; | |
2906 | ||
2907 | return size; | |
2908 | } | |
2909 | ||
2910 | /* | |
2911 | * We should update the current array size iff all caches updates succeed. This | |
2912 | * can only be done from the slab side. The slab mutex needs to be held when | |
2913 | * calling this. | |
2914 | */ | |
2915 | void memcg_update_array_size(int num) | |
2916 | { | |
2917 | if (num > memcg_limited_groups_array_size) | |
2918 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
2919 | } | |
2920 | ||
2921 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) | |
2922 | { | |
2923 | struct memcg_cache_params *cur_params = s->memcg_params; | |
2924 | ||
f35c3a8e | 2925 | VM_BUG_ON(!is_root_cache(s)); |
55007d84 GC |
2926 | |
2927 | if (num_groups > memcg_limited_groups_array_size) { | |
2928 | int i; | |
f8570263 | 2929 | struct memcg_cache_params *new_params; |
55007d84 GC |
2930 | ssize_t size = memcg_caches_array_size(num_groups); |
2931 | ||
2932 | size *= sizeof(void *); | |
90c7a79c | 2933 | size += offsetof(struct memcg_cache_params, memcg_caches); |
55007d84 | 2934 | |
f8570263 VD |
2935 | new_params = kzalloc(size, GFP_KERNEL); |
2936 | if (!new_params) | |
55007d84 | 2937 | return -ENOMEM; |
55007d84 | 2938 | |
f8570263 | 2939 | new_params->is_root_cache = true; |
55007d84 GC |
2940 | |
2941 | /* | |
2942 | * There is the chance it will be bigger than | |
2943 | * memcg_limited_groups_array_size, if we failed an allocation | |
2944 | * in a cache, in which case all caches updated before it, will | |
2945 | * have a bigger array. | |
2946 | * | |
2947 | * But if that is the case, the data after | |
2948 | * memcg_limited_groups_array_size is certainly unused | |
2949 | */ | |
2950 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
2951 | if (!cur_params->memcg_caches[i]) | |
2952 | continue; | |
f8570263 | 2953 | new_params->memcg_caches[i] = |
55007d84 GC |
2954 | cur_params->memcg_caches[i]; |
2955 | } | |
2956 | ||
2957 | /* | |
2958 | * Ideally, we would wait until all caches succeed, and only | |
2959 | * then free the old one. But this is not worth the extra | |
2960 | * pointer per-cache we'd have to have for this. | |
2961 | * | |
2962 | * It is not a big deal if some caches are left with a size | |
2963 | * bigger than the others. And all updates will reset this | |
2964 | * anyway. | |
2965 | */ | |
f8570263 VD |
2966 | rcu_assign_pointer(s->memcg_params, new_params); |
2967 | if (cur_params) | |
2968 | kfree_rcu(cur_params, rcu_head); | |
55007d84 GC |
2969 | } |
2970 | return 0; | |
2971 | } | |
2972 | ||
363a044f VD |
2973 | int memcg_alloc_cache_params(struct mem_cgroup *memcg, struct kmem_cache *s, |
2974 | struct kmem_cache *root_cache) | |
2633d7a0 | 2975 | { |
90c7a79c | 2976 | size_t size; |
2633d7a0 GC |
2977 | |
2978 | if (!memcg_kmem_enabled()) | |
2979 | return 0; | |
2980 | ||
90c7a79c AV |
2981 | if (!memcg) { |
2982 | size = offsetof(struct memcg_cache_params, memcg_caches); | |
55007d84 | 2983 | size += memcg_limited_groups_array_size * sizeof(void *); |
90c7a79c AV |
2984 | } else |
2985 | size = sizeof(struct memcg_cache_params); | |
55007d84 | 2986 | |
2633d7a0 GC |
2987 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
2988 | if (!s->memcg_params) | |
2989 | return -ENOMEM; | |
2990 | ||
943a451a | 2991 | if (memcg) { |
2633d7a0 | 2992 | s->memcg_params->memcg = memcg; |
943a451a | 2993 | s->memcg_params->root_cache = root_cache; |
051dd460 | 2994 | css_get(&memcg->css); |
4ba902b5 GC |
2995 | } else |
2996 | s->memcg_params->is_root_cache = true; | |
2997 | ||
2633d7a0 GC |
2998 | return 0; |
2999 | } | |
3000 | ||
363a044f VD |
3001 | void memcg_free_cache_params(struct kmem_cache *s) |
3002 | { | |
051dd460 VD |
3003 | if (!s->memcg_params) |
3004 | return; | |
3005 | if (!s->memcg_params->is_root_cache) | |
3006 | css_put(&s->memcg_params->memcg->css); | |
363a044f VD |
3007 | kfree(s->memcg_params); |
3008 | } | |
3009 | ||
776ed0f0 VD |
3010 | static void memcg_register_cache(struct mem_cgroup *memcg, |
3011 | struct kmem_cache *root_cache) | |
2633d7a0 | 3012 | { |
93f39eea VD |
3013 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by |
3014 | memcg_slab_mutex */ | |
bd673145 | 3015 | struct kmem_cache *cachep; |
d7f25f8a GC |
3016 | int id; |
3017 | ||
bd673145 VD |
3018 | lockdep_assert_held(&memcg_slab_mutex); |
3019 | ||
3020 | id = memcg_cache_id(memcg); | |
3021 | ||
3022 | /* | |
3023 | * Since per-memcg caches are created asynchronously on first | |
3024 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
3025 | * create the same cache, but only one of them may succeed. | |
3026 | */ | |
3027 | if (cache_from_memcg_idx(root_cache, id)) | |
1aa13254 VD |
3028 | return; |
3029 | ||
073ee1c6 | 3030 | cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1); |
776ed0f0 | 3031 | cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf); |
2edefe11 | 3032 | /* |
bd673145 VD |
3033 | * If we could not create a memcg cache, do not complain, because |
3034 | * that's not critical at all as we can always proceed with the root | |
3035 | * cache. | |
2edefe11 | 3036 | */ |
bd673145 VD |
3037 | if (!cachep) |
3038 | return; | |
2edefe11 | 3039 | |
bd673145 | 3040 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
1aa13254 | 3041 | |
d7f25f8a | 3042 | /* |
959c8963 VD |
3043 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
3044 | * barrier here to ensure nobody will see the kmem_cache partially | |
3045 | * initialized. | |
d7f25f8a | 3046 | */ |
959c8963 VD |
3047 | smp_wmb(); |
3048 | ||
bd673145 VD |
3049 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
3050 | root_cache->memcg_params->memcg_caches[id] = cachep; | |
1aa13254 | 3051 | } |
d7f25f8a | 3052 | |
776ed0f0 | 3053 | static void memcg_unregister_cache(struct kmem_cache *cachep) |
1aa13254 | 3054 | { |
bd673145 | 3055 | struct kmem_cache *root_cache; |
1aa13254 VD |
3056 | struct mem_cgroup *memcg; |
3057 | int id; | |
3058 | ||
bd673145 | 3059 | lockdep_assert_held(&memcg_slab_mutex); |
d7f25f8a | 3060 | |
bd673145 | 3061 | BUG_ON(is_root_cache(cachep)); |
2edefe11 | 3062 | |
bd673145 VD |
3063 | root_cache = cachep->memcg_params->root_cache; |
3064 | memcg = cachep->memcg_params->memcg; | |
96403da2 | 3065 | id = memcg_cache_id(memcg); |
d7f25f8a | 3066 | |
bd673145 VD |
3067 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
3068 | root_cache->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a | 3069 | |
bd673145 VD |
3070 | list_del(&cachep->memcg_params->list); |
3071 | ||
3072 | kmem_cache_destroy(cachep); | |
2633d7a0 GC |
3073 | } |
3074 | ||
0e9d92f2 GC |
3075 | /* |
3076 | * During the creation a new cache, we need to disable our accounting mechanism | |
3077 | * altogether. This is true even if we are not creating, but rather just | |
3078 | * enqueing new caches to be created. | |
3079 | * | |
3080 | * This is because that process will trigger allocations; some visible, like | |
3081 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3082 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3083 | * objects during debug. | |
3084 | * | |
3085 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3086 | * to it. This may not be a bounded recursion: since the first cache creation | |
3087 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3088 | * cache again, failing at the same point. | |
3089 | * | |
3090 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3091 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3092 | * inside the following two functions. | |
3093 | */ | |
3094 | static inline void memcg_stop_kmem_account(void) | |
3095 | { | |
3096 | VM_BUG_ON(!current->mm); | |
3097 | current->memcg_kmem_skip_account++; | |
3098 | } | |
3099 | ||
3100 | static inline void memcg_resume_kmem_account(void) | |
3101 | { | |
3102 | VM_BUG_ON(!current->mm); | |
3103 | current->memcg_kmem_skip_account--; | |
3104 | } | |
3105 | ||
776ed0f0 | 3106 | int __memcg_cleanup_cache_params(struct kmem_cache *s) |
7cf27982 GC |
3107 | { |
3108 | struct kmem_cache *c; | |
b8529907 | 3109 | int i, failed = 0; |
7cf27982 | 3110 | |
bd673145 | 3111 | mutex_lock(&memcg_slab_mutex); |
7a67d7ab QH |
3112 | for_each_memcg_cache_index(i) { |
3113 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
3114 | if (!c) |
3115 | continue; | |
3116 | ||
776ed0f0 | 3117 | memcg_unregister_cache(c); |
b8529907 VD |
3118 | |
3119 | if (cache_from_memcg_idx(s, i)) | |
3120 | failed++; | |
7cf27982 | 3121 | } |
bd673145 | 3122 | mutex_unlock(&memcg_slab_mutex); |
b8529907 | 3123 | return failed; |
7cf27982 GC |
3124 | } |
3125 | ||
776ed0f0 | 3126 | static void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3127 | { |
3128 | struct kmem_cache *cachep; | |
bd673145 | 3129 | struct memcg_cache_params *params, *tmp; |
1f458cbf GC |
3130 | |
3131 | if (!memcg_kmem_is_active(memcg)) | |
3132 | return; | |
3133 | ||
bd673145 VD |
3134 | mutex_lock(&memcg_slab_mutex); |
3135 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { | |
1f458cbf | 3136 | cachep = memcg_params_to_cache(params); |
bd673145 VD |
3137 | kmem_cache_shrink(cachep); |
3138 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
776ed0f0 | 3139 | memcg_unregister_cache(cachep); |
1f458cbf | 3140 | } |
bd673145 | 3141 | mutex_unlock(&memcg_slab_mutex); |
1f458cbf GC |
3142 | } |
3143 | ||
776ed0f0 | 3144 | struct memcg_register_cache_work { |
5722d094 VD |
3145 | struct mem_cgroup *memcg; |
3146 | struct kmem_cache *cachep; | |
3147 | struct work_struct work; | |
3148 | }; | |
3149 | ||
776ed0f0 | 3150 | static void memcg_register_cache_func(struct work_struct *w) |
d7f25f8a | 3151 | { |
776ed0f0 VD |
3152 | struct memcg_register_cache_work *cw = |
3153 | container_of(w, struct memcg_register_cache_work, work); | |
5722d094 VD |
3154 | struct mem_cgroup *memcg = cw->memcg; |
3155 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 3156 | |
bd673145 | 3157 | mutex_lock(&memcg_slab_mutex); |
776ed0f0 | 3158 | memcg_register_cache(memcg, cachep); |
bd673145 VD |
3159 | mutex_unlock(&memcg_slab_mutex); |
3160 | ||
5722d094 | 3161 | css_put(&memcg->css); |
d7f25f8a GC |
3162 | kfree(cw); |
3163 | } | |
3164 | ||
3165 | /* | |
3166 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3167 | */ |
776ed0f0 VD |
3168 | static void __memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3169 | struct kmem_cache *cachep) | |
d7f25f8a | 3170 | { |
776ed0f0 | 3171 | struct memcg_register_cache_work *cw; |
d7f25f8a | 3172 | |
776ed0f0 | 3173 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
ca0dde97 LZ |
3174 | if (cw == NULL) { |
3175 | css_put(&memcg->css); | |
d7f25f8a GC |
3176 | return; |
3177 | } | |
3178 | ||
3179 | cw->memcg = memcg; | |
3180 | cw->cachep = cachep; | |
3181 | ||
776ed0f0 | 3182 | INIT_WORK(&cw->work, memcg_register_cache_func); |
d7f25f8a GC |
3183 | schedule_work(&cw->work); |
3184 | } | |
3185 | ||
776ed0f0 VD |
3186 | static void memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3187 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
3188 | { |
3189 | /* | |
3190 | * We need to stop accounting when we kmalloc, because if the | |
3191 | * corresponding kmalloc cache is not yet created, the first allocation | |
776ed0f0 | 3192 | * in __memcg_schedule_register_cache will recurse. |
0e9d92f2 GC |
3193 | * |
3194 | * However, it is better to enclose the whole function. Depending on | |
3195 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3196 | * trigger an allocation. This too, will make us recurse. Because at | |
3197 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3198 | * the safest choice is to do it like this, wrapping the whole function. | |
3199 | */ | |
3200 | memcg_stop_kmem_account(); | |
776ed0f0 | 3201 | __memcg_schedule_register_cache(memcg, cachep); |
0e9d92f2 GC |
3202 | memcg_resume_kmem_account(); |
3203 | } | |
c67a8a68 VD |
3204 | |
3205 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) | |
3206 | { | |
3207 | int res; | |
3208 | ||
3209 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, | |
3210 | PAGE_SIZE << order); | |
3211 | if (!res) | |
3212 | atomic_add(1 << order, &cachep->memcg_params->nr_pages); | |
3213 | return res; | |
3214 | } | |
3215 | ||
3216 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) | |
3217 | { | |
3218 | memcg_uncharge_kmem(cachep->memcg_params->memcg, PAGE_SIZE << order); | |
3219 | atomic_sub(1 << order, &cachep->memcg_params->nr_pages); | |
3220 | } | |
3221 | ||
d7f25f8a GC |
3222 | /* |
3223 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3224 | * We try to use the current memcg's version of the cache. | |
3225 | * | |
3226 | * If the cache does not exist yet, if we are the first user of it, | |
3227 | * we either create it immediately, if possible, or create it asynchronously | |
3228 | * in a workqueue. | |
3229 | * In the latter case, we will let the current allocation go through with | |
3230 | * the original cache. | |
3231 | * | |
3232 | * Can't be called in interrupt context or from kernel threads. | |
3233 | * This function needs to be called with rcu_read_lock() held. | |
3234 | */ | |
3235 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3236 | gfp_t gfp) | |
3237 | { | |
3238 | struct mem_cgroup *memcg; | |
959c8963 | 3239 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
3240 | |
3241 | VM_BUG_ON(!cachep->memcg_params); | |
3242 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3243 | ||
0e9d92f2 GC |
3244 | if (!current->mm || current->memcg_kmem_skip_account) |
3245 | return cachep; | |
3246 | ||
d7f25f8a GC |
3247 | rcu_read_lock(); |
3248 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3249 | |
3250 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3251 | goto out; |
d7f25f8a | 3252 | |
959c8963 VD |
3253 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
3254 | if (likely(memcg_cachep)) { | |
3255 | cachep = memcg_cachep; | |
ca0dde97 | 3256 | goto out; |
d7f25f8a GC |
3257 | } |
3258 | ||
ca0dde97 | 3259 | /* The corresponding put will be done in the workqueue. */ |
ec903c0c | 3260 | if (!css_tryget_online(&memcg->css)) |
ca0dde97 LZ |
3261 | goto out; |
3262 | rcu_read_unlock(); | |
3263 | ||
3264 | /* | |
3265 | * If we are in a safe context (can wait, and not in interrupt | |
3266 | * context), we could be be predictable and return right away. | |
3267 | * This would guarantee that the allocation being performed | |
3268 | * already belongs in the new cache. | |
3269 | * | |
3270 | * However, there are some clashes that can arrive from locking. | |
3271 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
3272 | * memcg_create_kmem_cache, this means no further allocation |
3273 | * could happen with the slab_mutex held. So it's better to | |
3274 | * defer everything. | |
ca0dde97 | 3275 | */ |
776ed0f0 | 3276 | memcg_schedule_register_cache(memcg, cachep); |
ca0dde97 LZ |
3277 | return cachep; |
3278 | out: | |
3279 | rcu_read_unlock(); | |
3280 | return cachep; | |
d7f25f8a | 3281 | } |
d7f25f8a | 3282 | |
7ae1e1d0 GC |
3283 | /* |
3284 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3285 | * possible for the given order. But the page is not allocated yet, so we'll | |
3286 | * need a further commit step to do the final arrangements. | |
3287 | * | |
3288 | * It is possible for the task to switch cgroups in this mean time, so at | |
3289 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3290 | * the handle argument to return to the caller which cgroup we should commit | |
3291 | * against. We could also return the memcg directly and avoid the pointer | |
3292 | * passing, but a boolean return value gives better semantics considering | |
3293 | * the compiled-out case as well. | |
3294 | * | |
3295 | * Returning true means the allocation is possible. | |
3296 | */ | |
3297 | bool | |
3298 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3299 | { | |
3300 | struct mem_cgroup *memcg; | |
3301 | int ret; | |
3302 | ||
3303 | *_memcg = NULL; | |
6d42c232 GC |
3304 | |
3305 | /* | |
3306 | * Disabling accounting is only relevant for some specific memcg | |
3307 | * internal allocations. Therefore we would initially not have such | |
52383431 VD |
3308 | * check here, since direct calls to the page allocator that are |
3309 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen | |
3310 | * outside memcg core. We are mostly concerned with cache allocations, | |
3311 | * and by having this test at memcg_kmem_get_cache, we are already able | |
3312 | * to relay the allocation to the root cache and bypass the memcg cache | |
3313 | * altogether. | |
6d42c232 GC |
3314 | * |
3315 | * There is one exception, though: the SLUB allocator does not create | |
3316 | * large order caches, but rather service large kmallocs directly from | |
3317 | * the page allocator. Therefore, the following sequence when backed by | |
3318 | * the SLUB allocator: | |
3319 | * | |
f894ffa8 AM |
3320 | * memcg_stop_kmem_account(); |
3321 | * kmalloc(<large_number>) | |
3322 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3323 | * |
3324 | * would effectively ignore the fact that we should skip accounting, | |
3325 | * since it will drive us directly to this function without passing | |
3326 | * through the cache selector memcg_kmem_get_cache. Such large | |
3327 | * allocations are extremely rare but can happen, for instance, for the | |
3328 | * cache arrays. We bring this test here. | |
3329 | */ | |
3330 | if (!current->mm || current->memcg_kmem_skip_account) | |
3331 | return true; | |
3332 | ||
df381975 | 3333 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 GC |
3334 | |
3335 | if (!memcg_can_account_kmem(memcg)) { | |
3336 | css_put(&memcg->css); | |
3337 | return true; | |
3338 | } | |
3339 | ||
7ae1e1d0 GC |
3340 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3341 | if (!ret) | |
3342 | *_memcg = memcg; | |
7ae1e1d0 GC |
3343 | |
3344 | css_put(&memcg->css); | |
3345 | return (ret == 0); | |
3346 | } | |
3347 | ||
3348 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3349 | int order) | |
3350 | { | |
3351 | struct page_cgroup *pc; | |
3352 | ||
3353 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3354 | ||
3355 | /* The page allocation failed. Revert */ | |
3356 | if (!page) { | |
3357 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3358 | return; |
3359 | } | |
a840cda6 JW |
3360 | /* |
3361 | * The page is freshly allocated and not visible to any | |
3362 | * outside callers yet. Set up pc non-atomically. | |
3363 | */ | |
7ae1e1d0 | 3364 | pc = lookup_page_cgroup(page); |
7ae1e1d0 | 3365 | pc->mem_cgroup = memcg; |
a840cda6 | 3366 | pc->flags = PCG_USED; |
7ae1e1d0 GC |
3367 | } |
3368 | ||
3369 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3370 | { | |
3371 | struct mem_cgroup *memcg = NULL; | |
3372 | struct page_cgroup *pc; | |
3373 | ||
3374 | ||
3375 | pc = lookup_page_cgroup(page); | |
7ae1e1d0 GC |
3376 | if (!PageCgroupUsed(pc)) |
3377 | return; | |
3378 | ||
a840cda6 JW |
3379 | memcg = pc->mem_cgroup; |
3380 | pc->flags = 0; | |
7ae1e1d0 GC |
3381 | |
3382 | /* | |
3383 | * We trust that only if there is a memcg associated with the page, it | |
3384 | * is a valid allocation | |
3385 | */ | |
3386 | if (!memcg) | |
3387 | return; | |
3388 | ||
309381fe | 3389 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
7ae1e1d0 | 3390 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); |
7ae1e1d0 | 3391 | } |
1f458cbf | 3392 | #else |
776ed0f0 | 3393 | static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3394 | { |
3395 | } | |
7ae1e1d0 GC |
3396 | #endif /* CONFIG_MEMCG_KMEM */ |
3397 | ||
ca3e0214 KH |
3398 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3399 | ||
ca3e0214 KH |
3400 | /* |
3401 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3402 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3403 | * charge/uncharge will be never happen and move_account() is done under | |
3404 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3405 | */ |
e94c8a9c | 3406 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3407 | { |
3408 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3409 | struct page_cgroup *pc; |
b070e65c | 3410 | struct mem_cgroup *memcg; |
e94c8a9c | 3411 | int i; |
ca3e0214 | 3412 | |
3d37c4a9 KH |
3413 | if (mem_cgroup_disabled()) |
3414 | return; | |
b070e65c DR |
3415 | |
3416 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3417 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3418 | pc = head_pc + i; | |
b070e65c | 3419 | pc->mem_cgroup = memcg; |
0a31bc97 | 3420 | pc->flags = head_pc->flags; |
e94c8a9c | 3421 | } |
b070e65c DR |
3422 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3423 | HPAGE_PMD_NR); | |
ca3e0214 | 3424 | } |
12d27107 | 3425 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3426 | |
f817ed48 | 3427 | /** |
de3638d9 | 3428 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3429 | * @page: the page |
7ec99d62 | 3430 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3431 | * @pc: page_cgroup of the page. |
3432 | * @from: mem_cgroup which the page is moved from. | |
3433 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3434 | * | |
3435 | * The caller must confirm following. | |
08e552c6 | 3436 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3437 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3438 | * |
2f3479b1 KH |
3439 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3440 | * from old cgroup. | |
f817ed48 | 3441 | */ |
7ec99d62 JW |
3442 | static int mem_cgroup_move_account(struct page *page, |
3443 | unsigned int nr_pages, | |
3444 | struct page_cgroup *pc, | |
3445 | struct mem_cgroup *from, | |
2f3479b1 | 3446 | struct mem_cgroup *to) |
f817ed48 | 3447 | { |
de3638d9 JW |
3448 | unsigned long flags; |
3449 | int ret; | |
987eba66 | 3450 | |
f817ed48 | 3451 | VM_BUG_ON(from == to); |
309381fe | 3452 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3453 | /* |
3454 | * The page is isolated from LRU. So, collapse function | |
3455 | * will not handle this page. But page splitting can happen. | |
3456 | * Do this check under compound_page_lock(). The caller should | |
3457 | * hold it. | |
3458 | */ | |
3459 | ret = -EBUSY; | |
7ec99d62 | 3460 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3461 | goto out; |
3462 | ||
0a31bc97 JW |
3463 | /* |
3464 | * Prevent mem_cgroup_migrate() from looking at pc->mem_cgroup | |
3465 | * of its source page while we change it: page migration takes | |
3466 | * both pages off the LRU, but page cache replacement doesn't. | |
3467 | */ | |
3468 | if (!trylock_page(page)) | |
3469 | goto out; | |
de3638d9 JW |
3470 | |
3471 | ret = -EINVAL; | |
3472 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
0a31bc97 | 3473 | goto out_unlock; |
de3638d9 | 3474 | |
312734c0 | 3475 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3476 | |
0a31bc97 | 3477 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
3478 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
3479 | nr_pages); | |
3480 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3481 | nr_pages); | |
3482 | } | |
3ea67d06 | 3483 | |
59d1d256 JW |
3484 | if (PageWriteback(page)) { |
3485 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3486 | nr_pages); | |
3487 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3488 | nr_pages); | |
3489 | } | |
3ea67d06 | 3490 | |
0a31bc97 JW |
3491 | /* |
3492 | * It is safe to change pc->mem_cgroup here because the page | |
3493 | * is referenced, charged, and isolated - we can't race with | |
3494 | * uncharging, charging, migration, or LRU putback. | |
3495 | */ | |
d69b042f | 3496 | |
854ffa8d | 3497 | /* caller should have done css_get */ |
08e552c6 | 3498 | pc->mem_cgroup = to; |
312734c0 | 3499 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 | 3500 | ret = 0; |
0a31bc97 JW |
3501 | |
3502 | local_irq_disable(); | |
3503 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 3504 | memcg_check_events(to, page); |
0a31bc97 | 3505 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 3506 | memcg_check_events(from, page); |
0a31bc97 JW |
3507 | local_irq_enable(); |
3508 | out_unlock: | |
3509 | unlock_page(page); | |
de3638d9 | 3510 | out: |
f817ed48 KH |
3511 | return ret; |
3512 | } | |
3513 | ||
2ef37d3f MH |
3514 | /** |
3515 | * mem_cgroup_move_parent - moves page to the parent group | |
3516 | * @page: the page to move | |
3517 | * @pc: page_cgroup of the page | |
3518 | * @child: page's cgroup | |
3519 | * | |
3520 | * move charges to its parent or the root cgroup if the group has no | |
3521 | * parent (aka use_hierarchy==0). | |
3522 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3523 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3524 | * it signals a race with a page removal/uncharge or migration. In the | |
3525 | * first case the page is on the way out and it will vanish from the LRU | |
3526 | * on the next attempt and the call should be retried later. | |
3527 | * Isolation from the LRU fails only if page has been isolated from | |
3528 | * the LRU since we looked at it and that usually means either global | |
3529 | * reclaim or migration going on. The page will either get back to the | |
3530 | * LRU or vanish. | |
3531 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3532 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3533 | * disappear in the next attempt. | |
f817ed48 | 3534 | */ |
5564e88b JW |
3535 | static int mem_cgroup_move_parent(struct page *page, |
3536 | struct page_cgroup *pc, | |
6068bf01 | 3537 | struct mem_cgroup *child) |
f817ed48 | 3538 | { |
f817ed48 | 3539 | struct mem_cgroup *parent; |
7ec99d62 | 3540 | unsigned int nr_pages; |
4be4489f | 3541 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3542 | int ret; |
3543 | ||
d8423011 | 3544 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3545 | |
57f9fd7d DN |
3546 | ret = -EBUSY; |
3547 | if (!get_page_unless_zero(page)) | |
3548 | goto out; | |
3549 | if (isolate_lru_page(page)) | |
3550 | goto put; | |
52dbb905 | 3551 | |
7ec99d62 | 3552 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3553 | |
cc926f78 KH |
3554 | parent = parent_mem_cgroup(child); |
3555 | /* | |
3556 | * If no parent, move charges to root cgroup. | |
3557 | */ | |
3558 | if (!parent) | |
3559 | parent = root_mem_cgroup; | |
f817ed48 | 3560 | |
2ef37d3f | 3561 | if (nr_pages > 1) { |
309381fe | 3562 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
987eba66 | 3563 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3564 | } |
987eba66 | 3565 | |
cc926f78 | 3566 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3567 | pc, child, parent); |
cc926f78 KH |
3568 | if (!ret) |
3569 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3570 | |
7ec99d62 | 3571 | if (nr_pages > 1) |
987eba66 | 3572 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3573 | putback_lru_page(page); |
57f9fd7d | 3574 | put: |
40d58138 | 3575 | put_page(page); |
57f9fd7d | 3576 | out: |
f817ed48 KH |
3577 | return ret; |
3578 | } | |
3579 | ||
c255a458 | 3580 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
3581 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
3582 | bool charge) | |
d13d1443 | 3583 | { |
0a31bc97 JW |
3584 | int val = (charge) ? 1 : -1; |
3585 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 3586 | } |
02491447 DN |
3587 | |
3588 | /** | |
3589 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3590 | * @entry: swap entry to be moved | |
3591 | * @from: mem_cgroup which the entry is moved from | |
3592 | * @to: mem_cgroup which the entry is moved to | |
3593 | * | |
3594 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3595 | * as the mem_cgroup's id of @from. | |
3596 | * | |
3597 | * Returns 0 on success, -EINVAL on failure. | |
3598 | * | |
3599 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3600 | * both res and memsw, and called css_get(). | |
3601 | */ | |
3602 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3603 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3604 | { |
3605 | unsigned short old_id, new_id; | |
3606 | ||
34c00c31 LZ |
3607 | old_id = mem_cgroup_id(from); |
3608 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3609 | |
3610 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3611 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3612 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3613 | /* |
483c30b5 DN |
3614 | * This function is only called from task migration context now. |
3615 | * It postpones res_counter and refcount handling till the end | |
3616 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
3617 | * improvement. But we cannot postpone css_get(to) because if |
3618 | * the process that has been moved to @to does swap-in, the | |
3619 | * refcount of @to might be decreased to 0. | |
3620 | * | |
3621 | * We are in attach() phase, so the cgroup is guaranteed to be | |
3622 | * alive, so we can just call css_get(). | |
02491447 | 3623 | */ |
4050377b | 3624 | css_get(&to->css); |
02491447 DN |
3625 | return 0; |
3626 | } | |
3627 | return -EINVAL; | |
3628 | } | |
3629 | #else | |
3630 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3631 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3632 | { |
3633 | return -EINVAL; | |
3634 | } | |
8c7c6e34 | 3635 | #endif |
d13d1443 | 3636 | |
f212ad7c DN |
3637 | #ifdef CONFIG_DEBUG_VM |
3638 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3639 | { | |
3640 | struct page_cgroup *pc; | |
3641 | ||
3642 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3643 | /* |
3644 | * Can be NULL while feeding pages into the page allocator for | |
3645 | * the first time, i.e. during boot or memory hotplug; | |
3646 | * or when mem_cgroup_disabled(). | |
3647 | */ | |
f212ad7c DN |
3648 | if (likely(pc) && PageCgroupUsed(pc)) |
3649 | return pc; | |
3650 | return NULL; | |
3651 | } | |
3652 | ||
3653 | bool mem_cgroup_bad_page_check(struct page *page) | |
3654 | { | |
3655 | if (mem_cgroup_disabled()) | |
3656 | return false; | |
3657 | ||
3658 | return lookup_page_cgroup_used(page) != NULL; | |
3659 | } | |
3660 | ||
3661 | void mem_cgroup_print_bad_page(struct page *page) | |
3662 | { | |
3663 | struct page_cgroup *pc; | |
3664 | ||
3665 | pc = lookup_page_cgroup_used(page); | |
3666 | if (pc) { | |
d045197f AM |
3667 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
3668 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
3669 | } |
3670 | } | |
3671 | #endif | |
3672 | ||
d38d2a75 | 3673 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3674 | unsigned long long val) |
628f4235 | 3675 | { |
81d39c20 | 3676 | int retry_count; |
3c11ecf4 | 3677 | u64 memswlimit, memlimit; |
628f4235 | 3678 | int ret = 0; |
81d39c20 KH |
3679 | int children = mem_cgroup_count_children(memcg); |
3680 | u64 curusage, oldusage; | |
3c11ecf4 | 3681 | int enlarge; |
81d39c20 KH |
3682 | |
3683 | /* | |
3684 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3685 | * is depends on callers. We set our retry-count to be function | |
3686 | * of # of children which we should visit in this loop. | |
3687 | */ | |
3688 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3689 | ||
3690 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3691 | |
3c11ecf4 | 3692 | enlarge = 0; |
8c7c6e34 | 3693 | while (retry_count) { |
628f4235 KH |
3694 | if (signal_pending(current)) { |
3695 | ret = -EINTR; | |
3696 | break; | |
3697 | } | |
8c7c6e34 KH |
3698 | /* |
3699 | * Rather than hide all in some function, I do this in | |
3700 | * open coded manner. You see what this really does. | |
aaad153e | 3701 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3702 | */ |
3703 | mutex_lock(&set_limit_mutex); | |
3704 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3705 | if (memswlimit < val) { | |
3706 | ret = -EINVAL; | |
3707 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3708 | break; |
3709 | } | |
3c11ecf4 KH |
3710 | |
3711 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3712 | if (memlimit < val) | |
3713 | enlarge = 1; | |
3714 | ||
8c7c6e34 | 3715 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3716 | if (!ret) { |
3717 | if (memswlimit == val) | |
3718 | memcg->memsw_is_minimum = true; | |
3719 | else | |
3720 | memcg->memsw_is_minimum = false; | |
3721 | } | |
8c7c6e34 KH |
3722 | mutex_unlock(&set_limit_mutex); |
3723 | ||
3724 | if (!ret) | |
3725 | break; | |
3726 | ||
5660048c JW |
3727 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3728 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3729 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3730 | /* Usage is reduced ? */ | |
f894ffa8 | 3731 | if (curusage >= oldusage) |
81d39c20 KH |
3732 | retry_count--; |
3733 | else | |
3734 | oldusage = curusage; | |
8c7c6e34 | 3735 | } |
3c11ecf4 KH |
3736 | if (!ret && enlarge) |
3737 | memcg_oom_recover(memcg); | |
14797e23 | 3738 | |
8c7c6e34 KH |
3739 | return ret; |
3740 | } | |
3741 | ||
338c8431 LZ |
3742 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3743 | unsigned long long val) | |
8c7c6e34 | 3744 | { |
81d39c20 | 3745 | int retry_count; |
3c11ecf4 | 3746 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3747 | int children = mem_cgroup_count_children(memcg); |
3748 | int ret = -EBUSY; | |
3c11ecf4 | 3749 | int enlarge = 0; |
8c7c6e34 | 3750 | |
81d39c20 | 3751 | /* see mem_cgroup_resize_res_limit */ |
f894ffa8 | 3752 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
81d39c20 | 3753 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
8c7c6e34 KH |
3754 | while (retry_count) { |
3755 | if (signal_pending(current)) { | |
3756 | ret = -EINTR; | |
3757 | break; | |
3758 | } | |
3759 | /* | |
3760 | * Rather than hide all in some function, I do this in | |
3761 | * open coded manner. You see what this really does. | |
aaad153e | 3762 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3763 | */ |
3764 | mutex_lock(&set_limit_mutex); | |
3765 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3766 | if (memlimit > val) { | |
3767 | ret = -EINVAL; | |
3768 | mutex_unlock(&set_limit_mutex); | |
3769 | break; | |
3770 | } | |
3c11ecf4 KH |
3771 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3772 | if (memswlimit < val) | |
3773 | enlarge = 1; | |
8c7c6e34 | 3774 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3775 | if (!ret) { |
3776 | if (memlimit == val) | |
3777 | memcg->memsw_is_minimum = true; | |
3778 | else | |
3779 | memcg->memsw_is_minimum = false; | |
3780 | } | |
8c7c6e34 KH |
3781 | mutex_unlock(&set_limit_mutex); |
3782 | ||
3783 | if (!ret) | |
3784 | break; | |
3785 | ||
5660048c JW |
3786 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3787 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3788 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3789 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3790 | /* Usage is reduced ? */ |
8c7c6e34 | 3791 | if (curusage >= oldusage) |
628f4235 | 3792 | retry_count--; |
81d39c20 KH |
3793 | else |
3794 | oldusage = curusage; | |
628f4235 | 3795 | } |
3c11ecf4 KH |
3796 | if (!ret && enlarge) |
3797 | memcg_oom_recover(memcg); | |
628f4235 KH |
3798 | return ret; |
3799 | } | |
3800 | ||
0608f43d AM |
3801 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3802 | gfp_t gfp_mask, | |
3803 | unsigned long *total_scanned) | |
3804 | { | |
3805 | unsigned long nr_reclaimed = 0; | |
3806 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3807 | unsigned long reclaimed; | |
3808 | int loop = 0; | |
3809 | struct mem_cgroup_tree_per_zone *mctz; | |
3810 | unsigned long long excess; | |
3811 | unsigned long nr_scanned; | |
3812 | ||
3813 | if (order > 0) | |
3814 | return 0; | |
3815 | ||
3816 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3817 | /* | |
3818 | * This loop can run a while, specially if mem_cgroup's continuously | |
3819 | * keep exceeding their soft limit and putting the system under | |
3820 | * pressure | |
3821 | */ | |
3822 | do { | |
3823 | if (next_mz) | |
3824 | mz = next_mz; | |
3825 | else | |
3826 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3827 | if (!mz) | |
3828 | break; | |
3829 | ||
3830 | nr_scanned = 0; | |
3831 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3832 | gfp_mask, &nr_scanned); | |
3833 | nr_reclaimed += reclaimed; | |
3834 | *total_scanned += nr_scanned; | |
0a31bc97 | 3835 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3836 | |
3837 | /* | |
3838 | * If we failed to reclaim anything from this memory cgroup | |
3839 | * it is time to move on to the next cgroup | |
3840 | */ | |
3841 | next_mz = NULL; | |
3842 | if (!reclaimed) { | |
3843 | do { | |
3844 | /* | |
3845 | * Loop until we find yet another one. | |
3846 | * | |
3847 | * By the time we get the soft_limit lock | |
3848 | * again, someone might have aded the | |
3849 | * group back on the RB tree. Iterate to | |
3850 | * make sure we get a different mem. | |
3851 | * mem_cgroup_largest_soft_limit_node returns | |
3852 | * NULL if no other cgroup is present on | |
3853 | * the tree | |
3854 | */ | |
3855 | next_mz = | |
3856 | __mem_cgroup_largest_soft_limit_node(mctz); | |
3857 | if (next_mz == mz) | |
3858 | css_put(&next_mz->memcg->css); | |
3859 | else /* next_mz == NULL or other memcg */ | |
3860 | break; | |
3861 | } while (1); | |
3862 | } | |
cf2c8127 | 3863 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3864 | excess = res_counter_soft_limit_excess(&mz->memcg->res); |
3865 | /* | |
3866 | * One school of thought says that we should not add | |
3867 | * back the node to the tree if reclaim returns 0. | |
3868 | * But our reclaim could return 0, simply because due | |
3869 | * to priority we are exposing a smaller subset of | |
3870 | * memory to reclaim from. Consider this as a longer | |
3871 | * term TODO. | |
3872 | */ | |
3873 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3874 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3875 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3876 | css_put(&mz->memcg->css); |
3877 | loop++; | |
3878 | /* | |
3879 | * Could not reclaim anything and there are no more | |
3880 | * mem cgroups to try or we seem to be looping without | |
3881 | * reclaiming anything. | |
3882 | */ | |
3883 | if (!nr_reclaimed && | |
3884 | (next_mz == NULL || | |
3885 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3886 | break; | |
3887 | } while (!nr_reclaimed); | |
3888 | if (next_mz) | |
3889 | css_put(&next_mz->memcg->css); | |
3890 | return nr_reclaimed; | |
3891 | } | |
3892 | ||
2ef37d3f MH |
3893 | /** |
3894 | * mem_cgroup_force_empty_list - clears LRU of a group | |
3895 | * @memcg: group to clear | |
3896 | * @node: NUMA node | |
3897 | * @zid: zone id | |
3898 | * @lru: lru to to clear | |
3899 | * | |
3c935d18 | 3900 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
3901 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
3902 | * group. | |
cc847582 | 3903 | */ |
2ef37d3f | 3904 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3905 | int node, int zid, enum lru_list lru) |
cc847582 | 3906 | { |
bea8c150 | 3907 | struct lruvec *lruvec; |
2ef37d3f | 3908 | unsigned long flags; |
072c56c1 | 3909 | struct list_head *list; |
925b7673 JW |
3910 | struct page *busy; |
3911 | struct zone *zone; | |
072c56c1 | 3912 | |
08e552c6 | 3913 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
3914 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
3915 | list = &lruvec->lists[lru]; | |
cc847582 | 3916 | |
f817ed48 | 3917 | busy = NULL; |
2ef37d3f | 3918 | do { |
925b7673 | 3919 | struct page_cgroup *pc; |
5564e88b JW |
3920 | struct page *page; |
3921 | ||
08e552c6 | 3922 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3923 | if (list_empty(list)) { |
08e552c6 | 3924 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3925 | break; |
f817ed48 | 3926 | } |
925b7673 JW |
3927 | page = list_entry(list->prev, struct page, lru); |
3928 | if (busy == page) { | |
3929 | list_move(&page->lru, list); | |
648bcc77 | 3930 | busy = NULL; |
08e552c6 | 3931 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3932 | continue; |
3933 | } | |
08e552c6 | 3934 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3935 | |
925b7673 | 3936 | pc = lookup_page_cgroup(page); |
5564e88b | 3937 | |
3c935d18 | 3938 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 3939 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 3940 | busy = page; |
f817ed48 KH |
3941 | } else |
3942 | busy = NULL; | |
2a7a0e0f | 3943 | cond_resched(); |
2ef37d3f | 3944 | } while (!list_empty(list)); |
cc847582 KH |
3945 | } |
3946 | ||
3947 | /* | |
c26251f9 MH |
3948 | * make mem_cgroup's charge to be 0 if there is no task by moving |
3949 | * all the charges and pages to the parent. | |
cc847582 | 3950 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
3951 | * |
3952 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 3953 | */ |
ab5196c2 | 3954 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 3955 | { |
c26251f9 | 3956 | int node, zid; |
bea207c8 | 3957 | u64 usage; |
f817ed48 | 3958 | |
fce66477 | 3959 | do { |
52d4b9ac KH |
3960 | /* This is for making all *used* pages to be on LRU. */ |
3961 | lru_add_drain_all(); | |
c0ff4b85 | 3962 | drain_all_stock_sync(memcg); |
c0ff4b85 | 3963 | mem_cgroup_start_move(memcg); |
31aaea4a | 3964 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 3965 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3966 | enum lru_list lru; |
3967 | for_each_lru(lru) { | |
2ef37d3f | 3968 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3969 | node, zid, lru); |
f817ed48 | 3970 | } |
1ecaab2b | 3971 | } |
f817ed48 | 3972 | } |
c0ff4b85 R |
3973 | mem_cgroup_end_move(memcg); |
3974 | memcg_oom_recover(memcg); | |
52d4b9ac | 3975 | cond_resched(); |
f817ed48 | 3976 | |
2ef37d3f | 3977 | /* |
bea207c8 GC |
3978 | * Kernel memory may not necessarily be trackable to a specific |
3979 | * process. So they are not migrated, and therefore we can't | |
3980 | * expect their value to drop to 0 here. | |
3981 | * Having res filled up with kmem only is enough. | |
3982 | * | |
2ef37d3f MH |
3983 | * This is a safety check because mem_cgroup_force_empty_list |
3984 | * could have raced with mem_cgroup_replace_page_cache callers | |
3985 | * so the lru seemed empty but the page could have been added | |
3986 | * right after the check. RES_USAGE should be safe as we always | |
3987 | * charge before adding to the LRU. | |
3988 | */ | |
bea207c8 GC |
3989 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
3990 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
3991 | } while (usage > 0); | |
c26251f9 MH |
3992 | } |
3993 | ||
ea280e7b TH |
3994 | /* |
3995 | * Test whether @memcg has children, dead or alive. Note that this | |
3996 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3997 | * returns %true if there are child csses according to the cgroup | |
3998 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3999 | */ | |
b5f99b53 GC |
4000 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
4001 | { | |
ea280e7b TH |
4002 | bool ret; |
4003 | ||
696ac172 | 4004 | /* |
ea280e7b TH |
4005 | * The lock does not prevent addition or deletion of children, but |
4006 | * it prevents a new child from being initialized based on this | |
4007 | * parent in css_online(), so it's enough to decide whether | |
4008 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 4009 | */ |
ea280e7b TH |
4010 | lockdep_assert_held(&memcg_create_mutex); |
4011 | ||
4012 | rcu_read_lock(); | |
4013 | ret = css_next_child(NULL, &memcg->css); | |
4014 | rcu_read_unlock(); | |
4015 | return ret; | |
b5f99b53 GC |
4016 | } |
4017 | ||
c26251f9 MH |
4018 | /* |
4019 | * Reclaims as many pages from the given memcg as possible and moves | |
4020 | * the rest to the parent. | |
4021 | * | |
4022 | * Caller is responsible for holding css reference for memcg. | |
4023 | */ | |
4024 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4025 | { | |
4026 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 4027 | |
c1e862c1 KH |
4028 | /* we call try-to-free pages for make this cgroup empty */ |
4029 | lru_add_drain_all(); | |
f817ed48 | 4030 | /* try to free all pages in this cgroup */ |
569530fb | 4031 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4032 | int progress; |
c1e862c1 | 4033 | |
c26251f9 MH |
4034 | if (signal_pending(current)) |
4035 | return -EINTR; | |
4036 | ||
c0ff4b85 | 4037 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4038 | false); |
c1e862c1 | 4039 | if (!progress) { |
f817ed48 | 4040 | nr_retries--; |
c1e862c1 | 4041 | /* maybe some writeback is necessary */ |
8aa7e847 | 4042 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4043 | } |
f817ed48 KH |
4044 | |
4045 | } | |
ab5196c2 MH |
4046 | |
4047 | return 0; | |
cc847582 KH |
4048 | } |
4049 | ||
6770c64e TH |
4050 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
4051 | char *buf, size_t nbytes, | |
4052 | loff_t off) | |
c1e862c1 | 4053 | { |
6770c64e | 4054 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 4055 | |
d8423011 MH |
4056 | if (mem_cgroup_is_root(memcg)) |
4057 | return -EINVAL; | |
6770c64e | 4058 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
4059 | } |
4060 | ||
182446d0 TH |
4061 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
4062 | struct cftype *cft) | |
18f59ea7 | 4063 | { |
182446d0 | 4064 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
4065 | } |
4066 | ||
182446d0 TH |
4067 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
4068 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
4069 | { |
4070 | int retval = 0; | |
182446d0 | 4071 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4072 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 4073 | |
0999821b | 4074 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
4075 | |
4076 | if (memcg->use_hierarchy == val) | |
4077 | goto out; | |
4078 | ||
18f59ea7 | 4079 | /* |
af901ca1 | 4080 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
4081 | * in the child subtrees. If it is unset, then the change can |
4082 | * occur, provided the current cgroup has no children. | |
4083 | * | |
4084 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
4085 | * set if there are no children. | |
4086 | */ | |
c0ff4b85 | 4087 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 4088 | (val == 1 || val == 0)) { |
ea280e7b | 4089 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 4090 | memcg->use_hierarchy = val; |
18f59ea7 BS |
4091 | else |
4092 | retval = -EBUSY; | |
4093 | } else | |
4094 | retval = -EINVAL; | |
567fb435 GC |
4095 | |
4096 | out: | |
0999821b | 4097 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
4098 | |
4099 | return retval; | |
4100 | } | |
4101 | ||
ce00a967 JW |
4102 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
4103 | enum mem_cgroup_stat_index idx) | |
4104 | { | |
4105 | struct mem_cgroup *iter; | |
4106 | long val = 0; | |
4107 | ||
4108 | /* Per-cpu values can be negative, use a signed accumulator */ | |
4109 | for_each_mem_cgroup_tree(iter, memcg) | |
4110 | val += mem_cgroup_read_stat(iter, idx); | |
4111 | ||
4112 | if (val < 0) /* race ? */ | |
4113 | val = 0; | |
4114 | return val; | |
4115 | } | |
4116 | ||
4117 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
4118 | { | |
4119 | u64 val; | |
4120 | ||
4121 | if (!mem_cgroup_is_root(memcg)) { | |
4122 | if (!swap) | |
4123 | return res_counter_read_u64(&memcg->res, RES_USAGE); | |
4124 | else | |
4125 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
4126 | } | |
4127 | ||
4128 | /* | |
4129 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
4130 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
4131 | */ | |
4132 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
4133 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
4134 | ||
4135 | if (swap) | |
4136 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
4137 | ||
4138 | return val << PAGE_SHIFT; | |
4139 | } | |
4140 | ||
4141 | ||
791badbd | 4142 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 4143 | struct cftype *cft) |
8cdea7c0 | 4144 | { |
182446d0 | 4145 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
05b84301 JW |
4146 | enum res_type type = MEMFILE_TYPE(cft->private); |
4147 | int name = MEMFILE_ATTR(cft->private); | |
af36f906 | 4148 | |
8c7c6e34 KH |
4149 | switch (type) { |
4150 | case _MEM: | |
ce00a967 JW |
4151 | if (name == RES_USAGE) |
4152 | return mem_cgroup_usage(memcg, false); | |
05b84301 | 4153 | return res_counter_read_u64(&memcg->res, name); |
8c7c6e34 | 4154 | case _MEMSWAP: |
ce00a967 JW |
4155 | if (name == RES_USAGE) |
4156 | return mem_cgroup_usage(memcg, true); | |
05b84301 | 4157 | return res_counter_read_u64(&memcg->memsw, name); |
510fc4e1 | 4158 | case _KMEM: |
05b84301 | 4159 | return res_counter_read_u64(&memcg->kmem, name); |
510fc4e1 | 4160 | break; |
8c7c6e34 KH |
4161 | default: |
4162 | BUG(); | |
8c7c6e34 | 4163 | } |
8cdea7c0 | 4164 | } |
510fc4e1 | 4165 | |
510fc4e1 | 4166 | #ifdef CONFIG_MEMCG_KMEM |
d6441637 VD |
4167 | /* should be called with activate_kmem_mutex held */ |
4168 | static int __memcg_activate_kmem(struct mem_cgroup *memcg, | |
4169 | unsigned long long limit) | |
4170 | { | |
4171 | int err = 0; | |
4172 | int memcg_id; | |
4173 | ||
4174 | if (memcg_kmem_is_active(memcg)) | |
4175 | return 0; | |
4176 | ||
4177 | /* | |
4178 | * We are going to allocate memory for data shared by all memory | |
4179 | * cgroups so let's stop accounting here. | |
4180 | */ | |
4181 | memcg_stop_kmem_account(); | |
4182 | ||
510fc4e1 GC |
4183 | /* |
4184 | * For simplicity, we won't allow this to be disabled. It also can't | |
4185 | * be changed if the cgroup has children already, or if tasks had | |
4186 | * already joined. | |
4187 | * | |
4188 | * If tasks join before we set the limit, a person looking at | |
4189 | * kmem.usage_in_bytes will have no way to determine when it took | |
4190 | * place, which makes the value quite meaningless. | |
4191 | * | |
4192 | * After it first became limited, changes in the value of the limit are | |
4193 | * of course permitted. | |
510fc4e1 | 4194 | */ |
0999821b | 4195 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
4196 | if (cgroup_has_tasks(memcg->css.cgroup) || |
4197 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
4198 | err = -EBUSY; |
4199 | mutex_unlock(&memcg_create_mutex); | |
4200 | if (err) | |
4201 | goto out; | |
510fc4e1 | 4202 | |
d6441637 VD |
4203 | memcg_id = ida_simple_get(&kmem_limited_groups, |
4204 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
4205 | if (memcg_id < 0) { | |
4206 | err = memcg_id; | |
4207 | goto out; | |
4208 | } | |
4209 | ||
4210 | /* | |
4211 | * Make sure we have enough space for this cgroup in each root cache's | |
4212 | * memcg_params. | |
4213 | */ | |
bd673145 | 4214 | mutex_lock(&memcg_slab_mutex); |
d6441637 | 4215 | err = memcg_update_all_caches(memcg_id + 1); |
bd673145 | 4216 | mutex_unlock(&memcg_slab_mutex); |
d6441637 VD |
4217 | if (err) |
4218 | goto out_rmid; | |
4219 | ||
4220 | memcg->kmemcg_id = memcg_id; | |
4221 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
d6441637 VD |
4222 | |
4223 | /* | |
4224 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
4225 | * active bit set yet, so this should succeed. | |
4226 | */ | |
4227 | err = res_counter_set_limit(&memcg->kmem, limit); | |
4228 | VM_BUG_ON(err); | |
4229 | ||
4230 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
4231 | /* | |
4232 | * Setting the active bit after enabling static branching will | |
4233 | * guarantee no one starts accounting before all call sites are | |
4234 | * patched. | |
4235 | */ | |
4236 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 4237 | out: |
d6441637 VD |
4238 | memcg_resume_kmem_account(); |
4239 | return err; | |
4240 | ||
4241 | out_rmid: | |
4242 | ida_simple_remove(&kmem_limited_groups, memcg_id); | |
4243 | goto out; | |
4244 | } | |
4245 | ||
4246 | static int memcg_activate_kmem(struct mem_cgroup *memcg, | |
4247 | unsigned long long limit) | |
4248 | { | |
4249 | int ret; | |
4250 | ||
4251 | mutex_lock(&activate_kmem_mutex); | |
4252 | ret = __memcg_activate_kmem(memcg, limit); | |
4253 | mutex_unlock(&activate_kmem_mutex); | |
4254 | return ret; | |
4255 | } | |
4256 | ||
4257 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
4258 | unsigned long long val) | |
4259 | { | |
4260 | int ret; | |
4261 | ||
4262 | if (!memcg_kmem_is_active(memcg)) | |
4263 | ret = memcg_activate_kmem(memcg, val); | |
4264 | else | |
4265 | ret = res_counter_set_limit(&memcg->kmem, val); | |
510fc4e1 GC |
4266 | return ret; |
4267 | } | |
4268 | ||
55007d84 | 4269 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 4270 | { |
55007d84 | 4271 | int ret = 0; |
510fc4e1 | 4272 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 4273 | |
d6441637 VD |
4274 | if (!parent) |
4275 | return 0; | |
55007d84 | 4276 | |
d6441637 | 4277 | mutex_lock(&activate_kmem_mutex); |
55007d84 | 4278 | /* |
d6441637 VD |
4279 | * If the parent cgroup is not kmem-active now, it cannot be activated |
4280 | * after this point, because it has at least one child already. | |
55007d84 | 4281 | */ |
d6441637 VD |
4282 | if (memcg_kmem_is_active(parent)) |
4283 | ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX); | |
4284 | mutex_unlock(&activate_kmem_mutex); | |
55007d84 | 4285 | return ret; |
510fc4e1 | 4286 | } |
d6441637 VD |
4287 | #else |
4288 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
4289 | unsigned long long val) | |
4290 | { | |
4291 | return -EINVAL; | |
4292 | } | |
6d043990 | 4293 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 4294 | |
628f4235 KH |
4295 | /* |
4296 | * The user of this function is... | |
4297 | * RES_LIMIT. | |
4298 | */ | |
451af504 TH |
4299 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
4300 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 4301 | { |
451af504 | 4302 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
86ae53e1 GC |
4303 | enum res_type type; |
4304 | int name; | |
628f4235 KH |
4305 | unsigned long long val; |
4306 | int ret; | |
4307 | ||
451af504 TH |
4308 | buf = strstrip(buf); |
4309 | type = MEMFILE_TYPE(of_cft(of)->private); | |
4310 | name = MEMFILE_ATTR(of_cft(of)->private); | |
af36f906 | 4311 | |
8c7c6e34 | 4312 | switch (name) { |
628f4235 | 4313 | case RES_LIMIT: |
4b3bde4c BS |
4314 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4315 | ret = -EINVAL; | |
4316 | break; | |
4317 | } | |
628f4235 | 4318 | /* This function does all necessary parse...reuse it */ |
451af504 | 4319 | ret = res_counter_memparse_write_strategy(buf, &val); |
8c7c6e34 KH |
4320 | if (ret) |
4321 | break; | |
4322 | if (type == _MEM) | |
628f4235 | 4323 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 4324 | else if (type == _MEMSWAP) |
8c7c6e34 | 4325 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 4326 | else if (type == _KMEM) |
d6441637 | 4327 | ret = memcg_update_kmem_limit(memcg, val); |
510fc4e1 GC |
4328 | else |
4329 | return -EINVAL; | |
628f4235 | 4330 | break; |
296c81d8 | 4331 | case RES_SOFT_LIMIT: |
451af504 | 4332 | ret = res_counter_memparse_write_strategy(buf, &val); |
296c81d8 BS |
4333 | if (ret) |
4334 | break; | |
4335 | /* | |
4336 | * For memsw, soft limits are hard to implement in terms | |
4337 | * of semantics, for now, we support soft limits for | |
4338 | * control without swap | |
4339 | */ | |
4340 | if (type == _MEM) | |
4341 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
4342 | else | |
4343 | ret = -EINVAL; | |
4344 | break; | |
628f4235 KH |
4345 | default: |
4346 | ret = -EINVAL; /* should be BUG() ? */ | |
4347 | break; | |
4348 | } | |
451af504 | 4349 | return ret ?: nbytes; |
8cdea7c0 BS |
4350 | } |
4351 | ||
fee7b548 KH |
4352 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
4353 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
4354 | { | |
fee7b548 KH |
4355 | unsigned long long min_limit, min_memsw_limit, tmp; |
4356 | ||
4357 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4358 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
4359 | if (!memcg->use_hierarchy) |
4360 | goto out; | |
4361 | ||
5c9d535b TH |
4362 | while (memcg->css.parent) { |
4363 | memcg = mem_cgroup_from_css(memcg->css.parent); | |
fee7b548 KH |
4364 | if (!memcg->use_hierarchy) |
4365 | break; | |
4366 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4367 | min_limit = min(min_limit, tmp); | |
4368 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4369 | min_memsw_limit = min(min_memsw_limit, tmp); | |
4370 | } | |
4371 | out: | |
4372 | *mem_limit = min_limit; | |
4373 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
4374 | } |
4375 | ||
6770c64e TH |
4376 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
4377 | size_t nbytes, loff_t off) | |
c84872e1 | 4378 | { |
6770c64e | 4379 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
86ae53e1 GC |
4380 | int name; |
4381 | enum res_type type; | |
c84872e1 | 4382 | |
6770c64e TH |
4383 | type = MEMFILE_TYPE(of_cft(of)->private); |
4384 | name = MEMFILE_ATTR(of_cft(of)->private); | |
af36f906 | 4385 | |
8c7c6e34 | 4386 | switch (name) { |
29f2a4da | 4387 | case RES_MAX_USAGE: |
8c7c6e34 | 4388 | if (type == _MEM) |
c0ff4b85 | 4389 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 4390 | else if (type == _MEMSWAP) |
c0ff4b85 | 4391 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
4392 | else if (type == _KMEM) |
4393 | res_counter_reset_max(&memcg->kmem); | |
4394 | else | |
4395 | return -EINVAL; | |
29f2a4da PE |
4396 | break; |
4397 | case RES_FAILCNT: | |
8c7c6e34 | 4398 | if (type == _MEM) |
c0ff4b85 | 4399 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 4400 | else if (type == _MEMSWAP) |
c0ff4b85 | 4401 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
4402 | else if (type == _KMEM) |
4403 | res_counter_reset_failcnt(&memcg->kmem); | |
4404 | else | |
4405 | return -EINVAL; | |
29f2a4da PE |
4406 | break; |
4407 | } | |
f64c3f54 | 4408 | |
6770c64e | 4409 | return nbytes; |
c84872e1 PE |
4410 | } |
4411 | ||
182446d0 | 4412 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4413 | struct cftype *cft) |
4414 | { | |
182446d0 | 4415 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
4416 | } |
4417 | ||
02491447 | 4418 | #ifdef CONFIG_MMU |
182446d0 | 4419 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4420 | struct cftype *cft, u64 val) |
4421 | { | |
182446d0 | 4422 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
4423 | |
4424 | if (val >= (1 << NR_MOVE_TYPE)) | |
4425 | return -EINVAL; | |
ee5e8472 | 4426 | |
7dc74be0 | 4427 | /* |
ee5e8472 GC |
4428 | * No kind of locking is needed in here, because ->can_attach() will |
4429 | * check this value once in the beginning of the process, and then carry | |
4430 | * on with stale data. This means that changes to this value will only | |
4431 | * affect task migrations starting after the change. | |
7dc74be0 | 4432 | */ |
c0ff4b85 | 4433 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4434 | return 0; |
4435 | } | |
02491447 | 4436 | #else |
182446d0 | 4437 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
4438 | struct cftype *cft, u64 val) |
4439 | { | |
4440 | return -ENOSYS; | |
4441 | } | |
4442 | #endif | |
7dc74be0 | 4443 | |
406eb0c9 | 4444 | #ifdef CONFIG_NUMA |
2da8ca82 | 4445 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 4446 | { |
25485de6 GT |
4447 | struct numa_stat { |
4448 | const char *name; | |
4449 | unsigned int lru_mask; | |
4450 | }; | |
4451 | ||
4452 | static const struct numa_stat stats[] = { | |
4453 | { "total", LRU_ALL }, | |
4454 | { "file", LRU_ALL_FILE }, | |
4455 | { "anon", LRU_ALL_ANON }, | |
4456 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4457 | }; | |
4458 | const struct numa_stat *stat; | |
406eb0c9 | 4459 | int nid; |
25485de6 | 4460 | unsigned long nr; |
2da8ca82 | 4461 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 4462 | |
25485de6 GT |
4463 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
4464 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
4465 | seq_printf(m, "%s=%lu", stat->name, nr); | |
4466 | for_each_node_state(nid, N_MEMORY) { | |
4467 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4468 | stat->lru_mask); | |
4469 | seq_printf(m, " N%d=%lu", nid, nr); | |
4470 | } | |
4471 | seq_putc(m, '\n'); | |
406eb0c9 | 4472 | } |
406eb0c9 | 4473 | |
071aee13 YH |
4474 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
4475 | struct mem_cgroup *iter; | |
4476 | ||
4477 | nr = 0; | |
4478 | for_each_mem_cgroup_tree(iter, memcg) | |
4479 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
4480 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
4481 | for_each_node_state(nid, N_MEMORY) { | |
4482 | nr = 0; | |
4483 | for_each_mem_cgroup_tree(iter, memcg) | |
4484 | nr += mem_cgroup_node_nr_lru_pages( | |
4485 | iter, nid, stat->lru_mask); | |
4486 | seq_printf(m, " N%d=%lu", nid, nr); | |
4487 | } | |
4488 | seq_putc(m, '\n'); | |
406eb0c9 | 4489 | } |
406eb0c9 | 4490 | |
406eb0c9 YH |
4491 | return 0; |
4492 | } | |
4493 | #endif /* CONFIG_NUMA */ | |
4494 | ||
af7c4b0e JW |
4495 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
4496 | { | |
4497 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4498 | } | |
4499 | ||
2da8ca82 | 4500 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4501 | { |
2da8ca82 | 4502 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
af7c4b0e JW |
4503 | struct mem_cgroup *mi; |
4504 | unsigned int i; | |
406eb0c9 | 4505 | |
af7c4b0e | 4506 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4507 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4508 | continue; |
af7c4b0e JW |
4509 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4510 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4511 | } |
7b854121 | 4512 | |
af7c4b0e JW |
4513 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4514 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4515 | mem_cgroup_read_events(memcg, i)); | |
4516 | ||
4517 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4518 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4519 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4520 | ||
14067bb3 | 4521 | /* Hierarchical information */ |
fee7b548 KH |
4522 | { |
4523 | unsigned long long limit, memsw_limit; | |
d79154bb | 4524 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 4525 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 4526 | if (do_swap_account) |
78ccf5b5 JW |
4527 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4528 | memsw_limit); | |
fee7b548 | 4529 | } |
7f016ee8 | 4530 | |
af7c4b0e JW |
4531 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4532 | long long val = 0; | |
4533 | ||
bff6bb83 | 4534 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4535 | continue; |
af7c4b0e JW |
4536 | for_each_mem_cgroup_tree(mi, memcg) |
4537 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4538 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4539 | } | |
4540 | ||
4541 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4542 | unsigned long long val = 0; | |
4543 | ||
4544 | for_each_mem_cgroup_tree(mi, memcg) | |
4545 | val += mem_cgroup_read_events(mi, i); | |
4546 | seq_printf(m, "total_%s %llu\n", | |
4547 | mem_cgroup_events_names[i], val); | |
4548 | } | |
4549 | ||
4550 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4551 | unsigned long long val = 0; | |
4552 | ||
4553 | for_each_mem_cgroup_tree(mi, memcg) | |
4554 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4555 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4556 | } |
14067bb3 | 4557 | |
7f016ee8 | 4558 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4559 | { |
4560 | int nid, zid; | |
4561 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4562 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4563 | unsigned long recent_rotated[2] = {0, 0}; |
4564 | unsigned long recent_scanned[2] = {0, 0}; | |
4565 | ||
4566 | for_each_online_node(nid) | |
4567 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 4568 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 4569 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4570 | |
89abfab1 HD |
4571 | recent_rotated[0] += rstat->recent_rotated[0]; |
4572 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4573 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4574 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4575 | } |
78ccf5b5 JW |
4576 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4577 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4578 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4579 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4580 | } |
4581 | #endif | |
4582 | ||
d2ceb9b7 KH |
4583 | return 0; |
4584 | } | |
4585 | ||
182446d0 TH |
4586 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4587 | struct cftype *cft) | |
a7885eb8 | 4588 | { |
182446d0 | 4589 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4590 | |
1f4c025b | 4591 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4592 | } |
4593 | ||
182446d0 TH |
4594 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4595 | struct cftype *cft, u64 val) | |
a7885eb8 | 4596 | { |
182446d0 | 4597 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4598 | |
3dae7fec | 4599 | if (val > 100) |
a7885eb8 KM |
4600 | return -EINVAL; |
4601 | ||
14208b0e | 4602 | if (css->parent) |
3dae7fec JW |
4603 | memcg->swappiness = val; |
4604 | else | |
4605 | vm_swappiness = val; | |
068b38c1 | 4606 | |
a7885eb8 KM |
4607 | return 0; |
4608 | } | |
4609 | ||
2e72b634 KS |
4610 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4611 | { | |
4612 | struct mem_cgroup_threshold_ary *t; | |
4613 | u64 usage; | |
4614 | int i; | |
4615 | ||
4616 | rcu_read_lock(); | |
4617 | if (!swap) | |
2c488db2 | 4618 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4619 | else |
2c488db2 | 4620 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4621 | |
4622 | if (!t) | |
4623 | goto unlock; | |
4624 | ||
ce00a967 | 4625 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4626 | |
4627 | /* | |
748dad36 | 4628 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4629 | * If it's not true, a threshold was crossed after last |
4630 | * call of __mem_cgroup_threshold(). | |
4631 | */ | |
5407a562 | 4632 | i = t->current_threshold; |
2e72b634 KS |
4633 | |
4634 | /* | |
4635 | * Iterate backward over array of thresholds starting from | |
4636 | * current_threshold and check if a threshold is crossed. | |
4637 | * If none of thresholds below usage is crossed, we read | |
4638 | * only one element of the array here. | |
4639 | */ | |
4640 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4641 | eventfd_signal(t->entries[i].eventfd, 1); | |
4642 | ||
4643 | /* i = current_threshold + 1 */ | |
4644 | i++; | |
4645 | ||
4646 | /* | |
4647 | * Iterate forward over array of thresholds starting from | |
4648 | * current_threshold+1 and check if a threshold is crossed. | |
4649 | * If none of thresholds above usage is crossed, we read | |
4650 | * only one element of the array here. | |
4651 | */ | |
4652 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4653 | eventfd_signal(t->entries[i].eventfd, 1); | |
4654 | ||
4655 | /* Update current_threshold */ | |
5407a562 | 4656 | t->current_threshold = i - 1; |
2e72b634 KS |
4657 | unlock: |
4658 | rcu_read_unlock(); | |
4659 | } | |
4660 | ||
4661 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4662 | { | |
ad4ca5f4 KS |
4663 | while (memcg) { |
4664 | __mem_cgroup_threshold(memcg, false); | |
4665 | if (do_swap_account) | |
4666 | __mem_cgroup_threshold(memcg, true); | |
4667 | ||
4668 | memcg = parent_mem_cgroup(memcg); | |
4669 | } | |
2e72b634 KS |
4670 | } |
4671 | ||
4672 | static int compare_thresholds(const void *a, const void *b) | |
4673 | { | |
4674 | const struct mem_cgroup_threshold *_a = a; | |
4675 | const struct mem_cgroup_threshold *_b = b; | |
4676 | ||
2bff24a3 GT |
4677 | if (_a->threshold > _b->threshold) |
4678 | return 1; | |
4679 | ||
4680 | if (_a->threshold < _b->threshold) | |
4681 | return -1; | |
4682 | ||
4683 | return 0; | |
2e72b634 KS |
4684 | } |
4685 | ||
c0ff4b85 | 4686 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4687 | { |
4688 | struct mem_cgroup_eventfd_list *ev; | |
4689 | ||
2bcf2e92 MH |
4690 | spin_lock(&memcg_oom_lock); |
4691 | ||
c0ff4b85 | 4692 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4693 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4694 | |
4695 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4696 | return 0; |
4697 | } | |
4698 | ||
c0ff4b85 | 4699 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4700 | { |
7d74b06f KH |
4701 | struct mem_cgroup *iter; |
4702 | ||
c0ff4b85 | 4703 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4704 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4705 | } |
4706 | ||
59b6f873 | 4707 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4708 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4709 | { |
2c488db2 KS |
4710 | struct mem_cgroup_thresholds *thresholds; |
4711 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 4712 | u64 threshold, usage; |
2c488db2 | 4713 | int i, size, ret; |
2e72b634 KS |
4714 | |
4715 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4716 | if (ret) | |
4717 | return ret; | |
4718 | ||
4719 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4720 | |
05b84301 | 4721 | if (type == _MEM) { |
2c488db2 | 4722 | thresholds = &memcg->thresholds; |
ce00a967 | 4723 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4724 | } else if (type == _MEMSWAP) { |
2c488db2 | 4725 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4726 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4727 | } else |
2e72b634 KS |
4728 | BUG(); |
4729 | ||
2e72b634 | 4730 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4731 | if (thresholds->primary) |
2e72b634 KS |
4732 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4733 | ||
2c488db2 | 4734 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4735 | |
4736 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4737 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4738 | GFP_KERNEL); |
2c488db2 | 4739 | if (!new) { |
2e72b634 KS |
4740 | ret = -ENOMEM; |
4741 | goto unlock; | |
4742 | } | |
2c488db2 | 4743 | new->size = size; |
2e72b634 KS |
4744 | |
4745 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4746 | if (thresholds->primary) { |
4747 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4748 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4749 | } |
4750 | ||
2e72b634 | 4751 | /* Add new threshold */ |
2c488db2 KS |
4752 | new->entries[size - 1].eventfd = eventfd; |
4753 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4754 | |
4755 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4756 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4757 | compare_thresholds, NULL); |
4758 | ||
4759 | /* Find current threshold */ | |
2c488db2 | 4760 | new->current_threshold = -1; |
2e72b634 | 4761 | for (i = 0; i < size; i++) { |
748dad36 | 4762 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4763 | /* |
2c488db2 KS |
4764 | * new->current_threshold will not be used until |
4765 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4766 | * it here. |
4767 | */ | |
2c488db2 | 4768 | ++new->current_threshold; |
748dad36 SZ |
4769 | } else |
4770 | break; | |
2e72b634 KS |
4771 | } |
4772 | ||
2c488db2 KS |
4773 | /* Free old spare buffer and save old primary buffer as spare */ |
4774 | kfree(thresholds->spare); | |
4775 | thresholds->spare = thresholds->primary; | |
4776 | ||
4777 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4778 | |
907860ed | 4779 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4780 | synchronize_rcu(); |
4781 | ||
2e72b634 KS |
4782 | unlock: |
4783 | mutex_unlock(&memcg->thresholds_lock); | |
4784 | ||
4785 | return ret; | |
4786 | } | |
4787 | ||
59b6f873 | 4788 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4789 | struct eventfd_ctx *eventfd, const char *args) |
4790 | { | |
59b6f873 | 4791 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4792 | } |
4793 | ||
59b6f873 | 4794 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4795 | struct eventfd_ctx *eventfd, const char *args) |
4796 | { | |
59b6f873 | 4797 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4798 | } |
4799 | ||
59b6f873 | 4800 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4801 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4802 | { |
2c488db2 KS |
4803 | struct mem_cgroup_thresholds *thresholds; |
4804 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 4805 | u64 usage; |
2c488db2 | 4806 | int i, j, size; |
2e72b634 KS |
4807 | |
4808 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4809 | |
4810 | if (type == _MEM) { | |
2c488db2 | 4811 | thresholds = &memcg->thresholds; |
ce00a967 | 4812 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4813 | } else if (type == _MEMSWAP) { |
2c488db2 | 4814 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4815 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4816 | } else |
2e72b634 KS |
4817 | BUG(); |
4818 | ||
371528ca AV |
4819 | if (!thresholds->primary) |
4820 | goto unlock; | |
4821 | ||
2e72b634 KS |
4822 | /* Check if a threshold crossed before removing */ |
4823 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4824 | ||
4825 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4826 | size = 0; |
4827 | for (i = 0; i < thresholds->primary->size; i++) { | |
4828 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4829 | size++; |
4830 | } | |
4831 | ||
2c488db2 | 4832 | new = thresholds->spare; |
907860ed | 4833 | |
2e72b634 KS |
4834 | /* Set thresholds array to NULL if we don't have thresholds */ |
4835 | if (!size) { | |
2c488db2 KS |
4836 | kfree(new); |
4837 | new = NULL; | |
907860ed | 4838 | goto swap_buffers; |
2e72b634 KS |
4839 | } |
4840 | ||
2c488db2 | 4841 | new->size = size; |
2e72b634 KS |
4842 | |
4843 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4844 | new->current_threshold = -1; |
4845 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4846 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4847 | continue; |
4848 | ||
2c488db2 | 4849 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4850 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4851 | /* |
2c488db2 | 4852 | * new->current_threshold will not be used |
2e72b634 KS |
4853 | * until rcu_assign_pointer(), so it's safe to increment |
4854 | * it here. | |
4855 | */ | |
2c488db2 | 4856 | ++new->current_threshold; |
2e72b634 KS |
4857 | } |
4858 | j++; | |
4859 | } | |
4860 | ||
907860ed | 4861 | swap_buffers: |
2c488db2 KS |
4862 | /* Swap primary and spare array */ |
4863 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4864 | /* If all events are unregistered, free the spare array */ |
4865 | if (!new) { | |
4866 | kfree(thresholds->spare); | |
4867 | thresholds->spare = NULL; | |
4868 | } | |
4869 | ||
2c488db2 | 4870 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4871 | |
907860ed | 4872 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4873 | synchronize_rcu(); |
371528ca | 4874 | unlock: |
2e72b634 | 4875 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4876 | } |
c1e862c1 | 4877 | |
59b6f873 | 4878 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4879 | struct eventfd_ctx *eventfd) |
4880 | { | |
59b6f873 | 4881 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4882 | } |
4883 | ||
59b6f873 | 4884 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4885 | struct eventfd_ctx *eventfd) |
4886 | { | |
59b6f873 | 4887 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4888 | } |
4889 | ||
59b6f873 | 4890 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4891 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4892 | { |
9490ff27 | 4893 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4894 | |
9490ff27 KH |
4895 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4896 | if (!event) | |
4897 | return -ENOMEM; | |
4898 | ||
1af8efe9 | 4899 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4900 | |
4901 | event->eventfd = eventfd; | |
4902 | list_add(&event->list, &memcg->oom_notify); | |
4903 | ||
4904 | /* already in OOM ? */ | |
79dfdacc | 4905 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4906 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4907 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4908 | |
4909 | return 0; | |
4910 | } | |
4911 | ||
59b6f873 | 4912 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4913 | struct eventfd_ctx *eventfd) |
9490ff27 | 4914 | { |
9490ff27 | 4915 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4916 | |
1af8efe9 | 4917 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4918 | |
c0ff4b85 | 4919 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4920 | if (ev->eventfd == eventfd) { |
4921 | list_del(&ev->list); | |
4922 | kfree(ev); | |
4923 | } | |
4924 | } | |
4925 | ||
1af8efe9 | 4926 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4927 | } |
4928 | ||
2da8ca82 | 4929 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4930 | { |
2da8ca82 | 4931 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4932 | |
791badbd TH |
4933 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4934 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4935 | return 0; |
4936 | } | |
4937 | ||
182446d0 | 4938 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4939 | struct cftype *cft, u64 val) |
4940 | { | |
182446d0 | 4941 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4942 | |
4943 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4944 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4945 | return -EINVAL; |
4946 | ||
c0ff4b85 | 4947 | memcg->oom_kill_disable = val; |
4d845ebf | 4948 | if (!val) |
c0ff4b85 | 4949 | memcg_oom_recover(memcg); |
3dae7fec | 4950 | |
3c11ecf4 KH |
4951 | return 0; |
4952 | } | |
4953 | ||
c255a458 | 4954 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4955 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4956 | { |
55007d84 GC |
4957 | int ret; |
4958 | ||
2633d7a0 | 4959 | memcg->kmemcg_id = -1; |
55007d84 GC |
4960 | ret = memcg_propagate_kmem(memcg); |
4961 | if (ret) | |
4962 | return ret; | |
2633d7a0 | 4963 | |
1d62e436 | 4964 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4965 | } |
e5671dfa | 4966 | |
10d5ebf4 | 4967 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4968 | { |
1d62e436 | 4969 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
4970 | } |
4971 | ||
4972 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
4973 | { | |
4974 | if (!memcg_kmem_is_active(memcg)) | |
4975 | return; | |
4976 | ||
4977 | /* | |
4978 | * kmem charges can outlive the cgroup. In the case of slab | |
4979 | * pages, for instance, a page contain objects from various | |
4980 | * processes. As we prevent from taking a reference for every | |
4981 | * such allocation we have to be careful when doing uncharge | |
4982 | * (see memcg_uncharge_kmem) and here during offlining. | |
4983 | * | |
4984 | * The idea is that that only the _last_ uncharge which sees | |
4985 | * the dead memcg will drop the last reference. An additional | |
4986 | * reference is taken here before the group is marked dead | |
4987 | * which is then paired with css_put during uncharge resp. here. | |
4988 | * | |
4989 | * Although this might sound strange as this path is called from | |
ec903c0c TH |
4990 | * css_offline() when the referencemight have dropped down to 0 and |
4991 | * shouldn't be incremented anymore (css_tryget_online() would | |
4992 | * fail) we do not have other options because of the kmem | |
4993 | * allocations lifetime. | |
10d5ebf4 LZ |
4994 | */ |
4995 | css_get(&memcg->css); | |
7de37682 GC |
4996 | |
4997 | memcg_kmem_mark_dead(memcg); | |
4998 | ||
4999 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5000 | return; | |
5001 | ||
7de37682 | 5002 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5003 | css_put(&memcg->css); |
d1a4c0b3 | 5004 | } |
e5671dfa | 5005 | #else |
cbe128e3 | 5006 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5007 | { |
5008 | return 0; | |
5009 | } | |
d1a4c0b3 | 5010 | |
10d5ebf4 LZ |
5011 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
5012 | { | |
5013 | } | |
5014 | ||
5015 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
5016 | { |
5017 | } | |
e5671dfa GC |
5018 | #endif |
5019 | ||
3bc942f3 TH |
5020 | /* |
5021 | * DO NOT USE IN NEW FILES. | |
5022 | * | |
5023 | * "cgroup.event_control" implementation. | |
5024 | * | |
5025 | * This is way over-engineered. It tries to support fully configurable | |
5026 | * events for each user. Such level of flexibility is completely | |
5027 | * unnecessary especially in the light of the planned unified hierarchy. | |
5028 | * | |
5029 | * Please deprecate this and replace with something simpler if at all | |
5030 | * possible. | |
5031 | */ | |
5032 | ||
79bd9814 TH |
5033 | /* |
5034 | * Unregister event and free resources. | |
5035 | * | |
5036 | * Gets called from workqueue. | |
5037 | */ | |
3bc942f3 | 5038 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 5039 | { |
3bc942f3 TH |
5040 | struct mem_cgroup_event *event = |
5041 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 5042 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
5043 | |
5044 | remove_wait_queue(event->wqh, &event->wait); | |
5045 | ||
59b6f873 | 5046 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
5047 | |
5048 | /* Notify userspace the event is going away. */ | |
5049 | eventfd_signal(event->eventfd, 1); | |
5050 | ||
5051 | eventfd_ctx_put(event->eventfd); | |
5052 | kfree(event); | |
59b6f873 | 5053 | css_put(&memcg->css); |
79bd9814 TH |
5054 | } |
5055 | ||
5056 | /* | |
5057 | * Gets called on POLLHUP on eventfd when user closes it. | |
5058 | * | |
5059 | * Called with wqh->lock held and interrupts disabled. | |
5060 | */ | |
3bc942f3 TH |
5061 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
5062 | int sync, void *key) | |
79bd9814 | 5063 | { |
3bc942f3 TH |
5064 | struct mem_cgroup_event *event = |
5065 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 5066 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
5067 | unsigned long flags = (unsigned long)key; |
5068 | ||
5069 | if (flags & POLLHUP) { | |
5070 | /* | |
5071 | * If the event has been detached at cgroup removal, we | |
5072 | * can simply return knowing the other side will cleanup | |
5073 | * for us. | |
5074 | * | |
5075 | * We can't race against event freeing since the other | |
5076 | * side will require wqh->lock via remove_wait_queue(), | |
5077 | * which we hold. | |
5078 | */ | |
fba94807 | 5079 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
5080 | if (!list_empty(&event->list)) { |
5081 | list_del_init(&event->list); | |
5082 | /* | |
5083 | * We are in atomic context, but cgroup_event_remove() | |
5084 | * may sleep, so we have to call it in workqueue. | |
5085 | */ | |
5086 | schedule_work(&event->remove); | |
5087 | } | |
fba94807 | 5088 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
5089 | } |
5090 | ||
5091 | return 0; | |
5092 | } | |
5093 | ||
3bc942f3 | 5094 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
5095 | wait_queue_head_t *wqh, poll_table *pt) |
5096 | { | |
3bc942f3 TH |
5097 | struct mem_cgroup_event *event = |
5098 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
5099 | |
5100 | event->wqh = wqh; | |
5101 | add_wait_queue(wqh, &event->wait); | |
5102 | } | |
5103 | ||
5104 | /* | |
3bc942f3 TH |
5105 | * DO NOT USE IN NEW FILES. |
5106 | * | |
79bd9814 TH |
5107 | * Parse input and register new cgroup event handler. |
5108 | * | |
5109 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
5110 | * Interpretation of args is defined by control file implementation. | |
5111 | */ | |
451af504 TH |
5112 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
5113 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 5114 | { |
451af504 | 5115 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 5116 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5117 | struct mem_cgroup_event *event; |
79bd9814 TH |
5118 | struct cgroup_subsys_state *cfile_css; |
5119 | unsigned int efd, cfd; | |
5120 | struct fd efile; | |
5121 | struct fd cfile; | |
fba94807 | 5122 | const char *name; |
79bd9814 TH |
5123 | char *endp; |
5124 | int ret; | |
5125 | ||
451af504 TH |
5126 | buf = strstrip(buf); |
5127 | ||
5128 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
5129 | if (*endp != ' ') |
5130 | return -EINVAL; | |
451af504 | 5131 | buf = endp + 1; |
79bd9814 | 5132 | |
451af504 | 5133 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
5134 | if ((*endp != ' ') && (*endp != '\0')) |
5135 | return -EINVAL; | |
451af504 | 5136 | buf = endp + 1; |
79bd9814 TH |
5137 | |
5138 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
5139 | if (!event) | |
5140 | return -ENOMEM; | |
5141 | ||
59b6f873 | 5142 | event->memcg = memcg; |
79bd9814 | 5143 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
5144 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
5145 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
5146 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
5147 | |
5148 | efile = fdget(efd); | |
5149 | if (!efile.file) { | |
5150 | ret = -EBADF; | |
5151 | goto out_kfree; | |
5152 | } | |
5153 | ||
5154 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
5155 | if (IS_ERR(event->eventfd)) { | |
5156 | ret = PTR_ERR(event->eventfd); | |
5157 | goto out_put_efile; | |
5158 | } | |
5159 | ||
5160 | cfile = fdget(cfd); | |
5161 | if (!cfile.file) { | |
5162 | ret = -EBADF; | |
5163 | goto out_put_eventfd; | |
5164 | } | |
5165 | ||
5166 | /* the process need read permission on control file */ | |
5167 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
5168 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
5169 | if (ret < 0) | |
5170 | goto out_put_cfile; | |
5171 | ||
fba94807 TH |
5172 | /* |
5173 | * Determine the event callbacks and set them in @event. This used | |
5174 | * to be done via struct cftype but cgroup core no longer knows | |
5175 | * about these events. The following is crude but the whole thing | |
5176 | * is for compatibility anyway. | |
3bc942f3 TH |
5177 | * |
5178 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
5179 | */ |
5180 | name = cfile.file->f_dentry->d_name.name; | |
5181 | ||
5182 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
5183 | event->register_event = mem_cgroup_usage_register_event; | |
5184 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
5185 | } else if (!strcmp(name, "memory.oom_control")) { | |
5186 | event->register_event = mem_cgroup_oom_register_event; | |
5187 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
5188 | } else if (!strcmp(name, "memory.pressure_level")) { | |
5189 | event->register_event = vmpressure_register_event; | |
5190 | event->unregister_event = vmpressure_unregister_event; | |
5191 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
5192 | event->register_event = memsw_cgroup_usage_register_event; |
5193 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
5194 | } else { |
5195 | ret = -EINVAL; | |
5196 | goto out_put_cfile; | |
5197 | } | |
5198 | ||
79bd9814 | 5199 | /* |
b5557c4c TH |
5200 | * Verify @cfile should belong to @css. Also, remaining events are |
5201 | * automatically removed on cgroup destruction but the removal is | |
5202 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 5203 | */ |
ec903c0c TH |
5204 | cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent, |
5205 | &memory_cgrp_subsys); | |
79bd9814 | 5206 | ret = -EINVAL; |
5a17f543 | 5207 | if (IS_ERR(cfile_css)) |
79bd9814 | 5208 | goto out_put_cfile; |
5a17f543 TH |
5209 | if (cfile_css != css) { |
5210 | css_put(cfile_css); | |
79bd9814 | 5211 | goto out_put_cfile; |
5a17f543 | 5212 | } |
79bd9814 | 5213 | |
451af504 | 5214 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
5215 | if (ret) |
5216 | goto out_put_css; | |
5217 | ||
5218 | efile.file->f_op->poll(efile.file, &event->pt); | |
5219 | ||
fba94807 TH |
5220 | spin_lock(&memcg->event_list_lock); |
5221 | list_add(&event->list, &memcg->event_list); | |
5222 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
5223 | |
5224 | fdput(cfile); | |
5225 | fdput(efile); | |
5226 | ||
451af504 | 5227 | return nbytes; |
79bd9814 TH |
5228 | |
5229 | out_put_css: | |
b5557c4c | 5230 | css_put(css); |
79bd9814 TH |
5231 | out_put_cfile: |
5232 | fdput(cfile); | |
5233 | out_put_eventfd: | |
5234 | eventfd_ctx_put(event->eventfd); | |
5235 | out_put_efile: | |
5236 | fdput(efile); | |
5237 | out_kfree: | |
5238 | kfree(event); | |
5239 | ||
5240 | return ret; | |
5241 | } | |
5242 | ||
8cdea7c0 BS |
5243 | static struct cftype mem_cgroup_files[] = { |
5244 | { | |
0eea1030 | 5245 | .name = "usage_in_bytes", |
8c7c6e34 | 5246 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 5247 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5248 | }, |
c84872e1 PE |
5249 | { |
5250 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5251 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5252 | .write = mem_cgroup_reset, |
791badbd | 5253 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5254 | }, |
8cdea7c0 | 5255 | { |
0eea1030 | 5256 | .name = "limit_in_bytes", |
8c7c6e34 | 5257 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5258 | .write = mem_cgroup_write, |
791badbd | 5259 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5260 | }, |
296c81d8 BS |
5261 | { |
5262 | .name = "soft_limit_in_bytes", | |
5263 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5264 | .write = mem_cgroup_write, |
791badbd | 5265 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5266 | }, |
8cdea7c0 BS |
5267 | { |
5268 | .name = "failcnt", | |
8c7c6e34 | 5269 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5270 | .write = mem_cgroup_reset, |
791badbd | 5271 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5272 | }, |
d2ceb9b7 KH |
5273 | { |
5274 | .name = "stat", | |
2da8ca82 | 5275 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5276 | }, |
c1e862c1 KH |
5277 | { |
5278 | .name = "force_empty", | |
6770c64e | 5279 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5280 | }, |
18f59ea7 BS |
5281 | { |
5282 | .name = "use_hierarchy", | |
5283 | .write_u64 = mem_cgroup_hierarchy_write, | |
5284 | .read_u64 = mem_cgroup_hierarchy_read, | |
5285 | }, | |
79bd9814 | 5286 | { |
3bc942f3 | 5287 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5288 | .write = memcg_write_event_control, |
79bd9814 TH |
5289 | .flags = CFTYPE_NO_PREFIX, |
5290 | .mode = S_IWUGO, | |
5291 | }, | |
a7885eb8 KM |
5292 | { |
5293 | .name = "swappiness", | |
5294 | .read_u64 = mem_cgroup_swappiness_read, | |
5295 | .write_u64 = mem_cgroup_swappiness_write, | |
5296 | }, | |
7dc74be0 DN |
5297 | { |
5298 | .name = "move_charge_at_immigrate", | |
5299 | .read_u64 = mem_cgroup_move_charge_read, | |
5300 | .write_u64 = mem_cgroup_move_charge_write, | |
5301 | }, | |
9490ff27 KH |
5302 | { |
5303 | .name = "oom_control", | |
2da8ca82 | 5304 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5305 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5306 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5307 | }, | |
70ddf637 AV |
5308 | { |
5309 | .name = "pressure_level", | |
70ddf637 | 5310 | }, |
406eb0c9 YH |
5311 | #ifdef CONFIG_NUMA |
5312 | { | |
5313 | .name = "numa_stat", | |
2da8ca82 | 5314 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5315 | }, |
5316 | #endif | |
510fc4e1 GC |
5317 | #ifdef CONFIG_MEMCG_KMEM |
5318 | { | |
5319 | .name = "kmem.limit_in_bytes", | |
5320 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5321 | .write = mem_cgroup_write, |
791badbd | 5322 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5323 | }, |
5324 | { | |
5325 | .name = "kmem.usage_in_bytes", | |
5326 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5327 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5328 | }, |
5329 | { | |
5330 | .name = "kmem.failcnt", | |
5331 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5332 | .write = mem_cgroup_reset, |
791badbd | 5333 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5334 | }, |
5335 | { | |
5336 | .name = "kmem.max_usage_in_bytes", | |
5337 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5338 | .write = mem_cgroup_reset, |
791badbd | 5339 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5340 | }, |
749c5415 GC |
5341 | #ifdef CONFIG_SLABINFO |
5342 | { | |
5343 | .name = "kmem.slabinfo", | |
2da8ca82 | 5344 | .seq_show = mem_cgroup_slabinfo_read, |
749c5415 GC |
5345 | }, |
5346 | #endif | |
8c7c6e34 | 5347 | #endif |
6bc10349 | 5348 | { }, /* terminate */ |
af36f906 | 5349 | }; |
8c7c6e34 | 5350 | |
2d11085e MH |
5351 | #ifdef CONFIG_MEMCG_SWAP |
5352 | static struct cftype memsw_cgroup_files[] = { | |
5353 | { | |
5354 | .name = "memsw.usage_in_bytes", | |
5355 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 5356 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5357 | }, |
5358 | { | |
5359 | .name = "memsw.max_usage_in_bytes", | |
5360 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 5361 | .write = mem_cgroup_reset, |
791badbd | 5362 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5363 | }, |
5364 | { | |
5365 | .name = "memsw.limit_in_bytes", | |
5366 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 5367 | .write = mem_cgroup_write, |
791badbd | 5368 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5369 | }, |
5370 | { | |
5371 | .name = "memsw.failcnt", | |
5372 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 5373 | .write = mem_cgroup_reset, |
791badbd | 5374 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5375 | }, |
5376 | { }, /* terminate */ | |
5377 | }; | |
5378 | #endif | |
c0ff4b85 | 5379 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5380 | { |
5381 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 5382 | struct mem_cgroup_per_zone *mz; |
41e3355d | 5383 | int zone, tmp = node; |
1ecaab2b KH |
5384 | /* |
5385 | * This routine is called against possible nodes. | |
5386 | * But it's BUG to call kmalloc() against offline node. | |
5387 | * | |
5388 | * TODO: this routine can waste much memory for nodes which will | |
5389 | * never be onlined. It's better to use memory hotplug callback | |
5390 | * function. | |
5391 | */ | |
41e3355d KH |
5392 | if (!node_state(node, N_NORMAL_MEMORY)) |
5393 | tmp = -1; | |
17295c88 | 5394 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5395 | if (!pn) |
5396 | return 1; | |
1ecaab2b | 5397 | |
1ecaab2b KH |
5398 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
5399 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 5400 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
5401 | mz->usage_in_excess = 0; |
5402 | mz->on_tree = false; | |
d79154bb | 5403 | mz->memcg = memcg; |
1ecaab2b | 5404 | } |
54f72fe0 | 5405 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5406 | return 0; |
5407 | } | |
5408 | ||
c0ff4b85 | 5409 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5410 | { |
54f72fe0 | 5411 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
5412 | } |
5413 | ||
33327948 KH |
5414 | static struct mem_cgroup *mem_cgroup_alloc(void) |
5415 | { | |
d79154bb | 5416 | struct mem_cgroup *memcg; |
8ff69e2c | 5417 | size_t size; |
33327948 | 5418 | |
8ff69e2c VD |
5419 | size = sizeof(struct mem_cgroup); |
5420 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 5421 | |
8ff69e2c | 5422 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 5423 | if (!memcg) |
e7bbcdf3 DC |
5424 | return NULL; |
5425 | ||
d79154bb HD |
5426 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
5427 | if (!memcg->stat) | |
d2e61b8d | 5428 | goto out_free; |
d79154bb HD |
5429 | spin_lock_init(&memcg->pcp_counter_lock); |
5430 | return memcg; | |
d2e61b8d DC |
5431 | |
5432 | out_free: | |
8ff69e2c | 5433 | kfree(memcg); |
d2e61b8d | 5434 | return NULL; |
33327948 KH |
5435 | } |
5436 | ||
59927fb9 | 5437 | /* |
c8b2a36f GC |
5438 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
5439 | * (scanning all at force_empty is too costly...) | |
5440 | * | |
5441 | * Instead of clearing all references at force_empty, we remember | |
5442 | * the number of reference from swap_cgroup and free mem_cgroup when | |
5443 | * it goes down to 0. | |
5444 | * | |
5445 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 5446 | */ |
c8b2a36f GC |
5447 | |
5448 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 5449 | { |
c8b2a36f | 5450 | int node; |
59927fb9 | 5451 | |
bb4cc1a8 | 5452 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
5453 | |
5454 | for_each_node(node) | |
5455 | free_mem_cgroup_per_zone_info(memcg, node); | |
5456 | ||
5457 | free_percpu(memcg->stat); | |
5458 | ||
3f134619 GC |
5459 | /* |
5460 | * We need to make sure that (at least for now), the jump label | |
5461 | * destruction code runs outside of the cgroup lock. This is because | |
5462 | * get_online_cpus(), which is called from the static_branch update, | |
5463 | * can't be called inside the cgroup_lock. cpusets are the ones | |
5464 | * enforcing this dependency, so if they ever change, we might as well. | |
5465 | * | |
5466 | * schedule_work() will guarantee this happens. Be careful if you need | |
5467 | * to move this code around, and make sure it is outside | |
5468 | * the cgroup_lock. | |
5469 | */ | |
a8964b9b | 5470 | disarm_static_keys(memcg); |
8ff69e2c | 5471 | kfree(memcg); |
59927fb9 | 5472 | } |
3afe36b1 | 5473 | |
7bcc1bb1 DN |
5474 | /* |
5475 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
5476 | */ | |
e1aab161 | 5477 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 5478 | { |
c0ff4b85 | 5479 | if (!memcg->res.parent) |
7bcc1bb1 | 5480 | return NULL; |
c0ff4b85 | 5481 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 5482 | } |
e1aab161 | 5483 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 5484 | |
bb4cc1a8 AM |
5485 | static void __init mem_cgroup_soft_limit_tree_init(void) |
5486 | { | |
5487 | struct mem_cgroup_tree_per_node *rtpn; | |
5488 | struct mem_cgroup_tree_per_zone *rtpz; | |
5489 | int tmp, node, zone; | |
5490 | ||
5491 | for_each_node(node) { | |
5492 | tmp = node; | |
5493 | if (!node_state(node, N_NORMAL_MEMORY)) | |
5494 | tmp = -1; | |
5495 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
5496 | BUG_ON(!rtpn); | |
5497 | ||
5498 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5499 | ||
5500 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5501 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5502 | rtpz->rb_root = RB_ROOT; | |
5503 | spin_lock_init(&rtpz->lock); | |
5504 | } | |
5505 | } | |
5506 | } | |
5507 | ||
0eb253e2 | 5508 | static struct cgroup_subsys_state * __ref |
eb95419b | 5509 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 5510 | { |
d142e3e6 | 5511 | struct mem_cgroup *memcg; |
04046e1a | 5512 | long error = -ENOMEM; |
6d12e2d8 | 5513 | int node; |
8cdea7c0 | 5514 | |
c0ff4b85 R |
5515 | memcg = mem_cgroup_alloc(); |
5516 | if (!memcg) | |
04046e1a | 5517 | return ERR_PTR(error); |
78fb7466 | 5518 | |
3ed28fa1 | 5519 | for_each_node(node) |
c0ff4b85 | 5520 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5521 | goto free_out; |
f64c3f54 | 5522 | |
c077719b | 5523 | /* root ? */ |
eb95419b | 5524 | if (parent_css == NULL) { |
a41c58a6 | 5525 | root_mem_cgroup = memcg; |
d142e3e6 GC |
5526 | res_counter_init(&memcg->res, NULL); |
5527 | res_counter_init(&memcg->memsw, NULL); | |
5528 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 5529 | } |
28dbc4b6 | 5530 | |
d142e3e6 GC |
5531 | memcg->last_scanned_node = MAX_NUMNODES; |
5532 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
5533 | memcg->move_charge_at_immigrate = 0; |
5534 | mutex_init(&memcg->thresholds_lock); | |
5535 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5536 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5537 | INIT_LIST_HEAD(&memcg->event_list); |
5538 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
5539 | |
5540 | return &memcg->css; | |
5541 | ||
5542 | free_out: | |
5543 | __mem_cgroup_free(memcg); | |
5544 | return ERR_PTR(error); | |
5545 | } | |
5546 | ||
5547 | static int | |
eb95419b | 5548 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 5549 | { |
eb95419b | 5550 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 5551 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
d142e3e6 | 5552 | |
15a4c835 | 5553 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
5554 | return -ENOSPC; |
5555 | ||
63876986 | 5556 | if (!parent) |
d142e3e6 GC |
5557 | return 0; |
5558 | ||
0999821b | 5559 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
5560 | |
5561 | memcg->use_hierarchy = parent->use_hierarchy; | |
5562 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5563 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5564 | ||
5565 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
5566 | res_counter_init(&memcg->res, &parent->res); |
5567 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 5568 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 5569 | |
7bcc1bb1 | 5570 | /* |
8d76a979 LZ |
5571 | * No need to take a reference to the parent because cgroup |
5572 | * core guarantees its existence. | |
7bcc1bb1 | 5573 | */ |
18f59ea7 | 5574 | } else { |
ce00a967 JW |
5575 | res_counter_init(&memcg->res, NULL); |
5576 | res_counter_init(&memcg->memsw, NULL); | |
5577 | res_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
5578 | /* |
5579 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5580 | * much sense so let cgroup subsystem know about this | |
5581 | * unfortunate state in our controller. | |
5582 | */ | |
d142e3e6 | 5583 | if (parent != root_mem_cgroup) |
073219e9 | 5584 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5585 | } |
0999821b | 5586 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 5587 | |
073219e9 | 5588 | return memcg_init_kmem(memcg, &memory_cgrp_subsys); |
8cdea7c0 BS |
5589 | } |
5590 | ||
5f578161 MH |
5591 | /* |
5592 | * Announce all parents that a group from their hierarchy is gone. | |
5593 | */ | |
5594 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
5595 | { | |
5596 | struct mem_cgroup *parent = memcg; | |
5597 | ||
5598 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 5599 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
5600 | |
5601 | /* | |
5602 | * if the root memcg is not hierarchical we have to check it | |
5603 | * explicitely. | |
5604 | */ | |
5605 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 5606 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
5607 | } |
5608 | ||
eb95419b | 5609 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5610 | { |
eb95419b | 5611 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5612 | struct mem_cgroup_event *event, *tmp; |
4fb1a86f | 5613 | struct cgroup_subsys_state *iter; |
79bd9814 TH |
5614 | |
5615 | /* | |
5616 | * Unregister events and notify userspace. | |
5617 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5618 | * directory to avoid race between userspace and kernelspace. | |
5619 | */ | |
fba94807 TH |
5620 | spin_lock(&memcg->event_list_lock); |
5621 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5622 | list_del_init(&event->list); |
5623 | schedule_work(&event->remove); | |
5624 | } | |
fba94807 | 5625 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5626 | |
10d5ebf4 LZ |
5627 | kmem_cgroup_css_offline(memcg); |
5628 | ||
5f578161 | 5629 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
4fb1a86f FB |
5630 | |
5631 | /* | |
5632 | * This requires that offlining is serialized. Right now that is | |
5633 | * guaranteed because css_killed_work_fn() holds the cgroup_mutex. | |
5634 | */ | |
5635 | css_for_each_descendant_post(iter, css) | |
5636 | mem_cgroup_reparent_charges(mem_cgroup_from_css(iter)); | |
5637 | ||
776ed0f0 | 5638 | memcg_unregister_all_caches(memcg); |
33cb876e | 5639 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
5640 | } |
5641 | ||
eb95419b | 5642 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5643 | { |
eb95419b | 5644 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
96f1c58d JW |
5645 | /* |
5646 | * XXX: css_offline() would be where we should reparent all | |
5647 | * memory to prepare the cgroup for destruction. However, | |
ec903c0c | 5648 | * memcg does not do css_tryget_online() and res_counter charging |
96f1c58d JW |
5649 | * under the same RCU lock region, which means that charging |
5650 | * could race with offlining. Offlining only happens to | |
5651 | * cgroups with no tasks in them but charges can show up | |
5652 | * without any tasks from the swapin path when the target | |
5653 | * memcg is looked up from the swapout record and not from the | |
5654 | * current task as it usually is. A race like this can leak | |
5655 | * charges and put pages with stale cgroup pointers into | |
5656 | * circulation: | |
5657 | * | |
5658 | * #0 #1 | |
5659 | * lookup_swap_cgroup_id() | |
5660 | * rcu_read_lock() | |
5661 | * mem_cgroup_lookup() | |
ec903c0c | 5662 | * css_tryget_online() |
96f1c58d | 5663 | * rcu_read_unlock() |
ec903c0c | 5664 | * disable css_tryget_online() |
96f1c58d JW |
5665 | * call_rcu() |
5666 | * offline_css() | |
5667 | * reparent_charges() | |
5668 | * res_counter_charge() | |
5669 | * css_put() | |
5670 | * css_free() | |
5671 | * pc->mem_cgroup = dead memcg | |
5672 | * add page to lru | |
5673 | * | |
5674 | * The bulk of the charges are still moved in offline_css() to | |
5675 | * avoid pinning a lot of pages in case a long-term reference | |
5676 | * like a swapout record is deferring the css_free() to long | |
5677 | * after offlining. But this makes sure we catch any charges | |
5678 | * made after offlining: | |
5679 | */ | |
5680 | mem_cgroup_reparent_charges(memcg); | |
c268e994 | 5681 | |
10d5ebf4 | 5682 | memcg_destroy_kmem(memcg); |
465939a1 | 5683 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
5684 | } |
5685 | ||
1ced953b TH |
5686 | /** |
5687 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5688 | * @css: the target css | |
5689 | * | |
5690 | * Reset the states of the mem_cgroup associated with @css. This is | |
5691 | * invoked when the userland requests disabling on the default hierarchy | |
5692 | * but the memcg is pinned through dependency. The memcg should stop | |
5693 | * applying policies and should revert to the vanilla state as it may be | |
5694 | * made visible again. | |
5695 | * | |
5696 | * The current implementation only resets the essential configurations. | |
5697 | * This needs to be expanded to cover all the visible parts. | |
5698 | */ | |
5699 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5700 | { | |
5701 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5702 | ||
5703 | mem_cgroup_resize_limit(memcg, ULLONG_MAX); | |
5704 | mem_cgroup_resize_memsw_limit(memcg, ULLONG_MAX); | |
5705 | memcg_update_kmem_limit(memcg, ULLONG_MAX); | |
5706 | res_counter_set_soft_limit(&memcg->res, ULLONG_MAX); | |
5707 | } | |
5708 | ||
02491447 | 5709 | #ifdef CONFIG_MMU |
7dc74be0 | 5710 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5711 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5712 | { |
05b84301 | 5713 | int ret; |
9476db97 JW |
5714 | |
5715 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 5716 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 5717 | if (!ret) { |
854ffa8d | 5718 | mc.precharge += count; |
854ffa8d DN |
5719 | return ret; |
5720 | } | |
692e7c45 | 5721 | if (ret == -EINTR) { |
00501b53 | 5722 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
5723 | return ret; |
5724 | } | |
9476db97 JW |
5725 | |
5726 | /* Try charges one by one with reclaim */ | |
854ffa8d | 5727 | while (count--) { |
00501b53 | 5728 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
5729 | /* |
5730 | * In case of failure, any residual charges against | |
5731 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
5732 | * later on. However, cancel any charges that are |
5733 | * bypassed to root right away or they'll be lost. | |
9476db97 | 5734 | */ |
692e7c45 | 5735 | if (ret == -EINTR) |
00501b53 | 5736 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 5737 | if (ret) |
38c5d72f | 5738 | return ret; |
854ffa8d | 5739 | mc.precharge++; |
9476db97 | 5740 | cond_resched(); |
854ffa8d | 5741 | } |
9476db97 | 5742 | return 0; |
4ffef5fe DN |
5743 | } |
5744 | ||
5745 | /** | |
8d32ff84 | 5746 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5747 | * @vma: the vma the pte to be checked belongs |
5748 | * @addr: the address corresponding to the pte to be checked | |
5749 | * @ptent: the pte to be checked | |
02491447 | 5750 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5751 | * |
5752 | * Returns | |
5753 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5754 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5755 | * move charge. if @target is not NULL, the page is stored in target->page | |
5756 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5757 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5758 | * target for charge migration. if @target is not NULL, the entry is stored | |
5759 | * in target->ent. | |
4ffef5fe DN |
5760 | * |
5761 | * Called with pte lock held. | |
5762 | */ | |
4ffef5fe DN |
5763 | union mc_target { |
5764 | struct page *page; | |
02491447 | 5765 | swp_entry_t ent; |
4ffef5fe DN |
5766 | }; |
5767 | ||
4ffef5fe | 5768 | enum mc_target_type { |
8d32ff84 | 5769 | MC_TARGET_NONE = 0, |
4ffef5fe | 5770 | MC_TARGET_PAGE, |
02491447 | 5771 | MC_TARGET_SWAP, |
4ffef5fe DN |
5772 | }; |
5773 | ||
90254a65 DN |
5774 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5775 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5776 | { |
90254a65 | 5777 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5778 | |
90254a65 DN |
5779 | if (!page || !page_mapped(page)) |
5780 | return NULL; | |
5781 | if (PageAnon(page)) { | |
5782 | /* we don't move shared anon */ | |
4b91355e | 5783 | if (!move_anon()) |
90254a65 | 5784 | return NULL; |
87946a72 DN |
5785 | } else if (!move_file()) |
5786 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5787 | return NULL; |
5788 | if (!get_page_unless_zero(page)) | |
5789 | return NULL; | |
5790 | ||
5791 | return page; | |
5792 | } | |
5793 | ||
4b91355e | 5794 | #ifdef CONFIG_SWAP |
90254a65 DN |
5795 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5796 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5797 | { | |
90254a65 DN |
5798 | struct page *page = NULL; |
5799 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5800 | ||
5801 | if (!move_anon() || non_swap_entry(ent)) | |
5802 | return NULL; | |
4b91355e KH |
5803 | /* |
5804 | * Because lookup_swap_cache() updates some statistics counter, | |
5805 | * we call find_get_page() with swapper_space directly. | |
5806 | */ | |
33806f06 | 5807 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
5808 | if (do_swap_account) |
5809 | entry->val = ent.val; | |
5810 | ||
5811 | return page; | |
5812 | } | |
4b91355e KH |
5813 | #else |
5814 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5815 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5816 | { | |
5817 | return NULL; | |
5818 | } | |
5819 | #endif | |
90254a65 | 5820 | |
87946a72 DN |
5821 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5822 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5823 | { | |
5824 | struct page *page = NULL; | |
87946a72 DN |
5825 | struct address_space *mapping; |
5826 | pgoff_t pgoff; | |
5827 | ||
5828 | if (!vma->vm_file) /* anonymous vma */ | |
5829 | return NULL; | |
5830 | if (!move_file()) | |
5831 | return NULL; | |
5832 | ||
87946a72 DN |
5833 | mapping = vma->vm_file->f_mapping; |
5834 | if (pte_none(ptent)) | |
5835 | pgoff = linear_page_index(vma, addr); | |
5836 | else /* pte_file(ptent) is true */ | |
5837 | pgoff = pte_to_pgoff(ptent); | |
5838 | ||
5839 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5840 | #ifdef CONFIG_SWAP |
5841 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5842 | if (shmem_mapping(mapping)) { |
5843 | page = find_get_entry(mapping, pgoff); | |
5844 | if (radix_tree_exceptional_entry(page)) { | |
5845 | swp_entry_t swp = radix_to_swp_entry(page); | |
5846 | if (do_swap_account) | |
5847 | *entry = swp; | |
5848 | page = find_get_page(swap_address_space(swp), swp.val); | |
5849 | } | |
5850 | } else | |
5851 | page = find_get_page(mapping, pgoff); | |
5852 | #else | |
5853 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5854 | #endif |
87946a72 DN |
5855 | return page; |
5856 | } | |
5857 | ||
8d32ff84 | 5858 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5859 | unsigned long addr, pte_t ptent, union mc_target *target) |
5860 | { | |
5861 | struct page *page = NULL; | |
5862 | struct page_cgroup *pc; | |
8d32ff84 | 5863 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5864 | swp_entry_t ent = { .val = 0 }; |
5865 | ||
5866 | if (pte_present(ptent)) | |
5867 | page = mc_handle_present_pte(vma, addr, ptent); | |
5868 | else if (is_swap_pte(ptent)) | |
5869 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5870 | else if (pte_none(ptent) || pte_file(ptent)) |
5871 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5872 | |
5873 | if (!page && !ent.val) | |
8d32ff84 | 5874 | return ret; |
02491447 DN |
5875 | if (page) { |
5876 | pc = lookup_page_cgroup(page); | |
5877 | /* | |
0a31bc97 JW |
5878 | * Do only loose check w/o serialization. |
5879 | * mem_cgroup_move_account() checks the pc is valid or | |
5880 | * not under LRU exclusion. | |
02491447 DN |
5881 | */ |
5882 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5883 | ret = MC_TARGET_PAGE; | |
5884 | if (target) | |
5885 | target->page = page; | |
5886 | } | |
5887 | if (!ret || !target) | |
5888 | put_page(page); | |
5889 | } | |
90254a65 DN |
5890 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5891 | if (ent.val && !ret && | |
34c00c31 | 5892 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5893 | ret = MC_TARGET_SWAP; |
5894 | if (target) | |
5895 | target->ent = ent; | |
4ffef5fe | 5896 | } |
4ffef5fe DN |
5897 | return ret; |
5898 | } | |
5899 | ||
12724850 NH |
5900 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5901 | /* | |
5902 | * We don't consider swapping or file mapped pages because THP does not | |
5903 | * support them for now. | |
5904 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5905 | */ | |
5906 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5907 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5908 | { | |
5909 | struct page *page = NULL; | |
5910 | struct page_cgroup *pc; | |
5911 | enum mc_target_type ret = MC_TARGET_NONE; | |
5912 | ||
5913 | page = pmd_page(pmd); | |
309381fe | 5914 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
5915 | if (!move_anon()) |
5916 | return ret; | |
5917 | pc = lookup_page_cgroup(page); | |
5918 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5919 | ret = MC_TARGET_PAGE; | |
5920 | if (target) { | |
5921 | get_page(page); | |
5922 | target->page = page; | |
5923 | } | |
5924 | } | |
5925 | return ret; | |
5926 | } | |
5927 | #else | |
5928 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5929 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5930 | { | |
5931 | return MC_TARGET_NONE; | |
5932 | } | |
5933 | #endif | |
5934 | ||
4ffef5fe DN |
5935 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5936 | unsigned long addr, unsigned long end, | |
5937 | struct mm_walk *walk) | |
5938 | { | |
5939 | struct vm_area_struct *vma = walk->private; | |
5940 | pte_t *pte; | |
5941 | spinlock_t *ptl; | |
5942 | ||
bf929152 | 5943 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
5944 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5945 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5946 | spin_unlock(ptl); |
1a5a9906 | 5947 | return 0; |
12724850 | 5948 | } |
03319327 | 5949 | |
45f83cef AA |
5950 | if (pmd_trans_unstable(pmd)) |
5951 | return 0; | |
4ffef5fe DN |
5952 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5953 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5954 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5955 | mc.precharge++; /* increment precharge temporarily */ |
5956 | pte_unmap_unlock(pte - 1, ptl); | |
5957 | cond_resched(); | |
5958 | ||
7dc74be0 DN |
5959 | return 0; |
5960 | } | |
5961 | ||
4ffef5fe DN |
5962 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5963 | { | |
5964 | unsigned long precharge; | |
5965 | struct vm_area_struct *vma; | |
5966 | ||
dfe076b0 | 5967 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5968 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5969 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5970 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5971 | .mm = mm, | |
5972 | .private = vma, | |
5973 | }; | |
5974 | if (is_vm_hugetlb_page(vma)) | |
5975 | continue; | |
4ffef5fe DN |
5976 | walk_page_range(vma->vm_start, vma->vm_end, |
5977 | &mem_cgroup_count_precharge_walk); | |
5978 | } | |
dfe076b0 | 5979 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5980 | |
5981 | precharge = mc.precharge; | |
5982 | mc.precharge = 0; | |
5983 | ||
5984 | return precharge; | |
5985 | } | |
5986 | ||
4ffef5fe DN |
5987 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5988 | { | |
dfe076b0 DN |
5989 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5990 | ||
5991 | VM_BUG_ON(mc.moving_task); | |
5992 | mc.moving_task = current; | |
5993 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5994 | } |
5995 | ||
dfe076b0 DN |
5996 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5997 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5998 | { |
2bd9bb20 KH |
5999 | struct mem_cgroup *from = mc.from; |
6000 | struct mem_cgroup *to = mc.to; | |
4050377b | 6001 | int i; |
2bd9bb20 | 6002 | |
4ffef5fe | 6003 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 6004 | if (mc.precharge) { |
00501b53 | 6005 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
6006 | mc.precharge = 0; |
6007 | } | |
6008 | /* | |
6009 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6010 | * we must uncharge here. | |
6011 | */ | |
6012 | if (mc.moved_charge) { | |
00501b53 | 6013 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 6014 | mc.moved_charge = 0; |
4ffef5fe | 6015 | } |
483c30b5 DN |
6016 | /* we must fixup refcnts and charges */ |
6017 | if (mc.moved_swap) { | |
483c30b5 | 6018 | /* uncharge swap account from the old cgroup */ |
ce00a967 JW |
6019 | if (!mem_cgroup_is_root(mc.from)) |
6020 | res_counter_uncharge(&mc.from->memsw, | |
6021 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6022 | |
6023 | for (i = 0; i < mc.moved_swap; i++) | |
6024 | css_put(&mc.from->css); | |
483c30b5 | 6025 | |
05b84301 JW |
6026 | /* |
6027 | * we charged both to->res and to->memsw, so we should | |
6028 | * uncharge to->res. | |
6029 | */ | |
ce00a967 JW |
6030 | if (!mem_cgroup_is_root(mc.to)) |
6031 | res_counter_uncharge(&mc.to->res, | |
6032 | PAGE_SIZE * mc.moved_swap); | |
4050377b | 6033 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6034 | mc.moved_swap = 0; |
6035 | } | |
dfe076b0 DN |
6036 | memcg_oom_recover(from); |
6037 | memcg_oom_recover(to); | |
6038 | wake_up_all(&mc.waitq); | |
6039 | } | |
6040 | ||
6041 | static void mem_cgroup_clear_mc(void) | |
6042 | { | |
6043 | struct mem_cgroup *from = mc.from; | |
6044 | ||
6045 | /* | |
6046 | * we must clear moving_task before waking up waiters at the end of | |
6047 | * task migration. | |
6048 | */ | |
6049 | mc.moving_task = NULL; | |
6050 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6051 | spin_lock(&mc.lock); |
4ffef5fe DN |
6052 | mc.from = NULL; |
6053 | mc.to = NULL; | |
2bd9bb20 | 6054 | spin_unlock(&mc.lock); |
32047e2a | 6055 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6056 | } |
6057 | ||
eb95419b | 6058 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6059 | struct cgroup_taskset *tset) |
7dc74be0 | 6060 | { |
2f7ee569 | 6061 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6062 | int ret = 0; |
eb95419b | 6063 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 6064 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6065 | |
ee5e8472 GC |
6066 | /* |
6067 | * We are now commited to this value whatever it is. Changes in this | |
6068 | * tunable will only affect upcoming migrations, not the current one. | |
6069 | * So we need to save it, and keep it going. | |
6070 | */ | |
6071 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
6072 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
6073 | struct mm_struct *mm; |
6074 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
6075 | ||
c0ff4b85 | 6076 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
6077 | |
6078 | mm = get_task_mm(p); | |
6079 | if (!mm) | |
6080 | return 0; | |
7dc74be0 | 6081 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
6082 | if (mm->owner == p) { |
6083 | VM_BUG_ON(mc.from); | |
6084 | VM_BUG_ON(mc.to); | |
6085 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 6086 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 6087 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 6088 | mem_cgroup_start_move(from); |
2bd9bb20 | 6089 | spin_lock(&mc.lock); |
4ffef5fe | 6090 | mc.from = from; |
c0ff4b85 | 6091 | mc.to = memcg; |
ee5e8472 | 6092 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 6093 | spin_unlock(&mc.lock); |
dfe076b0 | 6094 | /* We set mc.moving_task later */ |
4ffef5fe DN |
6095 | |
6096 | ret = mem_cgroup_precharge_mc(mm); | |
6097 | if (ret) | |
6098 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
6099 | } |
6100 | mmput(mm); | |
7dc74be0 DN |
6101 | } |
6102 | return ret; | |
6103 | } | |
6104 | ||
eb95419b | 6105 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6106 | struct cgroup_taskset *tset) |
7dc74be0 | 6107 | { |
4ffef5fe | 6108 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
6109 | } |
6110 | ||
4ffef5fe DN |
6111 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6112 | unsigned long addr, unsigned long end, | |
6113 | struct mm_walk *walk) | |
7dc74be0 | 6114 | { |
4ffef5fe DN |
6115 | int ret = 0; |
6116 | struct vm_area_struct *vma = walk->private; | |
6117 | pte_t *pte; | |
6118 | spinlock_t *ptl; | |
12724850 NH |
6119 | enum mc_target_type target_type; |
6120 | union mc_target target; | |
6121 | struct page *page; | |
6122 | struct page_cgroup *pc; | |
4ffef5fe | 6123 | |
12724850 NH |
6124 | /* |
6125 | * We don't take compound_lock() here but no race with splitting thp | |
6126 | * happens because: | |
6127 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
6128 | * under splitting, which means there's no concurrent thp split, | |
6129 | * - if another thread runs into split_huge_page() just after we | |
6130 | * entered this if-block, the thread must wait for page table lock | |
6131 | * to be unlocked in __split_huge_page_splitting(), where the main | |
6132 | * part of thp split is not executed yet. | |
6133 | */ | |
bf929152 | 6134 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 6135 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6136 | spin_unlock(ptl); |
12724850 NH |
6137 | return 0; |
6138 | } | |
6139 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6140 | if (target_type == MC_TARGET_PAGE) { | |
6141 | page = target.page; | |
6142 | if (!isolate_lru_page(page)) { | |
6143 | pc = lookup_page_cgroup(page); | |
6144 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6145 | pc, mc.from, mc.to)) { |
12724850 NH |
6146 | mc.precharge -= HPAGE_PMD_NR; |
6147 | mc.moved_charge += HPAGE_PMD_NR; | |
6148 | } | |
6149 | putback_lru_page(page); | |
6150 | } | |
6151 | put_page(page); | |
6152 | } | |
bf929152 | 6153 | spin_unlock(ptl); |
1a5a9906 | 6154 | return 0; |
12724850 NH |
6155 | } |
6156 | ||
45f83cef AA |
6157 | if (pmd_trans_unstable(pmd)) |
6158 | return 0; | |
4ffef5fe DN |
6159 | retry: |
6160 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6161 | for (; addr != end; addr += PAGE_SIZE) { | |
6162 | pte_t ptent = *(pte++); | |
02491447 | 6163 | swp_entry_t ent; |
4ffef5fe DN |
6164 | |
6165 | if (!mc.precharge) | |
6166 | break; | |
6167 | ||
8d32ff84 | 6168 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6169 | case MC_TARGET_PAGE: |
6170 | page = target.page; | |
6171 | if (isolate_lru_page(page)) | |
6172 | goto put; | |
6173 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6174 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6175 | mc.from, mc.to)) { |
4ffef5fe | 6176 | mc.precharge--; |
854ffa8d DN |
6177 | /* we uncharge from mc.from later. */ |
6178 | mc.moved_charge++; | |
4ffef5fe DN |
6179 | } |
6180 | putback_lru_page(page); | |
8d32ff84 | 6181 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6182 | put_page(page); |
6183 | break; | |
02491447 DN |
6184 | case MC_TARGET_SWAP: |
6185 | ent = target.ent; | |
e91cbb42 | 6186 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6187 | mc.precharge--; |
483c30b5 DN |
6188 | /* we fixup refcnts and charges later. */ |
6189 | mc.moved_swap++; | |
6190 | } | |
02491447 | 6191 | break; |
4ffef5fe DN |
6192 | default: |
6193 | break; | |
6194 | } | |
6195 | } | |
6196 | pte_unmap_unlock(pte - 1, ptl); | |
6197 | cond_resched(); | |
6198 | ||
6199 | if (addr != end) { | |
6200 | /* | |
6201 | * We have consumed all precharges we got in can_attach(). | |
6202 | * We try charge one by one, but don't do any additional | |
6203 | * charges to mc.to if we have failed in charge once in attach() | |
6204 | * phase. | |
6205 | */ | |
854ffa8d | 6206 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6207 | if (!ret) |
6208 | goto retry; | |
6209 | } | |
6210 | ||
6211 | return ret; | |
6212 | } | |
6213 | ||
6214 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6215 | { | |
6216 | struct vm_area_struct *vma; | |
6217 | ||
6218 | lru_add_drain_all(); | |
dfe076b0 DN |
6219 | retry: |
6220 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6221 | /* | |
6222 | * Someone who are holding the mmap_sem might be waiting in | |
6223 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6224 | * and retry. Because we cancel precharges, we might not be able | |
6225 | * to move enough charges, but moving charge is a best-effort | |
6226 | * feature anyway, so it wouldn't be a big problem. | |
6227 | */ | |
6228 | __mem_cgroup_clear_mc(); | |
6229 | cond_resched(); | |
6230 | goto retry; | |
6231 | } | |
4ffef5fe DN |
6232 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6233 | int ret; | |
6234 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6235 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6236 | .mm = mm, | |
6237 | .private = vma, | |
6238 | }; | |
6239 | if (is_vm_hugetlb_page(vma)) | |
6240 | continue; | |
4ffef5fe DN |
6241 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6242 | &mem_cgroup_move_charge_walk); | |
6243 | if (ret) | |
6244 | /* | |
6245 | * means we have consumed all precharges and failed in | |
6246 | * doing additional charge. Just abandon here. | |
6247 | */ | |
6248 | break; | |
6249 | } | |
dfe076b0 | 6250 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6251 | } |
6252 | ||
eb95419b | 6253 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6254 | struct cgroup_taskset *tset) |
67e465a7 | 6255 | { |
2f7ee569 | 6256 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6257 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6258 | |
dfe076b0 | 6259 | if (mm) { |
a433658c KM |
6260 | if (mc.to) |
6261 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6262 | mmput(mm); |
6263 | } | |
a433658c KM |
6264 | if (mc.to) |
6265 | mem_cgroup_clear_mc(); | |
67e465a7 | 6266 | } |
5cfb80a7 | 6267 | #else /* !CONFIG_MMU */ |
eb95419b | 6268 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6269 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6270 | { |
6271 | return 0; | |
6272 | } | |
eb95419b | 6273 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6274 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6275 | { |
6276 | } | |
eb95419b | 6277 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6278 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6279 | { |
6280 | } | |
6281 | #endif | |
67e465a7 | 6282 | |
f00baae7 TH |
6283 | /* |
6284 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6285 | * to verify whether we're attached to the default hierarchy on each mount |
6286 | * attempt. | |
f00baae7 | 6287 | */ |
eb95419b | 6288 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6289 | { |
6290 | /* | |
aa6ec29b | 6291 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6292 | * guarantees that @root doesn't have any children, so turning it |
6293 | * on for the root memcg is enough. | |
6294 | */ | |
aa6ec29b | 6295 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 6296 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
6297 | } |
6298 | ||
073219e9 | 6299 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6300 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6301 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6302 | .css_offline = mem_cgroup_css_offline, |
6303 | .css_free = mem_cgroup_css_free, | |
1ced953b | 6304 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6305 | .can_attach = mem_cgroup_can_attach, |
6306 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6307 | .attach = mem_cgroup_move_task, |
f00baae7 | 6308 | .bind = mem_cgroup_bind, |
5577964e | 6309 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 6310 | .early_init = 0, |
8cdea7c0 | 6311 | }; |
c077719b | 6312 | |
c255a458 | 6313 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6314 | static int __init enable_swap_account(char *s) |
6315 | { | |
a2c8990a | 6316 | if (!strcmp(s, "1")) |
a42c390c | 6317 | really_do_swap_account = 1; |
a2c8990a | 6318 | else if (!strcmp(s, "0")) |
a42c390c MH |
6319 | really_do_swap_account = 0; |
6320 | return 1; | |
6321 | } | |
a2c8990a | 6322 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6323 | |
2d11085e MH |
6324 | static void __init memsw_file_init(void) |
6325 | { | |
2cf669a5 TH |
6326 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6327 | memsw_cgroup_files)); | |
6acc8b02 MH |
6328 | } |
6329 | ||
6330 | static void __init enable_swap_cgroup(void) | |
6331 | { | |
6332 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6333 | do_swap_account = 1; | |
6334 | memsw_file_init(); | |
6335 | } | |
2d11085e | 6336 | } |
6acc8b02 | 6337 | |
2d11085e | 6338 | #else |
6acc8b02 | 6339 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6340 | { |
6341 | } | |
c077719b | 6342 | #endif |
2d11085e | 6343 | |
0a31bc97 JW |
6344 | #ifdef CONFIG_MEMCG_SWAP |
6345 | /** | |
6346 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6347 | * @page: page whose memsw charge to transfer | |
6348 | * @entry: swap entry to move the charge to | |
6349 | * | |
6350 | * Transfer the memsw charge of @page to @entry. | |
6351 | */ | |
6352 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6353 | { | |
6354 | struct page_cgroup *pc; | |
6355 | unsigned short oldid; | |
6356 | ||
6357 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6358 | VM_BUG_ON_PAGE(page_count(page), page); | |
6359 | ||
6360 | if (!do_swap_account) | |
6361 | return; | |
6362 | ||
6363 | pc = lookup_page_cgroup(page); | |
6364 | ||
6365 | /* Readahead page, never charged */ | |
6366 | if (!PageCgroupUsed(pc)) | |
6367 | return; | |
6368 | ||
6369 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEMSW), page); | |
6370 | ||
6371 | oldid = swap_cgroup_record(entry, mem_cgroup_id(pc->mem_cgroup)); | |
6372 | VM_BUG_ON_PAGE(oldid, page); | |
6373 | ||
6374 | pc->flags &= ~PCG_MEMSW; | |
6375 | css_get(&pc->mem_cgroup->css); | |
6376 | mem_cgroup_swap_statistics(pc->mem_cgroup, true); | |
6377 | } | |
6378 | ||
6379 | /** | |
6380 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
6381 | * @entry: swap entry to uncharge | |
6382 | * | |
6383 | * Drop the memsw charge associated with @entry. | |
6384 | */ | |
6385 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
6386 | { | |
6387 | struct mem_cgroup *memcg; | |
6388 | unsigned short id; | |
6389 | ||
6390 | if (!do_swap_account) | |
6391 | return; | |
6392 | ||
6393 | id = swap_cgroup_record(entry, 0); | |
6394 | rcu_read_lock(); | |
6395 | memcg = mem_cgroup_lookup(id); | |
6396 | if (memcg) { | |
ce00a967 JW |
6397 | if (!mem_cgroup_is_root(memcg)) |
6398 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
0a31bc97 JW |
6399 | mem_cgroup_swap_statistics(memcg, false); |
6400 | css_put(&memcg->css); | |
6401 | } | |
6402 | rcu_read_unlock(); | |
6403 | } | |
6404 | #endif | |
6405 | ||
00501b53 JW |
6406 | /** |
6407 | * mem_cgroup_try_charge - try charging a page | |
6408 | * @page: page to charge | |
6409 | * @mm: mm context of the victim | |
6410 | * @gfp_mask: reclaim mode | |
6411 | * @memcgp: charged memcg return | |
6412 | * | |
6413 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6414 | * pages according to @gfp_mask if necessary. | |
6415 | * | |
6416 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
6417 | * Otherwise, an error code is returned. | |
6418 | * | |
6419 | * After page->mapping has been set up, the caller must finalize the | |
6420 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
6421 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
6422 | */ | |
6423 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
6424 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
6425 | { | |
6426 | struct mem_cgroup *memcg = NULL; | |
6427 | unsigned int nr_pages = 1; | |
6428 | int ret = 0; | |
6429 | ||
6430 | if (mem_cgroup_disabled()) | |
6431 | goto out; | |
6432 | ||
6433 | if (PageSwapCache(page)) { | |
6434 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
6435 | /* | |
6436 | * Every swap fault against a single page tries to charge the | |
6437 | * page, bail as early as possible. shmem_unuse() encounters | |
6438 | * already charged pages, too. The USED bit is protected by | |
6439 | * the page lock, which serializes swap cache removal, which | |
6440 | * in turn serializes uncharging. | |
6441 | */ | |
6442 | if (PageCgroupUsed(pc)) | |
6443 | goto out; | |
6444 | } | |
6445 | ||
6446 | if (PageTransHuge(page)) { | |
6447 | nr_pages <<= compound_order(page); | |
6448 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6449 | } | |
6450 | ||
6451 | if (do_swap_account && PageSwapCache(page)) | |
6452 | memcg = try_get_mem_cgroup_from_page(page); | |
6453 | if (!memcg) | |
6454 | memcg = get_mem_cgroup_from_mm(mm); | |
6455 | ||
6456 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6457 | ||
6458 | css_put(&memcg->css); | |
6459 | ||
6460 | if (ret == -EINTR) { | |
6461 | memcg = root_mem_cgroup; | |
6462 | ret = 0; | |
6463 | } | |
6464 | out: | |
6465 | *memcgp = memcg; | |
6466 | return ret; | |
6467 | } | |
6468 | ||
6469 | /** | |
6470 | * mem_cgroup_commit_charge - commit a page charge | |
6471 | * @page: page to charge | |
6472 | * @memcg: memcg to charge the page to | |
6473 | * @lrucare: page might be on LRU already | |
6474 | * | |
6475 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6476 | * after page->mapping has been set up. This must happen atomically | |
6477 | * as part of the page instantiation, i.e. under the page table lock | |
6478 | * for anonymous pages, under the page lock for page and swap cache. | |
6479 | * | |
6480 | * In addition, the page must not be on the LRU during the commit, to | |
6481 | * prevent racing with task migration. If it might be, use @lrucare. | |
6482 | * | |
6483 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6484 | */ | |
6485 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
6486 | bool lrucare) | |
6487 | { | |
6488 | unsigned int nr_pages = 1; | |
6489 | ||
6490 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6491 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6492 | ||
6493 | if (mem_cgroup_disabled()) | |
6494 | return; | |
6495 | /* | |
6496 | * Swap faults will attempt to charge the same page multiple | |
6497 | * times. But reuse_swap_page() might have removed the page | |
6498 | * from swapcache already, so we can't check PageSwapCache(). | |
6499 | */ | |
6500 | if (!memcg) | |
6501 | return; | |
6502 | ||
6abb5a86 JW |
6503 | commit_charge(page, memcg, lrucare); |
6504 | ||
00501b53 JW |
6505 | if (PageTransHuge(page)) { |
6506 | nr_pages <<= compound_order(page); | |
6507 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6508 | } | |
6509 | ||
6abb5a86 JW |
6510 | local_irq_disable(); |
6511 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
6512 | memcg_check_events(memcg, page); | |
6513 | local_irq_enable(); | |
00501b53 JW |
6514 | |
6515 | if (do_swap_account && PageSwapCache(page)) { | |
6516 | swp_entry_t entry = { .val = page_private(page) }; | |
6517 | /* | |
6518 | * The swap entry might not get freed for a long time, | |
6519 | * let's not wait for it. The page already received a | |
6520 | * memory+swap charge, drop the swap entry duplicate. | |
6521 | */ | |
6522 | mem_cgroup_uncharge_swap(entry); | |
6523 | } | |
6524 | } | |
6525 | ||
6526 | /** | |
6527 | * mem_cgroup_cancel_charge - cancel a page charge | |
6528 | * @page: page to charge | |
6529 | * @memcg: memcg to charge the page to | |
6530 | * | |
6531 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6532 | */ | |
6533 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
6534 | { | |
6535 | unsigned int nr_pages = 1; | |
6536 | ||
6537 | if (mem_cgroup_disabled()) | |
6538 | return; | |
6539 | /* | |
6540 | * Swap faults will attempt to charge the same page multiple | |
6541 | * times. But reuse_swap_page() might have removed the page | |
6542 | * from swapcache already, so we can't check PageSwapCache(). | |
6543 | */ | |
6544 | if (!memcg) | |
6545 | return; | |
6546 | ||
6547 | if (PageTransHuge(page)) { | |
6548 | nr_pages <<= compound_order(page); | |
6549 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6550 | } | |
6551 | ||
6552 | cancel_charge(memcg, nr_pages); | |
6553 | } | |
6554 | ||
747db954 JW |
6555 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
6556 | unsigned long nr_mem, unsigned long nr_memsw, | |
6557 | unsigned long nr_anon, unsigned long nr_file, | |
6558 | unsigned long nr_huge, struct page *dummy_page) | |
6559 | { | |
6560 | unsigned long flags; | |
6561 | ||
ce00a967 JW |
6562 | if (!mem_cgroup_is_root(memcg)) { |
6563 | if (nr_mem) | |
6564 | res_counter_uncharge(&memcg->res, | |
6565 | nr_mem * PAGE_SIZE); | |
6566 | if (nr_memsw) | |
6567 | res_counter_uncharge(&memcg->memsw, | |
6568 | nr_memsw * PAGE_SIZE); | |
6569 | memcg_oom_recover(memcg); | |
6570 | } | |
747db954 JW |
6571 | |
6572 | local_irq_save(flags); | |
6573 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
6574 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
6575 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
6576 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
6577 | __this_cpu_add(memcg->stat->nr_page_events, nr_anon + nr_file); | |
6578 | memcg_check_events(memcg, dummy_page); | |
6579 | local_irq_restore(flags); | |
6580 | } | |
6581 | ||
6582 | static void uncharge_list(struct list_head *page_list) | |
6583 | { | |
6584 | struct mem_cgroup *memcg = NULL; | |
6585 | unsigned long nr_memsw = 0; | |
6586 | unsigned long nr_anon = 0; | |
6587 | unsigned long nr_file = 0; | |
6588 | unsigned long nr_huge = 0; | |
6589 | unsigned long pgpgout = 0; | |
6590 | unsigned long nr_mem = 0; | |
6591 | struct list_head *next; | |
6592 | struct page *page; | |
6593 | ||
6594 | next = page_list->next; | |
6595 | do { | |
6596 | unsigned int nr_pages = 1; | |
6597 | struct page_cgroup *pc; | |
6598 | ||
6599 | page = list_entry(next, struct page, lru); | |
6600 | next = page->lru.next; | |
6601 | ||
6602 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6603 | VM_BUG_ON_PAGE(page_count(page), page); | |
6604 | ||
6605 | pc = lookup_page_cgroup(page); | |
6606 | if (!PageCgroupUsed(pc)) | |
6607 | continue; | |
6608 | ||
6609 | /* | |
6610 | * Nobody should be changing or seriously looking at | |
6611 | * pc->mem_cgroup and pc->flags at this point, we have | |
6612 | * fully exclusive access to the page. | |
6613 | */ | |
6614 | ||
6615 | if (memcg != pc->mem_cgroup) { | |
6616 | if (memcg) { | |
6617 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6618 | nr_anon, nr_file, nr_huge, page); | |
6619 | pgpgout = nr_mem = nr_memsw = 0; | |
6620 | nr_anon = nr_file = nr_huge = 0; | |
6621 | } | |
6622 | memcg = pc->mem_cgroup; | |
6623 | } | |
6624 | ||
6625 | if (PageTransHuge(page)) { | |
6626 | nr_pages <<= compound_order(page); | |
6627 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6628 | nr_huge += nr_pages; | |
6629 | } | |
6630 | ||
6631 | if (PageAnon(page)) | |
6632 | nr_anon += nr_pages; | |
6633 | else | |
6634 | nr_file += nr_pages; | |
6635 | ||
6636 | if (pc->flags & PCG_MEM) | |
6637 | nr_mem += nr_pages; | |
6638 | if (pc->flags & PCG_MEMSW) | |
6639 | nr_memsw += nr_pages; | |
6640 | pc->flags = 0; | |
6641 | ||
6642 | pgpgout++; | |
6643 | } while (next != page_list); | |
6644 | ||
6645 | if (memcg) | |
6646 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6647 | nr_anon, nr_file, nr_huge, page); | |
6648 | } | |
6649 | ||
0a31bc97 JW |
6650 | /** |
6651 | * mem_cgroup_uncharge - uncharge a page | |
6652 | * @page: page to uncharge | |
6653 | * | |
6654 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6655 | * mem_cgroup_commit_charge(). | |
6656 | */ | |
6657 | void mem_cgroup_uncharge(struct page *page) | |
6658 | { | |
0a31bc97 | 6659 | struct page_cgroup *pc; |
0a31bc97 JW |
6660 | |
6661 | if (mem_cgroup_disabled()) | |
6662 | return; | |
6663 | ||
747db954 | 6664 | /* Don't touch page->lru of any random page, pre-check: */ |
0a31bc97 | 6665 | pc = lookup_page_cgroup(page); |
0a31bc97 JW |
6666 | if (!PageCgroupUsed(pc)) |
6667 | return; | |
6668 | ||
747db954 JW |
6669 | INIT_LIST_HEAD(&page->lru); |
6670 | uncharge_list(&page->lru); | |
6671 | } | |
0a31bc97 | 6672 | |
747db954 JW |
6673 | /** |
6674 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6675 | * @page_list: list of pages to uncharge | |
6676 | * | |
6677 | * Uncharge a list of pages previously charged with | |
6678 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6679 | */ | |
6680 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6681 | { | |
6682 | if (mem_cgroup_disabled()) | |
6683 | return; | |
0a31bc97 | 6684 | |
747db954 JW |
6685 | if (!list_empty(page_list)) |
6686 | uncharge_list(page_list); | |
0a31bc97 JW |
6687 | } |
6688 | ||
6689 | /** | |
6690 | * mem_cgroup_migrate - migrate a charge to another page | |
6691 | * @oldpage: currently charged page | |
6692 | * @newpage: page to transfer the charge to | |
6693 | * @lrucare: both pages might be on the LRU already | |
6694 | * | |
6695 | * Migrate the charge from @oldpage to @newpage. | |
6696 | * | |
6697 | * Both pages must be locked, @newpage->mapping must be set up. | |
6698 | */ | |
6699 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
6700 | bool lrucare) | |
6701 | { | |
0a31bc97 JW |
6702 | struct page_cgroup *pc; |
6703 | int isolated; | |
6704 | ||
6705 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6706 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
6707 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
6708 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
6709 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
6710 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6711 | newpage); | |
0a31bc97 JW |
6712 | |
6713 | if (mem_cgroup_disabled()) | |
6714 | return; | |
6715 | ||
6716 | /* Page cache replacement: new page already charged? */ | |
6717 | pc = lookup_page_cgroup(newpage); | |
6718 | if (PageCgroupUsed(pc)) | |
6719 | return; | |
6720 | ||
6721 | /* Re-entrant migration: old page already uncharged? */ | |
6722 | pc = lookup_page_cgroup(oldpage); | |
6723 | if (!PageCgroupUsed(pc)) | |
6724 | return; | |
6725 | ||
6726 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEM), oldpage); | |
6727 | VM_BUG_ON_PAGE(do_swap_account && !(pc->flags & PCG_MEMSW), oldpage); | |
6728 | ||
0a31bc97 JW |
6729 | if (lrucare) |
6730 | lock_page_lru(oldpage, &isolated); | |
6731 | ||
6732 | pc->flags = 0; | |
6733 | ||
6734 | if (lrucare) | |
6735 | unlock_page_lru(oldpage, isolated); | |
6736 | ||
6abb5a86 | 6737 | commit_charge(newpage, pc->mem_cgroup, lrucare); |
0a31bc97 JW |
6738 | } |
6739 | ||
2d11085e | 6740 | /* |
1081312f MH |
6741 | * subsys_initcall() for memory controller. |
6742 | * | |
6743 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6744 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6745 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6746 | * be initialized from here. | |
2d11085e MH |
6747 | */ |
6748 | static int __init mem_cgroup_init(void) | |
6749 | { | |
6750 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6751 | enable_swap_cgroup(); |
bb4cc1a8 | 6752 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6753 | memcg_stock_init(); |
2d11085e MH |
6754 | return 0; |
6755 | } | |
6756 | subsys_initcall(mem_cgroup_init); |