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 AM |
756 | { |
757 | spin_lock(&mctz->lock); | |
cf2c8127 | 758 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
759 | spin_unlock(&mctz->lock); |
760 | } | |
761 | ||
762 | ||
763 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
764 | { | |
765 | unsigned long long excess; | |
766 | struct mem_cgroup_per_zone *mz; | |
767 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 768 | |
e231875b | 769 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
770 | /* |
771 | * Necessary to update all ancestors when hierarchy is used. | |
772 | * because their event counter is not touched. | |
773 | */ | |
774 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 775 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bb4cc1a8 AM |
776 | excess = res_counter_soft_limit_excess(&memcg->res); |
777 | /* | |
778 | * We have to update the tree if mz is on RB-tree or | |
779 | * mem is over its softlimit. | |
780 | */ | |
781 | if (excess || mz->on_tree) { | |
782 | spin_lock(&mctz->lock); | |
783 | /* if on-tree, remove it */ | |
784 | if (mz->on_tree) | |
cf2c8127 | 785 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
786 | /* |
787 | * Insert again. mz->usage_in_excess will be updated. | |
788 | * If excess is 0, no tree ops. | |
789 | */ | |
cf2c8127 | 790 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
bb4cc1a8 AM |
791 | spin_unlock(&mctz->lock); |
792 | } | |
793 | } | |
794 | } | |
795 | ||
796 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
797 | { | |
bb4cc1a8 | 798 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
799 | struct mem_cgroup_per_zone *mz; |
800 | int nid, zid; | |
bb4cc1a8 | 801 | |
e231875b JZ |
802 | for_each_node(nid) { |
803 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
804 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
805 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 806 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
807 | } |
808 | } | |
809 | } | |
810 | ||
811 | static struct mem_cgroup_per_zone * | |
812 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
813 | { | |
814 | struct rb_node *rightmost = NULL; | |
815 | struct mem_cgroup_per_zone *mz; | |
816 | ||
817 | retry: | |
818 | mz = NULL; | |
819 | rightmost = rb_last(&mctz->rb_root); | |
820 | if (!rightmost) | |
821 | goto done; /* Nothing to reclaim from */ | |
822 | ||
823 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
824 | /* | |
825 | * Remove the node now but someone else can add it back, | |
826 | * we will to add it back at the end of reclaim to its correct | |
827 | * position in the tree. | |
828 | */ | |
cf2c8127 | 829 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 | 830 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || |
ec903c0c | 831 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
832 | goto retry; |
833 | done: | |
834 | return mz; | |
835 | } | |
836 | ||
837 | static struct mem_cgroup_per_zone * | |
838 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
839 | { | |
840 | struct mem_cgroup_per_zone *mz; | |
841 | ||
842 | spin_lock(&mctz->lock); | |
843 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
844 | spin_unlock(&mctz->lock); | |
845 | return mz; | |
846 | } | |
847 | ||
711d3d2c KH |
848 | /* |
849 | * Implementation Note: reading percpu statistics for memcg. | |
850 | * | |
851 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
852 | * synchronization to implement "quick" read. There are trade-off between | |
853 | * reading cost and precision of value. Then, we may have a chance to implement | |
854 | * a periodic synchronizion of counter in memcg's counter. | |
855 | * | |
856 | * But this _read() function is used for user interface now. The user accounts | |
857 | * memory usage by memory cgroup and he _always_ requires exact value because | |
858 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
859 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
860 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
861 | * | |
862 | * If there are kernel internal actions which can make use of some not-exact | |
863 | * value, and reading all cpu value can be performance bottleneck in some | |
864 | * common workload, threashold and synchonization as vmstat[] should be | |
865 | * implemented. | |
866 | */ | |
c0ff4b85 | 867 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 868 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 869 | { |
7a159cc9 | 870 | long val = 0; |
c62b1a3b | 871 | int cpu; |
c62b1a3b | 872 | |
711d3d2c KH |
873 | get_online_cpus(); |
874 | for_each_online_cpu(cpu) | |
c0ff4b85 | 875 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 876 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
877 | spin_lock(&memcg->pcp_counter_lock); |
878 | val += memcg->nocpu_base.count[idx]; | |
879 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
880 | #endif |
881 | put_online_cpus(); | |
c62b1a3b KH |
882 | return val; |
883 | } | |
884 | ||
c0ff4b85 | 885 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
886 | bool charge) |
887 | { | |
888 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 889 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
890 | } |
891 | ||
c0ff4b85 | 892 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
893 | enum mem_cgroup_events_index idx) |
894 | { | |
895 | unsigned long val = 0; | |
896 | int cpu; | |
897 | ||
9c567512 | 898 | get_online_cpus(); |
e9f8974f | 899 | for_each_online_cpu(cpu) |
c0ff4b85 | 900 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 901 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
902 | spin_lock(&memcg->pcp_counter_lock); |
903 | val += memcg->nocpu_base.events[idx]; | |
904 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 905 | #endif |
9c567512 | 906 | put_online_cpus(); |
e9f8974f JW |
907 | return val; |
908 | } | |
909 | ||
c0ff4b85 | 910 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 911 | struct page *page, |
b2402857 | 912 | bool anon, int nr_pages) |
d52aa412 | 913 | { |
b2402857 KH |
914 | /* |
915 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
916 | * counted as CACHE even if it's on ANON LRU. | |
917 | */ | |
918 | if (anon) | |
919 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 920 | nr_pages); |
d52aa412 | 921 | else |
b2402857 | 922 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 923 | nr_pages); |
55e462b0 | 924 | |
b070e65c DR |
925 | if (PageTransHuge(page)) |
926 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
927 | nr_pages); | |
928 | ||
e401f176 KH |
929 | /* pagein of a big page is an event. So, ignore page size */ |
930 | if (nr_pages > 0) | |
c0ff4b85 | 931 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 932 | else { |
c0ff4b85 | 933 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
934 | nr_pages = -nr_pages; /* for event */ |
935 | } | |
e401f176 | 936 | |
13114716 | 937 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
938 | } |
939 | ||
e231875b | 940 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
941 | { |
942 | struct mem_cgroup_per_zone *mz; | |
943 | ||
944 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
945 | return mz->lru_size[lru]; | |
946 | } | |
947 | ||
e231875b JZ |
948 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
949 | int nid, | |
950 | unsigned int lru_mask) | |
bb2a0de9 | 951 | { |
e231875b | 952 | unsigned long nr = 0; |
889976db YH |
953 | int zid; |
954 | ||
e231875b | 955 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 956 | |
e231875b JZ |
957 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
958 | struct mem_cgroup_per_zone *mz; | |
959 | enum lru_list lru; | |
960 | ||
961 | for_each_lru(lru) { | |
962 | if (!(BIT(lru) & lru_mask)) | |
963 | continue; | |
964 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
965 | nr += mz->lru_size[lru]; | |
966 | } | |
967 | } | |
968 | return nr; | |
889976db | 969 | } |
bb2a0de9 | 970 | |
c0ff4b85 | 971 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 972 | unsigned int lru_mask) |
6d12e2d8 | 973 | { |
e231875b | 974 | unsigned long nr = 0; |
889976db | 975 | int nid; |
6d12e2d8 | 976 | |
31aaea4a | 977 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
978 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
979 | return nr; | |
d52aa412 KH |
980 | } |
981 | ||
f53d7ce3 JW |
982 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
983 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
984 | { |
985 | unsigned long val, next; | |
986 | ||
13114716 | 987 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 988 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 989 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
990 | if ((long)next - (long)val < 0) { |
991 | switch (target) { | |
992 | case MEM_CGROUP_TARGET_THRESH: | |
993 | next = val + THRESHOLDS_EVENTS_TARGET; | |
994 | break; | |
bb4cc1a8 AM |
995 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
996 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
997 | break; | |
f53d7ce3 JW |
998 | case MEM_CGROUP_TARGET_NUMAINFO: |
999 | next = val + NUMAINFO_EVENTS_TARGET; | |
1000 | break; | |
1001 | default: | |
1002 | break; | |
1003 | } | |
1004 | __this_cpu_write(memcg->stat->targets[target], next); | |
1005 | return true; | |
7a159cc9 | 1006 | } |
f53d7ce3 | 1007 | return false; |
d2265e6f KH |
1008 | } |
1009 | ||
1010 | /* | |
1011 | * Check events in order. | |
1012 | * | |
1013 | */ | |
c0ff4b85 | 1014 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 1015 | { |
4799401f | 1016 | preempt_disable(); |
d2265e6f | 1017 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1018 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1019 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1020 | bool do_softlimit; |
82b3f2a7 | 1021 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1022 | |
bb4cc1a8 AM |
1023 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1024 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1025 | #if MAX_NUMNODES > 1 |
1026 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1027 | MEM_CGROUP_TARGET_NUMAINFO); | |
1028 | #endif | |
1029 | preempt_enable(); | |
1030 | ||
c0ff4b85 | 1031 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1032 | if (unlikely(do_softlimit)) |
1033 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1034 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1035 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1036 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1037 | #endif |
f53d7ce3 JW |
1038 | } else |
1039 | preempt_enable(); | |
d2265e6f KH |
1040 | } |
1041 | ||
cf475ad2 | 1042 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1043 | { |
31a78f23 BS |
1044 | /* |
1045 | * mm_update_next_owner() may clear mm->owner to NULL | |
1046 | * if it races with swapoff, page migration, etc. | |
1047 | * So this can be called with p == NULL. | |
1048 | */ | |
1049 | if (unlikely(!p)) | |
1050 | return NULL; | |
1051 | ||
073219e9 | 1052 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1053 | } |
1054 | ||
df381975 | 1055 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1056 | { |
c0ff4b85 | 1057 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 1058 | |
54595fe2 KH |
1059 | rcu_read_lock(); |
1060 | do { | |
6f6acb00 MH |
1061 | /* |
1062 | * Page cache insertions can happen withou an | |
1063 | * actual mm context, e.g. during disk probing | |
1064 | * on boot, loopback IO, acct() writes etc. | |
1065 | */ | |
1066 | if (unlikely(!mm)) | |
df381975 | 1067 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1068 | else { |
1069 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1070 | if (unlikely(!memcg)) | |
1071 | memcg = root_mem_cgroup; | |
1072 | } | |
ec903c0c | 1073 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1074 | rcu_read_unlock(); |
c0ff4b85 | 1075 | return memcg; |
54595fe2 KH |
1076 | } |
1077 | ||
16248d8f MH |
1078 | /* |
1079 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1080 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1081 | * | |
1082 | * helper function to be used by mem_cgroup_iter | |
1083 | */ | |
1084 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
694fbc0f | 1085 | struct mem_cgroup *last_visited) |
16248d8f | 1086 | { |
492eb21b | 1087 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 1088 | |
bd8815a6 | 1089 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 1090 | skip_node: |
492eb21b | 1091 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
1092 | |
1093 | /* | |
1094 | * Even if we found a group we have to make sure it is | |
1095 | * alive. css && !memcg means that the groups should be | |
1096 | * skipped and we should continue the tree walk. | |
1097 | * last_visited css is safe to use because it is | |
1098 | * protected by css_get and the tree walk is rcu safe. | |
0eef6156 MH |
1099 | * |
1100 | * We do not take a reference on the root of the tree walk | |
1101 | * because we might race with the root removal when it would | |
1102 | * be the only node in the iterated hierarchy and mem_cgroup_iter | |
1103 | * would end up in an endless loop because it expects that at | |
1104 | * least one valid node will be returned. Root cannot disappear | |
1105 | * because caller of the iterator should hold it already so | |
1106 | * skipping css reference should be safe. | |
16248d8f | 1107 | */ |
492eb21b | 1108 | if (next_css) { |
ce48225f | 1109 | if ((next_css == &root->css) || |
ec903c0c TH |
1110 | ((next_css->flags & CSS_ONLINE) && |
1111 | css_tryget_online(next_css))) | |
d8ad3055 | 1112 | return mem_cgroup_from_css(next_css); |
0eef6156 MH |
1113 | |
1114 | prev_css = next_css; | |
1115 | goto skip_node; | |
16248d8f MH |
1116 | } |
1117 | ||
1118 | return NULL; | |
1119 | } | |
1120 | ||
519ebea3 JW |
1121 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1122 | { | |
1123 | /* | |
1124 | * When a group in the hierarchy below root is destroyed, the | |
1125 | * hierarchy iterator can no longer be trusted since it might | |
1126 | * have pointed to the destroyed group. Invalidate it. | |
1127 | */ | |
1128 | atomic_inc(&root->dead_count); | |
1129 | } | |
1130 | ||
1131 | static struct mem_cgroup * | |
1132 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1133 | struct mem_cgroup *root, | |
1134 | int *sequence) | |
1135 | { | |
1136 | struct mem_cgroup *position = NULL; | |
1137 | /* | |
1138 | * A cgroup destruction happens in two stages: offlining and | |
1139 | * release. They are separated by a RCU grace period. | |
1140 | * | |
1141 | * If the iterator is valid, we may still race with an | |
1142 | * offlining. The RCU lock ensures the object won't be | |
1143 | * released, tryget will fail if we lost the race. | |
1144 | */ | |
1145 | *sequence = atomic_read(&root->dead_count); | |
1146 | if (iter->last_dead_count == *sequence) { | |
1147 | smp_rmb(); | |
1148 | position = iter->last_visited; | |
ecc736fc MH |
1149 | |
1150 | /* | |
1151 | * We cannot take a reference to root because we might race | |
1152 | * with root removal and returning NULL would end up in | |
1153 | * an endless loop on the iterator user level when root | |
1154 | * would be returned all the time. | |
1155 | */ | |
1156 | if (position && position != root && | |
ec903c0c | 1157 | !css_tryget_online(&position->css)) |
519ebea3 JW |
1158 | position = NULL; |
1159 | } | |
1160 | return position; | |
1161 | } | |
1162 | ||
1163 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1164 | struct mem_cgroup *last_visited, | |
1165 | struct mem_cgroup *new_position, | |
ecc736fc | 1166 | struct mem_cgroup *root, |
519ebea3 JW |
1167 | int sequence) |
1168 | { | |
ecc736fc MH |
1169 | /* root reference counting symmetric to mem_cgroup_iter_load */ |
1170 | if (last_visited && last_visited != root) | |
519ebea3 JW |
1171 | css_put(&last_visited->css); |
1172 | /* | |
1173 | * We store the sequence count from the time @last_visited was | |
1174 | * loaded successfully instead of rereading it here so that we | |
1175 | * don't lose destruction events in between. We could have | |
1176 | * raced with the destruction of @new_position after all. | |
1177 | */ | |
1178 | iter->last_visited = new_position; | |
1179 | smp_wmb(); | |
1180 | iter->last_dead_count = sequence; | |
1181 | } | |
1182 | ||
5660048c JW |
1183 | /** |
1184 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1185 | * @root: hierarchy root | |
1186 | * @prev: previously returned memcg, NULL on first invocation | |
1187 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1188 | * | |
1189 | * Returns references to children of the hierarchy below @root, or | |
1190 | * @root itself, or %NULL after a full round-trip. | |
1191 | * | |
1192 | * Caller must pass the return value in @prev on subsequent | |
1193 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1194 | * to cancel a hierarchy walk before the round-trip is complete. | |
1195 | * | |
1196 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1197 | * divide up the memcgs in the hierarchy among all concurrent | |
1198 | * reclaimers operating on the same zone and priority. | |
1199 | */ | |
694fbc0f | 1200 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1201 | struct mem_cgroup *prev, |
694fbc0f | 1202 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1203 | { |
9f3a0d09 | 1204 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1205 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1206 | |
694fbc0f AM |
1207 | if (mem_cgroup_disabled()) |
1208 | return NULL; | |
5660048c | 1209 | |
9f3a0d09 JW |
1210 | if (!root) |
1211 | root = root_mem_cgroup; | |
7d74b06f | 1212 | |
9f3a0d09 | 1213 | if (prev && !reclaim) |
542f85f9 | 1214 | last_visited = prev; |
14067bb3 | 1215 | |
9f3a0d09 JW |
1216 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1217 | if (prev) | |
c40046f3 | 1218 | goto out_css_put; |
694fbc0f | 1219 | return root; |
9f3a0d09 | 1220 | } |
14067bb3 | 1221 | |
542f85f9 | 1222 | rcu_read_lock(); |
9f3a0d09 | 1223 | while (!memcg) { |
527a5ec9 | 1224 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1225 | int uninitialized_var(seq); |
711d3d2c | 1226 | |
527a5ec9 | 1227 | if (reclaim) { |
527a5ec9 JW |
1228 | struct mem_cgroup_per_zone *mz; |
1229 | ||
e231875b | 1230 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); |
527a5ec9 | 1231 | iter = &mz->reclaim_iter[reclaim->priority]; |
542f85f9 | 1232 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1233 | iter->last_visited = NULL; |
542f85f9 MH |
1234 | goto out_unlock; |
1235 | } | |
5f578161 | 1236 | |
519ebea3 | 1237 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1238 | } |
7d74b06f | 1239 | |
694fbc0f | 1240 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1241 | |
527a5ec9 | 1242 | if (reclaim) { |
ecc736fc MH |
1243 | mem_cgroup_iter_update(iter, last_visited, memcg, root, |
1244 | seq); | |
542f85f9 | 1245 | |
19f39402 | 1246 | if (!memcg) |
527a5ec9 JW |
1247 | iter->generation++; |
1248 | else if (!prev && memcg) | |
1249 | reclaim->generation = iter->generation; | |
1250 | } | |
9f3a0d09 | 1251 | |
694fbc0f | 1252 | if (prev && !memcg) |
542f85f9 | 1253 | goto out_unlock; |
9f3a0d09 | 1254 | } |
542f85f9 MH |
1255 | out_unlock: |
1256 | rcu_read_unlock(); | |
c40046f3 MH |
1257 | out_css_put: |
1258 | if (prev && prev != root) | |
1259 | css_put(&prev->css); | |
1260 | ||
9f3a0d09 | 1261 | return memcg; |
14067bb3 | 1262 | } |
7d74b06f | 1263 | |
5660048c JW |
1264 | /** |
1265 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1266 | * @root: hierarchy root | |
1267 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1268 | */ | |
1269 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1270 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1271 | { |
1272 | if (!root) | |
1273 | root = root_mem_cgroup; | |
1274 | if (prev && prev != root) | |
1275 | css_put(&prev->css); | |
1276 | } | |
7d74b06f | 1277 | |
9f3a0d09 JW |
1278 | /* |
1279 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1280 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1281 | * be used for reference counting. | |
1282 | */ | |
1283 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1284 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1285 | iter != NULL; \ |
527a5ec9 | 1286 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1287 | |
9f3a0d09 | 1288 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1289 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1290 | iter != NULL; \ |
527a5ec9 | 1291 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1292 | |
68ae564b | 1293 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1294 | { |
c0ff4b85 | 1295 | struct mem_cgroup *memcg; |
456f998e | 1296 | |
456f998e | 1297 | rcu_read_lock(); |
c0ff4b85 R |
1298 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1299 | if (unlikely(!memcg)) | |
456f998e YH |
1300 | goto out; |
1301 | ||
1302 | switch (idx) { | |
456f998e | 1303 | case PGFAULT: |
0e574a93 JW |
1304 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1305 | break; | |
1306 | case PGMAJFAULT: | |
1307 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1308 | break; |
1309 | default: | |
1310 | BUG(); | |
1311 | } | |
1312 | out: | |
1313 | rcu_read_unlock(); | |
1314 | } | |
68ae564b | 1315 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1316 | |
925b7673 JW |
1317 | /** |
1318 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1319 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1320 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1321 | * |
1322 | * Returns the lru list vector holding pages for the given @zone and | |
1323 | * @mem. This can be the global zone lruvec, if the memory controller | |
1324 | * is disabled. | |
1325 | */ | |
1326 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1327 | struct mem_cgroup *memcg) | |
1328 | { | |
1329 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1330 | struct lruvec *lruvec; |
925b7673 | 1331 | |
bea8c150 HD |
1332 | if (mem_cgroup_disabled()) { |
1333 | lruvec = &zone->lruvec; | |
1334 | goto out; | |
1335 | } | |
925b7673 | 1336 | |
e231875b | 1337 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1338 | lruvec = &mz->lruvec; |
1339 | out: | |
1340 | /* | |
1341 | * Since a node can be onlined after the mem_cgroup was created, | |
1342 | * we have to be prepared to initialize lruvec->zone here; | |
1343 | * and if offlined then reonlined, we need to reinitialize it. | |
1344 | */ | |
1345 | if (unlikely(lruvec->zone != zone)) | |
1346 | lruvec->zone = zone; | |
1347 | return lruvec; | |
925b7673 JW |
1348 | } |
1349 | ||
08e552c6 KH |
1350 | /* |
1351 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1352 | * Operations are called by routine of global LRU independently from memcg. | |
1353 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1354 | * | |
1355 | * Changes to pc->mem_cgroup happens when | |
1356 | * 1. charge | |
1357 | * 2. moving account | |
1358 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1359 | * It is added to LRU before charge. | |
1360 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1361 | * When moving account, the page is not on LRU. It's isolated. | |
1362 | */ | |
4f98a2fe | 1363 | |
925b7673 | 1364 | /** |
fa9add64 | 1365 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1366 | * @page: the page |
fa9add64 | 1367 | * @zone: zone of the page |
925b7673 | 1368 | */ |
fa9add64 | 1369 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1370 | { |
08e552c6 | 1371 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1372 | struct mem_cgroup *memcg; |
1373 | struct page_cgroup *pc; | |
bea8c150 | 1374 | struct lruvec *lruvec; |
6d12e2d8 | 1375 | |
bea8c150 HD |
1376 | if (mem_cgroup_disabled()) { |
1377 | lruvec = &zone->lruvec; | |
1378 | goto out; | |
1379 | } | |
925b7673 | 1380 | |
08e552c6 | 1381 | pc = lookup_page_cgroup(page); |
38c5d72f | 1382 | memcg = pc->mem_cgroup; |
7512102c HD |
1383 | |
1384 | /* | |
fa9add64 | 1385 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1386 | * an uncharged page off lru does nothing to secure |
1387 | * its former mem_cgroup from sudden removal. | |
1388 | * | |
1389 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1390 | * under page_cgroup lock: between them, they make all uses | |
1391 | * of pc->mem_cgroup safe. | |
1392 | */ | |
fa9add64 | 1393 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1394 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1395 | ||
e231875b | 1396 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1397 | lruvec = &mz->lruvec; |
1398 | out: | |
1399 | /* | |
1400 | * Since a node can be onlined after the mem_cgroup was created, | |
1401 | * we have to be prepared to initialize lruvec->zone here; | |
1402 | * and if offlined then reonlined, we need to reinitialize it. | |
1403 | */ | |
1404 | if (unlikely(lruvec->zone != zone)) | |
1405 | lruvec->zone = zone; | |
1406 | return lruvec; | |
08e552c6 | 1407 | } |
b69408e8 | 1408 | |
925b7673 | 1409 | /** |
fa9add64 HD |
1410 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1411 | * @lruvec: mem_cgroup per zone lru vector | |
1412 | * @lru: index of lru list the page is sitting on | |
1413 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1414 | * |
fa9add64 HD |
1415 | * This function must be called when a page is added to or removed from an |
1416 | * lru list. | |
3f58a829 | 1417 | */ |
fa9add64 HD |
1418 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1419 | int nr_pages) | |
3f58a829 MK |
1420 | { |
1421 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1422 | unsigned long *lru_size; |
3f58a829 MK |
1423 | |
1424 | if (mem_cgroup_disabled()) | |
1425 | return; | |
1426 | ||
fa9add64 HD |
1427 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1428 | lru_size = mz->lru_size + lru; | |
1429 | *lru_size += nr_pages; | |
1430 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1431 | } |
544122e5 | 1432 | |
3e92041d | 1433 | /* |
c0ff4b85 | 1434 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1435 | * hierarchy subtree |
1436 | */ | |
c3ac9a8a JW |
1437 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1438 | struct mem_cgroup *memcg) | |
3e92041d | 1439 | { |
91c63734 JW |
1440 | if (root_memcg == memcg) |
1441 | return true; | |
3a981f48 | 1442 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1443 | return false; |
b47f77b5 | 1444 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1445 | } |
1446 | ||
1447 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1448 | struct mem_cgroup *memcg) | |
1449 | { | |
1450 | bool ret; | |
1451 | ||
91c63734 | 1452 | rcu_read_lock(); |
c3ac9a8a | 1453 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1454 | rcu_read_unlock(); |
1455 | return ret; | |
3e92041d MH |
1456 | } |
1457 | ||
ffbdccf5 DR |
1458 | bool task_in_mem_cgroup(struct task_struct *task, |
1459 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1460 | { |
0b7f569e | 1461 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1462 | struct task_struct *p; |
ffbdccf5 | 1463 | bool ret; |
4c4a2214 | 1464 | |
158e0a2d | 1465 | p = find_lock_task_mm(task); |
de077d22 | 1466 | if (p) { |
df381975 | 1467 | curr = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1468 | task_unlock(p); |
1469 | } else { | |
1470 | /* | |
1471 | * All threads may have already detached their mm's, but the oom | |
1472 | * killer still needs to detect if they have already been oom | |
1473 | * killed to prevent needlessly killing additional tasks. | |
1474 | */ | |
ffbdccf5 | 1475 | rcu_read_lock(); |
de077d22 DR |
1476 | curr = mem_cgroup_from_task(task); |
1477 | if (curr) | |
1478 | css_get(&curr->css); | |
ffbdccf5 | 1479 | rcu_read_unlock(); |
de077d22 | 1480 | } |
d31f56db | 1481 | /* |
c0ff4b85 | 1482 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1483 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1484 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1485 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1486 | */ |
c0ff4b85 | 1487 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1488 | css_put(&curr->css); |
4c4a2214 DR |
1489 | return ret; |
1490 | } | |
1491 | ||
c56d5c7d | 1492 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1493 | { |
9b272977 | 1494 | unsigned long inactive_ratio; |
14797e23 | 1495 | unsigned long inactive; |
9b272977 | 1496 | unsigned long active; |
c772be93 | 1497 | unsigned long gb; |
14797e23 | 1498 | |
4d7dcca2 HD |
1499 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1500 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1501 | |
c772be93 KM |
1502 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1503 | if (gb) | |
1504 | inactive_ratio = int_sqrt(10 * gb); | |
1505 | else | |
1506 | inactive_ratio = 1; | |
1507 | ||
9b272977 | 1508 | return inactive * inactive_ratio < active; |
14797e23 KM |
1509 | } |
1510 | ||
6d61ef40 BS |
1511 | #define mem_cgroup_from_res_counter(counter, member) \ |
1512 | container_of(counter, struct mem_cgroup, member) | |
1513 | ||
19942822 | 1514 | /** |
9d11ea9f | 1515 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1516 | * @memcg: the memory cgroup |
19942822 | 1517 | * |
9d11ea9f | 1518 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1519 | * pages. |
19942822 | 1520 | */ |
c0ff4b85 | 1521 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1522 | { |
9d11ea9f JW |
1523 | unsigned long long margin; |
1524 | ||
c0ff4b85 | 1525 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1526 | if (do_swap_account) |
c0ff4b85 | 1527 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1528 | return margin >> PAGE_SHIFT; |
19942822 JW |
1529 | } |
1530 | ||
1f4c025b | 1531 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1532 | { |
a7885eb8 | 1533 | /* root ? */ |
14208b0e | 1534 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1535 | return vm_swappiness; |
1536 | ||
bf1ff263 | 1537 | return memcg->swappiness; |
a7885eb8 KM |
1538 | } |
1539 | ||
619d094b KH |
1540 | /* |
1541 | * memcg->moving_account is used for checking possibility that some thread is | |
1542 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1543 | * a memcg, other threads should check memcg->moving_account under | |
1544 | * rcu_read_lock(), like this: | |
1545 | * | |
1546 | * CPU-A CPU-B | |
1547 | * rcu_read_lock() | |
1548 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1549 | * take heavy locks. | |
1550 | * synchronize_rcu() update something. | |
1551 | * rcu_read_unlock() | |
1552 | * start move here. | |
1553 | */ | |
4331f7d3 KH |
1554 | |
1555 | /* for quick checking without looking up memcg */ | |
1556 | atomic_t memcg_moving __read_mostly; | |
1557 | ||
c0ff4b85 | 1558 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1559 | { |
4331f7d3 | 1560 | atomic_inc(&memcg_moving); |
619d094b | 1561 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1562 | synchronize_rcu(); |
1563 | } | |
1564 | ||
c0ff4b85 | 1565 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1566 | { |
619d094b KH |
1567 | /* |
1568 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1569 | * We check NULL in callee rather than caller. | |
1570 | */ | |
4331f7d3 KH |
1571 | if (memcg) { |
1572 | atomic_dec(&memcg_moving); | |
619d094b | 1573 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1574 | } |
32047e2a | 1575 | } |
619d094b | 1576 | |
32047e2a | 1577 | /* |
bdcbb659 | 1578 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1579 | * |
bdcbb659 QH |
1580 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1581 | * moving cgroups. This is for waiting at high-memory pressure | |
1582 | * caused by "move". | |
32047e2a | 1583 | */ |
c0ff4b85 | 1584 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1585 | { |
2bd9bb20 KH |
1586 | struct mem_cgroup *from; |
1587 | struct mem_cgroup *to; | |
4b534334 | 1588 | bool ret = false; |
2bd9bb20 KH |
1589 | /* |
1590 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1591 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1592 | */ | |
1593 | spin_lock(&mc.lock); | |
1594 | from = mc.from; | |
1595 | to = mc.to; | |
1596 | if (!from) | |
1597 | goto unlock; | |
3e92041d | 1598 | |
c0ff4b85 R |
1599 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1600 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1601 | unlock: |
1602 | spin_unlock(&mc.lock); | |
4b534334 KH |
1603 | return ret; |
1604 | } | |
1605 | ||
c0ff4b85 | 1606 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1607 | { |
1608 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1609 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1610 | DEFINE_WAIT(wait); |
1611 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1612 | /* moving charge context might have finished. */ | |
1613 | if (mc.moving_task) | |
1614 | schedule(); | |
1615 | finish_wait(&mc.waitq, &wait); | |
1616 | return true; | |
1617 | } | |
1618 | } | |
1619 | return false; | |
1620 | } | |
1621 | ||
312734c0 KH |
1622 | /* |
1623 | * Take this lock when | |
1624 | * - a code tries to modify page's memcg while it's USED. | |
1625 | * - a code tries to modify page state accounting in a memcg. | |
312734c0 KH |
1626 | */ |
1627 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1628 | unsigned long *flags) | |
1629 | { | |
1630 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1631 | } | |
1632 | ||
1633 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1634 | unsigned long *flags) | |
1635 | { | |
1636 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1637 | } | |
1638 | ||
58cf188e | 1639 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1640 | /** |
58cf188e | 1641 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1642 | * @memcg: The memory cgroup that went over limit |
1643 | * @p: Task that is going to be killed | |
1644 | * | |
1645 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1646 | * enabled | |
1647 | */ | |
1648 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1649 | { | |
e61734c5 | 1650 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1651 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1652 | struct mem_cgroup *iter; |
1653 | unsigned int i; | |
e222432b | 1654 | |
58cf188e | 1655 | if (!p) |
e222432b BS |
1656 | return; |
1657 | ||
08088cb9 | 1658 | mutex_lock(&oom_info_lock); |
e222432b BS |
1659 | rcu_read_lock(); |
1660 | ||
e61734c5 TH |
1661 | pr_info("Task in "); |
1662 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1663 | pr_info(" killed as a result of limit of "); | |
1664 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1665 | pr_info("\n"); | |
e222432b | 1666 | |
e222432b BS |
1667 | rcu_read_unlock(); |
1668 | ||
d045197f | 1669 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1670 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1671 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1672 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1673 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1674 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1675 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1676 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1677 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1678 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1679 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1680 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1681 | |
1682 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1683 | pr_info("Memory cgroup stats for "); |
1684 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1685 | pr_cont(":"); |
1686 | ||
1687 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1688 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1689 | continue; | |
1690 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1691 | K(mem_cgroup_read_stat(iter, i))); | |
1692 | } | |
1693 | ||
1694 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1695 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1696 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1697 | ||
1698 | pr_cont("\n"); | |
1699 | } | |
08088cb9 | 1700 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1701 | } |
1702 | ||
81d39c20 KH |
1703 | /* |
1704 | * This function returns the number of memcg under hierarchy tree. Returns | |
1705 | * 1(self count) if no children. | |
1706 | */ | |
c0ff4b85 | 1707 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1708 | { |
1709 | int num = 0; | |
7d74b06f KH |
1710 | struct mem_cgroup *iter; |
1711 | ||
c0ff4b85 | 1712 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1713 | num++; |
81d39c20 KH |
1714 | return num; |
1715 | } | |
1716 | ||
a63d83f4 DR |
1717 | /* |
1718 | * Return the memory (and swap, if configured) limit for a memcg. | |
1719 | */ | |
9cbb78bb | 1720 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1721 | { |
1722 | u64 limit; | |
a63d83f4 | 1723 | |
f3e8eb70 | 1724 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1725 | |
a63d83f4 | 1726 | /* |
9a5a8f19 | 1727 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1728 | */ |
9a5a8f19 MH |
1729 | if (mem_cgroup_swappiness(memcg)) { |
1730 | u64 memsw; | |
1731 | ||
1732 | limit += total_swap_pages << PAGE_SHIFT; | |
1733 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1734 | ||
1735 | /* | |
1736 | * If memsw is finite and limits the amount of swap space | |
1737 | * available to this memcg, return that limit. | |
1738 | */ | |
1739 | limit = min(limit, memsw); | |
1740 | } | |
1741 | ||
1742 | return limit; | |
a63d83f4 DR |
1743 | } |
1744 | ||
19965460 DR |
1745 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1746 | int order) | |
9cbb78bb DR |
1747 | { |
1748 | struct mem_cgroup *iter; | |
1749 | unsigned long chosen_points = 0; | |
1750 | unsigned long totalpages; | |
1751 | unsigned int points = 0; | |
1752 | struct task_struct *chosen = NULL; | |
1753 | ||
876aafbf | 1754 | /* |
465adcf1 DR |
1755 | * If current has a pending SIGKILL or is exiting, then automatically |
1756 | * select it. The goal is to allow it to allocate so that it may | |
1757 | * quickly exit and free its memory. | |
876aafbf | 1758 | */ |
465adcf1 | 1759 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1760 | set_thread_flag(TIF_MEMDIE); |
1761 | return; | |
1762 | } | |
1763 | ||
1764 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1765 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1766 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1767 | struct css_task_iter it; |
9cbb78bb DR |
1768 | struct task_struct *task; |
1769 | ||
72ec7029 TH |
1770 | css_task_iter_start(&iter->css, &it); |
1771 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1772 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1773 | false)) { | |
1774 | case OOM_SCAN_SELECT: | |
1775 | if (chosen) | |
1776 | put_task_struct(chosen); | |
1777 | chosen = task; | |
1778 | chosen_points = ULONG_MAX; | |
1779 | get_task_struct(chosen); | |
1780 | /* fall through */ | |
1781 | case OOM_SCAN_CONTINUE: | |
1782 | continue; | |
1783 | case OOM_SCAN_ABORT: | |
72ec7029 | 1784 | css_task_iter_end(&it); |
9cbb78bb DR |
1785 | mem_cgroup_iter_break(memcg, iter); |
1786 | if (chosen) | |
1787 | put_task_struct(chosen); | |
1788 | return; | |
1789 | case OOM_SCAN_OK: | |
1790 | break; | |
1791 | }; | |
1792 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1793 | if (!points || points < chosen_points) |
1794 | continue; | |
1795 | /* Prefer thread group leaders for display purposes */ | |
1796 | if (points == chosen_points && | |
1797 | thread_group_leader(chosen)) | |
1798 | continue; | |
1799 | ||
1800 | if (chosen) | |
1801 | put_task_struct(chosen); | |
1802 | chosen = task; | |
1803 | chosen_points = points; | |
1804 | get_task_struct(chosen); | |
9cbb78bb | 1805 | } |
72ec7029 | 1806 | css_task_iter_end(&it); |
9cbb78bb DR |
1807 | } |
1808 | ||
1809 | if (!chosen) | |
1810 | return; | |
1811 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1812 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1813 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1814 | } |
1815 | ||
5660048c JW |
1816 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1817 | gfp_t gfp_mask, | |
1818 | unsigned long flags) | |
1819 | { | |
1820 | unsigned long total = 0; | |
1821 | bool noswap = false; | |
1822 | int loop; | |
1823 | ||
1824 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1825 | noswap = true; | |
1826 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1827 | noswap = true; | |
1828 | ||
1829 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1830 | if (loop) | |
1831 | drain_all_stock_async(memcg); | |
1832 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1833 | /* | |
1834 | * Allow limit shrinkers, which are triggered directly | |
1835 | * by userspace, to catch signals and stop reclaim | |
1836 | * after minimal progress, regardless of the margin. | |
1837 | */ | |
1838 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1839 | break; | |
1840 | if (mem_cgroup_margin(memcg)) | |
1841 | break; | |
1842 | /* | |
1843 | * If nothing was reclaimed after two attempts, there | |
1844 | * may be no reclaimable pages in this hierarchy. | |
1845 | */ | |
1846 | if (loop && !total) | |
1847 | break; | |
1848 | } | |
1849 | return total; | |
1850 | } | |
1851 | ||
4d0c066d KH |
1852 | /** |
1853 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1854 | * @memcg: the target memcg |
4d0c066d KH |
1855 | * @nid: the node ID to be checked. |
1856 | * @noswap : specify true here if the user wants flle only information. | |
1857 | * | |
1858 | * This function returns whether the specified memcg contains any | |
1859 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1860 | * pages in the node. | |
1861 | */ | |
c0ff4b85 | 1862 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1863 | int nid, bool noswap) |
1864 | { | |
c0ff4b85 | 1865 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1866 | return true; |
1867 | if (noswap || !total_swap_pages) | |
1868 | return false; | |
c0ff4b85 | 1869 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1870 | return true; |
1871 | return false; | |
1872 | ||
1873 | } | |
bb4cc1a8 | 1874 | #if MAX_NUMNODES > 1 |
889976db YH |
1875 | |
1876 | /* | |
1877 | * Always updating the nodemask is not very good - even if we have an empty | |
1878 | * list or the wrong list here, we can start from some node and traverse all | |
1879 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1880 | * | |
1881 | */ | |
c0ff4b85 | 1882 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1883 | { |
1884 | int nid; | |
453a9bf3 KH |
1885 | /* |
1886 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1887 | * pagein/pageout changes since the last update. | |
1888 | */ | |
c0ff4b85 | 1889 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1890 | return; |
c0ff4b85 | 1891 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1892 | return; |
1893 | ||
889976db | 1894 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1895 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1896 | |
31aaea4a | 1897 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1898 | |
c0ff4b85 R |
1899 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1900 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1901 | } |
453a9bf3 | 1902 | |
c0ff4b85 R |
1903 | atomic_set(&memcg->numainfo_events, 0); |
1904 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1905 | } |
1906 | ||
1907 | /* | |
1908 | * Selecting a node where we start reclaim from. Because what we need is just | |
1909 | * reducing usage counter, start from anywhere is O,K. Considering | |
1910 | * memory reclaim from current node, there are pros. and cons. | |
1911 | * | |
1912 | * Freeing memory from current node means freeing memory from a node which | |
1913 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1914 | * hit limits, it will see a contention on a node. But freeing from remote | |
1915 | * node means more costs for memory reclaim because of memory latency. | |
1916 | * | |
1917 | * Now, we use round-robin. Better algorithm is welcomed. | |
1918 | */ | |
c0ff4b85 | 1919 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1920 | { |
1921 | int node; | |
1922 | ||
c0ff4b85 R |
1923 | mem_cgroup_may_update_nodemask(memcg); |
1924 | node = memcg->last_scanned_node; | |
889976db | 1925 | |
c0ff4b85 | 1926 | node = next_node(node, memcg->scan_nodes); |
889976db | 1927 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1928 | node = first_node(memcg->scan_nodes); |
889976db YH |
1929 | /* |
1930 | * We call this when we hit limit, not when pages are added to LRU. | |
1931 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1932 | * memcg is too small and all pages are not on LRU. In that case, | |
1933 | * we use curret node. | |
1934 | */ | |
1935 | if (unlikely(node == MAX_NUMNODES)) | |
1936 | node = numa_node_id(); | |
1937 | ||
c0ff4b85 | 1938 | memcg->last_scanned_node = node; |
889976db YH |
1939 | return node; |
1940 | } | |
1941 | ||
bb4cc1a8 AM |
1942 | /* |
1943 | * Check all nodes whether it contains reclaimable pages or not. | |
1944 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1945 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1946 | * enough new information. We need to do double check. | |
1947 | */ | |
1948 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1949 | { | |
1950 | int nid; | |
1951 | ||
1952 | /* | |
1953 | * quick check...making use of scan_node. | |
1954 | * We can skip unused nodes. | |
1955 | */ | |
1956 | if (!nodes_empty(memcg->scan_nodes)) { | |
1957 | for (nid = first_node(memcg->scan_nodes); | |
1958 | nid < MAX_NUMNODES; | |
1959 | nid = next_node(nid, memcg->scan_nodes)) { | |
1960 | ||
1961 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1962 | return true; | |
1963 | } | |
1964 | } | |
1965 | /* | |
1966 | * Check rest of nodes. | |
1967 | */ | |
1968 | for_each_node_state(nid, N_MEMORY) { | |
1969 | if (node_isset(nid, memcg->scan_nodes)) | |
1970 | continue; | |
1971 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1972 | return true; | |
1973 | } | |
1974 | return false; | |
1975 | } | |
1976 | ||
889976db | 1977 | #else |
c0ff4b85 | 1978 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1979 | { |
1980 | return 0; | |
1981 | } | |
4d0c066d | 1982 | |
bb4cc1a8 AM |
1983 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
1984 | { | |
1985 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
1986 | } | |
889976db YH |
1987 | #endif |
1988 | ||
0608f43d AM |
1989 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1990 | struct zone *zone, | |
1991 | gfp_t gfp_mask, | |
1992 | unsigned long *total_scanned) | |
1993 | { | |
1994 | struct mem_cgroup *victim = NULL; | |
1995 | int total = 0; | |
1996 | int loop = 0; | |
1997 | unsigned long excess; | |
1998 | unsigned long nr_scanned; | |
1999 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2000 | .zone = zone, | |
2001 | .priority = 0, | |
2002 | }; | |
2003 | ||
2004 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; | |
2005 | ||
2006 | while (1) { | |
2007 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
2008 | if (!victim) { | |
2009 | loop++; | |
2010 | if (loop >= 2) { | |
2011 | /* | |
2012 | * If we have not been able to reclaim | |
2013 | * anything, it might because there are | |
2014 | * no reclaimable pages under this hierarchy | |
2015 | */ | |
2016 | if (!total) | |
2017 | break; | |
2018 | /* | |
2019 | * We want to do more targeted reclaim. | |
2020 | * excess >> 2 is not to excessive so as to | |
2021 | * reclaim too much, nor too less that we keep | |
2022 | * coming back to reclaim from this cgroup | |
2023 | */ | |
2024 | if (total >= (excess >> 2) || | |
2025 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
2026 | break; | |
2027 | } | |
2028 | continue; | |
2029 | } | |
2030 | if (!mem_cgroup_reclaimable(victim, false)) | |
2031 | continue; | |
2032 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
2033 | zone, &nr_scanned); | |
2034 | *total_scanned += nr_scanned; | |
2035 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
2036 | break; | |
6d61ef40 | 2037 | } |
0608f43d AM |
2038 | mem_cgroup_iter_break(root_memcg, victim); |
2039 | return total; | |
6d61ef40 BS |
2040 | } |
2041 | ||
0056f4e6 JW |
2042 | #ifdef CONFIG_LOCKDEP |
2043 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
2044 | .name = "memcg_oom_lock", | |
2045 | }; | |
2046 | #endif | |
2047 | ||
fb2a6fc5 JW |
2048 | static DEFINE_SPINLOCK(memcg_oom_lock); |
2049 | ||
867578cb KH |
2050 | /* |
2051 | * Check OOM-Killer is already running under our hierarchy. | |
2052 | * If someone is running, return false. | |
2053 | */ | |
fb2a6fc5 | 2054 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2055 | { |
79dfdacc | 2056 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2057 | |
fb2a6fc5 JW |
2058 | spin_lock(&memcg_oom_lock); |
2059 | ||
9f3a0d09 | 2060 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2061 | if (iter->oom_lock) { |
79dfdacc MH |
2062 | /* |
2063 | * this subtree of our hierarchy is already locked | |
2064 | * so we cannot give a lock. | |
2065 | */ | |
79dfdacc | 2066 | failed = iter; |
9f3a0d09 JW |
2067 | mem_cgroup_iter_break(memcg, iter); |
2068 | break; | |
23751be0 JW |
2069 | } else |
2070 | iter->oom_lock = true; | |
7d74b06f | 2071 | } |
867578cb | 2072 | |
fb2a6fc5 JW |
2073 | if (failed) { |
2074 | /* | |
2075 | * OK, we failed to lock the whole subtree so we have | |
2076 | * to clean up what we set up to the failing subtree | |
2077 | */ | |
2078 | for_each_mem_cgroup_tree(iter, memcg) { | |
2079 | if (iter == failed) { | |
2080 | mem_cgroup_iter_break(memcg, iter); | |
2081 | break; | |
2082 | } | |
2083 | iter->oom_lock = false; | |
79dfdacc | 2084 | } |
0056f4e6 JW |
2085 | } else |
2086 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2087 | |
2088 | spin_unlock(&memcg_oom_lock); | |
2089 | ||
2090 | return !failed; | |
a636b327 | 2091 | } |
0b7f569e | 2092 | |
fb2a6fc5 | 2093 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2094 | { |
7d74b06f KH |
2095 | struct mem_cgroup *iter; |
2096 | ||
fb2a6fc5 | 2097 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 2098 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 2099 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2100 | iter->oom_lock = false; |
fb2a6fc5 | 2101 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2102 | } |
2103 | ||
c0ff4b85 | 2104 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2105 | { |
2106 | struct mem_cgroup *iter; | |
2107 | ||
c0ff4b85 | 2108 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2109 | atomic_inc(&iter->under_oom); |
2110 | } | |
2111 | ||
c0ff4b85 | 2112 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2113 | { |
2114 | struct mem_cgroup *iter; | |
2115 | ||
867578cb KH |
2116 | /* |
2117 | * When a new child is created while the hierarchy is under oom, | |
2118 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2119 | * atomic_add_unless() here. | |
2120 | */ | |
c0ff4b85 | 2121 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2122 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2123 | } |
2124 | ||
867578cb KH |
2125 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2126 | ||
dc98df5a | 2127 | struct oom_wait_info { |
d79154bb | 2128 | struct mem_cgroup *memcg; |
dc98df5a KH |
2129 | wait_queue_t wait; |
2130 | }; | |
2131 | ||
2132 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2133 | unsigned mode, int sync, void *arg) | |
2134 | { | |
d79154bb HD |
2135 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2136 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2137 | struct oom_wait_info *oom_wait_info; |
2138 | ||
2139 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2140 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2141 | |
dc98df5a | 2142 | /* |
d79154bb | 2143 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2144 | * Then we can use css_is_ancestor without taking care of RCU. |
2145 | */ | |
c0ff4b85 R |
2146 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2147 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2148 | return 0; |
dc98df5a KH |
2149 | return autoremove_wake_function(wait, mode, sync, arg); |
2150 | } | |
2151 | ||
c0ff4b85 | 2152 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2153 | { |
3812c8c8 | 2154 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2155 | /* for filtering, pass "memcg" as argument. */ |
2156 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2157 | } |
2158 | ||
c0ff4b85 | 2159 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2160 | { |
c0ff4b85 R |
2161 | if (memcg && atomic_read(&memcg->under_oom)) |
2162 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2163 | } |
2164 | ||
3812c8c8 | 2165 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2166 | { |
3812c8c8 JW |
2167 | if (!current->memcg_oom.may_oom) |
2168 | return; | |
867578cb | 2169 | /* |
49426420 JW |
2170 | * We are in the middle of the charge context here, so we |
2171 | * don't want to block when potentially sitting on a callstack | |
2172 | * that holds all kinds of filesystem and mm locks. | |
2173 | * | |
2174 | * Also, the caller may handle a failed allocation gracefully | |
2175 | * (like optional page cache readahead) and so an OOM killer | |
2176 | * invocation might not even be necessary. | |
2177 | * | |
2178 | * That's why we don't do anything here except remember the | |
2179 | * OOM context and then deal with it at the end of the page | |
2180 | * fault when the stack is unwound, the locks are released, | |
2181 | * and when we know whether the fault was overall successful. | |
867578cb | 2182 | */ |
49426420 JW |
2183 | css_get(&memcg->css); |
2184 | current->memcg_oom.memcg = memcg; | |
2185 | current->memcg_oom.gfp_mask = mask; | |
2186 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2187 | } |
2188 | ||
2189 | /** | |
2190 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2191 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2192 | * |
49426420 JW |
2193 | * This has to be called at the end of a page fault if the memcg OOM |
2194 | * handler was enabled. | |
3812c8c8 | 2195 | * |
49426420 | 2196 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2197 | * sleep on a waitqueue until the userspace task resolves the |
2198 | * situation. Sleeping directly in the charge context with all kinds | |
2199 | * of locks held is not a good idea, instead we remember an OOM state | |
2200 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2201 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2202 | * |
2203 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2204 | * completed, %false otherwise. |
3812c8c8 | 2205 | */ |
49426420 | 2206 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2207 | { |
49426420 | 2208 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2209 | struct oom_wait_info owait; |
49426420 | 2210 | bool locked; |
3812c8c8 JW |
2211 | |
2212 | /* OOM is global, do not handle */ | |
3812c8c8 | 2213 | if (!memcg) |
49426420 | 2214 | return false; |
3812c8c8 | 2215 | |
49426420 JW |
2216 | if (!handle) |
2217 | goto cleanup; | |
3812c8c8 JW |
2218 | |
2219 | owait.memcg = memcg; | |
2220 | owait.wait.flags = 0; | |
2221 | owait.wait.func = memcg_oom_wake_function; | |
2222 | owait.wait.private = current; | |
2223 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2224 | |
3812c8c8 | 2225 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2226 | mem_cgroup_mark_under_oom(memcg); |
2227 | ||
2228 | locked = mem_cgroup_oom_trylock(memcg); | |
2229 | ||
2230 | if (locked) | |
2231 | mem_cgroup_oom_notify(memcg); | |
2232 | ||
2233 | if (locked && !memcg->oom_kill_disable) { | |
2234 | mem_cgroup_unmark_under_oom(memcg); | |
2235 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2236 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2237 | current->memcg_oom.order); | |
2238 | } else { | |
3812c8c8 | 2239 | schedule(); |
49426420 JW |
2240 | mem_cgroup_unmark_under_oom(memcg); |
2241 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2242 | } | |
2243 | ||
2244 | if (locked) { | |
fb2a6fc5 JW |
2245 | mem_cgroup_oom_unlock(memcg); |
2246 | /* | |
2247 | * There is no guarantee that an OOM-lock contender | |
2248 | * sees the wakeups triggered by the OOM kill | |
2249 | * uncharges. Wake any sleepers explicitely. | |
2250 | */ | |
2251 | memcg_oom_recover(memcg); | |
2252 | } | |
49426420 JW |
2253 | cleanup: |
2254 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2255 | css_put(&memcg->css); |
867578cb | 2256 | return true; |
0b7f569e KH |
2257 | } |
2258 | ||
d69b042f | 2259 | /* |
b5ffc856 | 2260 | * Used to update mapped file or writeback or other statistics. |
32047e2a KH |
2261 | * |
2262 | * Notes: Race condition | |
2263 | * | |
b5ffc856 | 2264 | * We usually use lock_page_cgroup() for accessing page_cgroup member but |
32047e2a KH |
2265 | * it tends to be costly. But considering some conditions, we doesn't need |
2266 | * to do so _always_. | |
2267 | * | |
2268 | * Considering "charge", lock_page_cgroup() is not required because all | |
2269 | * file-stat operations happen after a page is attached to radix-tree. There | |
2270 | * are no race with "charge". | |
2271 | * | |
2272 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2273 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2274 | * if there are race with "uncharge". Statistics itself is properly handled | |
2275 | * by flags. | |
2276 | * | |
2277 | * Considering "move", this is an only case we see a race. To make the race | |
b5ffc856 QH |
2278 | * small, we check memcg->moving_account and detect there are possibility |
2279 | * of race or not. If there is, we take a lock. | |
d69b042f | 2280 | */ |
26174efd | 2281 | |
89c06bd5 KH |
2282 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2283 | bool *locked, unsigned long *flags) | |
2284 | { | |
2285 | struct mem_cgroup *memcg; | |
2286 | struct page_cgroup *pc; | |
2287 | ||
2288 | pc = lookup_page_cgroup(page); | |
2289 | again: | |
2290 | memcg = pc->mem_cgroup; | |
2291 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2292 | return; | |
2293 | /* | |
2294 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2295 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2296 | * rcu_read_lock(), any calls to move_account will be delayed until |
bdcbb659 | 2297 | * rcu_read_unlock(). |
89c06bd5 | 2298 | */ |
bdcbb659 QH |
2299 | VM_BUG_ON(!rcu_read_lock_held()); |
2300 | if (atomic_read(&memcg->moving_account) <= 0) | |
89c06bd5 KH |
2301 | return; |
2302 | ||
2303 | move_lock_mem_cgroup(memcg, flags); | |
2304 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2305 | move_unlock_mem_cgroup(memcg, flags); | |
2306 | goto again; | |
2307 | } | |
2308 | *locked = true; | |
2309 | } | |
2310 | ||
2311 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2312 | { | |
2313 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2314 | ||
2315 | /* | |
2316 | * It's guaranteed that pc->mem_cgroup never changes while | |
2317 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2318 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2319 | */ |
2320 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2321 | } | |
2322 | ||
2a7106f2 | 2323 | void mem_cgroup_update_page_stat(struct page *page, |
68b4876d | 2324 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2325 | { |
c0ff4b85 | 2326 | struct mem_cgroup *memcg; |
32047e2a | 2327 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2328 | unsigned long uninitialized_var(flags); |
d69b042f | 2329 | |
cfa44946 | 2330 | if (mem_cgroup_disabled()) |
d69b042f | 2331 | return; |
89c06bd5 | 2332 | |
658b72c5 | 2333 | VM_BUG_ON(!rcu_read_lock_held()); |
c0ff4b85 R |
2334 | memcg = pc->mem_cgroup; |
2335 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2336 | return; |
26174efd | 2337 | |
c0ff4b85 | 2338 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2339 | } |
26174efd | 2340 | |
cdec2e42 KH |
2341 | /* |
2342 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2343 | * TODO: maybe necessary to use big numbers in big irons. | |
2344 | */ | |
7ec99d62 | 2345 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2346 | struct memcg_stock_pcp { |
2347 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2348 | unsigned int nr_pages; |
cdec2e42 | 2349 | struct work_struct work; |
26fe6168 | 2350 | unsigned long flags; |
a0db00fc | 2351 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2352 | }; |
2353 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2354 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2355 | |
a0956d54 SS |
2356 | /** |
2357 | * consume_stock: Try to consume stocked charge on this cpu. | |
2358 | * @memcg: memcg to consume from. | |
2359 | * @nr_pages: how many pages to charge. | |
2360 | * | |
2361 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2362 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2363 | * service an allocation will refill the stock. | |
2364 | * | |
2365 | * returns true if successful, false otherwise. | |
cdec2e42 | 2366 | */ |
a0956d54 | 2367 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2368 | { |
2369 | struct memcg_stock_pcp *stock; | |
2370 | bool ret = true; | |
2371 | ||
a0956d54 SS |
2372 | if (nr_pages > CHARGE_BATCH) |
2373 | return false; | |
2374 | ||
cdec2e42 | 2375 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2376 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2377 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2378 | else /* need to call res_counter_charge */ |
2379 | ret = false; | |
2380 | put_cpu_var(memcg_stock); | |
2381 | return ret; | |
2382 | } | |
2383 | ||
2384 | /* | |
2385 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2386 | */ | |
2387 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2388 | { | |
2389 | struct mem_cgroup *old = stock->cached; | |
2390 | ||
11c9ea4e JW |
2391 | if (stock->nr_pages) { |
2392 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2393 | ||
2394 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2395 | if (do_swap_account) |
11c9ea4e JW |
2396 | res_counter_uncharge(&old->memsw, bytes); |
2397 | stock->nr_pages = 0; | |
cdec2e42 KH |
2398 | } |
2399 | stock->cached = NULL; | |
cdec2e42 KH |
2400 | } |
2401 | ||
2402 | /* | |
2403 | * This must be called under preempt disabled or must be called by | |
2404 | * a thread which is pinned to local cpu. | |
2405 | */ | |
2406 | static void drain_local_stock(struct work_struct *dummy) | |
2407 | { | |
7c8e0181 | 2408 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2409 | drain_stock(stock); |
26fe6168 | 2410 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2411 | } |
2412 | ||
e4777496 MH |
2413 | static void __init memcg_stock_init(void) |
2414 | { | |
2415 | int cpu; | |
2416 | ||
2417 | for_each_possible_cpu(cpu) { | |
2418 | struct memcg_stock_pcp *stock = | |
2419 | &per_cpu(memcg_stock, cpu); | |
2420 | INIT_WORK(&stock->work, drain_local_stock); | |
2421 | } | |
2422 | } | |
2423 | ||
cdec2e42 KH |
2424 | /* |
2425 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2426 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2427 | */ |
c0ff4b85 | 2428 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2429 | { |
2430 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2431 | ||
c0ff4b85 | 2432 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2433 | drain_stock(stock); |
c0ff4b85 | 2434 | stock->cached = memcg; |
cdec2e42 | 2435 | } |
11c9ea4e | 2436 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2437 | put_cpu_var(memcg_stock); |
2438 | } | |
2439 | ||
2440 | /* | |
c0ff4b85 | 2441 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2442 | * of the hierarchy under it. sync flag says whether we should block |
2443 | * until the work is done. | |
cdec2e42 | 2444 | */ |
c0ff4b85 | 2445 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2446 | { |
26fe6168 | 2447 | int cpu, curcpu; |
d38144b7 | 2448 | |
cdec2e42 | 2449 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2450 | get_online_cpus(); |
5af12d0e | 2451 | curcpu = get_cpu(); |
cdec2e42 KH |
2452 | for_each_online_cpu(cpu) { |
2453 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2454 | struct mem_cgroup *memcg; |
26fe6168 | 2455 | |
c0ff4b85 R |
2456 | memcg = stock->cached; |
2457 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2458 | continue; |
c0ff4b85 | 2459 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2460 | continue; |
d1a05b69 MH |
2461 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2462 | if (cpu == curcpu) | |
2463 | drain_local_stock(&stock->work); | |
2464 | else | |
2465 | schedule_work_on(cpu, &stock->work); | |
2466 | } | |
cdec2e42 | 2467 | } |
5af12d0e | 2468 | put_cpu(); |
d38144b7 MH |
2469 | |
2470 | if (!sync) | |
2471 | goto out; | |
2472 | ||
2473 | for_each_online_cpu(cpu) { | |
2474 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2475 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2476 | flush_work(&stock->work); |
2477 | } | |
2478 | out: | |
f894ffa8 | 2479 | put_online_cpus(); |
d38144b7 MH |
2480 | } |
2481 | ||
2482 | /* | |
2483 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2484 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2485 | * expects some charges will be back to res_counter later but cannot wait for | |
2486 | * it. | |
2487 | */ | |
c0ff4b85 | 2488 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2489 | { |
9f50fad6 MH |
2490 | /* |
2491 | * If someone calls draining, avoid adding more kworker runs. | |
2492 | */ | |
2493 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2494 | return; | |
c0ff4b85 | 2495 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2496 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2497 | } |
2498 | ||
2499 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2500 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2501 | { |
2502 | /* called when force_empty is called */ | |
9f50fad6 | 2503 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2504 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2505 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2506 | } |
2507 | ||
711d3d2c KH |
2508 | /* |
2509 | * This function drains percpu counter value from DEAD cpu and | |
2510 | * move it to local cpu. Note that this function can be preempted. | |
2511 | */ | |
c0ff4b85 | 2512 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2513 | { |
2514 | int i; | |
2515 | ||
c0ff4b85 | 2516 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2517 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2518 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2519 | |
c0ff4b85 R |
2520 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2521 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2522 | } |
e9f8974f | 2523 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2524 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2525 | |
c0ff4b85 R |
2526 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2527 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2528 | } |
c0ff4b85 | 2529 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2530 | } |
2531 | ||
0db0628d | 2532 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2533 | unsigned long action, |
2534 | void *hcpu) | |
2535 | { | |
2536 | int cpu = (unsigned long)hcpu; | |
2537 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2538 | struct mem_cgroup *iter; |
cdec2e42 | 2539 | |
619d094b | 2540 | if (action == CPU_ONLINE) |
1489ebad | 2541 | return NOTIFY_OK; |
1489ebad | 2542 | |
d833049b | 2543 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2544 | return NOTIFY_OK; |
711d3d2c | 2545 | |
9f3a0d09 | 2546 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2547 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2548 | ||
cdec2e42 KH |
2549 | stock = &per_cpu(memcg_stock, cpu); |
2550 | drain_stock(stock); | |
2551 | return NOTIFY_OK; | |
2552 | } | |
2553 | ||
6d1fdc48 JW |
2554 | /** |
2555 | * mem_cgroup_try_charge - try charging a memcg | |
2556 | * @memcg: memcg to charge | |
2557 | * @nr_pages: number of pages to charge | |
38c5d72f | 2558 | * |
6d1fdc48 JW |
2559 | * Returns 0 if @memcg was charged successfully, -EINTR if the charge |
2560 | * was bypassed to root_mem_cgroup, and -ENOMEM if the charge failed. | |
8a9f3ccd | 2561 | */ |
6d1fdc48 JW |
2562 | static int mem_cgroup_try_charge(struct mem_cgroup *memcg, |
2563 | gfp_t gfp_mask, | |
0029e19e | 2564 | unsigned int nr_pages) |
8a9f3ccd | 2565 | { |
7ec99d62 | 2566 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2567 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 JW |
2568 | struct mem_cgroup *mem_over_limit; |
2569 | struct res_counter *fail_res; | |
2570 | unsigned long nr_reclaimed; | |
2571 | unsigned long flags = 0; | |
2572 | unsigned long long size; | |
05b84301 | 2573 | int ret = 0; |
a636b327 | 2574 | |
6539cc05 | 2575 | retry: |
b6b6cc72 MH |
2576 | if (consume_stock(memcg, nr_pages)) |
2577 | goto done; | |
8a9f3ccd | 2578 | |
6539cc05 JW |
2579 | size = batch * PAGE_SIZE; |
2580 | if (!res_counter_charge(&memcg->res, size, &fail_res)) { | |
2581 | if (!do_swap_account) | |
2582 | goto done_restock; | |
2583 | if (!res_counter_charge(&memcg->memsw, size, &fail_res)) | |
2584 | goto done_restock; | |
2585 | res_counter_uncharge(&memcg->res, size); | |
2586 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); | |
2587 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2588 | } else | |
2589 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
7a81b88c | 2590 | |
6539cc05 JW |
2591 | if (batch > nr_pages) { |
2592 | batch = nr_pages; | |
2593 | goto retry; | |
2594 | } | |
6d61ef40 | 2595 | |
06b078fc JW |
2596 | /* |
2597 | * Unlike in global OOM situations, memcg is not in a physical | |
2598 | * memory shortage. Allow dying and OOM-killed tasks to | |
2599 | * bypass the last charges so that they can exit quickly and | |
2600 | * free their memory. | |
2601 | */ | |
2602 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2603 | fatal_signal_pending(current) || | |
2604 | current->flags & PF_EXITING)) | |
2605 | goto bypass; | |
2606 | ||
2607 | if (unlikely(task_in_memcg_oom(current))) | |
2608 | goto nomem; | |
2609 | ||
6539cc05 JW |
2610 | if (!(gfp_mask & __GFP_WAIT)) |
2611 | goto nomem; | |
4b534334 | 2612 | |
6539cc05 JW |
2613 | nr_reclaimed = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
2614 | ||
61e02c74 | 2615 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2616 | goto retry; |
28c34c29 JW |
2617 | |
2618 | if (gfp_mask & __GFP_NORETRY) | |
2619 | goto nomem; | |
6539cc05 JW |
2620 | /* |
2621 | * Even though the limit is exceeded at this point, reclaim | |
2622 | * may have been able to free some pages. Retry the charge | |
2623 | * before killing the task. | |
2624 | * | |
2625 | * Only for regular pages, though: huge pages are rather | |
2626 | * unlikely to succeed so close to the limit, and we fall back | |
2627 | * to regular pages anyway in case of failure. | |
2628 | */ | |
61e02c74 | 2629 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2630 | goto retry; |
2631 | /* | |
2632 | * At task move, charge accounts can be doubly counted. So, it's | |
2633 | * better to wait until the end of task_move if something is going on. | |
2634 | */ | |
2635 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2636 | goto retry; | |
2637 | ||
9b130619 JW |
2638 | if (nr_retries--) |
2639 | goto retry; | |
2640 | ||
06b078fc JW |
2641 | if (gfp_mask & __GFP_NOFAIL) |
2642 | goto bypass; | |
2643 | ||
6539cc05 JW |
2644 | if (fatal_signal_pending(current)) |
2645 | goto bypass; | |
2646 | ||
61e02c74 | 2647 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2648 | nomem: |
6d1fdc48 | 2649 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2650 | return -ENOMEM; |
867578cb | 2651 | bypass: |
05b84301 JW |
2652 | memcg = root_mem_cgroup; |
2653 | ret = -EINTR; | |
2654 | goto retry; | |
6539cc05 JW |
2655 | |
2656 | done_restock: | |
2657 | if (batch > nr_pages) | |
2658 | refill_stock(memcg, batch - nr_pages); | |
2659 | done: | |
05b84301 | 2660 | return ret; |
7a81b88c | 2661 | } |
8a9f3ccd | 2662 | |
6d1fdc48 JW |
2663 | /** |
2664 | * mem_cgroup_try_charge_mm - try charging a mm | |
2665 | * @mm: mm_struct to charge | |
2666 | * @nr_pages: number of pages to charge | |
2667 | * @oom: trigger OOM if reclaim fails | |
2668 | * | |
2669 | * Returns the charged mem_cgroup associated with the given mm_struct or | |
2670 | * NULL the charge failed. | |
2671 | */ | |
2672 | static struct mem_cgroup *mem_cgroup_try_charge_mm(struct mm_struct *mm, | |
2673 | gfp_t gfp_mask, | |
0029e19e | 2674 | unsigned int nr_pages) |
6d1fdc48 JW |
2675 | |
2676 | { | |
2677 | struct mem_cgroup *memcg; | |
2678 | int ret; | |
2679 | ||
2680 | memcg = get_mem_cgroup_from_mm(mm); | |
0029e19e | 2681 | ret = mem_cgroup_try_charge(memcg, gfp_mask, nr_pages); |
6d1fdc48 JW |
2682 | css_put(&memcg->css); |
2683 | if (ret == -EINTR) | |
2684 | memcg = root_mem_cgroup; | |
2685 | else if (ret) | |
2686 | memcg = NULL; | |
2687 | ||
2688 | return memcg; | |
2689 | } | |
2690 | ||
a3032a2c DN |
2691 | /* |
2692 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2693 | * This function is for that and do uncharge, put css's refcnt. | |
2694 | * gotten by try_charge(). | |
2695 | */ | |
c0ff4b85 | 2696 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2697 | unsigned int nr_pages) |
a3032a2c | 2698 | { |
05b84301 | 2699 | unsigned long bytes = nr_pages * PAGE_SIZE; |
e7018b8d | 2700 | |
05b84301 JW |
2701 | res_counter_uncharge(&memcg->res, bytes); |
2702 | if (do_swap_account) | |
2703 | res_counter_uncharge(&memcg->memsw, bytes); | |
854ffa8d DN |
2704 | } |
2705 | ||
d01dd17f KH |
2706 | /* |
2707 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2708 | * This is useful when moving usage to parent cgroup. | |
2709 | */ | |
2710 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2711 | unsigned int nr_pages) | |
2712 | { | |
2713 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2714 | ||
d01dd17f KH |
2715 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); |
2716 | if (do_swap_account) | |
2717 | res_counter_uncharge_until(&memcg->memsw, | |
2718 | memcg->memsw.parent, bytes); | |
2719 | } | |
2720 | ||
a3b2d692 KH |
2721 | /* |
2722 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2723 | * rcu_read_lock(). The caller is responsible for calling |
2724 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2725 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2726 | */ |
2727 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2728 | { | |
a3b2d692 KH |
2729 | /* ID 0 is unused ID */ |
2730 | if (!id) | |
2731 | return NULL; | |
34c00c31 | 2732 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2733 | } |
2734 | ||
e42d9d5d | 2735 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2736 | { |
c0ff4b85 | 2737 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2738 | struct page_cgroup *pc; |
a3b2d692 | 2739 | unsigned short id; |
b5a84319 KH |
2740 | swp_entry_t ent; |
2741 | ||
309381fe | 2742 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2743 | |
3c776e64 | 2744 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2745 | lock_page_cgroup(pc); |
a3b2d692 | 2746 | if (PageCgroupUsed(pc)) { |
c0ff4b85 | 2747 | memcg = pc->mem_cgroup; |
ec903c0c | 2748 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2749 | memcg = NULL; |
e42d9d5d | 2750 | } else if (PageSwapCache(page)) { |
3c776e64 | 2751 | ent.val = page_private(page); |
9fb4b7cc | 2752 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2753 | rcu_read_lock(); |
c0ff4b85 | 2754 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2755 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2756 | memcg = NULL; |
a3b2d692 | 2757 | rcu_read_unlock(); |
3c776e64 | 2758 | } |
c0bd3f63 | 2759 | unlock_page_cgroup(pc); |
c0ff4b85 | 2760 | return memcg; |
b5a84319 KH |
2761 | } |
2762 | ||
c0ff4b85 | 2763 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2764 | struct page *page, |
7ec99d62 | 2765 | unsigned int nr_pages, |
9ce70c02 HD |
2766 | enum charge_type ctype, |
2767 | bool lrucare) | |
7a81b88c | 2768 | { |
ce587e65 | 2769 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2770 | struct zone *uninitialized_var(zone); |
fa9add64 | 2771 | struct lruvec *lruvec; |
9ce70c02 | 2772 | bool was_on_lru = false; |
b2402857 | 2773 | bool anon; |
9ce70c02 | 2774 | |
ca3e0214 | 2775 | lock_page_cgroup(pc); |
309381fe | 2776 | VM_BUG_ON_PAGE(PageCgroupUsed(pc), page); |
ca3e0214 KH |
2777 | /* |
2778 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2779 | * accessed by any other context at this point. | |
2780 | */ | |
9ce70c02 HD |
2781 | |
2782 | /* | |
2783 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2784 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2785 | */ | |
2786 | if (lrucare) { | |
2787 | zone = page_zone(page); | |
2788 | spin_lock_irq(&zone->lru_lock); | |
2789 | if (PageLRU(page)) { | |
fa9add64 | 2790 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2791 | ClearPageLRU(page); |
fa9add64 | 2792 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2793 | was_on_lru = true; |
2794 | } | |
2795 | } | |
2796 | ||
c0ff4b85 | 2797 | pc->mem_cgroup = memcg; |
b2402857 | 2798 | SetPageCgroupUsed(pc); |
3be91277 | 2799 | |
9ce70c02 HD |
2800 | if (lrucare) { |
2801 | if (was_on_lru) { | |
fa9add64 | 2802 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
309381fe | 2803 | VM_BUG_ON_PAGE(PageLRU(page), page); |
9ce70c02 | 2804 | SetPageLRU(page); |
fa9add64 | 2805 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2806 | } |
2807 | spin_unlock_irq(&zone->lru_lock); | |
2808 | } | |
2809 | ||
41326c17 | 2810 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2811 | anon = true; |
2812 | else | |
2813 | anon = false; | |
2814 | ||
b070e65c | 2815 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
52d4b9ac | 2816 | unlock_page_cgroup(pc); |
9ce70c02 | 2817 | |
430e4863 | 2818 | /* |
bb4cc1a8 AM |
2819 | * "charge_statistics" updated event counter. Then, check it. |
2820 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2821 | * if they exceeds softlimit. | |
430e4863 | 2822 | */ |
c0ff4b85 | 2823 | memcg_check_events(memcg, page); |
7a81b88c | 2824 | } |
66e1707b | 2825 | |
7cf27982 GC |
2826 | static DEFINE_MUTEX(set_limit_mutex); |
2827 | ||
7ae1e1d0 | 2828 | #ifdef CONFIG_MEMCG_KMEM |
bd673145 VD |
2829 | /* |
2830 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or | |
2831 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. | |
2832 | */ | |
2833 | static DEFINE_MUTEX(memcg_slab_mutex); | |
2834 | ||
d6441637 VD |
2835 | static DEFINE_MUTEX(activate_kmem_mutex); |
2836 | ||
7ae1e1d0 GC |
2837 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) |
2838 | { | |
2839 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
6de64beb | 2840 | memcg_kmem_is_active(memcg); |
7ae1e1d0 GC |
2841 | } |
2842 | ||
1f458cbf GC |
2843 | /* |
2844 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2845 | * in the memcg_cache_params struct. | |
2846 | */ | |
2847 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2848 | { | |
2849 | struct kmem_cache *cachep; | |
2850 | ||
2851 | VM_BUG_ON(p->is_root_cache); | |
2852 | cachep = p->root_cache; | |
7a67d7ab | 2853 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2854 | } |
2855 | ||
749c5415 | 2856 | #ifdef CONFIG_SLABINFO |
2da8ca82 | 2857 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
749c5415 | 2858 | { |
2da8ca82 | 2859 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
749c5415 GC |
2860 | struct memcg_cache_params *params; |
2861 | ||
2862 | if (!memcg_can_account_kmem(memcg)) | |
2863 | return -EIO; | |
2864 | ||
2865 | print_slabinfo_header(m); | |
2866 | ||
bd673145 | 2867 | mutex_lock(&memcg_slab_mutex); |
749c5415 GC |
2868 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) |
2869 | cache_show(memcg_params_to_cache(params), m); | |
bd673145 | 2870 | mutex_unlock(&memcg_slab_mutex); |
749c5415 GC |
2871 | |
2872 | return 0; | |
2873 | } | |
2874 | #endif | |
2875 | ||
c67a8a68 | 2876 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
7ae1e1d0 GC |
2877 | { |
2878 | struct res_counter *fail_res; | |
7ae1e1d0 | 2879 | int ret = 0; |
7ae1e1d0 GC |
2880 | |
2881 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2882 | if (ret) | |
2883 | return ret; | |
2884 | ||
0029e19e | 2885 | ret = mem_cgroup_try_charge(memcg, gfp, size >> PAGE_SHIFT); |
7ae1e1d0 GC |
2886 | if (ret == -EINTR) { |
2887 | /* | |
6d1fdc48 | 2888 | * mem_cgroup_try_charge() chosed to bypass to root due to |
7ae1e1d0 GC |
2889 | * OOM kill or fatal signal. Since our only options are to |
2890 | * either fail the allocation or charge it to this cgroup, do | |
2891 | * it as a temporary condition. But we can't fail. From a | |
2892 | * kmem/slab perspective, the cache has already been selected, | |
2893 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
2894 | * our minds. | |
2895 | * | |
2896 | * This condition will only trigger if the task entered | |
2897 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
6d1fdc48 | 2898 | * mem_cgroup_try_charge() above. Tasks that were already |
7ae1e1d0 GC |
2899 | * dying when the allocation triggers should have been already |
2900 | * directed to the root cgroup in memcontrol.h | |
2901 | */ | |
2902 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
2903 | if (do_swap_account) | |
2904 | res_counter_charge_nofail(&memcg->memsw, size, | |
2905 | &fail_res); | |
2906 | ret = 0; | |
2907 | } else if (ret) | |
2908 | res_counter_uncharge(&memcg->kmem, size); | |
2909 | ||
2910 | return ret; | |
2911 | } | |
2912 | ||
c67a8a68 | 2913 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) |
7ae1e1d0 | 2914 | { |
7ae1e1d0 GC |
2915 | res_counter_uncharge(&memcg->res, size); |
2916 | if (do_swap_account) | |
2917 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
2918 | |
2919 | /* Not down to 0 */ | |
2920 | if (res_counter_uncharge(&memcg->kmem, size)) | |
2921 | return; | |
2922 | ||
10d5ebf4 LZ |
2923 | /* |
2924 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
2925 | * this last uncharge is racing with the offlining code or it is | |
2926 | * outliving the memcg existence. | |
2927 | * | |
2928 | * The memory barrier imposed by test&clear is paired with the | |
2929 | * explicit one in memcg_kmem_mark_dead(). | |
2930 | */ | |
7de37682 | 2931 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 2932 | css_put(&memcg->css); |
7ae1e1d0 GC |
2933 | } |
2934 | ||
2633d7a0 GC |
2935 | /* |
2936 | * helper for acessing a memcg's index. It will be used as an index in the | |
2937 | * child cache array in kmem_cache, and also to derive its name. This function | |
2938 | * will return -1 when this is not a kmem-limited memcg. | |
2939 | */ | |
2940 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2941 | { | |
2942 | return memcg ? memcg->kmemcg_id : -1; | |
2943 | } | |
2944 | ||
55007d84 GC |
2945 | static size_t memcg_caches_array_size(int num_groups) |
2946 | { | |
2947 | ssize_t size; | |
2948 | if (num_groups <= 0) | |
2949 | return 0; | |
2950 | ||
2951 | size = 2 * num_groups; | |
2952 | if (size < MEMCG_CACHES_MIN_SIZE) | |
2953 | size = MEMCG_CACHES_MIN_SIZE; | |
2954 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2955 | size = MEMCG_CACHES_MAX_SIZE; | |
2956 | ||
2957 | return size; | |
2958 | } | |
2959 | ||
2960 | /* | |
2961 | * We should update the current array size iff all caches updates succeed. This | |
2962 | * can only be done from the slab side. The slab mutex needs to be held when | |
2963 | * calling this. | |
2964 | */ | |
2965 | void memcg_update_array_size(int num) | |
2966 | { | |
2967 | if (num > memcg_limited_groups_array_size) | |
2968 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
2969 | } | |
2970 | ||
2971 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) | |
2972 | { | |
2973 | struct memcg_cache_params *cur_params = s->memcg_params; | |
2974 | ||
f35c3a8e | 2975 | VM_BUG_ON(!is_root_cache(s)); |
55007d84 GC |
2976 | |
2977 | if (num_groups > memcg_limited_groups_array_size) { | |
2978 | int i; | |
f8570263 | 2979 | struct memcg_cache_params *new_params; |
55007d84 GC |
2980 | ssize_t size = memcg_caches_array_size(num_groups); |
2981 | ||
2982 | size *= sizeof(void *); | |
90c7a79c | 2983 | size += offsetof(struct memcg_cache_params, memcg_caches); |
55007d84 | 2984 | |
f8570263 VD |
2985 | new_params = kzalloc(size, GFP_KERNEL); |
2986 | if (!new_params) | |
55007d84 | 2987 | return -ENOMEM; |
55007d84 | 2988 | |
f8570263 | 2989 | new_params->is_root_cache = true; |
55007d84 GC |
2990 | |
2991 | /* | |
2992 | * There is the chance it will be bigger than | |
2993 | * memcg_limited_groups_array_size, if we failed an allocation | |
2994 | * in a cache, in which case all caches updated before it, will | |
2995 | * have a bigger array. | |
2996 | * | |
2997 | * But if that is the case, the data after | |
2998 | * memcg_limited_groups_array_size is certainly unused | |
2999 | */ | |
3000 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3001 | if (!cur_params->memcg_caches[i]) | |
3002 | continue; | |
f8570263 | 3003 | new_params->memcg_caches[i] = |
55007d84 GC |
3004 | cur_params->memcg_caches[i]; |
3005 | } | |
3006 | ||
3007 | /* | |
3008 | * Ideally, we would wait until all caches succeed, and only | |
3009 | * then free the old one. But this is not worth the extra | |
3010 | * pointer per-cache we'd have to have for this. | |
3011 | * | |
3012 | * It is not a big deal if some caches are left with a size | |
3013 | * bigger than the others. And all updates will reset this | |
3014 | * anyway. | |
3015 | */ | |
f8570263 VD |
3016 | rcu_assign_pointer(s->memcg_params, new_params); |
3017 | if (cur_params) | |
3018 | kfree_rcu(cur_params, rcu_head); | |
55007d84 GC |
3019 | } |
3020 | return 0; | |
3021 | } | |
3022 | ||
363a044f VD |
3023 | int memcg_alloc_cache_params(struct mem_cgroup *memcg, struct kmem_cache *s, |
3024 | struct kmem_cache *root_cache) | |
2633d7a0 | 3025 | { |
90c7a79c | 3026 | size_t size; |
2633d7a0 GC |
3027 | |
3028 | if (!memcg_kmem_enabled()) | |
3029 | return 0; | |
3030 | ||
90c7a79c AV |
3031 | if (!memcg) { |
3032 | size = offsetof(struct memcg_cache_params, memcg_caches); | |
55007d84 | 3033 | size += memcg_limited_groups_array_size * sizeof(void *); |
90c7a79c AV |
3034 | } else |
3035 | size = sizeof(struct memcg_cache_params); | |
55007d84 | 3036 | |
2633d7a0 GC |
3037 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3038 | if (!s->memcg_params) | |
3039 | return -ENOMEM; | |
3040 | ||
943a451a | 3041 | if (memcg) { |
2633d7a0 | 3042 | s->memcg_params->memcg = memcg; |
943a451a | 3043 | s->memcg_params->root_cache = root_cache; |
051dd460 | 3044 | css_get(&memcg->css); |
4ba902b5 GC |
3045 | } else |
3046 | s->memcg_params->is_root_cache = true; | |
3047 | ||
2633d7a0 GC |
3048 | return 0; |
3049 | } | |
3050 | ||
363a044f VD |
3051 | void memcg_free_cache_params(struct kmem_cache *s) |
3052 | { | |
051dd460 VD |
3053 | if (!s->memcg_params) |
3054 | return; | |
3055 | if (!s->memcg_params->is_root_cache) | |
3056 | css_put(&s->memcg_params->memcg->css); | |
363a044f VD |
3057 | kfree(s->memcg_params); |
3058 | } | |
3059 | ||
776ed0f0 VD |
3060 | static void memcg_register_cache(struct mem_cgroup *memcg, |
3061 | struct kmem_cache *root_cache) | |
2633d7a0 | 3062 | { |
93f39eea VD |
3063 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by |
3064 | memcg_slab_mutex */ | |
bd673145 | 3065 | struct kmem_cache *cachep; |
d7f25f8a GC |
3066 | int id; |
3067 | ||
bd673145 VD |
3068 | lockdep_assert_held(&memcg_slab_mutex); |
3069 | ||
3070 | id = memcg_cache_id(memcg); | |
3071 | ||
3072 | /* | |
3073 | * Since per-memcg caches are created asynchronously on first | |
3074 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
3075 | * create the same cache, but only one of them may succeed. | |
3076 | */ | |
3077 | if (cache_from_memcg_idx(root_cache, id)) | |
1aa13254 VD |
3078 | return; |
3079 | ||
073ee1c6 | 3080 | cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1); |
776ed0f0 | 3081 | cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf); |
2edefe11 | 3082 | /* |
bd673145 VD |
3083 | * If we could not create a memcg cache, do not complain, because |
3084 | * that's not critical at all as we can always proceed with the root | |
3085 | * cache. | |
2edefe11 | 3086 | */ |
bd673145 VD |
3087 | if (!cachep) |
3088 | return; | |
2edefe11 | 3089 | |
bd673145 | 3090 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
1aa13254 | 3091 | |
d7f25f8a | 3092 | /* |
959c8963 VD |
3093 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
3094 | * barrier here to ensure nobody will see the kmem_cache partially | |
3095 | * initialized. | |
d7f25f8a | 3096 | */ |
959c8963 VD |
3097 | smp_wmb(); |
3098 | ||
bd673145 VD |
3099 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
3100 | root_cache->memcg_params->memcg_caches[id] = cachep; | |
1aa13254 | 3101 | } |
d7f25f8a | 3102 | |
776ed0f0 | 3103 | static void memcg_unregister_cache(struct kmem_cache *cachep) |
1aa13254 | 3104 | { |
bd673145 | 3105 | struct kmem_cache *root_cache; |
1aa13254 VD |
3106 | struct mem_cgroup *memcg; |
3107 | int id; | |
3108 | ||
bd673145 | 3109 | lockdep_assert_held(&memcg_slab_mutex); |
d7f25f8a | 3110 | |
bd673145 | 3111 | BUG_ON(is_root_cache(cachep)); |
2edefe11 | 3112 | |
bd673145 VD |
3113 | root_cache = cachep->memcg_params->root_cache; |
3114 | memcg = cachep->memcg_params->memcg; | |
96403da2 | 3115 | id = memcg_cache_id(memcg); |
d7f25f8a | 3116 | |
bd673145 VD |
3117 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
3118 | root_cache->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a | 3119 | |
bd673145 VD |
3120 | list_del(&cachep->memcg_params->list); |
3121 | ||
3122 | kmem_cache_destroy(cachep); | |
2633d7a0 GC |
3123 | } |
3124 | ||
0e9d92f2 GC |
3125 | /* |
3126 | * During the creation a new cache, we need to disable our accounting mechanism | |
3127 | * altogether. This is true even if we are not creating, but rather just | |
3128 | * enqueing new caches to be created. | |
3129 | * | |
3130 | * This is because that process will trigger allocations; some visible, like | |
3131 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3132 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3133 | * objects during debug. | |
3134 | * | |
3135 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3136 | * to it. This may not be a bounded recursion: since the first cache creation | |
3137 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3138 | * cache again, failing at the same point. | |
3139 | * | |
3140 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3141 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3142 | * inside the following two functions. | |
3143 | */ | |
3144 | static inline void memcg_stop_kmem_account(void) | |
3145 | { | |
3146 | VM_BUG_ON(!current->mm); | |
3147 | current->memcg_kmem_skip_account++; | |
3148 | } | |
3149 | ||
3150 | static inline void memcg_resume_kmem_account(void) | |
3151 | { | |
3152 | VM_BUG_ON(!current->mm); | |
3153 | current->memcg_kmem_skip_account--; | |
3154 | } | |
3155 | ||
776ed0f0 | 3156 | int __memcg_cleanup_cache_params(struct kmem_cache *s) |
7cf27982 GC |
3157 | { |
3158 | struct kmem_cache *c; | |
b8529907 | 3159 | int i, failed = 0; |
7cf27982 | 3160 | |
bd673145 | 3161 | mutex_lock(&memcg_slab_mutex); |
7a67d7ab QH |
3162 | for_each_memcg_cache_index(i) { |
3163 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
3164 | if (!c) |
3165 | continue; | |
3166 | ||
776ed0f0 | 3167 | memcg_unregister_cache(c); |
b8529907 VD |
3168 | |
3169 | if (cache_from_memcg_idx(s, i)) | |
3170 | failed++; | |
7cf27982 | 3171 | } |
bd673145 | 3172 | mutex_unlock(&memcg_slab_mutex); |
b8529907 | 3173 | return failed; |
7cf27982 GC |
3174 | } |
3175 | ||
776ed0f0 | 3176 | static void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3177 | { |
3178 | struct kmem_cache *cachep; | |
bd673145 | 3179 | struct memcg_cache_params *params, *tmp; |
1f458cbf GC |
3180 | |
3181 | if (!memcg_kmem_is_active(memcg)) | |
3182 | return; | |
3183 | ||
bd673145 VD |
3184 | mutex_lock(&memcg_slab_mutex); |
3185 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { | |
1f458cbf | 3186 | cachep = memcg_params_to_cache(params); |
bd673145 VD |
3187 | kmem_cache_shrink(cachep); |
3188 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
776ed0f0 | 3189 | memcg_unregister_cache(cachep); |
1f458cbf | 3190 | } |
bd673145 | 3191 | mutex_unlock(&memcg_slab_mutex); |
1f458cbf GC |
3192 | } |
3193 | ||
776ed0f0 | 3194 | struct memcg_register_cache_work { |
5722d094 VD |
3195 | struct mem_cgroup *memcg; |
3196 | struct kmem_cache *cachep; | |
3197 | struct work_struct work; | |
3198 | }; | |
3199 | ||
776ed0f0 | 3200 | static void memcg_register_cache_func(struct work_struct *w) |
d7f25f8a | 3201 | { |
776ed0f0 VD |
3202 | struct memcg_register_cache_work *cw = |
3203 | container_of(w, struct memcg_register_cache_work, work); | |
5722d094 VD |
3204 | struct mem_cgroup *memcg = cw->memcg; |
3205 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 3206 | |
bd673145 | 3207 | mutex_lock(&memcg_slab_mutex); |
776ed0f0 | 3208 | memcg_register_cache(memcg, cachep); |
bd673145 VD |
3209 | mutex_unlock(&memcg_slab_mutex); |
3210 | ||
5722d094 | 3211 | css_put(&memcg->css); |
d7f25f8a GC |
3212 | kfree(cw); |
3213 | } | |
3214 | ||
3215 | /* | |
3216 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3217 | */ |
776ed0f0 VD |
3218 | static void __memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3219 | struct kmem_cache *cachep) | |
d7f25f8a | 3220 | { |
776ed0f0 | 3221 | struct memcg_register_cache_work *cw; |
d7f25f8a | 3222 | |
776ed0f0 | 3223 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
ca0dde97 LZ |
3224 | if (cw == NULL) { |
3225 | css_put(&memcg->css); | |
d7f25f8a GC |
3226 | return; |
3227 | } | |
3228 | ||
3229 | cw->memcg = memcg; | |
3230 | cw->cachep = cachep; | |
3231 | ||
776ed0f0 | 3232 | INIT_WORK(&cw->work, memcg_register_cache_func); |
d7f25f8a GC |
3233 | schedule_work(&cw->work); |
3234 | } | |
3235 | ||
776ed0f0 VD |
3236 | static void memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3237 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
3238 | { |
3239 | /* | |
3240 | * We need to stop accounting when we kmalloc, because if the | |
3241 | * corresponding kmalloc cache is not yet created, the first allocation | |
776ed0f0 | 3242 | * in __memcg_schedule_register_cache will recurse. |
0e9d92f2 GC |
3243 | * |
3244 | * However, it is better to enclose the whole function. Depending on | |
3245 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3246 | * trigger an allocation. This too, will make us recurse. Because at | |
3247 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3248 | * the safest choice is to do it like this, wrapping the whole function. | |
3249 | */ | |
3250 | memcg_stop_kmem_account(); | |
776ed0f0 | 3251 | __memcg_schedule_register_cache(memcg, cachep); |
0e9d92f2 GC |
3252 | memcg_resume_kmem_account(); |
3253 | } | |
c67a8a68 VD |
3254 | |
3255 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) | |
3256 | { | |
3257 | int res; | |
3258 | ||
3259 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, | |
3260 | PAGE_SIZE << order); | |
3261 | if (!res) | |
3262 | atomic_add(1 << order, &cachep->memcg_params->nr_pages); | |
3263 | return res; | |
3264 | } | |
3265 | ||
3266 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) | |
3267 | { | |
3268 | memcg_uncharge_kmem(cachep->memcg_params->memcg, PAGE_SIZE << order); | |
3269 | atomic_sub(1 << order, &cachep->memcg_params->nr_pages); | |
3270 | } | |
3271 | ||
d7f25f8a GC |
3272 | /* |
3273 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3274 | * We try to use the current memcg's version of the cache. | |
3275 | * | |
3276 | * If the cache does not exist yet, if we are the first user of it, | |
3277 | * we either create it immediately, if possible, or create it asynchronously | |
3278 | * in a workqueue. | |
3279 | * In the latter case, we will let the current allocation go through with | |
3280 | * the original cache. | |
3281 | * | |
3282 | * Can't be called in interrupt context or from kernel threads. | |
3283 | * This function needs to be called with rcu_read_lock() held. | |
3284 | */ | |
3285 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3286 | gfp_t gfp) | |
3287 | { | |
3288 | struct mem_cgroup *memcg; | |
959c8963 | 3289 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
3290 | |
3291 | VM_BUG_ON(!cachep->memcg_params); | |
3292 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3293 | ||
0e9d92f2 GC |
3294 | if (!current->mm || current->memcg_kmem_skip_account) |
3295 | return cachep; | |
3296 | ||
d7f25f8a GC |
3297 | rcu_read_lock(); |
3298 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3299 | |
3300 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3301 | goto out; |
d7f25f8a | 3302 | |
959c8963 VD |
3303 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
3304 | if (likely(memcg_cachep)) { | |
3305 | cachep = memcg_cachep; | |
ca0dde97 | 3306 | goto out; |
d7f25f8a GC |
3307 | } |
3308 | ||
ca0dde97 | 3309 | /* The corresponding put will be done in the workqueue. */ |
ec903c0c | 3310 | if (!css_tryget_online(&memcg->css)) |
ca0dde97 LZ |
3311 | goto out; |
3312 | rcu_read_unlock(); | |
3313 | ||
3314 | /* | |
3315 | * If we are in a safe context (can wait, and not in interrupt | |
3316 | * context), we could be be predictable and return right away. | |
3317 | * This would guarantee that the allocation being performed | |
3318 | * already belongs in the new cache. | |
3319 | * | |
3320 | * However, there are some clashes that can arrive from locking. | |
3321 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
3322 | * memcg_create_kmem_cache, this means no further allocation |
3323 | * could happen with the slab_mutex held. So it's better to | |
3324 | * defer everything. | |
ca0dde97 | 3325 | */ |
776ed0f0 | 3326 | memcg_schedule_register_cache(memcg, cachep); |
ca0dde97 LZ |
3327 | return cachep; |
3328 | out: | |
3329 | rcu_read_unlock(); | |
3330 | return cachep; | |
d7f25f8a | 3331 | } |
d7f25f8a | 3332 | |
7ae1e1d0 GC |
3333 | /* |
3334 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3335 | * possible for the given order. But the page is not allocated yet, so we'll | |
3336 | * need a further commit step to do the final arrangements. | |
3337 | * | |
3338 | * It is possible for the task to switch cgroups in this mean time, so at | |
3339 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3340 | * the handle argument to return to the caller which cgroup we should commit | |
3341 | * against. We could also return the memcg directly and avoid the pointer | |
3342 | * passing, but a boolean return value gives better semantics considering | |
3343 | * the compiled-out case as well. | |
3344 | * | |
3345 | * Returning true means the allocation is possible. | |
3346 | */ | |
3347 | bool | |
3348 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3349 | { | |
3350 | struct mem_cgroup *memcg; | |
3351 | int ret; | |
3352 | ||
3353 | *_memcg = NULL; | |
6d42c232 GC |
3354 | |
3355 | /* | |
3356 | * Disabling accounting is only relevant for some specific memcg | |
3357 | * internal allocations. Therefore we would initially not have such | |
52383431 VD |
3358 | * check here, since direct calls to the page allocator that are |
3359 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen | |
3360 | * outside memcg core. We are mostly concerned with cache allocations, | |
3361 | * and by having this test at memcg_kmem_get_cache, we are already able | |
3362 | * to relay the allocation to the root cache and bypass the memcg cache | |
3363 | * altogether. | |
6d42c232 GC |
3364 | * |
3365 | * There is one exception, though: the SLUB allocator does not create | |
3366 | * large order caches, but rather service large kmallocs directly from | |
3367 | * the page allocator. Therefore, the following sequence when backed by | |
3368 | * the SLUB allocator: | |
3369 | * | |
f894ffa8 AM |
3370 | * memcg_stop_kmem_account(); |
3371 | * kmalloc(<large_number>) | |
3372 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3373 | * |
3374 | * would effectively ignore the fact that we should skip accounting, | |
3375 | * since it will drive us directly to this function without passing | |
3376 | * through the cache selector memcg_kmem_get_cache. Such large | |
3377 | * allocations are extremely rare but can happen, for instance, for the | |
3378 | * cache arrays. We bring this test here. | |
3379 | */ | |
3380 | if (!current->mm || current->memcg_kmem_skip_account) | |
3381 | return true; | |
3382 | ||
df381975 | 3383 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 GC |
3384 | |
3385 | if (!memcg_can_account_kmem(memcg)) { | |
3386 | css_put(&memcg->css); | |
3387 | return true; | |
3388 | } | |
3389 | ||
7ae1e1d0 GC |
3390 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3391 | if (!ret) | |
3392 | *_memcg = memcg; | |
7ae1e1d0 GC |
3393 | |
3394 | css_put(&memcg->css); | |
3395 | return (ret == 0); | |
3396 | } | |
3397 | ||
3398 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3399 | int order) | |
3400 | { | |
3401 | struct page_cgroup *pc; | |
3402 | ||
3403 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3404 | ||
3405 | /* The page allocation failed. Revert */ | |
3406 | if (!page) { | |
3407 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3408 | return; |
3409 | } | |
a840cda6 JW |
3410 | /* |
3411 | * The page is freshly allocated and not visible to any | |
3412 | * outside callers yet. Set up pc non-atomically. | |
3413 | */ | |
7ae1e1d0 | 3414 | pc = lookup_page_cgroup(page); |
7ae1e1d0 | 3415 | pc->mem_cgroup = memcg; |
a840cda6 | 3416 | pc->flags = PCG_USED; |
7ae1e1d0 GC |
3417 | } |
3418 | ||
3419 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3420 | { | |
3421 | struct mem_cgroup *memcg = NULL; | |
3422 | struct page_cgroup *pc; | |
3423 | ||
3424 | ||
3425 | pc = lookup_page_cgroup(page); | |
7ae1e1d0 GC |
3426 | if (!PageCgroupUsed(pc)) |
3427 | return; | |
3428 | ||
a840cda6 JW |
3429 | memcg = pc->mem_cgroup; |
3430 | pc->flags = 0; | |
7ae1e1d0 GC |
3431 | |
3432 | /* | |
3433 | * We trust that only if there is a memcg associated with the page, it | |
3434 | * is a valid allocation | |
3435 | */ | |
3436 | if (!memcg) | |
3437 | return; | |
3438 | ||
309381fe | 3439 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
7ae1e1d0 | 3440 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); |
7ae1e1d0 | 3441 | } |
1f458cbf | 3442 | #else |
776ed0f0 | 3443 | static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3444 | { |
3445 | } | |
7ae1e1d0 GC |
3446 | #endif /* CONFIG_MEMCG_KMEM */ |
3447 | ||
ca3e0214 KH |
3448 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3449 | ||
a0db00fc | 3450 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3451 | /* |
3452 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3453 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3454 | * charge/uncharge will be never happen and move_account() is done under | |
3455 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3456 | */ |
e94c8a9c | 3457 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3458 | { |
3459 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3460 | struct page_cgroup *pc; |
b070e65c | 3461 | struct mem_cgroup *memcg; |
e94c8a9c | 3462 | int i; |
ca3e0214 | 3463 | |
3d37c4a9 KH |
3464 | if (mem_cgroup_disabled()) |
3465 | return; | |
b070e65c DR |
3466 | |
3467 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3468 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3469 | pc = head_pc + i; | |
b070e65c | 3470 | pc->mem_cgroup = memcg; |
e94c8a9c KH |
3471 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3472 | } | |
b070e65c DR |
3473 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3474 | HPAGE_PMD_NR); | |
ca3e0214 | 3475 | } |
12d27107 | 3476 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3477 | |
f817ed48 | 3478 | /** |
de3638d9 | 3479 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3480 | * @page: the page |
7ec99d62 | 3481 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3482 | * @pc: page_cgroup of the page. |
3483 | * @from: mem_cgroup which the page is moved from. | |
3484 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3485 | * | |
3486 | * The caller must confirm following. | |
08e552c6 | 3487 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3488 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3489 | * |
2f3479b1 KH |
3490 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3491 | * from old cgroup. | |
f817ed48 | 3492 | */ |
7ec99d62 JW |
3493 | static int mem_cgroup_move_account(struct page *page, |
3494 | unsigned int nr_pages, | |
3495 | struct page_cgroup *pc, | |
3496 | struct mem_cgroup *from, | |
2f3479b1 | 3497 | struct mem_cgroup *to) |
f817ed48 | 3498 | { |
de3638d9 JW |
3499 | unsigned long flags; |
3500 | int ret; | |
b2402857 | 3501 | bool anon = PageAnon(page); |
987eba66 | 3502 | |
f817ed48 | 3503 | VM_BUG_ON(from == to); |
309381fe | 3504 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3505 | /* |
3506 | * The page is isolated from LRU. So, collapse function | |
3507 | * will not handle this page. But page splitting can happen. | |
3508 | * Do this check under compound_page_lock(). The caller should | |
3509 | * hold it. | |
3510 | */ | |
3511 | ret = -EBUSY; | |
7ec99d62 | 3512 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3513 | goto out; |
3514 | ||
3515 | lock_page_cgroup(pc); | |
3516 | ||
3517 | ret = -EINVAL; | |
3518 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3519 | goto unlock; | |
3520 | ||
312734c0 | 3521 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3522 | |
59d1d256 JW |
3523 | if (!anon && page_mapped(page)) { |
3524 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3525 | nr_pages); | |
3526 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3527 | nr_pages); | |
3528 | } | |
3ea67d06 | 3529 | |
59d1d256 JW |
3530 | if (PageWriteback(page)) { |
3531 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3532 | nr_pages); | |
3533 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3534 | nr_pages); | |
3535 | } | |
3ea67d06 | 3536 | |
b070e65c | 3537 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
d69b042f | 3538 | |
854ffa8d | 3539 | /* caller should have done css_get */ |
08e552c6 | 3540 | pc->mem_cgroup = to; |
b070e65c | 3541 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
312734c0 | 3542 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3543 | ret = 0; |
3544 | unlock: | |
57f9fd7d | 3545 | unlock_page_cgroup(pc); |
d2265e6f KH |
3546 | /* |
3547 | * check events | |
3548 | */ | |
5564e88b JW |
3549 | memcg_check_events(to, page); |
3550 | memcg_check_events(from, page); | |
de3638d9 | 3551 | out: |
f817ed48 KH |
3552 | return ret; |
3553 | } | |
3554 | ||
2ef37d3f MH |
3555 | /** |
3556 | * mem_cgroup_move_parent - moves page to the parent group | |
3557 | * @page: the page to move | |
3558 | * @pc: page_cgroup of the page | |
3559 | * @child: page's cgroup | |
3560 | * | |
3561 | * move charges to its parent or the root cgroup if the group has no | |
3562 | * parent (aka use_hierarchy==0). | |
3563 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3564 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3565 | * it signals a race with a page removal/uncharge or migration. In the | |
3566 | * first case the page is on the way out and it will vanish from the LRU | |
3567 | * on the next attempt and the call should be retried later. | |
3568 | * Isolation from the LRU fails only if page has been isolated from | |
3569 | * the LRU since we looked at it and that usually means either global | |
3570 | * reclaim or migration going on. The page will either get back to the | |
3571 | * LRU or vanish. | |
3572 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3573 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3574 | * disappear in the next attempt. | |
f817ed48 | 3575 | */ |
5564e88b JW |
3576 | static int mem_cgroup_move_parent(struct page *page, |
3577 | struct page_cgroup *pc, | |
6068bf01 | 3578 | struct mem_cgroup *child) |
f817ed48 | 3579 | { |
f817ed48 | 3580 | struct mem_cgroup *parent; |
7ec99d62 | 3581 | unsigned int nr_pages; |
4be4489f | 3582 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3583 | int ret; |
3584 | ||
d8423011 | 3585 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3586 | |
57f9fd7d DN |
3587 | ret = -EBUSY; |
3588 | if (!get_page_unless_zero(page)) | |
3589 | goto out; | |
3590 | if (isolate_lru_page(page)) | |
3591 | goto put; | |
52dbb905 | 3592 | |
7ec99d62 | 3593 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3594 | |
cc926f78 KH |
3595 | parent = parent_mem_cgroup(child); |
3596 | /* | |
3597 | * If no parent, move charges to root cgroup. | |
3598 | */ | |
3599 | if (!parent) | |
3600 | parent = root_mem_cgroup; | |
f817ed48 | 3601 | |
2ef37d3f | 3602 | if (nr_pages > 1) { |
309381fe | 3603 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
987eba66 | 3604 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3605 | } |
987eba66 | 3606 | |
cc926f78 | 3607 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3608 | pc, child, parent); |
cc926f78 KH |
3609 | if (!ret) |
3610 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3611 | |
7ec99d62 | 3612 | if (nr_pages > 1) |
987eba66 | 3613 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3614 | putback_lru_page(page); |
57f9fd7d | 3615 | put: |
40d58138 | 3616 | put_page(page); |
57f9fd7d | 3617 | out: |
f817ed48 KH |
3618 | return ret; |
3619 | } | |
3620 | ||
d715ae08 | 3621 | int mem_cgroup_charge_anon(struct page *page, |
1bec6b33 | 3622 | struct mm_struct *mm, gfp_t gfp_mask) |
7a81b88c | 3623 | { |
7ec99d62 | 3624 | unsigned int nr_pages = 1; |
6d1fdc48 | 3625 | struct mem_cgroup *memcg; |
ec168510 | 3626 | |
1bec6b33 JW |
3627 | if (mem_cgroup_disabled()) |
3628 | return 0; | |
3629 | ||
3630 | VM_BUG_ON_PAGE(page_mapped(page), page); | |
3631 | VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page); | |
3632 | VM_BUG_ON(!mm); | |
3633 | ||
37c2ac78 | 3634 | if (PageTransHuge(page)) { |
7ec99d62 | 3635 | nr_pages <<= compound_order(page); |
309381fe | 3636 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
37c2ac78 | 3637 | } |
7a81b88c | 3638 | |
0029e19e | 3639 | memcg = mem_cgroup_try_charge_mm(mm, gfp_mask, nr_pages); |
6d1fdc48 JW |
3640 | if (!memcg) |
3641 | return -ENOMEM; | |
1bec6b33 JW |
3642 | __mem_cgroup_commit_charge(memcg, page, nr_pages, |
3643 | MEM_CGROUP_CHARGE_TYPE_ANON, false); | |
8a9f3ccd | 3644 | return 0; |
8a9f3ccd BS |
3645 | } |
3646 | ||
54595fe2 KH |
3647 | /* |
3648 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3649 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3650 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3651 | * "commit()" or removed by "cancel()" |
3652 | */ | |
0435a2fd JW |
3653 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3654 | struct page *page, | |
3655 | gfp_t mask, | |
3656 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3657 | { |
6d1fdc48 | 3658 | struct mem_cgroup *memcg = NULL; |
90deb788 | 3659 | struct page_cgroup *pc; |
54595fe2 | 3660 | int ret; |
8c7c6e34 | 3661 | |
90deb788 JW |
3662 | pc = lookup_page_cgroup(page); |
3663 | /* | |
3664 | * Every swap fault against a single page tries to charge the | |
3665 | * page, bail as early as possible. shmem_unuse() encounters | |
3666 | * already charged pages, too. The USED bit is protected by | |
3667 | * the page lock, which serializes swap cache removal, which | |
3668 | * in turn serializes uncharging. | |
3669 | */ | |
3670 | if (PageCgroupUsed(pc)) | |
6d1fdc48 JW |
3671 | goto out; |
3672 | if (do_swap_account) | |
3673 | memcg = try_get_mem_cgroup_from_page(page); | |
c0ff4b85 | 3674 | if (!memcg) |
6d1fdc48 | 3675 | memcg = get_mem_cgroup_from_mm(mm); |
0029e19e | 3676 | ret = mem_cgroup_try_charge(memcg, mask, 1); |
c0ff4b85 | 3677 | css_put(&memcg->css); |
38c5d72f | 3678 | if (ret == -EINTR) |
6d1fdc48 JW |
3679 | memcg = root_mem_cgroup; |
3680 | else if (ret) | |
3681 | return ret; | |
3682 | out: | |
3683 | *memcgp = memcg; | |
3684 | return 0; | |
8c7c6e34 KH |
3685 | } |
3686 | ||
0435a2fd JW |
3687 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
3688 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
3689 | { | |
6d1fdc48 JW |
3690 | if (mem_cgroup_disabled()) { |
3691 | *memcgp = NULL; | |
0435a2fd | 3692 | return 0; |
6d1fdc48 | 3693 | } |
bdf4f4d2 JW |
3694 | /* |
3695 | * A racing thread's fault, or swapoff, may have already | |
3696 | * updated the pte, and even removed page from swap cache: in | |
3697 | * those cases unuse_pte()'s pte_same() test will fail; but | |
3698 | * there's also a KSM case which does need to charge the page. | |
3699 | */ | |
3700 | if (!PageSwapCache(page)) { | |
6d1fdc48 | 3701 | struct mem_cgroup *memcg; |
bdf4f4d2 | 3702 | |
0029e19e | 3703 | memcg = mem_cgroup_try_charge_mm(mm, gfp_mask, 1); |
6d1fdc48 JW |
3704 | if (!memcg) |
3705 | return -ENOMEM; | |
3706 | *memcgp = memcg; | |
3707 | return 0; | |
bdf4f4d2 | 3708 | } |
0435a2fd JW |
3709 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
3710 | } | |
3711 | ||
827a03d2 JW |
3712 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
3713 | { | |
3714 | if (mem_cgroup_disabled()) | |
3715 | return; | |
3716 | if (!memcg) | |
3717 | return; | |
3718 | __mem_cgroup_cancel_charge(memcg, 1); | |
3719 | } | |
3720 | ||
83aae4c7 | 3721 | static void |
72835c86 | 3722 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 3723 | enum charge_type ctype) |
7a81b88c | 3724 | { |
f8d66542 | 3725 | if (mem_cgroup_disabled()) |
7a81b88c | 3726 | return; |
72835c86 | 3727 | if (!memcg) |
7a81b88c | 3728 | return; |
5a6475a4 | 3729 | |
ce587e65 | 3730 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
3731 | /* |
3732 | * Now swap is on-memory. This means this page may be | |
3733 | * counted both as mem and swap....double count. | |
03f3c433 KH |
3734 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
3735 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
3736 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 3737 | */ |
03f3c433 | 3738 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 3739 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 3740 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 3741 | } |
7a81b88c KH |
3742 | } |
3743 | ||
72835c86 JW |
3744 | void mem_cgroup_commit_charge_swapin(struct page *page, |
3745 | struct mem_cgroup *memcg) | |
83aae4c7 | 3746 | { |
72835c86 | 3747 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 3748 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
3749 | } |
3750 | ||
d715ae08 | 3751 | int mem_cgroup_charge_file(struct page *page, struct mm_struct *mm, |
827a03d2 | 3752 | gfp_t gfp_mask) |
7a81b88c | 3753 | { |
827a03d2 | 3754 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
6d1fdc48 | 3755 | struct mem_cgroup *memcg; |
827a03d2 JW |
3756 | int ret; |
3757 | ||
f8d66542 | 3758 | if (mem_cgroup_disabled()) |
827a03d2 JW |
3759 | return 0; |
3760 | if (PageCompound(page)) | |
3761 | return 0; | |
3762 | ||
6d1fdc48 | 3763 | if (PageSwapCache(page)) { /* shmem */ |
0435a2fd JW |
3764 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
3765 | gfp_mask, &memcg); | |
6d1fdc48 JW |
3766 | if (ret) |
3767 | return ret; | |
3768 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
3769 | return 0; | |
827a03d2 | 3770 | } |
6d1fdc48 | 3771 | |
0029e19e | 3772 | memcg = mem_cgroup_try_charge_mm(mm, gfp_mask, 1); |
6f6acb00 MH |
3773 | if (!memcg) |
3774 | return -ENOMEM; | |
6d1fdc48 JW |
3775 | __mem_cgroup_commit_charge(memcg, page, 1, type, false); |
3776 | return 0; | |
7a81b88c KH |
3777 | } |
3778 | ||
c0ff4b85 | 3779 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
3780 | unsigned int nr_pages, |
3781 | const enum charge_type ctype) | |
569b846d KH |
3782 | { |
3783 | struct memcg_batch_info *batch = NULL; | |
3784 | bool uncharge_memsw = true; | |
7ec99d62 | 3785 | |
569b846d KH |
3786 | /* If swapout, usage of swap doesn't decrease */ |
3787 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
3788 | uncharge_memsw = false; | |
569b846d KH |
3789 | |
3790 | batch = ¤t->memcg_batch; | |
3791 | /* | |
3792 | * In usual, we do css_get() when we remember memcg pointer. | |
3793 | * But in this case, we keep res->usage until end of a series of | |
3794 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
3795 | */ | |
3796 | if (!batch->memcg) | |
c0ff4b85 | 3797 | batch->memcg = memcg; |
3c11ecf4 KH |
3798 | /* |
3799 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 3800 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
3801 | * the same cgroup and we have chance to coalesce uncharges. |
3802 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
3803 | * because we want to do uncharge as soon as possible. | |
3804 | */ | |
3805 | ||
3806 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
3807 | goto direct_uncharge; | |
3808 | ||
7ec99d62 | 3809 | if (nr_pages > 1) |
ec168510 AA |
3810 | goto direct_uncharge; |
3811 | ||
569b846d KH |
3812 | /* |
3813 | * In typical case, batch->memcg == mem. This means we can | |
3814 | * merge a series of uncharges to an uncharge of res_counter. | |
3815 | * If not, we uncharge res_counter ony by one. | |
3816 | */ | |
c0ff4b85 | 3817 | if (batch->memcg != memcg) |
569b846d KH |
3818 | goto direct_uncharge; |
3819 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 3820 | batch->nr_pages++; |
569b846d | 3821 | if (uncharge_memsw) |
7ffd4ca7 | 3822 | batch->memsw_nr_pages++; |
569b846d KH |
3823 | return; |
3824 | direct_uncharge: | |
c0ff4b85 | 3825 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 3826 | if (uncharge_memsw) |
c0ff4b85 R |
3827 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
3828 | if (unlikely(batch->memcg != memcg)) | |
3829 | memcg_oom_recover(memcg); | |
569b846d | 3830 | } |
7a81b88c | 3831 | |
8a9f3ccd | 3832 | /* |
69029cd5 | 3833 | * uncharge if !page_mapped(page) |
8a9f3ccd | 3834 | */ |
8c7c6e34 | 3835 | static struct mem_cgroup * |
0030f535 JW |
3836 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
3837 | bool end_migration) | |
8a9f3ccd | 3838 | { |
c0ff4b85 | 3839 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
3840 | unsigned int nr_pages = 1; |
3841 | struct page_cgroup *pc; | |
b2402857 | 3842 | bool anon; |
8a9f3ccd | 3843 | |
f8d66542 | 3844 | if (mem_cgroup_disabled()) |
8c7c6e34 | 3845 | return NULL; |
4077960e | 3846 | |
37c2ac78 | 3847 | if (PageTransHuge(page)) { |
7ec99d62 | 3848 | nr_pages <<= compound_order(page); |
309381fe | 3849 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
37c2ac78 | 3850 | } |
8697d331 | 3851 | /* |
3c541e14 | 3852 | * Check if our page_cgroup is valid |
8697d331 | 3853 | */ |
52d4b9ac | 3854 | pc = lookup_page_cgroup(page); |
cfa44946 | 3855 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 3856 | return NULL; |
b9c565d5 | 3857 | |
52d4b9ac | 3858 | lock_page_cgroup(pc); |
d13d1443 | 3859 | |
c0ff4b85 | 3860 | memcg = pc->mem_cgroup; |
8c7c6e34 | 3861 | |
d13d1443 KH |
3862 | if (!PageCgroupUsed(pc)) |
3863 | goto unlock_out; | |
3864 | ||
b2402857 KH |
3865 | anon = PageAnon(page); |
3866 | ||
d13d1443 | 3867 | switch (ctype) { |
41326c17 | 3868 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
3869 | /* |
3870 | * Generally PageAnon tells if it's the anon statistics to be | |
3871 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
3872 | * used before page reached the stage of being marked PageAnon. | |
3873 | */ | |
b2402857 KH |
3874 | anon = true; |
3875 | /* fallthrough */ | |
8a9478ca | 3876 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 3877 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
3878 | if (page_mapped(page)) |
3879 | goto unlock_out; | |
3880 | /* | |
3881 | * Pages under migration may not be uncharged. But | |
3882 | * end_migration() /must/ be the one uncharging the | |
3883 | * unused post-migration page and so it has to call | |
3884 | * here with the migration bit still set. See the | |
3885 | * res_counter handling below. | |
3886 | */ | |
3887 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
3888 | goto unlock_out; |
3889 | break; | |
3890 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
3891 | if (!PageAnon(page)) { /* Shared memory */ | |
3892 | if (page->mapping && !page_is_file_cache(page)) | |
3893 | goto unlock_out; | |
3894 | } else if (page_mapped(page)) /* Anon */ | |
3895 | goto unlock_out; | |
3896 | break; | |
3897 | default: | |
3898 | break; | |
52d4b9ac | 3899 | } |
d13d1443 | 3900 | |
b070e65c | 3901 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
04046e1a | 3902 | |
52d4b9ac | 3903 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3904 | /* |
3905 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3906 | * freed from LRU. This is safe because uncharged page is expected not | |
3907 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3908 | * special functions. | |
3909 | */ | |
b9c565d5 | 3910 | |
52d4b9ac | 3911 | unlock_page_cgroup(pc); |
f75ca962 | 3912 | /* |
c0ff4b85 | 3913 | * even after unlock, we have memcg->res.usage here and this memcg |
4050377b | 3914 | * will never be freed, so it's safe to call css_get(). |
f75ca962 | 3915 | */ |
c0ff4b85 | 3916 | memcg_check_events(memcg, page); |
f75ca962 | 3917 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 | 3918 | mem_cgroup_swap_statistics(memcg, true); |
4050377b | 3919 | css_get(&memcg->css); |
f75ca962 | 3920 | } |
0030f535 JW |
3921 | /* |
3922 | * Migration does not charge the res_counter for the | |
3923 | * replacement page, so leave it alone when phasing out the | |
3924 | * page that is unused after the migration. | |
3925 | */ | |
05b84301 | 3926 | if (!end_migration) |
c0ff4b85 | 3927 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 3928 | |
c0ff4b85 | 3929 | return memcg; |
d13d1443 KH |
3930 | |
3931 | unlock_out: | |
3932 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3933 | return NULL; |
3c541e14 BS |
3934 | } |
3935 | ||
69029cd5 KH |
3936 | void mem_cgroup_uncharge_page(struct page *page) |
3937 | { | |
52d4b9ac KH |
3938 | /* early check. */ |
3939 | if (page_mapped(page)) | |
3940 | return; | |
309381fe | 3941 | VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page); |
28ccddf7 JW |
3942 | /* |
3943 | * If the page is in swap cache, uncharge should be deferred | |
3944 | * to the swap path, which also properly accounts swap usage | |
3945 | * and handles memcg lifetime. | |
3946 | * | |
3947 | * Note that this check is not stable and reclaim may add the | |
3948 | * page to swap cache at any time after this. However, if the | |
3949 | * page is not in swap cache by the time page->mapcount hits | |
3950 | * 0, there won't be any page table references to the swap | |
3951 | * slot, and reclaim will free it and not actually write the | |
3952 | * page to disk. | |
3953 | */ | |
0c59b89c JW |
3954 | if (PageSwapCache(page)) |
3955 | return; | |
0030f535 | 3956 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
3957 | } |
3958 | ||
3959 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3960 | { | |
309381fe SL |
3961 | VM_BUG_ON_PAGE(page_mapped(page), page); |
3962 | VM_BUG_ON_PAGE(page->mapping, page); | |
0030f535 | 3963 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
3964 | } |
3965 | ||
569b846d KH |
3966 | /* |
3967 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3968 | * In that cases, pages are freed continuously and we can expect pages | |
3969 | * are in the same memcg. All these calls itself limits the number of | |
3970 | * pages freed at once, then uncharge_start/end() is called properly. | |
3971 | * This may be called prural(2) times in a context, | |
3972 | */ | |
3973 | ||
3974 | void mem_cgroup_uncharge_start(void) | |
3975 | { | |
3976 | current->memcg_batch.do_batch++; | |
3977 | /* We can do nest. */ | |
3978 | if (current->memcg_batch.do_batch == 1) { | |
3979 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3980 | current->memcg_batch.nr_pages = 0; |
3981 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3982 | } |
3983 | } | |
3984 | ||
3985 | void mem_cgroup_uncharge_end(void) | |
3986 | { | |
3987 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3988 | ||
3989 | if (!batch->do_batch) | |
3990 | return; | |
3991 | ||
3992 | batch->do_batch--; | |
3993 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3994 | return; | |
3995 | ||
3996 | if (!batch->memcg) | |
3997 | return; | |
3998 | /* | |
3999 | * This "batch->memcg" is valid without any css_get/put etc... | |
4000 | * bacause we hide charges behind us. | |
4001 | */ | |
7ffd4ca7 JW |
4002 | if (batch->nr_pages) |
4003 | res_counter_uncharge(&batch->memcg->res, | |
4004 | batch->nr_pages * PAGE_SIZE); | |
4005 | if (batch->memsw_nr_pages) | |
4006 | res_counter_uncharge(&batch->memcg->memsw, | |
4007 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4008 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4009 | /* forget this pointer (for sanity check) */ |
4010 | batch->memcg = NULL; | |
4011 | } | |
4012 | ||
e767e056 | 4013 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4014 | /* |
e767e056 | 4015 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4016 | * memcg information is recorded to swap_cgroup of "ent" |
4017 | */ | |
8a9478ca KH |
4018 | void |
4019 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4020 | { |
4021 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4022 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4023 | ||
4024 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4025 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4026 | ||
0030f535 | 4027 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4028 | |
f75ca962 KH |
4029 | /* |
4030 | * record memcg information, if swapout && memcg != NULL, | |
4050377b | 4031 | * css_get() was called in uncharge(). |
f75ca962 KH |
4032 | */ |
4033 | if (do_swap_account && swapout && memcg) | |
34c00c31 | 4034 | swap_cgroup_record(ent, mem_cgroup_id(memcg)); |
8c7c6e34 | 4035 | } |
e767e056 | 4036 | #endif |
8c7c6e34 | 4037 | |
c255a458 | 4038 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4039 | /* |
4040 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4041 | * uncharge "memsw" account. | |
4042 | */ | |
4043 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4044 | { |
8c7c6e34 | 4045 | struct mem_cgroup *memcg; |
a3b2d692 | 4046 | unsigned short id; |
8c7c6e34 KH |
4047 | |
4048 | if (!do_swap_account) | |
4049 | return; | |
4050 | ||
a3b2d692 KH |
4051 | id = swap_cgroup_record(ent, 0); |
4052 | rcu_read_lock(); | |
4053 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4054 | if (memcg) { |
a3b2d692 | 4055 | /* |
ec903c0c TH |
4056 | * We uncharge this because swap is freed. This memcg can |
4057 | * be obsolete one. We avoid calling css_tryget_online(). | |
a3b2d692 | 4058 | */ |
05b84301 | 4059 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4060 | mem_cgroup_swap_statistics(memcg, false); |
4050377b | 4061 | css_put(&memcg->css); |
8c7c6e34 | 4062 | } |
a3b2d692 | 4063 | rcu_read_unlock(); |
d13d1443 | 4064 | } |
02491447 DN |
4065 | |
4066 | /** | |
4067 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4068 | * @entry: swap entry to be moved | |
4069 | * @from: mem_cgroup which the entry is moved from | |
4070 | * @to: mem_cgroup which the entry is moved to | |
4071 | * | |
4072 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4073 | * as the mem_cgroup's id of @from. | |
4074 | * | |
4075 | * Returns 0 on success, -EINVAL on failure. | |
4076 | * | |
4077 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4078 | * both res and memsw, and called css_get(). | |
4079 | */ | |
4080 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4081 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4082 | { |
4083 | unsigned short old_id, new_id; | |
4084 | ||
34c00c31 LZ |
4085 | old_id = mem_cgroup_id(from); |
4086 | new_id = mem_cgroup_id(to); | |
02491447 DN |
4087 | |
4088 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4089 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4090 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4091 | /* |
483c30b5 DN |
4092 | * This function is only called from task migration context now. |
4093 | * It postpones res_counter and refcount handling till the end | |
4094 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
4095 | * improvement. But we cannot postpone css_get(to) because if |
4096 | * the process that has been moved to @to does swap-in, the | |
4097 | * refcount of @to might be decreased to 0. | |
4098 | * | |
4099 | * We are in attach() phase, so the cgroup is guaranteed to be | |
4100 | * alive, so we can just call css_get(). | |
02491447 | 4101 | */ |
4050377b | 4102 | css_get(&to->css); |
02491447 DN |
4103 | return 0; |
4104 | } | |
4105 | return -EINVAL; | |
4106 | } | |
4107 | #else | |
4108 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4109 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4110 | { |
4111 | return -EINVAL; | |
4112 | } | |
8c7c6e34 | 4113 | #endif |
d13d1443 | 4114 | |
ae41be37 | 4115 | /* |
01b1ae63 KH |
4116 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4117 | * page belongs to. | |
ae41be37 | 4118 | */ |
0030f535 JW |
4119 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4120 | struct mem_cgroup **memcgp) | |
ae41be37 | 4121 | { |
c0ff4b85 | 4122 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4123 | unsigned int nr_pages = 1; |
7ec99d62 | 4124 | struct page_cgroup *pc; |
ac39cf8c | 4125 | enum charge_type ctype; |
8869b8f6 | 4126 | |
72835c86 | 4127 | *memcgp = NULL; |
56039efa | 4128 | |
f8d66542 | 4129 | if (mem_cgroup_disabled()) |
0030f535 | 4130 | return; |
4077960e | 4131 | |
b32967ff MG |
4132 | if (PageTransHuge(page)) |
4133 | nr_pages <<= compound_order(page); | |
4134 | ||
52d4b9ac KH |
4135 | pc = lookup_page_cgroup(page); |
4136 | lock_page_cgroup(pc); | |
4137 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4138 | memcg = pc->mem_cgroup; |
4139 | css_get(&memcg->css); | |
ac39cf8c | 4140 | /* |
4141 | * At migrating an anonymous page, its mapcount goes down | |
4142 | * to 0 and uncharge() will be called. But, even if it's fully | |
4143 | * unmapped, migration may fail and this page has to be | |
4144 | * charged again. We set MIGRATION flag here and delay uncharge | |
4145 | * until end_migration() is called | |
4146 | * | |
4147 | * Corner Case Thinking | |
4148 | * A) | |
4149 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4150 | * while migration was ongoing. | |
4151 | * If unmap finds the old page, uncharge() of it will be delayed | |
4152 | * until end_migration(). If unmap finds a new page, it's | |
4153 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4154 | * finds swap_migration_entry, the new page will not be mapped | |
4155 | * and end_migration() will find it(mapcount==0). | |
4156 | * | |
4157 | * B) | |
4158 | * When the old page was mapped but migraion fails, the kernel | |
4159 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4160 | * if mapcount goes down to 0. We can do remap successfully | |
4161 | * without charging it again. | |
4162 | * | |
4163 | * C) | |
4164 | * The "old" page is under lock_page() until the end of | |
4165 | * migration, so, the old page itself will not be swapped-out. | |
4166 | * If the new page is swapped out before end_migraton, our | |
4167 | * hook to usual swap-out path will catch the event. | |
4168 | */ | |
4169 | if (PageAnon(page)) | |
4170 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4171 | } |
52d4b9ac | 4172 | unlock_page_cgroup(pc); |
ac39cf8c | 4173 | /* |
4174 | * If the page is not charged at this point, | |
4175 | * we return here. | |
4176 | */ | |
c0ff4b85 | 4177 | if (!memcg) |
0030f535 | 4178 | return; |
01b1ae63 | 4179 | |
72835c86 | 4180 | *memcgp = memcg; |
ac39cf8c | 4181 | /* |
4182 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4183 | * is called before end_migration, we can catch all events on this new | |
4184 | * page. In the case new page is migrated but not remapped, new page's | |
4185 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4186 | */ | |
ac39cf8c | 4187 | if (PageAnon(page)) |
41326c17 | 4188 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4189 | else |
62ba7442 | 4190 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4191 | /* |
4192 | * The page is committed to the memcg, but it's not actually | |
4193 | * charged to the res_counter since we plan on replacing the | |
4194 | * old one and only one page is going to be left afterwards. | |
4195 | */ | |
b32967ff | 4196 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4197 | } |
8869b8f6 | 4198 | |
69029cd5 | 4199 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4200 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4201 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4202 | { |
ac39cf8c | 4203 | struct page *used, *unused; |
01b1ae63 | 4204 | struct page_cgroup *pc; |
b2402857 | 4205 | bool anon; |
01b1ae63 | 4206 | |
c0ff4b85 | 4207 | if (!memcg) |
01b1ae63 | 4208 | return; |
b25ed609 | 4209 | |
50de1dd9 | 4210 | if (!migration_ok) { |
ac39cf8c | 4211 | used = oldpage; |
4212 | unused = newpage; | |
01b1ae63 | 4213 | } else { |
ac39cf8c | 4214 | used = newpage; |
01b1ae63 KH |
4215 | unused = oldpage; |
4216 | } | |
0030f535 | 4217 | anon = PageAnon(used); |
7d188958 JW |
4218 | __mem_cgroup_uncharge_common(unused, |
4219 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4220 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4221 | true); | |
0030f535 | 4222 | css_put(&memcg->css); |
69029cd5 | 4223 | /* |
ac39cf8c | 4224 | * We disallowed uncharge of pages under migration because mapcount |
4225 | * of the page goes down to zero, temporarly. | |
4226 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4227 | */ |
ac39cf8c | 4228 | pc = lookup_page_cgroup(oldpage); |
4229 | lock_page_cgroup(pc); | |
4230 | ClearPageCgroupMigration(pc); | |
4231 | unlock_page_cgroup(pc); | |
ac39cf8c | 4232 | |
01b1ae63 | 4233 | /* |
ac39cf8c | 4234 | * If a page is a file cache, radix-tree replacement is very atomic |
4235 | * and we can skip this check. When it was an Anon page, its mapcount | |
4236 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4237 | * uncharged yet. There are several case but page->mapcount check | |
4238 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4239 | * check. (see prepare_charge() also) | |
69029cd5 | 4240 | */ |
b2402857 | 4241 | if (anon) |
ac39cf8c | 4242 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4243 | } |
78fb7466 | 4244 | |
ab936cbc KH |
4245 | /* |
4246 | * At replace page cache, newpage is not under any memcg but it's on | |
4247 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4248 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4249 | */ | |
4250 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4251 | struct page *newpage) | |
4252 | { | |
bde05d1c | 4253 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4254 | struct page_cgroup *pc; |
ab936cbc | 4255 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4256 | |
4257 | if (mem_cgroup_disabled()) | |
4258 | return; | |
4259 | ||
4260 | pc = lookup_page_cgroup(oldpage); | |
4261 | /* fix accounting on old pages */ | |
4262 | lock_page_cgroup(pc); | |
bde05d1c HD |
4263 | if (PageCgroupUsed(pc)) { |
4264 | memcg = pc->mem_cgroup; | |
b070e65c | 4265 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
bde05d1c HD |
4266 | ClearPageCgroupUsed(pc); |
4267 | } | |
ab936cbc KH |
4268 | unlock_page_cgroup(pc); |
4269 | ||
bde05d1c HD |
4270 | /* |
4271 | * When called from shmem_replace_page(), in some cases the | |
4272 | * oldpage has already been charged, and in some cases not. | |
4273 | */ | |
4274 | if (!memcg) | |
4275 | return; | |
ab936cbc KH |
4276 | /* |
4277 | * Even if newpage->mapping was NULL before starting replacement, | |
4278 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4279 | * LRU while we overwrite pc->mem_cgroup. | |
4280 | */ | |
ce587e65 | 4281 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4282 | } |
4283 | ||
f212ad7c DN |
4284 | #ifdef CONFIG_DEBUG_VM |
4285 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4286 | { | |
4287 | struct page_cgroup *pc; | |
4288 | ||
4289 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4290 | /* |
4291 | * Can be NULL while feeding pages into the page allocator for | |
4292 | * the first time, i.e. during boot or memory hotplug; | |
4293 | * or when mem_cgroup_disabled(). | |
4294 | */ | |
f212ad7c DN |
4295 | if (likely(pc) && PageCgroupUsed(pc)) |
4296 | return pc; | |
4297 | return NULL; | |
4298 | } | |
4299 | ||
4300 | bool mem_cgroup_bad_page_check(struct page *page) | |
4301 | { | |
4302 | if (mem_cgroup_disabled()) | |
4303 | return false; | |
4304 | ||
4305 | return lookup_page_cgroup_used(page) != NULL; | |
4306 | } | |
4307 | ||
4308 | void mem_cgroup_print_bad_page(struct page *page) | |
4309 | { | |
4310 | struct page_cgroup *pc; | |
4311 | ||
4312 | pc = lookup_page_cgroup_used(page); | |
4313 | if (pc) { | |
d045197f AM |
4314 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4315 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4316 | } |
4317 | } | |
4318 | #endif | |
4319 | ||
d38d2a75 | 4320 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4321 | unsigned long long val) |
628f4235 | 4322 | { |
81d39c20 | 4323 | int retry_count; |
3c11ecf4 | 4324 | u64 memswlimit, memlimit; |
628f4235 | 4325 | int ret = 0; |
81d39c20 KH |
4326 | int children = mem_cgroup_count_children(memcg); |
4327 | u64 curusage, oldusage; | |
3c11ecf4 | 4328 | int enlarge; |
81d39c20 KH |
4329 | |
4330 | /* | |
4331 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4332 | * is depends on callers. We set our retry-count to be function | |
4333 | * of # of children which we should visit in this loop. | |
4334 | */ | |
4335 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4336 | ||
4337 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4338 | |
3c11ecf4 | 4339 | enlarge = 0; |
8c7c6e34 | 4340 | while (retry_count) { |
628f4235 KH |
4341 | if (signal_pending(current)) { |
4342 | ret = -EINTR; | |
4343 | break; | |
4344 | } | |
8c7c6e34 KH |
4345 | /* |
4346 | * Rather than hide all in some function, I do this in | |
4347 | * open coded manner. You see what this really does. | |
aaad153e | 4348 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4349 | */ |
4350 | mutex_lock(&set_limit_mutex); | |
4351 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4352 | if (memswlimit < val) { | |
4353 | ret = -EINVAL; | |
4354 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4355 | break; |
4356 | } | |
3c11ecf4 KH |
4357 | |
4358 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4359 | if (memlimit < val) | |
4360 | enlarge = 1; | |
4361 | ||
8c7c6e34 | 4362 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4363 | if (!ret) { |
4364 | if (memswlimit == val) | |
4365 | memcg->memsw_is_minimum = true; | |
4366 | else | |
4367 | memcg->memsw_is_minimum = false; | |
4368 | } | |
8c7c6e34 KH |
4369 | mutex_unlock(&set_limit_mutex); |
4370 | ||
4371 | if (!ret) | |
4372 | break; | |
4373 | ||
5660048c JW |
4374 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4375 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4376 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4377 | /* Usage is reduced ? */ | |
f894ffa8 | 4378 | if (curusage >= oldusage) |
81d39c20 KH |
4379 | retry_count--; |
4380 | else | |
4381 | oldusage = curusage; | |
8c7c6e34 | 4382 | } |
3c11ecf4 KH |
4383 | if (!ret && enlarge) |
4384 | memcg_oom_recover(memcg); | |
14797e23 | 4385 | |
8c7c6e34 KH |
4386 | return ret; |
4387 | } | |
4388 | ||
338c8431 LZ |
4389 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4390 | unsigned long long val) | |
8c7c6e34 | 4391 | { |
81d39c20 | 4392 | int retry_count; |
3c11ecf4 | 4393 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4394 | int children = mem_cgroup_count_children(memcg); |
4395 | int ret = -EBUSY; | |
3c11ecf4 | 4396 | int enlarge = 0; |
8c7c6e34 | 4397 | |
81d39c20 | 4398 | /* see mem_cgroup_resize_res_limit */ |
f894ffa8 | 4399 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
81d39c20 | 4400 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
8c7c6e34 KH |
4401 | while (retry_count) { |
4402 | if (signal_pending(current)) { | |
4403 | ret = -EINTR; | |
4404 | break; | |
4405 | } | |
4406 | /* | |
4407 | * Rather than hide all in some function, I do this in | |
4408 | * open coded manner. You see what this really does. | |
aaad153e | 4409 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4410 | */ |
4411 | mutex_lock(&set_limit_mutex); | |
4412 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4413 | if (memlimit > val) { | |
4414 | ret = -EINVAL; | |
4415 | mutex_unlock(&set_limit_mutex); | |
4416 | break; | |
4417 | } | |
3c11ecf4 KH |
4418 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4419 | if (memswlimit < val) | |
4420 | enlarge = 1; | |
8c7c6e34 | 4421 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4422 | if (!ret) { |
4423 | if (memlimit == val) | |
4424 | memcg->memsw_is_minimum = true; | |
4425 | else | |
4426 | memcg->memsw_is_minimum = false; | |
4427 | } | |
8c7c6e34 KH |
4428 | mutex_unlock(&set_limit_mutex); |
4429 | ||
4430 | if (!ret) | |
4431 | break; | |
4432 | ||
5660048c JW |
4433 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4434 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4435 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4436 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4437 | /* Usage is reduced ? */ |
8c7c6e34 | 4438 | if (curusage >= oldusage) |
628f4235 | 4439 | retry_count--; |
81d39c20 KH |
4440 | else |
4441 | oldusage = curusage; | |
628f4235 | 4442 | } |
3c11ecf4 KH |
4443 | if (!ret && enlarge) |
4444 | memcg_oom_recover(memcg); | |
628f4235 KH |
4445 | return ret; |
4446 | } | |
4447 | ||
0608f43d AM |
4448 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
4449 | gfp_t gfp_mask, | |
4450 | unsigned long *total_scanned) | |
4451 | { | |
4452 | unsigned long nr_reclaimed = 0; | |
4453 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
4454 | unsigned long reclaimed; | |
4455 | int loop = 0; | |
4456 | struct mem_cgroup_tree_per_zone *mctz; | |
4457 | unsigned long long excess; | |
4458 | unsigned long nr_scanned; | |
4459 | ||
4460 | if (order > 0) | |
4461 | return 0; | |
4462 | ||
4463 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
4464 | /* | |
4465 | * This loop can run a while, specially if mem_cgroup's continuously | |
4466 | * keep exceeding their soft limit and putting the system under | |
4467 | * pressure | |
4468 | */ | |
4469 | do { | |
4470 | if (next_mz) | |
4471 | mz = next_mz; | |
4472 | else | |
4473 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
4474 | if (!mz) | |
4475 | break; | |
4476 | ||
4477 | nr_scanned = 0; | |
4478 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
4479 | gfp_mask, &nr_scanned); | |
4480 | nr_reclaimed += reclaimed; | |
4481 | *total_scanned += nr_scanned; | |
4482 | spin_lock(&mctz->lock); | |
4483 | ||
4484 | /* | |
4485 | * If we failed to reclaim anything from this memory cgroup | |
4486 | * it is time to move on to the next cgroup | |
4487 | */ | |
4488 | next_mz = NULL; | |
4489 | if (!reclaimed) { | |
4490 | do { | |
4491 | /* | |
4492 | * Loop until we find yet another one. | |
4493 | * | |
4494 | * By the time we get the soft_limit lock | |
4495 | * again, someone might have aded the | |
4496 | * group back on the RB tree. Iterate to | |
4497 | * make sure we get a different mem. | |
4498 | * mem_cgroup_largest_soft_limit_node returns | |
4499 | * NULL if no other cgroup is present on | |
4500 | * the tree | |
4501 | */ | |
4502 | next_mz = | |
4503 | __mem_cgroup_largest_soft_limit_node(mctz); | |
4504 | if (next_mz == mz) | |
4505 | css_put(&next_mz->memcg->css); | |
4506 | else /* next_mz == NULL or other memcg */ | |
4507 | break; | |
4508 | } while (1); | |
4509 | } | |
cf2c8127 | 4510 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
4511 | excess = res_counter_soft_limit_excess(&mz->memcg->res); |
4512 | /* | |
4513 | * One school of thought says that we should not add | |
4514 | * back the node to the tree if reclaim returns 0. | |
4515 | * But our reclaim could return 0, simply because due | |
4516 | * to priority we are exposing a smaller subset of | |
4517 | * memory to reclaim from. Consider this as a longer | |
4518 | * term TODO. | |
4519 | */ | |
4520 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 4521 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0608f43d AM |
4522 | spin_unlock(&mctz->lock); |
4523 | css_put(&mz->memcg->css); | |
4524 | loop++; | |
4525 | /* | |
4526 | * Could not reclaim anything and there are no more | |
4527 | * mem cgroups to try or we seem to be looping without | |
4528 | * reclaiming anything. | |
4529 | */ | |
4530 | if (!nr_reclaimed && | |
4531 | (next_mz == NULL || | |
4532 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4533 | break; | |
4534 | } while (!nr_reclaimed); | |
4535 | if (next_mz) | |
4536 | css_put(&next_mz->memcg->css); | |
4537 | return nr_reclaimed; | |
4538 | } | |
4539 | ||
2ef37d3f MH |
4540 | /** |
4541 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4542 | * @memcg: group to clear | |
4543 | * @node: NUMA node | |
4544 | * @zid: zone id | |
4545 | * @lru: lru to to clear | |
4546 | * | |
3c935d18 | 4547 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4548 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4549 | * group. | |
cc847582 | 4550 | */ |
2ef37d3f | 4551 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4552 | int node, int zid, enum lru_list lru) |
cc847582 | 4553 | { |
bea8c150 | 4554 | struct lruvec *lruvec; |
2ef37d3f | 4555 | unsigned long flags; |
072c56c1 | 4556 | struct list_head *list; |
925b7673 JW |
4557 | struct page *busy; |
4558 | struct zone *zone; | |
072c56c1 | 4559 | |
08e552c6 | 4560 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4561 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4562 | list = &lruvec->lists[lru]; | |
cc847582 | 4563 | |
f817ed48 | 4564 | busy = NULL; |
2ef37d3f | 4565 | do { |
925b7673 | 4566 | struct page_cgroup *pc; |
5564e88b JW |
4567 | struct page *page; |
4568 | ||
08e552c6 | 4569 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4570 | if (list_empty(list)) { |
08e552c6 | 4571 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4572 | break; |
f817ed48 | 4573 | } |
925b7673 JW |
4574 | page = list_entry(list->prev, struct page, lru); |
4575 | if (busy == page) { | |
4576 | list_move(&page->lru, list); | |
648bcc77 | 4577 | busy = NULL; |
08e552c6 | 4578 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4579 | continue; |
4580 | } | |
08e552c6 | 4581 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4582 | |
925b7673 | 4583 | pc = lookup_page_cgroup(page); |
5564e88b | 4584 | |
3c935d18 | 4585 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4586 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4587 | busy = page; |
f817ed48 KH |
4588 | } else |
4589 | busy = NULL; | |
2a7a0e0f | 4590 | cond_resched(); |
2ef37d3f | 4591 | } while (!list_empty(list)); |
cc847582 KH |
4592 | } |
4593 | ||
4594 | /* | |
c26251f9 MH |
4595 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4596 | * all the charges and pages to the parent. | |
cc847582 | 4597 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4598 | * |
4599 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4600 | */ |
ab5196c2 | 4601 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4602 | { |
c26251f9 | 4603 | int node, zid; |
bea207c8 | 4604 | u64 usage; |
f817ed48 | 4605 | |
fce66477 | 4606 | do { |
52d4b9ac KH |
4607 | /* This is for making all *used* pages to be on LRU. */ |
4608 | lru_add_drain_all(); | |
c0ff4b85 | 4609 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4610 | mem_cgroup_start_move(memcg); |
31aaea4a | 4611 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4612 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4613 | enum lru_list lru; |
4614 | for_each_lru(lru) { | |
2ef37d3f | 4615 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4616 | node, zid, lru); |
f817ed48 | 4617 | } |
1ecaab2b | 4618 | } |
f817ed48 | 4619 | } |
c0ff4b85 R |
4620 | mem_cgroup_end_move(memcg); |
4621 | memcg_oom_recover(memcg); | |
52d4b9ac | 4622 | cond_resched(); |
f817ed48 | 4623 | |
2ef37d3f | 4624 | /* |
bea207c8 GC |
4625 | * Kernel memory may not necessarily be trackable to a specific |
4626 | * process. So they are not migrated, and therefore we can't | |
4627 | * expect their value to drop to 0 here. | |
4628 | * Having res filled up with kmem only is enough. | |
4629 | * | |
2ef37d3f MH |
4630 | * This is a safety check because mem_cgroup_force_empty_list |
4631 | * could have raced with mem_cgroup_replace_page_cache callers | |
4632 | * so the lru seemed empty but the page could have been added | |
4633 | * right after the check. RES_USAGE should be safe as we always | |
4634 | * charge before adding to the LRU. | |
4635 | */ | |
bea207c8 GC |
4636 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4637 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4638 | } while (usage > 0); | |
c26251f9 MH |
4639 | } |
4640 | ||
ea280e7b TH |
4641 | /* |
4642 | * Test whether @memcg has children, dead or alive. Note that this | |
4643 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
4644 | * returns %true if there are child csses according to the cgroup | |
4645 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
4646 | */ | |
b5f99b53 GC |
4647 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
4648 | { | |
ea280e7b TH |
4649 | bool ret; |
4650 | ||
696ac172 | 4651 | /* |
ea280e7b TH |
4652 | * The lock does not prevent addition or deletion of children, but |
4653 | * it prevents a new child from being initialized based on this | |
4654 | * parent in css_online(), so it's enough to decide whether | |
4655 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 4656 | */ |
ea280e7b TH |
4657 | lockdep_assert_held(&memcg_create_mutex); |
4658 | ||
4659 | rcu_read_lock(); | |
4660 | ret = css_next_child(NULL, &memcg->css); | |
4661 | rcu_read_unlock(); | |
4662 | return ret; | |
b5f99b53 GC |
4663 | } |
4664 | ||
c26251f9 MH |
4665 | /* |
4666 | * Reclaims as many pages from the given memcg as possible and moves | |
4667 | * the rest to the parent. | |
4668 | * | |
4669 | * Caller is responsible for holding css reference for memcg. | |
4670 | */ | |
4671 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4672 | { | |
4673 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 4674 | |
c1e862c1 KH |
4675 | /* we call try-to-free pages for make this cgroup empty */ |
4676 | lru_add_drain_all(); | |
f817ed48 | 4677 | /* try to free all pages in this cgroup */ |
569530fb | 4678 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4679 | int progress; |
c1e862c1 | 4680 | |
c26251f9 MH |
4681 | if (signal_pending(current)) |
4682 | return -EINTR; | |
4683 | ||
c0ff4b85 | 4684 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4685 | false); |
c1e862c1 | 4686 | if (!progress) { |
f817ed48 | 4687 | nr_retries--; |
c1e862c1 | 4688 | /* maybe some writeback is necessary */ |
8aa7e847 | 4689 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4690 | } |
f817ed48 KH |
4691 | |
4692 | } | |
ab5196c2 MH |
4693 | |
4694 | return 0; | |
cc847582 KH |
4695 | } |
4696 | ||
6770c64e TH |
4697 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
4698 | char *buf, size_t nbytes, | |
4699 | loff_t off) | |
c1e862c1 | 4700 | { |
6770c64e | 4701 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 4702 | |
d8423011 MH |
4703 | if (mem_cgroup_is_root(memcg)) |
4704 | return -EINVAL; | |
6770c64e | 4705 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
4706 | } |
4707 | ||
182446d0 TH |
4708 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
4709 | struct cftype *cft) | |
18f59ea7 | 4710 | { |
182446d0 | 4711 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
4712 | } |
4713 | ||
182446d0 TH |
4714 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
4715 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
4716 | { |
4717 | int retval = 0; | |
182446d0 | 4718 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4719 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 4720 | |
0999821b | 4721 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
4722 | |
4723 | if (memcg->use_hierarchy == val) | |
4724 | goto out; | |
4725 | ||
18f59ea7 | 4726 | /* |
af901ca1 | 4727 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
4728 | * in the child subtrees. If it is unset, then the change can |
4729 | * occur, provided the current cgroup has no children. | |
4730 | * | |
4731 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
4732 | * set if there are no children. | |
4733 | */ | |
c0ff4b85 | 4734 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 4735 | (val == 1 || val == 0)) { |
ea280e7b | 4736 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 4737 | memcg->use_hierarchy = val; |
18f59ea7 BS |
4738 | else |
4739 | retval = -EBUSY; | |
4740 | } else | |
4741 | retval = -EINVAL; | |
567fb435 GC |
4742 | |
4743 | out: | |
0999821b | 4744 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
4745 | |
4746 | return retval; | |
4747 | } | |
4748 | ||
791badbd | 4749 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 4750 | struct cftype *cft) |
8cdea7c0 | 4751 | { |
182446d0 | 4752 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
05b84301 JW |
4753 | enum res_type type = MEMFILE_TYPE(cft->private); |
4754 | int name = MEMFILE_ATTR(cft->private); | |
af36f906 | 4755 | |
8c7c6e34 KH |
4756 | switch (type) { |
4757 | case _MEM: | |
05b84301 | 4758 | return res_counter_read_u64(&memcg->res, name); |
8c7c6e34 | 4759 | case _MEMSWAP: |
05b84301 | 4760 | return res_counter_read_u64(&memcg->memsw, name); |
510fc4e1 | 4761 | case _KMEM: |
05b84301 | 4762 | return res_counter_read_u64(&memcg->kmem, name); |
510fc4e1 | 4763 | break; |
8c7c6e34 KH |
4764 | default: |
4765 | BUG(); | |
8c7c6e34 | 4766 | } |
8cdea7c0 | 4767 | } |
510fc4e1 | 4768 | |
510fc4e1 | 4769 | #ifdef CONFIG_MEMCG_KMEM |
d6441637 VD |
4770 | /* should be called with activate_kmem_mutex held */ |
4771 | static int __memcg_activate_kmem(struct mem_cgroup *memcg, | |
4772 | unsigned long long limit) | |
4773 | { | |
4774 | int err = 0; | |
4775 | int memcg_id; | |
4776 | ||
4777 | if (memcg_kmem_is_active(memcg)) | |
4778 | return 0; | |
4779 | ||
4780 | /* | |
4781 | * We are going to allocate memory for data shared by all memory | |
4782 | * cgroups so let's stop accounting here. | |
4783 | */ | |
4784 | memcg_stop_kmem_account(); | |
4785 | ||
510fc4e1 GC |
4786 | /* |
4787 | * For simplicity, we won't allow this to be disabled. It also can't | |
4788 | * be changed if the cgroup has children already, or if tasks had | |
4789 | * already joined. | |
4790 | * | |
4791 | * If tasks join before we set the limit, a person looking at | |
4792 | * kmem.usage_in_bytes will have no way to determine when it took | |
4793 | * place, which makes the value quite meaningless. | |
4794 | * | |
4795 | * After it first became limited, changes in the value of the limit are | |
4796 | * of course permitted. | |
510fc4e1 | 4797 | */ |
0999821b | 4798 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
4799 | if (cgroup_has_tasks(memcg->css.cgroup) || |
4800 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
4801 | err = -EBUSY; |
4802 | mutex_unlock(&memcg_create_mutex); | |
4803 | if (err) | |
4804 | goto out; | |
510fc4e1 | 4805 | |
d6441637 VD |
4806 | memcg_id = ida_simple_get(&kmem_limited_groups, |
4807 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
4808 | if (memcg_id < 0) { | |
4809 | err = memcg_id; | |
4810 | goto out; | |
4811 | } | |
4812 | ||
4813 | /* | |
4814 | * Make sure we have enough space for this cgroup in each root cache's | |
4815 | * memcg_params. | |
4816 | */ | |
bd673145 | 4817 | mutex_lock(&memcg_slab_mutex); |
d6441637 | 4818 | err = memcg_update_all_caches(memcg_id + 1); |
bd673145 | 4819 | mutex_unlock(&memcg_slab_mutex); |
d6441637 VD |
4820 | if (err) |
4821 | goto out_rmid; | |
4822 | ||
4823 | memcg->kmemcg_id = memcg_id; | |
4824 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
d6441637 VD |
4825 | |
4826 | /* | |
4827 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
4828 | * active bit set yet, so this should succeed. | |
4829 | */ | |
4830 | err = res_counter_set_limit(&memcg->kmem, limit); | |
4831 | VM_BUG_ON(err); | |
4832 | ||
4833 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
4834 | /* | |
4835 | * Setting the active bit after enabling static branching will | |
4836 | * guarantee no one starts accounting before all call sites are | |
4837 | * patched. | |
4838 | */ | |
4839 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 4840 | out: |
d6441637 VD |
4841 | memcg_resume_kmem_account(); |
4842 | return err; | |
4843 | ||
4844 | out_rmid: | |
4845 | ida_simple_remove(&kmem_limited_groups, memcg_id); | |
4846 | goto out; | |
4847 | } | |
4848 | ||
4849 | static int memcg_activate_kmem(struct mem_cgroup *memcg, | |
4850 | unsigned long long limit) | |
4851 | { | |
4852 | int ret; | |
4853 | ||
4854 | mutex_lock(&activate_kmem_mutex); | |
4855 | ret = __memcg_activate_kmem(memcg, limit); | |
4856 | mutex_unlock(&activate_kmem_mutex); | |
4857 | return ret; | |
4858 | } | |
4859 | ||
4860 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
4861 | unsigned long long val) | |
4862 | { | |
4863 | int ret; | |
4864 | ||
4865 | if (!memcg_kmem_is_active(memcg)) | |
4866 | ret = memcg_activate_kmem(memcg, val); | |
4867 | else | |
4868 | ret = res_counter_set_limit(&memcg->kmem, val); | |
510fc4e1 GC |
4869 | return ret; |
4870 | } | |
4871 | ||
55007d84 | 4872 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 4873 | { |
55007d84 | 4874 | int ret = 0; |
510fc4e1 | 4875 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 4876 | |
d6441637 VD |
4877 | if (!parent) |
4878 | return 0; | |
55007d84 | 4879 | |
d6441637 | 4880 | mutex_lock(&activate_kmem_mutex); |
55007d84 | 4881 | /* |
d6441637 VD |
4882 | * If the parent cgroup is not kmem-active now, it cannot be activated |
4883 | * after this point, because it has at least one child already. | |
55007d84 | 4884 | */ |
d6441637 VD |
4885 | if (memcg_kmem_is_active(parent)) |
4886 | ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX); | |
4887 | mutex_unlock(&activate_kmem_mutex); | |
55007d84 | 4888 | return ret; |
510fc4e1 | 4889 | } |
d6441637 VD |
4890 | #else |
4891 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
4892 | unsigned long long val) | |
4893 | { | |
4894 | return -EINVAL; | |
4895 | } | |
6d043990 | 4896 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 4897 | |
628f4235 KH |
4898 | /* |
4899 | * The user of this function is... | |
4900 | * RES_LIMIT. | |
4901 | */ | |
451af504 TH |
4902 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
4903 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 4904 | { |
451af504 | 4905 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
86ae53e1 GC |
4906 | enum res_type type; |
4907 | int name; | |
628f4235 KH |
4908 | unsigned long long val; |
4909 | int ret; | |
4910 | ||
451af504 TH |
4911 | buf = strstrip(buf); |
4912 | type = MEMFILE_TYPE(of_cft(of)->private); | |
4913 | name = MEMFILE_ATTR(of_cft(of)->private); | |
af36f906 | 4914 | |
8c7c6e34 | 4915 | switch (name) { |
628f4235 | 4916 | case RES_LIMIT: |
4b3bde4c BS |
4917 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4918 | ret = -EINVAL; | |
4919 | break; | |
4920 | } | |
628f4235 | 4921 | /* This function does all necessary parse...reuse it */ |
451af504 | 4922 | ret = res_counter_memparse_write_strategy(buf, &val); |
8c7c6e34 KH |
4923 | if (ret) |
4924 | break; | |
4925 | if (type == _MEM) | |
628f4235 | 4926 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 4927 | else if (type == _MEMSWAP) |
8c7c6e34 | 4928 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 4929 | else if (type == _KMEM) |
d6441637 | 4930 | ret = memcg_update_kmem_limit(memcg, val); |
510fc4e1 GC |
4931 | else |
4932 | return -EINVAL; | |
628f4235 | 4933 | break; |
296c81d8 | 4934 | case RES_SOFT_LIMIT: |
451af504 | 4935 | ret = res_counter_memparse_write_strategy(buf, &val); |
296c81d8 BS |
4936 | if (ret) |
4937 | break; | |
4938 | /* | |
4939 | * For memsw, soft limits are hard to implement in terms | |
4940 | * of semantics, for now, we support soft limits for | |
4941 | * control without swap | |
4942 | */ | |
4943 | if (type == _MEM) | |
4944 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
4945 | else | |
4946 | ret = -EINVAL; | |
4947 | break; | |
628f4235 KH |
4948 | default: |
4949 | ret = -EINVAL; /* should be BUG() ? */ | |
4950 | break; | |
4951 | } | |
451af504 | 4952 | return ret ?: nbytes; |
8cdea7c0 BS |
4953 | } |
4954 | ||
fee7b548 KH |
4955 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
4956 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
4957 | { | |
fee7b548 KH |
4958 | unsigned long long min_limit, min_memsw_limit, tmp; |
4959 | ||
4960 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4961 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
4962 | if (!memcg->use_hierarchy) |
4963 | goto out; | |
4964 | ||
5c9d535b TH |
4965 | while (memcg->css.parent) { |
4966 | memcg = mem_cgroup_from_css(memcg->css.parent); | |
fee7b548 KH |
4967 | if (!memcg->use_hierarchy) |
4968 | break; | |
4969 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4970 | min_limit = min(min_limit, tmp); | |
4971 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4972 | min_memsw_limit = min(min_memsw_limit, tmp); | |
4973 | } | |
4974 | out: | |
4975 | *mem_limit = min_limit; | |
4976 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
4977 | } |
4978 | ||
6770c64e TH |
4979 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
4980 | size_t nbytes, loff_t off) | |
c84872e1 | 4981 | { |
6770c64e | 4982 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
86ae53e1 GC |
4983 | int name; |
4984 | enum res_type type; | |
c84872e1 | 4985 | |
6770c64e TH |
4986 | type = MEMFILE_TYPE(of_cft(of)->private); |
4987 | name = MEMFILE_ATTR(of_cft(of)->private); | |
af36f906 | 4988 | |
8c7c6e34 | 4989 | switch (name) { |
29f2a4da | 4990 | case RES_MAX_USAGE: |
8c7c6e34 | 4991 | if (type == _MEM) |
c0ff4b85 | 4992 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 4993 | else if (type == _MEMSWAP) |
c0ff4b85 | 4994 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
4995 | else if (type == _KMEM) |
4996 | res_counter_reset_max(&memcg->kmem); | |
4997 | else | |
4998 | return -EINVAL; | |
29f2a4da PE |
4999 | break; |
5000 | case RES_FAILCNT: | |
8c7c6e34 | 5001 | if (type == _MEM) |
c0ff4b85 | 5002 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5003 | else if (type == _MEMSWAP) |
c0ff4b85 | 5004 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5005 | else if (type == _KMEM) |
5006 | res_counter_reset_failcnt(&memcg->kmem); | |
5007 | else | |
5008 | return -EINVAL; | |
29f2a4da PE |
5009 | break; |
5010 | } | |
f64c3f54 | 5011 | |
6770c64e | 5012 | return nbytes; |
c84872e1 PE |
5013 | } |
5014 | ||
182446d0 | 5015 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5016 | struct cftype *cft) |
5017 | { | |
182446d0 | 5018 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
5019 | } |
5020 | ||
02491447 | 5021 | #ifdef CONFIG_MMU |
182446d0 | 5022 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5023 | struct cftype *cft, u64 val) |
5024 | { | |
182446d0 | 5025 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
5026 | |
5027 | if (val >= (1 << NR_MOVE_TYPE)) | |
5028 | return -EINVAL; | |
ee5e8472 | 5029 | |
7dc74be0 | 5030 | /* |
ee5e8472 GC |
5031 | * No kind of locking is needed in here, because ->can_attach() will |
5032 | * check this value once in the beginning of the process, and then carry | |
5033 | * on with stale data. This means that changes to this value will only | |
5034 | * affect task migrations starting after the change. | |
7dc74be0 | 5035 | */ |
c0ff4b85 | 5036 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5037 | return 0; |
5038 | } | |
02491447 | 5039 | #else |
182446d0 | 5040 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
5041 | struct cftype *cft, u64 val) |
5042 | { | |
5043 | return -ENOSYS; | |
5044 | } | |
5045 | #endif | |
7dc74be0 | 5046 | |
406eb0c9 | 5047 | #ifdef CONFIG_NUMA |
2da8ca82 | 5048 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 5049 | { |
25485de6 GT |
5050 | struct numa_stat { |
5051 | const char *name; | |
5052 | unsigned int lru_mask; | |
5053 | }; | |
5054 | ||
5055 | static const struct numa_stat stats[] = { | |
5056 | { "total", LRU_ALL }, | |
5057 | { "file", LRU_ALL_FILE }, | |
5058 | { "anon", LRU_ALL_ANON }, | |
5059 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
5060 | }; | |
5061 | const struct numa_stat *stat; | |
406eb0c9 | 5062 | int nid; |
25485de6 | 5063 | unsigned long nr; |
2da8ca82 | 5064 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 5065 | |
25485de6 GT |
5066 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
5067 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
5068 | seq_printf(m, "%s=%lu", stat->name, nr); | |
5069 | for_each_node_state(nid, N_MEMORY) { | |
5070 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
5071 | stat->lru_mask); | |
5072 | seq_printf(m, " N%d=%lu", nid, nr); | |
5073 | } | |
5074 | seq_putc(m, '\n'); | |
406eb0c9 | 5075 | } |
406eb0c9 | 5076 | |
071aee13 YH |
5077 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
5078 | struct mem_cgroup *iter; | |
5079 | ||
5080 | nr = 0; | |
5081 | for_each_mem_cgroup_tree(iter, memcg) | |
5082 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
5083 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
5084 | for_each_node_state(nid, N_MEMORY) { | |
5085 | nr = 0; | |
5086 | for_each_mem_cgroup_tree(iter, memcg) | |
5087 | nr += mem_cgroup_node_nr_lru_pages( | |
5088 | iter, nid, stat->lru_mask); | |
5089 | seq_printf(m, " N%d=%lu", nid, nr); | |
5090 | } | |
5091 | seq_putc(m, '\n'); | |
406eb0c9 | 5092 | } |
406eb0c9 | 5093 | |
406eb0c9 YH |
5094 | return 0; |
5095 | } | |
5096 | #endif /* CONFIG_NUMA */ | |
5097 | ||
af7c4b0e JW |
5098 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5099 | { | |
5100 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5101 | } | |
5102 | ||
2da8ca82 | 5103 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 5104 | { |
2da8ca82 | 5105 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
af7c4b0e JW |
5106 | struct mem_cgroup *mi; |
5107 | unsigned int i; | |
406eb0c9 | 5108 | |
af7c4b0e | 5109 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5110 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5111 | continue; |
af7c4b0e JW |
5112 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5113 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5114 | } |
7b854121 | 5115 | |
af7c4b0e JW |
5116 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5117 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5118 | mem_cgroup_read_events(memcg, i)); | |
5119 | ||
5120 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5121 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5122 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5123 | ||
14067bb3 | 5124 | /* Hierarchical information */ |
fee7b548 KH |
5125 | { |
5126 | unsigned long long limit, memsw_limit; | |
d79154bb | 5127 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5128 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5129 | if (do_swap_account) |
78ccf5b5 JW |
5130 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5131 | memsw_limit); | |
fee7b548 | 5132 | } |
7f016ee8 | 5133 | |
af7c4b0e JW |
5134 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5135 | long long val = 0; | |
5136 | ||
bff6bb83 | 5137 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5138 | continue; |
af7c4b0e JW |
5139 | for_each_mem_cgroup_tree(mi, memcg) |
5140 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5141 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5142 | } | |
5143 | ||
5144 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5145 | unsigned long long val = 0; | |
5146 | ||
5147 | for_each_mem_cgroup_tree(mi, memcg) | |
5148 | val += mem_cgroup_read_events(mi, i); | |
5149 | seq_printf(m, "total_%s %llu\n", | |
5150 | mem_cgroup_events_names[i], val); | |
5151 | } | |
5152 | ||
5153 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5154 | unsigned long long val = 0; | |
5155 | ||
5156 | for_each_mem_cgroup_tree(mi, memcg) | |
5157 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5158 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5159 | } |
14067bb3 | 5160 | |
7f016ee8 | 5161 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5162 | { |
5163 | int nid, zid; | |
5164 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5165 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5166 | unsigned long recent_rotated[2] = {0, 0}; |
5167 | unsigned long recent_scanned[2] = {0, 0}; | |
5168 | ||
5169 | for_each_online_node(nid) | |
5170 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 5171 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 5172 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5173 | |
89abfab1 HD |
5174 | recent_rotated[0] += rstat->recent_rotated[0]; |
5175 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5176 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5177 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5178 | } |
78ccf5b5 JW |
5179 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5180 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5181 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5182 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5183 | } |
5184 | #endif | |
5185 | ||
d2ceb9b7 KH |
5186 | return 0; |
5187 | } | |
5188 | ||
182446d0 TH |
5189 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
5190 | struct cftype *cft) | |
a7885eb8 | 5191 | { |
182446d0 | 5192 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 5193 | |
1f4c025b | 5194 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5195 | } |
5196 | ||
182446d0 TH |
5197 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
5198 | struct cftype *cft, u64 val) | |
a7885eb8 | 5199 | { |
182446d0 | 5200 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 5201 | |
3dae7fec | 5202 | if (val > 100) |
a7885eb8 KM |
5203 | return -EINVAL; |
5204 | ||
14208b0e | 5205 | if (css->parent) |
3dae7fec JW |
5206 | memcg->swappiness = val; |
5207 | else | |
5208 | vm_swappiness = val; | |
068b38c1 | 5209 | |
a7885eb8 KM |
5210 | return 0; |
5211 | } | |
5212 | ||
2e72b634 KS |
5213 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5214 | { | |
5215 | struct mem_cgroup_threshold_ary *t; | |
5216 | u64 usage; | |
5217 | int i; | |
5218 | ||
5219 | rcu_read_lock(); | |
5220 | if (!swap) | |
2c488db2 | 5221 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5222 | else |
2c488db2 | 5223 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5224 | |
5225 | if (!t) | |
5226 | goto unlock; | |
5227 | ||
05b84301 JW |
5228 | if (!swap) |
5229 | usage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
5230 | else | |
5231 | usage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
2e72b634 KS |
5232 | |
5233 | /* | |
748dad36 | 5234 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5235 | * If it's not true, a threshold was crossed after last |
5236 | * call of __mem_cgroup_threshold(). | |
5237 | */ | |
5407a562 | 5238 | i = t->current_threshold; |
2e72b634 KS |
5239 | |
5240 | /* | |
5241 | * Iterate backward over array of thresholds starting from | |
5242 | * current_threshold and check if a threshold is crossed. | |
5243 | * If none of thresholds below usage is crossed, we read | |
5244 | * only one element of the array here. | |
5245 | */ | |
5246 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5247 | eventfd_signal(t->entries[i].eventfd, 1); | |
5248 | ||
5249 | /* i = current_threshold + 1 */ | |
5250 | i++; | |
5251 | ||
5252 | /* | |
5253 | * Iterate forward over array of thresholds starting from | |
5254 | * current_threshold+1 and check if a threshold is crossed. | |
5255 | * If none of thresholds above usage is crossed, we read | |
5256 | * only one element of the array here. | |
5257 | */ | |
5258 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5259 | eventfd_signal(t->entries[i].eventfd, 1); | |
5260 | ||
5261 | /* Update current_threshold */ | |
5407a562 | 5262 | t->current_threshold = i - 1; |
2e72b634 KS |
5263 | unlock: |
5264 | rcu_read_unlock(); | |
5265 | } | |
5266 | ||
5267 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5268 | { | |
ad4ca5f4 KS |
5269 | while (memcg) { |
5270 | __mem_cgroup_threshold(memcg, false); | |
5271 | if (do_swap_account) | |
5272 | __mem_cgroup_threshold(memcg, true); | |
5273 | ||
5274 | memcg = parent_mem_cgroup(memcg); | |
5275 | } | |
2e72b634 KS |
5276 | } |
5277 | ||
5278 | static int compare_thresholds(const void *a, const void *b) | |
5279 | { | |
5280 | const struct mem_cgroup_threshold *_a = a; | |
5281 | const struct mem_cgroup_threshold *_b = b; | |
5282 | ||
2bff24a3 GT |
5283 | if (_a->threshold > _b->threshold) |
5284 | return 1; | |
5285 | ||
5286 | if (_a->threshold < _b->threshold) | |
5287 | return -1; | |
5288 | ||
5289 | return 0; | |
2e72b634 KS |
5290 | } |
5291 | ||
c0ff4b85 | 5292 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5293 | { |
5294 | struct mem_cgroup_eventfd_list *ev; | |
5295 | ||
2bcf2e92 MH |
5296 | spin_lock(&memcg_oom_lock); |
5297 | ||
c0ff4b85 | 5298 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 5299 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
5300 | |
5301 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
5302 | return 0; |
5303 | } | |
5304 | ||
c0ff4b85 | 5305 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5306 | { |
7d74b06f KH |
5307 | struct mem_cgroup *iter; |
5308 | ||
c0ff4b85 | 5309 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5310 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5311 | } |
5312 | ||
59b6f873 | 5313 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 5314 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 5315 | { |
2c488db2 KS |
5316 | struct mem_cgroup_thresholds *thresholds; |
5317 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 5318 | u64 threshold, usage; |
2c488db2 | 5319 | int i, size, ret; |
2e72b634 KS |
5320 | |
5321 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5322 | if (ret) | |
5323 | return ret; | |
5324 | ||
5325 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5326 | |
05b84301 | 5327 | if (type == _MEM) { |
2c488db2 | 5328 | thresholds = &memcg->thresholds; |
05b84301 JW |
5329 | usage = res_counter_read_u64(&memcg->res, RES_USAGE); |
5330 | } else if (type == _MEMSWAP) { | |
2c488db2 | 5331 | thresholds = &memcg->memsw_thresholds; |
05b84301 JW |
5332 | usage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
5333 | } else | |
2e72b634 KS |
5334 | BUG(); |
5335 | ||
2e72b634 | 5336 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 5337 | if (thresholds->primary) |
2e72b634 KS |
5338 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5339 | ||
2c488db2 | 5340 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5341 | |
5342 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5343 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5344 | GFP_KERNEL); |
2c488db2 | 5345 | if (!new) { |
2e72b634 KS |
5346 | ret = -ENOMEM; |
5347 | goto unlock; | |
5348 | } | |
2c488db2 | 5349 | new->size = size; |
2e72b634 KS |
5350 | |
5351 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5352 | if (thresholds->primary) { |
5353 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5354 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5355 | } |
5356 | ||
2e72b634 | 5357 | /* Add new threshold */ |
2c488db2 KS |
5358 | new->entries[size - 1].eventfd = eventfd; |
5359 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5360 | |
5361 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5362 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5363 | compare_thresholds, NULL); |
5364 | ||
5365 | /* Find current threshold */ | |
2c488db2 | 5366 | new->current_threshold = -1; |
2e72b634 | 5367 | for (i = 0; i < size; i++) { |
748dad36 | 5368 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5369 | /* |
2c488db2 KS |
5370 | * new->current_threshold will not be used until |
5371 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5372 | * it here. |
5373 | */ | |
2c488db2 | 5374 | ++new->current_threshold; |
748dad36 SZ |
5375 | } else |
5376 | break; | |
2e72b634 KS |
5377 | } |
5378 | ||
2c488db2 KS |
5379 | /* Free old spare buffer and save old primary buffer as spare */ |
5380 | kfree(thresholds->spare); | |
5381 | thresholds->spare = thresholds->primary; | |
5382 | ||
5383 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5384 | |
907860ed | 5385 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5386 | synchronize_rcu(); |
5387 | ||
2e72b634 KS |
5388 | unlock: |
5389 | mutex_unlock(&memcg->thresholds_lock); | |
5390 | ||
5391 | return ret; | |
5392 | } | |
5393 | ||
59b6f873 | 5394 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5395 | struct eventfd_ctx *eventfd, const char *args) |
5396 | { | |
59b6f873 | 5397 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
5398 | } |
5399 | ||
59b6f873 | 5400 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5401 | struct eventfd_ctx *eventfd, const char *args) |
5402 | { | |
59b6f873 | 5403 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
5404 | } |
5405 | ||
59b6f873 | 5406 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 5407 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 5408 | { |
2c488db2 KS |
5409 | struct mem_cgroup_thresholds *thresholds; |
5410 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 5411 | u64 usage; |
2c488db2 | 5412 | int i, j, size; |
2e72b634 KS |
5413 | |
5414 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
5415 | |
5416 | if (type == _MEM) { | |
2c488db2 | 5417 | thresholds = &memcg->thresholds; |
05b84301 JW |
5418 | usage = res_counter_read_u64(&memcg->res, RES_USAGE); |
5419 | } else if (type == _MEMSWAP) { | |
2c488db2 | 5420 | thresholds = &memcg->memsw_thresholds; |
05b84301 JW |
5421 | usage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
5422 | } else | |
2e72b634 KS |
5423 | BUG(); |
5424 | ||
371528ca AV |
5425 | if (!thresholds->primary) |
5426 | goto unlock; | |
5427 | ||
2e72b634 KS |
5428 | /* Check if a threshold crossed before removing */ |
5429 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5430 | ||
5431 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5432 | size = 0; |
5433 | for (i = 0; i < thresholds->primary->size; i++) { | |
5434 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5435 | size++; |
5436 | } | |
5437 | ||
2c488db2 | 5438 | new = thresholds->spare; |
907860ed | 5439 | |
2e72b634 KS |
5440 | /* Set thresholds array to NULL if we don't have thresholds */ |
5441 | if (!size) { | |
2c488db2 KS |
5442 | kfree(new); |
5443 | new = NULL; | |
907860ed | 5444 | goto swap_buffers; |
2e72b634 KS |
5445 | } |
5446 | ||
2c488db2 | 5447 | new->size = size; |
2e72b634 KS |
5448 | |
5449 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5450 | new->current_threshold = -1; |
5451 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5452 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5453 | continue; |
5454 | ||
2c488db2 | 5455 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5456 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5457 | /* |
2c488db2 | 5458 | * new->current_threshold will not be used |
2e72b634 KS |
5459 | * until rcu_assign_pointer(), so it's safe to increment |
5460 | * it here. | |
5461 | */ | |
2c488db2 | 5462 | ++new->current_threshold; |
2e72b634 KS |
5463 | } |
5464 | j++; | |
5465 | } | |
5466 | ||
907860ed | 5467 | swap_buffers: |
2c488db2 KS |
5468 | /* Swap primary and spare array */ |
5469 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5470 | /* If all events are unregistered, free the spare array */ |
5471 | if (!new) { | |
5472 | kfree(thresholds->spare); | |
5473 | thresholds->spare = NULL; | |
5474 | } | |
5475 | ||
2c488db2 | 5476 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5477 | |
907860ed | 5478 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5479 | synchronize_rcu(); |
371528ca | 5480 | unlock: |
2e72b634 | 5481 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5482 | } |
c1e862c1 | 5483 | |
59b6f873 | 5484 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5485 | struct eventfd_ctx *eventfd) |
5486 | { | |
59b6f873 | 5487 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
5488 | } |
5489 | ||
59b6f873 | 5490 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5491 | struct eventfd_ctx *eventfd) |
5492 | { | |
59b6f873 | 5493 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
5494 | } |
5495 | ||
59b6f873 | 5496 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 5497 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 5498 | { |
9490ff27 | 5499 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 5500 | |
9490ff27 KH |
5501 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
5502 | if (!event) | |
5503 | return -ENOMEM; | |
5504 | ||
1af8efe9 | 5505 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5506 | |
5507 | event->eventfd = eventfd; | |
5508 | list_add(&event->list, &memcg->oom_notify); | |
5509 | ||
5510 | /* already in OOM ? */ | |
79dfdacc | 5511 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5512 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5513 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5514 | |
5515 | return 0; | |
5516 | } | |
5517 | ||
59b6f873 | 5518 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 5519 | struct eventfd_ctx *eventfd) |
9490ff27 | 5520 | { |
9490ff27 | 5521 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 5522 | |
1af8efe9 | 5523 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5524 | |
c0ff4b85 | 5525 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5526 | if (ev->eventfd == eventfd) { |
5527 | list_del(&ev->list); | |
5528 | kfree(ev); | |
5529 | } | |
5530 | } | |
5531 | ||
1af8efe9 | 5532 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5533 | } |
5534 | ||
2da8ca82 | 5535 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 5536 | { |
2da8ca82 | 5537 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 5538 | |
791badbd TH |
5539 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
5540 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
5541 | return 0; |
5542 | } | |
5543 | ||
182446d0 | 5544 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5545 | struct cftype *cft, u64 val) |
5546 | { | |
182446d0 | 5547 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
5548 | |
5549 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 5550 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
5551 | return -EINVAL; |
5552 | ||
c0ff4b85 | 5553 | memcg->oom_kill_disable = val; |
4d845ebf | 5554 | if (!val) |
c0ff4b85 | 5555 | memcg_oom_recover(memcg); |
3dae7fec | 5556 | |
3c11ecf4 KH |
5557 | return 0; |
5558 | } | |
5559 | ||
c255a458 | 5560 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5561 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5562 | { |
55007d84 GC |
5563 | int ret; |
5564 | ||
2633d7a0 | 5565 | memcg->kmemcg_id = -1; |
55007d84 GC |
5566 | ret = memcg_propagate_kmem(memcg); |
5567 | if (ret) | |
5568 | return ret; | |
2633d7a0 | 5569 | |
1d62e436 | 5570 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5571 | } |
e5671dfa | 5572 | |
10d5ebf4 | 5573 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 5574 | { |
1d62e436 | 5575 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
5576 | } |
5577 | ||
5578 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
5579 | { | |
5580 | if (!memcg_kmem_is_active(memcg)) | |
5581 | return; | |
5582 | ||
5583 | /* | |
5584 | * kmem charges can outlive the cgroup. In the case of slab | |
5585 | * pages, for instance, a page contain objects from various | |
5586 | * processes. As we prevent from taking a reference for every | |
5587 | * such allocation we have to be careful when doing uncharge | |
5588 | * (see memcg_uncharge_kmem) and here during offlining. | |
5589 | * | |
5590 | * The idea is that that only the _last_ uncharge which sees | |
5591 | * the dead memcg will drop the last reference. An additional | |
5592 | * reference is taken here before the group is marked dead | |
5593 | * which is then paired with css_put during uncharge resp. here. | |
5594 | * | |
5595 | * Although this might sound strange as this path is called from | |
ec903c0c TH |
5596 | * css_offline() when the referencemight have dropped down to 0 and |
5597 | * shouldn't be incremented anymore (css_tryget_online() would | |
5598 | * fail) we do not have other options because of the kmem | |
5599 | * allocations lifetime. | |
10d5ebf4 LZ |
5600 | */ |
5601 | css_get(&memcg->css); | |
7de37682 GC |
5602 | |
5603 | memcg_kmem_mark_dead(memcg); | |
5604 | ||
5605 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5606 | return; | |
5607 | ||
7de37682 | 5608 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5609 | css_put(&memcg->css); |
d1a4c0b3 | 5610 | } |
e5671dfa | 5611 | #else |
cbe128e3 | 5612 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5613 | { |
5614 | return 0; | |
5615 | } | |
d1a4c0b3 | 5616 | |
10d5ebf4 LZ |
5617 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
5618 | { | |
5619 | } | |
5620 | ||
5621 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
5622 | { |
5623 | } | |
e5671dfa GC |
5624 | #endif |
5625 | ||
3bc942f3 TH |
5626 | /* |
5627 | * DO NOT USE IN NEW FILES. | |
5628 | * | |
5629 | * "cgroup.event_control" implementation. | |
5630 | * | |
5631 | * This is way over-engineered. It tries to support fully configurable | |
5632 | * events for each user. Such level of flexibility is completely | |
5633 | * unnecessary especially in the light of the planned unified hierarchy. | |
5634 | * | |
5635 | * Please deprecate this and replace with something simpler if at all | |
5636 | * possible. | |
5637 | */ | |
5638 | ||
79bd9814 TH |
5639 | /* |
5640 | * Unregister event and free resources. | |
5641 | * | |
5642 | * Gets called from workqueue. | |
5643 | */ | |
3bc942f3 | 5644 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 5645 | { |
3bc942f3 TH |
5646 | struct mem_cgroup_event *event = |
5647 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 5648 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
5649 | |
5650 | remove_wait_queue(event->wqh, &event->wait); | |
5651 | ||
59b6f873 | 5652 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
5653 | |
5654 | /* Notify userspace the event is going away. */ | |
5655 | eventfd_signal(event->eventfd, 1); | |
5656 | ||
5657 | eventfd_ctx_put(event->eventfd); | |
5658 | kfree(event); | |
59b6f873 | 5659 | css_put(&memcg->css); |
79bd9814 TH |
5660 | } |
5661 | ||
5662 | /* | |
5663 | * Gets called on POLLHUP on eventfd when user closes it. | |
5664 | * | |
5665 | * Called with wqh->lock held and interrupts disabled. | |
5666 | */ | |
3bc942f3 TH |
5667 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
5668 | int sync, void *key) | |
79bd9814 | 5669 | { |
3bc942f3 TH |
5670 | struct mem_cgroup_event *event = |
5671 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 5672 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
5673 | unsigned long flags = (unsigned long)key; |
5674 | ||
5675 | if (flags & POLLHUP) { | |
5676 | /* | |
5677 | * If the event has been detached at cgroup removal, we | |
5678 | * can simply return knowing the other side will cleanup | |
5679 | * for us. | |
5680 | * | |
5681 | * We can't race against event freeing since the other | |
5682 | * side will require wqh->lock via remove_wait_queue(), | |
5683 | * which we hold. | |
5684 | */ | |
fba94807 | 5685 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
5686 | if (!list_empty(&event->list)) { |
5687 | list_del_init(&event->list); | |
5688 | /* | |
5689 | * We are in atomic context, but cgroup_event_remove() | |
5690 | * may sleep, so we have to call it in workqueue. | |
5691 | */ | |
5692 | schedule_work(&event->remove); | |
5693 | } | |
fba94807 | 5694 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
5695 | } |
5696 | ||
5697 | return 0; | |
5698 | } | |
5699 | ||
3bc942f3 | 5700 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
5701 | wait_queue_head_t *wqh, poll_table *pt) |
5702 | { | |
3bc942f3 TH |
5703 | struct mem_cgroup_event *event = |
5704 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
5705 | |
5706 | event->wqh = wqh; | |
5707 | add_wait_queue(wqh, &event->wait); | |
5708 | } | |
5709 | ||
5710 | /* | |
3bc942f3 TH |
5711 | * DO NOT USE IN NEW FILES. |
5712 | * | |
79bd9814 TH |
5713 | * Parse input and register new cgroup event handler. |
5714 | * | |
5715 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
5716 | * Interpretation of args is defined by control file implementation. | |
5717 | */ | |
451af504 TH |
5718 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
5719 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 5720 | { |
451af504 | 5721 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 5722 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5723 | struct mem_cgroup_event *event; |
79bd9814 TH |
5724 | struct cgroup_subsys_state *cfile_css; |
5725 | unsigned int efd, cfd; | |
5726 | struct fd efile; | |
5727 | struct fd cfile; | |
fba94807 | 5728 | const char *name; |
79bd9814 TH |
5729 | char *endp; |
5730 | int ret; | |
5731 | ||
451af504 TH |
5732 | buf = strstrip(buf); |
5733 | ||
5734 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
5735 | if (*endp != ' ') |
5736 | return -EINVAL; | |
451af504 | 5737 | buf = endp + 1; |
79bd9814 | 5738 | |
451af504 | 5739 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
5740 | if ((*endp != ' ') && (*endp != '\0')) |
5741 | return -EINVAL; | |
451af504 | 5742 | buf = endp + 1; |
79bd9814 TH |
5743 | |
5744 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
5745 | if (!event) | |
5746 | return -ENOMEM; | |
5747 | ||
59b6f873 | 5748 | event->memcg = memcg; |
79bd9814 | 5749 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
5750 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
5751 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
5752 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
5753 | |
5754 | efile = fdget(efd); | |
5755 | if (!efile.file) { | |
5756 | ret = -EBADF; | |
5757 | goto out_kfree; | |
5758 | } | |
5759 | ||
5760 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
5761 | if (IS_ERR(event->eventfd)) { | |
5762 | ret = PTR_ERR(event->eventfd); | |
5763 | goto out_put_efile; | |
5764 | } | |
5765 | ||
5766 | cfile = fdget(cfd); | |
5767 | if (!cfile.file) { | |
5768 | ret = -EBADF; | |
5769 | goto out_put_eventfd; | |
5770 | } | |
5771 | ||
5772 | /* the process need read permission on control file */ | |
5773 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
5774 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
5775 | if (ret < 0) | |
5776 | goto out_put_cfile; | |
5777 | ||
fba94807 TH |
5778 | /* |
5779 | * Determine the event callbacks and set them in @event. This used | |
5780 | * to be done via struct cftype but cgroup core no longer knows | |
5781 | * about these events. The following is crude but the whole thing | |
5782 | * is for compatibility anyway. | |
3bc942f3 TH |
5783 | * |
5784 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
5785 | */ |
5786 | name = cfile.file->f_dentry->d_name.name; | |
5787 | ||
5788 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
5789 | event->register_event = mem_cgroup_usage_register_event; | |
5790 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
5791 | } else if (!strcmp(name, "memory.oom_control")) { | |
5792 | event->register_event = mem_cgroup_oom_register_event; | |
5793 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
5794 | } else if (!strcmp(name, "memory.pressure_level")) { | |
5795 | event->register_event = vmpressure_register_event; | |
5796 | event->unregister_event = vmpressure_unregister_event; | |
5797 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
5798 | event->register_event = memsw_cgroup_usage_register_event; |
5799 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
5800 | } else { |
5801 | ret = -EINVAL; | |
5802 | goto out_put_cfile; | |
5803 | } | |
5804 | ||
79bd9814 | 5805 | /* |
b5557c4c TH |
5806 | * Verify @cfile should belong to @css. Also, remaining events are |
5807 | * automatically removed on cgroup destruction but the removal is | |
5808 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 5809 | */ |
ec903c0c TH |
5810 | cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent, |
5811 | &memory_cgrp_subsys); | |
79bd9814 | 5812 | ret = -EINVAL; |
5a17f543 | 5813 | if (IS_ERR(cfile_css)) |
79bd9814 | 5814 | goto out_put_cfile; |
5a17f543 TH |
5815 | if (cfile_css != css) { |
5816 | css_put(cfile_css); | |
79bd9814 | 5817 | goto out_put_cfile; |
5a17f543 | 5818 | } |
79bd9814 | 5819 | |
451af504 | 5820 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
5821 | if (ret) |
5822 | goto out_put_css; | |
5823 | ||
5824 | efile.file->f_op->poll(efile.file, &event->pt); | |
5825 | ||
fba94807 TH |
5826 | spin_lock(&memcg->event_list_lock); |
5827 | list_add(&event->list, &memcg->event_list); | |
5828 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
5829 | |
5830 | fdput(cfile); | |
5831 | fdput(efile); | |
5832 | ||
451af504 | 5833 | return nbytes; |
79bd9814 TH |
5834 | |
5835 | out_put_css: | |
b5557c4c | 5836 | css_put(css); |
79bd9814 TH |
5837 | out_put_cfile: |
5838 | fdput(cfile); | |
5839 | out_put_eventfd: | |
5840 | eventfd_ctx_put(event->eventfd); | |
5841 | out_put_efile: | |
5842 | fdput(efile); | |
5843 | out_kfree: | |
5844 | kfree(event); | |
5845 | ||
5846 | return ret; | |
5847 | } | |
5848 | ||
8cdea7c0 BS |
5849 | static struct cftype mem_cgroup_files[] = { |
5850 | { | |
0eea1030 | 5851 | .name = "usage_in_bytes", |
8c7c6e34 | 5852 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 5853 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5854 | }, |
c84872e1 PE |
5855 | { |
5856 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5857 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5858 | .write = mem_cgroup_reset, |
791badbd | 5859 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5860 | }, |
8cdea7c0 | 5861 | { |
0eea1030 | 5862 | .name = "limit_in_bytes", |
8c7c6e34 | 5863 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5864 | .write = mem_cgroup_write, |
791badbd | 5865 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5866 | }, |
296c81d8 BS |
5867 | { |
5868 | .name = "soft_limit_in_bytes", | |
5869 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5870 | .write = mem_cgroup_write, |
791badbd | 5871 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5872 | }, |
8cdea7c0 BS |
5873 | { |
5874 | .name = "failcnt", | |
8c7c6e34 | 5875 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5876 | .write = mem_cgroup_reset, |
791badbd | 5877 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5878 | }, |
d2ceb9b7 KH |
5879 | { |
5880 | .name = "stat", | |
2da8ca82 | 5881 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5882 | }, |
c1e862c1 KH |
5883 | { |
5884 | .name = "force_empty", | |
6770c64e | 5885 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5886 | }, |
18f59ea7 BS |
5887 | { |
5888 | .name = "use_hierarchy", | |
5889 | .write_u64 = mem_cgroup_hierarchy_write, | |
5890 | .read_u64 = mem_cgroup_hierarchy_read, | |
5891 | }, | |
79bd9814 | 5892 | { |
3bc942f3 | 5893 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5894 | .write = memcg_write_event_control, |
79bd9814 TH |
5895 | .flags = CFTYPE_NO_PREFIX, |
5896 | .mode = S_IWUGO, | |
5897 | }, | |
a7885eb8 KM |
5898 | { |
5899 | .name = "swappiness", | |
5900 | .read_u64 = mem_cgroup_swappiness_read, | |
5901 | .write_u64 = mem_cgroup_swappiness_write, | |
5902 | }, | |
7dc74be0 DN |
5903 | { |
5904 | .name = "move_charge_at_immigrate", | |
5905 | .read_u64 = mem_cgroup_move_charge_read, | |
5906 | .write_u64 = mem_cgroup_move_charge_write, | |
5907 | }, | |
9490ff27 KH |
5908 | { |
5909 | .name = "oom_control", | |
2da8ca82 | 5910 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5911 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5912 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5913 | }, | |
70ddf637 AV |
5914 | { |
5915 | .name = "pressure_level", | |
70ddf637 | 5916 | }, |
406eb0c9 YH |
5917 | #ifdef CONFIG_NUMA |
5918 | { | |
5919 | .name = "numa_stat", | |
2da8ca82 | 5920 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5921 | }, |
5922 | #endif | |
510fc4e1 GC |
5923 | #ifdef CONFIG_MEMCG_KMEM |
5924 | { | |
5925 | .name = "kmem.limit_in_bytes", | |
5926 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5927 | .write = mem_cgroup_write, |
791badbd | 5928 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5929 | }, |
5930 | { | |
5931 | .name = "kmem.usage_in_bytes", | |
5932 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5933 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5934 | }, |
5935 | { | |
5936 | .name = "kmem.failcnt", | |
5937 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5938 | .write = mem_cgroup_reset, |
791badbd | 5939 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5940 | }, |
5941 | { | |
5942 | .name = "kmem.max_usage_in_bytes", | |
5943 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5944 | .write = mem_cgroup_reset, |
791badbd | 5945 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5946 | }, |
749c5415 GC |
5947 | #ifdef CONFIG_SLABINFO |
5948 | { | |
5949 | .name = "kmem.slabinfo", | |
2da8ca82 | 5950 | .seq_show = mem_cgroup_slabinfo_read, |
749c5415 GC |
5951 | }, |
5952 | #endif | |
8c7c6e34 | 5953 | #endif |
6bc10349 | 5954 | { }, /* terminate */ |
af36f906 | 5955 | }; |
8c7c6e34 | 5956 | |
2d11085e MH |
5957 | #ifdef CONFIG_MEMCG_SWAP |
5958 | static struct cftype memsw_cgroup_files[] = { | |
5959 | { | |
5960 | .name = "memsw.usage_in_bytes", | |
5961 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 5962 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5963 | }, |
5964 | { | |
5965 | .name = "memsw.max_usage_in_bytes", | |
5966 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 5967 | .write = mem_cgroup_reset, |
791badbd | 5968 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5969 | }, |
5970 | { | |
5971 | .name = "memsw.limit_in_bytes", | |
5972 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 5973 | .write = mem_cgroup_write, |
791badbd | 5974 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5975 | }, |
5976 | { | |
5977 | .name = "memsw.failcnt", | |
5978 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 5979 | .write = mem_cgroup_reset, |
791badbd | 5980 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5981 | }, |
5982 | { }, /* terminate */ | |
5983 | }; | |
5984 | #endif | |
c0ff4b85 | 5985 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5986 | { |
5987 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 5988 | struct mem_cgroup_per_zone *mz; |
41e3355d | 5989 | int zone, tmp = node; |
1ecaab2b KH |
5990 | /* |
5991 | * This routine is called against possible nodes. | |
5992 | * But it's BUG to call kmalloc() against offline node. | |
5993 | * | |
5994 | * TODO: this routine can waste much memory for nodes which will | |
5995 | * never be onlined. It's better to use memory hotplug callback | |
5996 | * function. | |
5997 | */ | |
41e3355d KH |
5998 | if (!node_state(node, N_NORMAL_MEMORY)) |
5999 | tmp = -1; | |
17295c88 | 6000 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
6001 | if (!pn) |
6002 | return 1; | |
1ecaab2b | 6003 | |
1ecaab2b KH |
6004 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
6005 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 6006 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
6007 | mz->usage_in_excess = 0; |
6008 | mz->on_tree = false; | |
d79154bb | 6009 | mz->memcg = memcg; |
1ecaab2b | 6010 | } |
54f72fe0 | 6011 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
6012 | return 0; |
6013 | } | |
6014 | ||
c0ff4b85 | 6015 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 6016 | { |
54f72fe0 | 6017 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
6018 | } |
6019 | ||
33327948 KH |
6020 | static struct mem_cgroup *mem_cgroup_alloc(void) |
6021 | { | |
d79154bb | 6022 | struct mem_cgroup *memcg; |
8ff69e2c | 6023 | size_t size; |
33327948 | 6024 | |
8ff69e2c VD |
6025 | size = sizeof(struct mem_cgroup); |
6026 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 6027 | |
8ff69e2c | 6028 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 6029 | if (!memcg) |
e7bbcdf3 DC |
6030 | return NULL; |
6031 | ||
d79154bb HD |
6032 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
6033 | if (!memcg->stat) | |
d2e61b8d | 6034 | goto out_free; |
d79154bb HD |
6035 | spin_lock_init(&memcg->pcp_counter_lock); |
6036 | return memcg; | |
d2e61b8d DC |
6037 | |
6038 | out_free: | |
8ff69e2c | 6039 | kfree(memcg); |
d2e61b8d | 6040 | return NULL; |
33327948 KH |
6041 | } |
6042 | ||
59927fb9 | 6043 | /* |
c8b2a36f GC |
6044 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
6045 | * (scanning all at force_empty is too costly...) | |
6046 | * | |
6047 | * Instead of clearing all references at force_empty, we remember | |
6048 | * the number of reference from swap_cgroup and free mem_cgroup when | |
6049 | * it goes down to 0. | |
6050 | * | |
6051 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 6052 | */ |
c8b2a36f GC |
6053 | |
6054 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 6055 | { |
c8b2a36f | 6056 | int node; |
59927fb9 | 6057 | |
bb4cc1a8 | 6058 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
6059 | |
6060 | for_each_node(node) | |
6061 | free_mem_cgroup_per_zone_info(memcg, node); | |
6062 | ||
6063 | free_percpu(memcg->stat); | |
6064 | ||
3f134619 GC |
6065 | /* |
6066 | * We need to make sure that (at least for now), the jump label | |
6067 | * destruction code runs outside of the cgroup lock. This is because | |
6068 | * get_online_cpus(), which is called from the static_branch update, | |
6069 | * can't be called inside the cgroup_lock. cpusets are the ones | |
6070 | * enforcing this dependency, so if they ever change, we might as well. | |
6071 | * | |
6072 | * schedule_work() will guarantee this happens. Be careful if you need | |
6073 | * to move this code around, and make sure it is outside | |
6074 | * the cgroup_lock. | |
6075 | */ | |
a8964b9b | 6076 | disarm_static_keys(memcg); |
8ff69e2c | 6077 | kfree(memcg); |
59927fb9 | 6078 | } |
3afe36b1 | 6079 | |
7bcc1bb1 DN |
6080 | /* |
6081 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6082 | */ | |
e1aab161 | 6083 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6084 | { |
c0ff4b85 | 6085 | if (!memcg->res.parent) |
7bcc1bb1 | 6086 | return NULL; |
c0ff4b85 | 6087 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6088 | } |
e1aab161 | 6089 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6090 | |
bb4cc1a8 AM |
6091 | static void __init mem_cgroup_soft_limit_tree_init(void) |
6092 | { | |
6093 | struct mem_cgroup_tree_per_node *rtpn; | |
6094 | struct mem_cgroup_tree_per_zone *rtpz; | |
6095 | int tmp, node, zone; | |
6096 | ||
6097 | for_each_node(node) { | |
6098 | tmp = node; | |
6099 | if (!node_state(node, N_NORMAL_MEMORY)) | |
6100 | tmp = -1; | |
6101 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
6102 | BUG_ON(!rtpn); | |
6103 | ||
6104 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
6105 | ||
6106 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
6107 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
6108 | rtpz->rb_root = RB_ROOT; | |
6109 | spin_lock_init(&rtpz->lock); | |
6110 | } | |
6111 | } | |
6112 | } | |
6113 | ||
0eb253e2 | 6114 | static struct cgroup_subsys_state * __ref |
eb95419b | 6115 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 6116 | { |
d142e3e6 | 6117 | struct mem_cgroup *memcg; |
04046e1a | 6118 | long error = -ENOMEM; |
6d12e2d8 | 6119 | int node; |
8cdea7c0 | 6120 | |
c0ff4b85 R |
6121 | memcg = mem_cgroup_alloc(); |
6122 | if (!memcg) | |
04046e1a | 6123 | return ERR_PTR(error); |
78fb7466 | 6124 | |
3ed28fa1 | 6125 | for_each_node(node) |
c0ff4b85 | 6126 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6127 | goto free_out; |
f64c3f54 | 6128 | |
c077719b | 6129 | /* root ? */ |
eb95419b | 6130 | if (parent_css == NULL) { |
a41c58a6 | 6131 | root_mem_cgroup = memcg; |
d142e3e6 GC |
6132 | res_counter_init(&memcg->res, NULL); |
6133 | res_counter_init(&memcg->memsw, NULL); | |
6134 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6135 | } |
28dbc4b6 | 6136 | |
d142e3e6 GC |
6137 | memcg->last_scanned_node = MAX_NUMNODES; |
6138 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
6139 | memcg->move_charge_at_immigrate = 0; |
6140 | mutex_init(&memcg->thresholds_lock); | |
6141 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 6142 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
6143 | INIT_LIST_HEAD(&memcg->event_list); |
6144 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
6145 | |
6146 | return &memcg->css; | |
6147 | ||
6148 | free_out: | |
6149 | __mem_cgroup_free(memcg); | |
6150 | return ERR_PTR(error); | |
6151 | } | |
6152 | ||
6153 | static int | |
eb95419b | 6154 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 6155 | { |
eb95419b | 6156 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 6157 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
d142e3e6 | 6158 | |
15a4c835 | 6159 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
6160 | return -ENOSPC; |
6161 | ||
63876986 | 6162 | if (!parent) |
d142e3e6 GC |
6163 | return 0; |
6164 | ||
0999821b | 6165 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6166 | |
6167 | memcg->use_hierarchy = parent->use_hierarchy; | |
6168 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6169 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6170 | ||
6171 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6172 | res_counter_init(&memcg->res, &parent->res); |
6173 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6174 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6175 | |
7bcc1bb1 | 6176 | /* |
8d76a979 LZ |
6177 | * No need to take a reference to the parent because cgroup |
6178 | * core guarantees its existence. | |
7bcc1bb1 | 6179 | */ |
18f59ea7 | 6180 | } else { |
05b84301 JW |
6181 | res_counter_init(&memcg->res, &root_mem_cgroup->res); |
6182 | res_counter_init(&memcg->memsw, &root_mem_cgroup->memsw); | |
6183 | res_counter_init(&memcg->kmem, &root_mem_cgroup->kmem); | |
8c7f6edb TH |
6184 | /* |
6185 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6186 | * much sense so let cgroup subsystem know about this | |
6187 | * unfortunate state in our controller. | |
6188 | */ | |
d142e3e6 | 6189 | if (parent != root_mem_cgroup) |
073219e9 | 6190 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 6191 | } |
0999821b | 6192 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 6193 | |
073219e9 | 6194 | return memcg_init_kmem(memcg, &memory_cgrp_subsys); |
8cdea7c0 BS |
6195 | } |
6196 | ||
5f578161 MH |
6197 | /* |
6198 | * Announce all parents that a group from their hierarchy is gone. | |
6199 | */ | |
6200 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6201 | { | |
6202 | struct mem_cgroup *parent = memcg; | |
6203 | ||
6204 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 6205 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
6206 | |
6207 | /* | |
6208 | * if the root memcg is not hierarchical we have to check it | |
6209 | * explicitely. | |
6210 | */ | |
6211 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 6212 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
6213 | } |
6214 | ||
eb95419b | 6215 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 6216 | { |
eb95419b | 6217 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 6218 | struct mem_cgroup_event *event, *tmp; |
4fb1a86f | 6219 | struct cgroup_subsys_state *iter; |
79bd9814 TH |
6220 | |
6221 | /* | |
6222 | * Unregister events and notify userspace. | |
6223 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
6224 | * directory to avoid race between userspace and kernelspace. | |
6225 | */ | |
fba94807 TH |
6226 | spin_lock(&memcg->event_list_lock); |
6227 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
6228 | list_del_init(&event->list); |
6229 | schedule_work(&event->remove); | |
6230 | } | |
fba94807 | 6231 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 6232 | |
10d5ebf4 LZ |
6233 | kmem_cgroup_css_offline(memcg); |
6234 | ||
5f578161 | 6235 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
4fb1a86f FB |
6236 | |
6237 | /* | |
6238 | * This requires that offlining is serialized. Right now that is | |
6239 | * guaranteed because css_killed_work_fn() holds the cgroup_mutex. | |
6240 | */ | |
6241 | css_for_each_descendant_post(iter, css) | |
6242 | mem_cgroup_reparent_charges(mem_cgroup_from_css(iter)); | |
6243 | ||
776ed0f0 | 6244 | memcg_unregister_all_caches(memcg); |
33cb876e | 6245 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
6246 | } |
6247 | ||
eb95419b | 6248 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 6249 | { |
eb95419b | 6250 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
96f1c58d JW |
6251 | /* |
6252 | * XXX: css_offline() would be where we should reparent all | |
6253 | * memory to prepare the cgroup for destruction. However, | |
ec903c0c | 6254 | * memcg does not do css_tryget_online() and res_counter charging |
96f1c58d JW |
6255 | * under the same RCU lock region, which means that charging |
6256 | * could race with offlining. Offlining only happens to | |
6257 | * cgroups with no tasks in them but charges can show up | |
6258 | * without any tasks from the swapin path when the target | |
6259 | * memcg is looked up from the swapout record and not from the | |
6260 | * current task as it usually is. A race like this can leak | |
6261 | * charges and put pages with stale cgroup pointers into | |
6262 | * circulation: | |
6263 | * | |
6264 | * #0 #1 | |
6265 | * lookup_swap_cgroup_id() | |
6266 | * rcu_read_lock() | |
6267 | * mem_cgroup_lookup() | |
ec903c0c | 6268 | * css_tryget_online() |
96f1c58d | 6269 | * rcu_read_unlock() |
ec903c0c | 6270 | * disable css_tryget_online() |
96f1c58d JW |
6271 | * call_rcu() |
6272 | * offline_css() | |
6273 | * reparent_charges() | |
6274 | * res_counter_charge() | |
6275 | * css_put() | |
6276 | * css_free() | |
6277 | * pc->mem_cgroup = dead memcg | |
6278 | * add page to lru | |
6279 | * | |
6280 | * The bulk of the charges are still moved in offline_css() to | |
6281 | * avoid pinning a lot of pages in case a long-term reference | |
6282 | * like a swapout record is deferring the css_free() to long | |
6283 | * after offlining. But this makes sure we catch any charges | |
6284 | * made after offlining: | |
6285 | */ | |
6286 | mem_cgroup_reparent_charges(memcg); | |
c268e994 | 6287 | |
10d5ebf4 | 6288 | memcg_destroy_kmem(memcg); |
465939a1 | 6289 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
6290 | } |
6291 | ||
1ced953b TH |
6292 | /** |
6293 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
6294 | * @css: the target css | |
6295 | * | |
6296 | * Reset the states of the mem_cgroup associated with @css. This is | |
6297 | * invoked when the userland requests disabling on the default hierarchy | |
6298 | * but the memcg is pinned through dependency. The memcg should stop | |
6299 | * applying policies and should revert to the vanilla state as it may be | |
6300 | * made visible again. | |
6301 | * | |
6302 | * The current implementation only resets the essential configurations. | |
6303 | * This needs to be expanded to cover all the visible parts. | |
6304 | */ | |
6305 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
6306 | { | |
6307 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6308 | ||
6309 | mem_cgroup_resize_limit(memcg, ULLONG_MAX); | |
6310 | mem_cgroup_resize_memsw_limit(memcg, ULLONG_MAX); | |
6311 | memcg_update_kmem_limit(memcg, ULLONG_MAX); | |
6312 | res_counter_set_soft_limit(&memcg->res, ULLONG_MAX); | |
6313 | } | |
6314 | ||
02491447 | 6315 | #ifdef CONFIG_MMU |
7dc74be0 | 6316 | /* Handlers for move charge at task migration. */ |
854ffa8d | 6317 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 6318 | { |
05b84301 | 6319 | int ret; |
9476db97 JW |
6320 | |
6321 | /* Try a single bulk charge without reclaim first */ | |
6322 | ret = mem_cgroup_try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); | |
6323 | if (!ret) { | |
854ffa8d | 6324 | mc.precharge += count; |
854ffa8d DN |
6325 | return ret; |
6326 | } | |
692e7c45 JW |
6327 | if (ret == -EINTR) { |
6328 | __mem_cgroup_cancel_charge(root_mem_cgroup, count); | |
6329 | return ret; | |
6330 | } | |
9476db97 JW |
6331 | |
6332 | /* Try charges one by one with reclaim */ | |
854ffa8d | 6333 | while (count--) { |
9476db97 | 6334 | ret = mem_cgroup_try_charge(mc.to, |
0029e19e | 6335 | GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
6336 | /* |
6337 | * In case of failure, any residual charges against | |
6338 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
6339 | * later on. However, cancel any charges that are |
6340 | * bypassed to root right away or they'll be lost. | |
9476db97 | 6341 | */ |
692e7c45 JW |
6342 | if (ret == -EINTR) |
6343 | __mem_cgroup_cancel_charge(root_mem_cgroup, 1); | |
38c5d72f | 6344 | if (ret) |
38c5d72f | 6345 | return ret; |
854ffa8d | 6346 | mc.precharge++; |
9476db97 | 6347 | cond_resched(); |
854ffa8d | 6348 | } |
9476db97 | 6349 | return 0; |
4ffef5fe DN |
6350 | } |
6351 | ||
6352 | /** | |
8d32ff84 | 6353 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6354 | * @vma: the vma the pte to be checked belongs |
6355 | * @addr: the address corresponding to the pte to be checked | |
6356 | * @ptent: the pte to be checked | |
02491447 | 6357 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6358 | * |
6359 | * Returns | |
6360 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6361 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6362 | * move charge. if @target is not NULL, the page is stored in target->page | |
6363 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6364 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6365 | * target for charge migration. if @target is not NULL, the entry is stored | |
6366 | * in target->ent. | |
4ffef5fe DN |
6367 | * |
6368 | * Called with pte lock held. | |
6369 | */ | |
4ffef5fe DN |
6370 | union mc_target { |
6371 | struct page *page; | |
02491447 | 6372 | swp_entry_t ent; |
4ffef5fe DN |
6373 | }; |
6374 | ||
4ffef5fe | 6375 | enum mc_target_type { |
8d32ff84 | 6376 | MC_TARGET_NONE = 0, |
4ffef5fe | 6377 | MC_TARGET_PAGE, |
02491447 | 6378 | MC_TARGET_SWAP, |
4ffef5fe DN |
6379 | }; |
6380 | ||
90254a65 DN |
6381 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6382 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6383 | { |
90254a65 | 6384 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6385 | |
90254a65 DN |
6386 | if (!page || !page_mapped(page)) |
6387 | return NULL; | |
6388 | if (PageAnon(page)) { | |
6389 | /* we don't move shared anon */ | |
4b91355e | 6390 | if (!move_anon()) |
90254a65 | 6391 | return NULL; |
87946a72 DN |
6392 | } else if (!move_file()) |
6393 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6394 | return NULL; |
6395 | if (!get_page_unless_zero(page)) | |
6396 | return NULL; | |
6397 | ||
6398 | return page; | |
6399 | } | |
6400 | ||
4b91355e | 6401 | #ifdef CONFIG_SWAP |
90254a65 DN |
6402 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6403 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6404 | { | |
90254a65 DN |
6405 | struct page *page = NULL; |
6406 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6407 | ||
6408 | if (!move_anon() || non_swap_entry(ent)) | |
6409 | return NULL; | |
4b91355e KH |
6410 | /* |
6411 | * Because lookup_swap_cache() updates some statistics counter, | |
6412 | * we call find_get_page() with swapper_space directly. | |
6413 | */ | |
33806f06 | 6414 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6415 | if (do_swap_account) |
6416 | entry->val = ent.val; | |
6417 | ||
6418 | return page; | |
6419 | } | |
4b91355e KH |
6420 | #else |
6421 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6422 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6423 | { | |
6424 | return NULL; | |
6425 | } | |
6426 | #endif | |
90254a65 | 6427 | |
87946a72 DN |
6428 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6429 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6430 | { | |
6431 | struct page *page = NULL; | |
87946a72 DN |
6432 | struct address_space *mapping; |
6433 | pgoff_t pgoff; | |
6434 | ||
6435 | if (!vma->vm_file) /* anonymous vma */ | |
6436 | return NULL; | |
6437 | if (!move_file()) | |
6438 | return NULL; | |
6439 | ||
87946a72 DN |
6440 | mapping = vma->vm_file->f_mapping; |
6441 | if (pte_none(ptent)) | |
6442 | pgoff = linear_page_index(vma, addr); | |
6443 | else /* pte_file(ptent) is true */ | |
6444 | pgoff = pte_to_pgoff(ptent); | |
6445 | ||
6446 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6447 | #ifdef CONFIG_SWAP |
6448 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
6449 | if (shmem_mapping(mapping)) { |
6450 | page = find_get_entry(mapping, pgoff); | |
6451 | if (radix_tree_exceptional_entry(page)) { | |
6452 | swp_entry_t swp = radix_to_swp_entry(page); | |
6453 | if (do_swap_account) | |
6454 | *entry = swp; | |
6455 | page = find_get_page(swap_address_space(swp), swp.val); | |
6456 | } | |
6457 | } else | |
6458 | page = find_get_page(mapping, pgoff); | |
6459 | #else | |
6460 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 6461 | #endif |
87946a72 DN |
6462 | return page; |
6463 | } | |
6464 | ||
8d32ff84 | 6465 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6466 | unsigned long addr, pte_t ptent, union mc_target *target) |
6467 | { | |
6468 | struct page *page = NULL; | |
6469 | struct page_cgroup *pc; | |
8d32ff84 | 6470 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6471 | swp_entry_t ent = { .val = 0 }; |
6472 | ||
6473 | if (pte_present(ptent)) | |
6474 | page = mc_handle_present_pte(vma, addr, ptent); | |
6475 | else if (is_swap_pte(ptent)) | |
6476 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6477 | else if (pte_none(ptent) || pte_file(ptent)) |
6478 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6479 | |
6480 | if (!page && !ent.val) | |
8d32ff84 | 6481 | return ret; |
02491447 DN |
6482 | if (page) { |
6483 | pc = lookup_page_cgroup(page); | |
6484 | /* | |
6485 | * Do only loose check w/o page_cgroup lock. | |
6486 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6487 | * the lock. | |
6488 | */ | |
6489 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6490 | ret = MC_TARGET_PAGE; | |
6491 | if (target) | |
6492 | target->page = page; | |
6493 | } | |
6494 | if (!ret || !target) | |
6495 | put_page(page); | |
6496 | } | |
90254a65 DN |
6497 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6498 | if (ent.val && !ret && | |
34c00c31 | 6499 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6500 | ret = MC_TARGET_SWAP; |
6501 | if (target) | |
6502 | target->ent = ent; | |
4ffef5fe | 6503 | } |
4ffef5fe DN |
6504 | return ret; |
6505 | } | |
6506 | ||
12724850 NH |
6507 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6508 | /* | |
6509 | * We don't consider swapping or file mapped pages because THP does not | |
6510 | * support them for now. | |
6511 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6512 | */ | |
6513 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6514 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6515 | { | |
6516 | struct page *page = NULL; | |
6517 | struct page_cgroup *pc; | |
6518 | enum mc_target_type ret = MC_TARGET_NONE; | |
6519 | ||
6520 | page = pmd_page(pmd); | |
309381fe | 6521 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
6522 | if (!move_anon()) |
6523 | return ret; | |
6524 | pc = lookup_page_cgroup(page); | |
6525 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6526 | ret = MC_TARGET_PAGE; | |
6527 | if (target) { | |
6528 | get_page(page); | |
6529 | target->page = page; | |
6530 | } | |
6531 | } | |
6532 | return ret; | |
6533 | } | |
6534 | #else | |
6535 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6536 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6537 | { | |
6538 | return MC_TARGET_NONE; | |
6539 | } | |
6540 | #endif | |
6541 | ||
4ffef5fe DN |
6542 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6543 | unsigned long addr, unsigned long end, | |
6544 | struct mm_walk *walk) | |
6545 | { | |
6546 | struct vm_area_struct *vma = walk->private; | |
6547 | pte_t *pte; | |
6548 | spinlock_t *ptl; | |
6549 | ||
bf929152 | 6550 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
6551 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
6552 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 6553 | spin_unlock(ptl); |
1a5a9906 | 6554 | return 0; |
12724850 | 6555 | } |
03319327 | 6556 | |
45f83cef AA |
6557 | if (pmd_trans_unstable(pmd)) |
6558 | return 0; | |
4ffef5fe DN |
6559 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6560 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6561 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6562 | mc.precharge++; /* increment precharge temporarily */ |
6563 | pte_unmap_unlock(pte - 1, ptl); | |
6564 | cond_resched(); | |
6565 | ||
7dc74be0 DN |
6566 | return 0; |
6567 | } | |
6568 | ||
4ffef5fe DN |
6569 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6570 | { | |
6571 | unsigned long precharge; | |
6572 | struct vm_area_struct *vma; | |
6573 | ||
dfe076b0 | 6574 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6575 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6576 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6577 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6578 | .mm = mm, | |
6579 | .private = vma, | |
6580 | }; | |
6581 | if (is_vm_hugetlb_page(vma)) | |
6582 | continue; | |
4ffef5fe DN |
6583 | walk_page_range(vma->vm_start, vma->vm_end, |
6584 | &mem_cgroup_count_precharge_walk); | |
6585 | } | |
dfe076b0 | 6586 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6587 | |
6588 | precharge = mc.precharge; | |
6589 | mc.precharge = 0; | |
6590 | ||
6591 | return precharge; | |
6592 | } | |
6593 | ||
4ffef5fe DN |
6594 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6595 | { | |
dfe076b0 DN |
6596 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6597 | ||
6598 | VM_BUG_ON(mc.moving_task); | |
6599 | mc.moving_task = current; | |
6600 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6601 | } |
6602 | ||
dfe076b0 DN |
6603 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6604 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6605 | { |
2bd9bb20 KH |
6606 | struct mem_cgroup *from = mc.from; |
6607 | struct mem_cgroup *to = mc.to; | |
4050377b | 6608 | int i; |
2bd9bb20 | 6609 | |
4ffef5fe | 6610 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6611 | if (mc.precharge) { |
6612 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6613 | mc.precharge = 0; | |
6614 | } | |
6615 | /* | |
6616 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6617 | * we must uncharge here. | |
6618 | */ | |
6619 | if (mc.moved_charge) { | |
6620 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6621 | mc.moved_charge = 0; | |
4ffef5fe | 6622 | } |
483c30b5 DN |
6623 | /* we must fixup refcnts and charges */ |
6624 | if (mc.moved_swap) { | |
483c30b5 | 6625 | /* uncharge swap account from the old cgroup */ |
05b84301 JW |
6626 | res_counter_uncharge(&mc.from->memsw, |
6627 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6628 | |
6629 | for (i = 0; i < mc.moved_swap; i++) | |
6630 | css_put(&mc.from->css); | |
483c30b5 | 6631 | |
05b84301 JW |
6632 | /* |
6633 | * we charged both to->res and to->memsw, so we should | |
6634 | * uncharge to->res. | |
6635 | */ | |
6636 | res_counter_uncharge(&mc.to->res, | |
6637 | PAGE_SIZE * mc.moved_swap); | |
4050377b | 6638 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6639 | mc.moved_swap = 0; |
6640 | } | |
dfe076b0 DN |
6641 | memcg_oom_recover(from); |
6642 | memcg_oom_recover(to); | |
6643 | wake_up_all(&mc.waitq); | |
6644 | } | |
6645 | ||
6646 | static void mem_cgroup_clear_mc(void) | |
6647 | { | |
6648 | struct mem_cgroup *from = mc.from; | |
6649 | ||
6650 | /* | |
6651 | * we must clear moving_task before waking up waiters at the end of | |
6652 | * task migration. | |
6653 | */ | |
6654 | mc.moving_task = NULL; | |
6655 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6656 | spin_lock(&mc.lock); |
4ffef5fe DN |
6657 | mc.from = NULL; |
6658 | mc.to = NULL; | |
2bd9bb20 | 6659 | spin_unlock(&mc.lock); |
32047e2a | 6660 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6661 | } |
6662 | ||
eb95419b | 6663 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6664 | struct cgroup_taskset *tset) |
7dc74be0 | 6665 | { |
2f7ee569 | 6666 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6667 | int ret = 0; |
eb95419b | 6668 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 6669 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6670 | |
ee5e8472 GC |
6671 | /* |
6672 | * We are now commited to this value whatever it is. Changes in this | |
6673 | * tunable will only affect upcoming migrations, not the current one. | |
6674 | * So we need to save it, and keep it going. | |
6675 | */ | |
6676 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
6677 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
6678 | struct mm_struct *mm; |
6679 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
6680 | ||
c0ff4b85 | 6681 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
6682 | |
6683 | mm = get_task_mm(p); | |
6684 | if (!mm) | |
6685 | return 0; | |
7dc74be0 | 6686 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
6687 | if (mm->owner == p) { |
6688 | VM_BUG_ON(mc.from); | |
6689 | VM_BUG_ON(mc.to); | |
6690 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 6691 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 6692 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 6693 | mem_cgroup_start_move(from); |
2bd9bb20 | 6694 | spin_lock(&mc.lock); |
4ffef5fe | 6695 | mc.from = from; |
c0ff4b85 | 6696 | mc.to = memcg; |
ee5e8472 | 6697 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 6698 | spin_unlock(&mc.lock); |
dfe076b0 | 6699 | /* We set mc.moving_task later */ |
4ffef5fe DN |
6700 | |
6701 | ret = mem_cgroup_precharge_mc(mm); | |
6702 | if (ret) | |
6703 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
6704 | } |
6705 | mmput(mm); | |
7dc74be0 DN |
6706 | } |
6707 | return ret; | |
6708 | } | |
6709 | ||
eb95419b | 6710 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6711 | struct cgroup_taskset *tset) |
7dc74be0 | 6712 | { |
4ffef5fe | 6713 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
6714 | } |
6715 | ||
4ffef5fe DN |
6716 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6717 | unsigned long addr, unsigned long end, | |
6718 | struct mm_walk *walk) | |
7dc74be0 | 6719 | { |
4ffef5fe DN |
6720 | int ret = 0; |
6721 | struct vm_area_struct *vma = walk->private; | |
6722 | pte_t *pte; | |
6723 | spinlock_t *ptl; | |
12724850 NH |
6724 | enum mc_target_type target_type; |
6725 | union mc_target target; | |
6726 | struct page *page; | |
6727 | struct page_cgroup *pc; | |
4ffef5fe | 6728 | |
12724850 NH |
6729 | /* |
6730 | * We don't take compound_lock() here but no race with splitting thp | |
6731 | * happens because: | |
6732 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
6733 | * under splitting, which means there's no concurrent thp split, | |
6734 | * - if another thread runs into split_huge_page() just after we | |
6735 | * entered this if-block, the thread must wait for page table lock | |
6736 | * to be unlocked in __split_huge_page_splitting(), where the main | |
6737 | * part of thp split is not executed yet. | |
6738 | */ | |
bf929152 | 6739 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 6740 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 6741 | spin_unlock(ptl); |
12724850 NH |
6742 | return 0; |
6743 | } | |
6744 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6745 | if (target_type == MC_TARGET_PAGE) { | |
6746 | page = target.page; | |
6747 | if (!isolate_lru_page(page)) { | |
6748 | pc = lookup_page_cgroup(page); | |
6749 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6750 | pc, mc.from, mc.to)) { |
12724850 NH |
6751 | mc.precharge -= HPAGE_PMD_NR; |
6752 | mc.moved_charge += HPAGE_PMD_NR; | |
6753 | } | |
6754 | putback_lru_page(page); | |
6755 | } | |
6756 | put_page(page); | |
6757 | } | |
bf929152 | 6758 | spin_unlock(ptl); |
1a5a9906 | 6759 | return 0; |
12724850 NH |
6760 | } |
6761 | ||
45f83cef AA |
6762 | if (pmd_trans_unstable(pmd)) |
6763 | return 0; | |
4ffef5fe DN |
6764 | retry: |
6765 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6766 | for (; addr != end; addr += PAGE_SIZE) { | |
6767 | pte_t ptent = *(pte++); | |
02491447 | 6768 | swp_entry_t ent; |
4ffef5fe DN |
6769 | |
6770 | if (!mc.precharge) | |
6771 | break; | |
6772 | ||
8d32ff84 | 6773 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6774 | case MC_TARGET_PAGE: |
6775 | page = target.page; | |
6776 | if (isolate_lru_page(page)) | |
6777 | goto put; | |
6778 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6779 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6780 | mc.from, mc.to)) { |
4ffef5fe | 6781 | mc.precharge--; |
854ffa8d DN |
6782 | /* we uncharge from mc.from later. */ |
6783 | mc.moved_charge++; | |
4ffef5fe DN |
6784 | } |
6785 | putback_lru_page(page); | |
8d32ff84 | 6786 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6787 | put_page(page); |
6788 | break; | |
02491447 DN |
6789 | case MC_TARGET_SWAP: |
6790 | ent = target.ent; | |
e91cbb42 | 6791 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6792 | mc.precharge--; |
483c30b5 DN |
6793 | /* we fixup refcnts and charges later. */ |
6794 | mc.moved_swap++; | |
6795 | } | |
02491447 | 6796 | break; |
4ffef5fe DN |
6797 | default: |
6798 | break; | |
6799 | } | |
6800 | } | |
6801 | pte_unmap_unlock(pte - 1, ptl); | |
6802 | cond_resched(); | |
6803 | ||
6804 | if (addr != end) { | |
6805 | /* | |
6806 | * We have consumed all precharges we got in can_attach(). | |
6807 | * We try charge one by one, but don't do any additional | |
6808 | * charges to mc.to if we have failed in charge once in attach() | |
6809 | * phase. | |
6810 | */ | |
854ffa8d | 6811 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6812 | if (!ret) |
6813 | goto retry; | |
6814 | } | |
6815 | ||
6816 | return ret; | |
6817 | } | |
6818 | ||
6819 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6820 | { | |
6821 | struct vm_area_struct *vma; | |
6822 | ||
6823 | lru_add_drain_all(); | |
dfe076b0 DN |
6824 | retry: |
6825 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6826 | /* | |
6827 | * Someone who are holding the mmap_sem might be waiting in | |
6828 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6829 | * and retry. Because we cancel precharges, we might not be able | |
6830 | * to move enough charges, but moving charge is a best-effort | |
6831 | * feature anyway, so it wouldn't be a big problem. | |
6832 | */ | |
6833 | __mem_cgroup_clear_mc(); | |
6834 | cond_resched(); | |
6835 | goto retry; | |
6836 | } | |
4ffef5fe DN |
6837 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6838 | int ret; | |
6839 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6840 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6841 | .mm = mm, | |
6842 | .private = vma, | |
6843 | }; | |
6844 | if (is_vm_hugetlb_page(vma)) | |
6845 | continue; | |
4ffef5fe DN |
6846 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6847 | &mem_cgroup_move_charge_walk); | |
6848 | if (ret) | |
6849 | /* | |
6850 | * means we have consumed all precharges and failed in | |
6851 | * doing additional charge. Just abandon here. | |
6852 | */ | |
6853 | break; | |
6854 | } | |
dfe076b0 | 6855 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6856 | } |
6857 | ||
eb95419b | 6858 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6859 | struct cgroup_taskset *tset) |
67e465a7 | 6860 | { |
2f7ee569 | 6861 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6862 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6863 | |
dfe076b0 | 6864 | if (mm) { |
a433658c KM |
6865 | if (mc.to) |
6866 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6867 | mmput(mm); |
6868 | } | |
a433658c KM |
6869 | if (mc.to) |
6870 | mem_cgroup_clear_mc(); | |
67e465a7 | 6871 | } |
5cfb80a7 | 6872 | #else /* !CONFIG_MMU */ |
eb95419b | 6873 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6874 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6875 | { |
6876 | return 0; | |
6877 | } | |
eb95419b | 6878 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6879 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6880 | { |
6881 | } | |
eb95419b | 6882 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6883 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6884 | { |
6885 | } | |
6886 | #endif | |
67e465a7 | 6887 | |
f00baae7 TH |
6888 | /* |
6889 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6890 | * to verify whether we're attached to the default hierarchy on each mount |
6891 | * attempt. | |
f00baae7 | 6892 | */ |
eb95419b | 6893 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6894 | { |
6895 | /* | |
aa6ec29b | 6896 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6897 | * guarantees that @root doesn't have any children, so turning it |
6898 | * on for the root memcg is enough. | |
6899 | */ | |
aa6ec29b | 6900 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 6901 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
6902 | } |
6903 | ||
073219e9 | 6904 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6905 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6906 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6907 | .css_offline = mem_cgroup_css_offline, |
6908 | .css_free = mem_cgroup_css_free, | |
1ced953b | 6909 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6910 | .can_attach = mem_cgroup_can_attach, |
6911 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6912 | .attach = mem_cgroup_move_task, |
f00baae7 | 6913 | .bind = mem_cgroup_bind, |
5577964e | 6914 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 6915 | .early_init = 0, |
8cdea7c0 | 6916 | }; |
c077719b | 6917 | |
c255a458 | 6918 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6919 | static int __init enable_swap_account(char *s) |
6920 | { | |
a2c8990a | 6921 | if (!strcmp(s, "1")) |
a42c390c | 6922 | really_do_swap_account = 1; |
a2c8990a | 6923 | else if (!strcmp(s, "0")) |
a42c390c MH |
6924 | really_do_swap_account = 0; |
6925 | return 1; | |
6926 | } | |
a2c8990a | 6927 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6928 | |
2d11085e MH |
6929 | static void __init memsw_file_init(void) |
6930 | { | |
2cf669a5 TH |
6931 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6932 | memsw_cgroup_files)); | |
6acc8b02 MH |
6933 | } |
6934 | ||
6935 | static void __init enable_swap_cgroup(void) | |
6936 | { | |
6937 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6938 | do_swap_account = 1; | |
6939 | memsw_file_init(); | |
6940 | } | |
2d11085e | 6941 | } |
6acc8b02 | 6942 | |
2d11085e | 6943 | #else |
6acc8b02 | 6944 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6945 | { |
6946 | } | |
c077719b | 6947 | #endif |
2d11085e MH |
6948 | |
6949 | /* | |
1081312f MH |
6950 | * subsys_initcall() for memory controller. |
6951 | * | |
6952 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6953 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6954 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6955 | * be initialized from here. | |
2d11085e MH |
6956 | */ |
6957 | static int __init mem_cgroup_init(void) | |
6958 | { | |
6959 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6960 | enable_swap_cgroup(); |
bb4cc1a8 | 6961 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6962 | memcg_stock_init(); |
2d11085e MH |
6963 | return 0; |
6964 | } | |
6965 | subsys_initcall(mem_cgroup_init); |