]>
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 | |
83 | static int really_do_swap_account __initdata = 0; | |
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 GC |
359 | #if defined(CONFIG_MEMCG_KMEM) |
360 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
361 | struct list_head memcg_slab_caches; | |
362 | /* Not a spinlock, we can take a lot of time walking the list */ | |
363 | struct mutex slab_caches_mutex; | |
364 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
365 | int kmemcg_id; | |
366 | #endif | |
45cf7ebd GC |
367 | |
368 | int last_scanned_node; | |
369 | #if MAX_NUMNODES > 1 | |
370 | nodemask_t scan_nodes; | |
371 | atomic_t numainfo_events; | |
372 | atomic_t numainfo_updating; | |
373 | #endif | |
70ddf637 | 374 | |
fba94807 TH |
375 | /* List of events which userspace want to receive */ |
376 | struct list_head event_list; | |
377 | spinlock_t event_list_lock; | |
378 | ||
54f72fe0 JW |
379 | struct mem_cgroup_per_node *nodeinfo[0]; |
380 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
381 | }; |
382 | ||
510fc4e1 GC |
383 | /* internal only representation about the status of kmem accounting. */ |
384 | enum { | |
6de64beb | 385 | KMEM_ACCOUNTED_ACTIVE, /* accounted by this cgroup itself */ |
7de37682 | 386 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
387 | }; |
388 | ||
510fc4e1 GC |
389 | #ifdef CONFIG_MEMCG_KMEM |
390 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
391 | { | |
392 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
393 | } | |
7de37682 GC |
394 | |
395 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
396 | { | |
397 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
398 | } | |
399 | ||
400 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) | |
401 | { | |
10d5ebf4 LZ |
402 | /* |
403 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
404 | * will call css_put() if it sees the memcg is dead. | |
405 | */ | |
406 | smp_wmb(); | |
7de37682 GC |
407 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
408 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
409 | } | |
410 | ||
411 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
412 | { | |
413 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
414 | &memcg->kmem_account_flags); | |
415 | } | |
510fc4e1 GC |
416 | #endif |
417 | ||
7dc74be0 DN |
418 | /* Stuffs for move charges at task migration. */ |
419 | /* | |
ee5e8472 GC |
420 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
421 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
422 | */ |
423 | enum move_type { | |
4ffef5fe | 424 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 425 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
426 | NR_MOVE_TYPE, |
427 | }; | |
428 | ||
4ffef5fe DN |
429 | /* "mc" and its members are protected by cgroup_mutex */ |
430 | static struct move_charge_struct { | |
b1dd693e | 431 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
432 | struct mem_cgroup *from; |
433 | struct mem_cgroup *to; | |
ee5e8472 | 434 | unsigned long immigrate_flags; |
4ffef5fe | 435 | unsigned long precharge; |
854ffa8d | 436 | unsigned long moved_charge; |
483c30b5 | 437 | unsigned long moved_swap; |
8033b97c DN |
438 | struct task_struct *moving_task; /* a task moving charges */ |
439 | wait_queue_head_t waitq; /* a waitq for other context */ | |
440 | } mc = { | |
2bd9bb20 | 441 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
442 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
443 | }; | |
4ffef5fe | 444 | |
90254a65 DN |
445 | static bool move_anon(void) |
446 | { | |
ee5e8472 | 447 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
448 | } |
449 | ||
87946a72 DN |
450 | static bool move_file(void) |
451 | { | |
ee5e8472 | 452 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
453 | } |
454 | ||
4e416953 BS |
455 | /* |
456 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
457 | * limit reclaim to prevent infinite loops, if they ever occur. | |
458 | */ | |
a0db00fc | 459 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 460 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 461 | |
217bc319 KH |
462 | enum charge_type { |
463 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 464 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 465 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 466 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
467 | NR_CHARGE_TYPE, |
468 | }; | |
469 | ||
8c7c6e34 | 470 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
471 | enum res_type { |
472 | _MEM, | |
473 | _MEMSWAP, | |
474 | _OOM_TYPE, | |
510fc4e1 | 475 | _KMEM, |
86ae53e1 GC |
476 | }; |
477 | ||
a0db00fc KS |
478 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
479 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 480 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
481 | /* Used for OOM nofiier */ |
482 | #define OOM_CONTROL (0) | |
8c7c6e34 | 483 | |
75822b44 BS |
484 | /* |
485 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
486 | */ | |
487 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
488 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
489 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
490 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
491 | ||
0999821b GC |
492 | /* |
493 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
494 | * As a consequence, any change that needs to protect against new child cgroups | |
495 | * appearing has to hold it as well. | |
496 | */ | |
497 | static DEFINE_MUTEX(memcg_create_mutex); | |
498 | ||
b2145145 WL |
499 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
500 | { | |
a7c6d554 | 501 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
502 | } |
503 | ||
70ddf637 AV |
504 | /* Some nice accessors for the vmpressure. */ |
505 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
506 | { | |
507 | if (!memcg) | |
508 | memcg = root_mem_cgroup; | |
509 | return &memcg->vmpressure; | |
510 | } | |
511 | ||
512 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
513 | { | |
514 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
515 | } | |
516 | ||
7ffc0edc MH |
517 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
518 | { | |
519 | return (memcg == root_mem_cgroup); | |
520 | } | |
521 | ||
4219b2da LZ |
522 | /* |
523 | * We restrict the id in the range of [1, 65535], so it can fit into | |
524 | * an unsigned short. | |
525 | */ | |
526 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
527 | ||
34c00c31 LZ |
528 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
529 | { | |
530 | /* | |
531 | * The ID of the root cgroup is 0, but memcg treat 0 as an | |
532 | * invalid ID, so we return (cgroup_id + 1). | |
533 | */ | |
534 | return memcg->css.cgroup->id + 1; | |
535 | } | |
536 | ||
537 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
538 | { | |
539 | struct cgroup_subsys_state *css; | |
540 | ||
073219e9 | 541 | css = css_from_id(id - 1, &memory_cgrp_subsys); |
34c00c31 LZ |
542 | return mem_cgroup_from_css(css); |
543 | } | |
544 | ||
e1aab161 | 545 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 546 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 547 | |
e1aab161 GC |
548 | void sock_update_memcg(struct sock *sk) |
549 | { | |
376be5ff | 550 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 551 | struct mem_cgroup *memcg; |
3f134619 | 552 | struct cg_proto *cg_proto; |
e1aab161 GC |
553 | |
554 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
555 | ||
f3f511e1 GC |
556 | /* Socket cloning can throw us here with sk_cgrp already |
557 | * filled. It won't however, necessarily happen from | |
558 | * process context. So the test for root memcg given | |
559 | * the current task's memcg won't help us in this case. | |
560 | * | |
561 | * Respecting the original socket's memcg is a better | |
562 | * decision in this case. | |
563 | */ | |
564 | if (sk->sk_cgrp) { | |
565 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 566 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
567 | return; |
568 | } | |
569 | ||
e1aab161 GC |
570 | rcu_read_lock(); |
571 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 572 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae LZ |
573 | if (!mem_cgroup_is_root(memcg) && |
574 | memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) { | |
3f134619 | 575 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
576 | } |
577 | rcu_read_unlock(); | |
578 | } | |
579 | } | |
580 | EXPORT_SYMBOL(sock_update_memcg); | |
581 | ||
582 | void sock_release_memcg(struct sock *sk) | |
583 | { | |
376be5ff | 584 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
585 | struct mem_cgroup *memcg; |
586 | WARN_ON(!sk->sk_cgrp->memcg); | |
587 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 588 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
589 | } |
590 | } | |
d1a4c0b3 GC |
591 | |
592 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
593 | { | |
594 | if (!memcg || mem_cgroup_is_root(memcg)) | |
595 | return NULL; | |
596 | ||
2e685cad | 597 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
598 | } |
599 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 600 | |
3f134619 GC |
601 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
602 | { | |
2e685cad | 603 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
604 | return; |
605 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
606 | } | |
607 | #else | |
608 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
609 | { | |
610 | } | |
611 | #endif | |
612 | ||
a8964b9b | 613 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
614 | /* |
615 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
b8627835 LZ |
616 | * The main reason for not using cgroup id for this: |
617 | * this works better in sparse environments, where we have a lot of memcgs, | |
618 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
619 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
620 | * 200 entry array for that. | |
55007d84 GC |
621 | * |
622 | * The current size of the caches array is stored in | |
623 | * memcg_limited_groups_array_size. It will double each time we have to | |
624 | * increase it. | |
625 | */ | |
626 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
627 | int memcg_limited_groups_array_size; |
628 | ||
55007d84 GC |
629 | /* |
630 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
631 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
632 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
633 | * tunable, but that is strictly not necessary. | |
634 | * | |
b8627835 | 635 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
636 | * this constant directly from cgroup, but it is understandable that this is |
637 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 638 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
639 | * increase ours as well if it increases. |
640 | */ | |
641 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 642 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 643 | |
d7f25f8a GC |
644 | /* |
645 | * A lot of the calls to the cache allocation functions are expected to be | |
646 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
647 | * conditional to this static branch, we'll have to allow modules that does | |
648 | * kmem_cache_alloc and the such to see this symbol as well | |
649 | */ | |
a8964b9b | 650 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 651 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
652 | |
653 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
654 | { | |
55007d84 | 655 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 656 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
657 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
658 | } | |
bea207c8 GC |
659 | /* |
660 | * This check can't live in kmem destruction function, | |
661 | * since the charges will outlive the cgroup | |
662 | */ | |
663 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
664 | } |
665 | #else | |
666 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
667 | { | |
668 | } | |
669 | #endif /* CONFIG_MEMCG_KMEM */ | |
670 | ||
671 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
672 | { | |
673 | disarm_sock_keys(memcg); | |
674 | disarm_kmem_keys(memcg); | |
675 | } | |
676 | ||
c0ff4b85 | 677 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 678 | |
f64c3f54 | 679 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 680 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 681 | { |
45cf7ebd | 682 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
54f72fe0 | 683 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
684 | } |
685 | ||
c0ff4b85 | 686 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 687 | { |
c0ff4b85 | 688 | return &memcg->css; |
d324236b WF |
689 | } |
690 | ||
f64c3f54 | 691 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 692 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 693 | { |
97a6c37b JW |
694 | int nid = page_to_nid(page); |
695 | int zid = page_zonenum(page); | |
f64c3f54 | 696 | |
c0ff4b85 | 697 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
698 | } |
699 | ||
bb4cc1a8 AM |
700 | static struct mem_cgroup_tree_per_zone * |
701 | soft_limit_tree_node_zone(int nid, int zid) | |
702 | { | |
703 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
704 | } | |
705 | ||
706 | static struct mem_cgroup_tree_per_zone * | |
707 | soft_limit_tree_from_page(struct page *page) | |
708 | { | |
709 | int nid = page_to_nid(page); | |
710 | int zid = page_zonenum(page); | |
711 | ||
712 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
713 | } | |
714 | ||
715 | static void | |
716 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, | |
717 | struct mem_cgroup_per_zone *mz, | |
718 | struct mem_cgroup_tree_per_zone *mctz, | |
719 | unsigned long long new_usage_in_excess) | |
720 | { | |
721 | struct rb_node **p = &mctz->rb_root.rb_node; | |
722 | struct rb_node *parent = NULL; | |
723 | struct mem_cgroup_per_zone *mz_node; | |
724 | ||
725 | if (mz->on_tree) | |
726 | return; | |
727 | ||
728 | mz->usage_in_excess = new_usage_in_excess; | |
729 | if (!mz->usage_in_excess) | |
730 | return; | |
731 | while (*p) { | |
732 | parent = *p; | |
733 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
734 | tree_node); | |
735 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
736 | p = &(*p)->rb_left; | |
737 | /* | |
738 | * We can't avoid mem cgroups that are over their soft | |
739 | * limit by the same amount | |
740 | */ | |
741 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
742 | p = &(*p)->rb_right; | |
743 | } | |
744 | rb_link_node(&mz->tree_node, parent, p); | |
745 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
746 | mz->on_tree = true; | |
747 | } | |
748 | ||
749 | static void | |
750 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, | |
751 | struct mem_cgroup_per_zone *mz, | |
752 | struct mem_cgroup_tree_per_zone *mctz) | |
753 | { | |
754 | if (!mz->on_tree) | |
755 | return; | |
756 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
757 | mz->on_tree = false; | |
758 | } | |
759 | ||
760 | static void | |
761 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, | |
762 | struct mem_cgroup_per_zone *mz, | |
763 | struct mem_cgroup_tree_per_zone *mctz) | |
764 | { | |
765 | spin_lock(&mctz->lock); | |
766 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
767 | spin_unlock(&mctz->lock); | |
768 | } | |
769 | ||
770 | ||
771 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
772 | { | |
773 | unsigned long long excess; | |
774 | struct mem_cgroup_per_zone *mz; | |
775 | struct mem_cgroup_tree_per_zone *mctz; | |
776 | int nid = page_to_nid(page); | |
777 | int zid = page_zonenum(page); | |
778 | mctz = soft_limit_tree_from_page(page); | |
779 | ||
780 | /* | |
781 | * Necessary to update all ancestors when hierarchy is used. | |
782 | * because their event counter is not touched. | |
783 | */ | |
784 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
785 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
786 | excess = res_counter_soft_limit_excess(&memcg->res); | |
787 | /* | |
788 | * We have to update the tree if mz is on RB-tree or | |
789 | * mem is over its softlimit. | |
790 | */ | |
791 | if (excess || mz->on_tree) { | |
792 | spin_lock(&mctz->lock); | |
793 | /* if on-tree, remove it */ | |
794 | if (mz->on_tree) | |
795 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
796 | /* | |
797 | * Insert again. mz->usage_in_excess will be updated. | |
798 | * If excess is 0, no tree ops. | |
799 | */ | |
800 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); | |
801 | spin_unlock(&mctz->lock); | |
802 | } | |
803 | } | |
804 | } | |
805 | ||
806 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
807 | { | |
808 | int node, zone; | |
809 | struct mem_cgroup_per_zone *mz; | |
810 | struct mem_cgroup_tree_per_zone *mctz; | |
811 | ||
812 | for_each_node(node) { | |
813 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
814 | mz = mem_cgroup_zoneinfo(memcg, node, zone); | |
815 | mctz = soft_limit_tree_node_zone(node, zone); | |
816 | mem_cgroup_remove_exceeded(memcg, mz, mctz); | |
817 | } | |
818 | } | |
819 | } | |
820 | ||
821 | static struct mem_cgroup_per_zone * | |
822 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
823 | { | |
824 | struct rb_node *rightmost = NULL; | |
825 | struct mem_cgroup_per_zone *mz; | |
826 | ||
827 | retry: | |
828 | mz = NULL; | |
829 | rightmost = rb_last(&mctz->rb_root); | |
830 | if (!rightmost) | |
831 | goto done; /* Nothing to reclaim from */ | |
832 | ||
833 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
834 | /* | |
835 | * Remove the node now but someone else can add it back, | |
836 | * we will to add it back at the end of reclaim to its correct | |
837 | * position in the tree. | |
838 | */ | |
839 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); | |
840 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
841 | !css_tryget(&mz->memcg->css)) | |
842 | goto retry; | |
843 | done: | |
844 | return mz; | |
845 | } | |
846 | ||
847 | static struct mem_cgroup_per_zone * | |
848 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
849 | { | |
850 | struct mem_cgroup_per_zone *mz; | |
851 | ||
852 | spin_lock(&mctz->lock); | |
853 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
854 | spin_unlock(&mctz->lock); | |
855 | return mz; | |
856 | } | |
857 | ||
711d3d2c KH |
858 | /* |
859 | * Implementation Note: reading percpu statistics for memcg. | |
860 | * | |
861 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
862 | * synchronization to implement "quick" read. There are trade-off between | |
863 | * reading cost and precision of value. Then, we may have a chance to implement | |
864 | * a periodic synchronizion of counter in memcg's counter. | |
865 | * | |
866 | * But this _read() function is used for user interface now. The user accounts | |
867 | * memory usage by memory cgroup and he _always_ requires exact value because | |
868 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
869 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
870 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
871 | * | |
872 | * If there are kernel internal actions which can make use of some not-exact | |
873 | * value, and reading all cpu value can be performance bottleneck in some | |
874 | * common workload, threashold and synchonization as vmstat[] should be | |
875 | * implemented. | |
876 | */ | |
c0ff4b85 | 877 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 878 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 879 | { |
7a159cc9 | 880 | long val = 0; |
c62b1a3b | 881 | int cpu; |
c62b1a3b | 882 | |
711d3d2c KH |
883 | get_online_cpus(); |
884 | for_each_online_cpu(cpu) | |
c0ff4b85 | 885 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 886 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
887 | spin_lock(&memcg->pcp_counter_lock); |
888 | val += memcg->nocpu_base.count[idx]; | |
889 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
890 | #endif |
891 | put_online_cpus(); | |
c62b1a3b KH |
892 | return val; |
893 | } | |
894 | ||
c0ff4b85 | 895 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
896 | bool charge) |
897 | { | |
898 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 899 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
900 | } |
901 | ||
c0ff4b85 | 902 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
903 | enum mem_cgroup_events_index idx) |
904 | { | |
905 | unsigned long val = 0; | |
906 | int cpu; | |
907 | ||
9c567512 | 908 | get_online_cpus(); |
e9f8974f | 909 | for_each_online_cpu(cpu) |
c0ff4b85 | 910 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 911 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
912 | spin_lock(&memcg->pcp_counter_lock); |
913 | val += memcg->nocpu_base.events[idx]; | |
914 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 915 | #endif |
9c567512 | 916 | put_online_cpus(); |
e9f8974f JW |
917 | return val; |
918 | } | |
919 | ||
c0ff4b85 | 920 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 921 | struct page *page, |
b2402857 | 922 | bool anon, int nr_pages) |
d52aa412 | 923 | { |
b2402857 KH |
924 | /* |
925 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
926 | * counted as CACHE even if it's on ANON LRU. | |
927 | */ | |
928 | if (anon) | |
929 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 930 | nr_pages); |
d52aa412 | 931 | else |
b2402857 | 932 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 933 | nr_pages); |
55e462b0 | 934 | |
b070e65c DR |
935 | if (PageTransHuge(page)) |
936 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
937 | nr_pages); | |
938 | ||
e401f176 KH |
939 | /* pagein of a big page is an event. So, ignore page size */ |
940 | if (nr_pages > 0) | |
c0ff4b85 | 941 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 942 | else { |
c0ff4b85 | 943 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
944 | nr_pages = -nr_pages; /* for event */ |
945 | } | |
e401f176 | 946 | |
13114716 | 947 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
948 | } |
949 | ||
bb2a0de9 | 950 | unsigned long |
4d7dcca2 | 951 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
952 | { |
953 | struct mem_cgroup_per_zone *mz; | |
954 | ||
955 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
956 | return mz->lru_size[lru]; | |
957 | } | |
958 | ||
959 | static unsigned long | |
c0ff4b85 | 960 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 961 | unsigned int lru_mask) |
889976db YH |
962 | { |
963 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 964 | enum lru_list lru; |
bb2a0de9 KH |
965 | unsigned long ret = 0; |
966 | ||
c0ff4b85 | 967 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 968 | |
f156ab93 HD |
969 | for_each_lru(lru) { |
970 | if (BIT(lru) & lru_mask) | |
971 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
972 | } |
973 | return ret; | |
974 | } | |
975 | ||
976 | static unsigned long | |
c0ff4b85 | 977 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
978 | int nid, unsigned int lru_mask) |
979 | { | |
889976db YH |
980 | u64 total = 0; |
981 | int zid; | |
982 | ||
bb2a0de9 | 983 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
984 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
985 | nid, zid, lru_mask); | |
bb2a0de9 | 986 | |
889976db YH |
987 | return total; |
988 | } | |
bb2a0de9 | 989 | |
c0ff4b85 | 990 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 991 | unsigned int lru_mask) |
6d12e2d8 | 992 | { |
889976db | 993 | int nid; |
6d12e2d8 KH |
994 | u64 total = 0; |
995 | ||
31aaea4a | 996 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 997 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 998 | return total; |
d52aa412 KH |
999 | } |
1000 | ||
f53d7ce3 JW |
1001 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
1002 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
1003 | { |
1004 | unsigned long val, next; | |
1005 | ||
13114716 | 1006 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 1007 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 1008 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
1009 | if ((long)next - (long)val < 0) { |
1010 | switch (target) { | |
1011 | case MEM_CGROUP_TARGET_THRESH: | |
1012 | next = val + THRESHOLDS_EVENTS_TARGET; | |
1013 | break; | |
bb4cc1a8 AM |
1014 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
1015 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
1016 | break; | |
f53d7ce3 JW |
1017 | case MEM_CGROUP_TARGET_NUMAINFO: |
1018 | next = val + NUMAINFO_EVENTS_TARGET; | |
1019 | break; | |
1020 | default: | |
1021 | break; | |
1022 | } | |
1023 | __this_cpu_write(memcg->stat->targets[target], next); | |
1024 | return true; | |
7a159cc9 | 1025 | } |
f53d7ce3 | 1026 | return false; |
d2265e6f KH |
1027 | } |
1028 | ||
1029 | /* | |
1030 | * Check events in order. | |
1031 | * | |
1032 | */ | |
c0ff4b85 | 1033 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 1034 | { |
4799401f | 1035 | preempt_disable(); |
d2265e6f | 1036 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1037 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1038 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1039 | bool do_softlimit; |
82b3f2a7 | 1040 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1041 | |
bb4cc1a8 AM |
1042 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1043 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1044 | #if MAX_NUMNODES > 1 |
1045 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1046 | MEM_CGROUP_TARGET_NUMAINFO); | |
1047 | #endif | |
1048 | preempt_enable(); | |
1049 | ||
c0ff4b85 | 1050 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1051 | if (unlikely(do_softlimit)) |
1052 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1053 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1054 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1055 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1056 | #endif |
f53d7ce3 JW |
1057 | } else |
1058 | preempt_enable(); | |
d2265e6f KH |
1059 | } |
1060 | ||
cf475ad2 | 1061 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1062 | { |
31a78f23 BS |
1063 | /* |
1064 | * mm_update_next_owner() may clear mm->owner to NULL | |
1065 | * if it races with swapoff, page migration, etc. | |
1066 | * So this can be called with p == NULL. | |
1067 | */ | |
1068 | if (unlikely(!p)) | |
1069 | return NULL; | |
1070 | ||
073219e9 | 1071 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1072 | } |
1073 | ||
a433658c | 1074 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1075 | { |
c0ff4b85 | 1076 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
1077 | |
1078 | if (!mm) | |
1079 | return NULL; | |
54595fe2 KH |
1080 | /* |
1081 | * Because we have no locks, mm->owner's may be being moved to other | |
1082 | * cgroup. We use css_tryget() here even if this looks | |
1083 | * pessimistic (rather than adding locks here). | |
1084 | */ | |
1085 | rcu_read_lock(); | |
1086 | do { | |
c0ff4b85 R |
1087 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1088 | if (unlikely(!memcg)) | |
54595fe2 | 1089 | break; |
c0ff4b85 | 1090 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1091 | rcu_read_unlock(); |
c0ff4b85 | 1092 | return memcg; |
54595fe2 KH |
1093 | } |
1094 | ||
16248d8f MH |
1095 | /* |
1096 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1097 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1098 | * | |
1099 | * helper function to be used by mem_cgroup_iter | |
1100 | */ | |
1101 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
694fbc0f | 1102 | struct mem_cgroup *last_visited) |
16248d8f | 1103 | { |
492eb21b | 1104 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 1105 | |
bd8815a6 | 1106 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 1107 | skip_node: |
492eb21b | 1108 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
1109 | |
1110 | /* | |
1111 | * Even if we found a group we have to make sure it is | |
1112 | * alive. css && !memcg means that the groups should be | |
1113 | * skipped and we should continue the tree walk. | |
1114 | * last_visited css is safe to use because it is | |
1115 | * protected by css_get and the tree walk is rcu safe. | |
0eef6156 MH |
1116 | * |
1117 | * We do not take a reference on the root of the tree walk | |
1118 | * because we might race with the root removal when it would | |
1119 | * be the only node in the iterated hierarchy and mem_cgroup_iter | |
1120 | * would end up in an endless loop because it expects that at | |
1121 | * least one valid node will be returned. Root cannot disappear | |
1122 | * because caller of the iterator should hold it already so | |
1123 | * skipping css reference should be safe. | |
16248d8f | 1124 | */ |
492eb21b | 1125 | if (next_css) { |
ce48225f HD |
1126 | if ((next_css == &root->css) || |
1127 | ((next_css->flags & CSS_ONLINE) && css_tryget(next_css))) | |
d8ad3055 | 1128 | return mem_cgroup_from_css(next_css); |
0eef6156 MH |
1129 | |
1130 | prev_css = next_css; | |
1131 | goto skip_node; | |
16248d8f MH |
1132 | } |
1133 | ||
1134 | return NULL; | |
1135 | } | |
1136 | ||
519ebea3 JW |
1137 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1138 | { | |
1139 | /* | |
1140 | * When a group in the hierarchy below root is destroyed, the | |
1141 | * hierarchy iterator can no longer be trusted since it might | |
1142 | * have pointed to the destroyed group. Invalidate it. | |
1143 | */ | |
1144 | atomic_inc(&root->dead_count); | |
1145 | } | |
1146 | ||
1147 | static struct mem_cgroup * | |
1148 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1149 | struct mem_cgroup *root, | |
1150 | int *sequence) | |
1151 | { | |
1152 | struct mem_cgroup *position = NULL; | |
1153 | /* | |
1154 | * A cgroup destruction happens in two stages: offlining and | |
1155 | * release. They are separated by a RCU grace period. | |
1156 | * | |
1157 | * If the iterator is valid, we may still race with an | |
1158 | * offlining. The RCU lock ensures the object won't be | |
1159 | * released, tryget will fail if we lost the race. | |
1160 | */ | |
1161 | *sequence = atomic_read(&root->dead_count); | |
1162 | if (iter->last_dead_count == *sequence) { | |
1163 | smp_rmb(); | |
1164 | position = iter->last_visited; | |
ecc736fc MH |
1165 | |
1166 | /* | |
1167 | * We cannot take a reference to root because we might race | |
1168 | * with root removal and returning NULL would end up in | |
1169 | * an endless loop on the iterator user level when root | |
1170 | * would be returned all the time. | |
1171 | */ | |
1172 | if (position && position != root && | |
1173 | !css_tryget(&position->css)) | |
519ebea3 JW |
1174 | position = NULL; |
1175 | } | |
1176 | return position; | |
1177 | } | |
1178 | ||
1179 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1180 | struct mem_cgroup *last_visited, | |
1181 | struct mem_cgroup *new_position, | |
ecc736fc | 1182 | struct mem_cgroup *root, |
519ebea3 JW |
1183 | int sequence) |
1184 | { | |
ecc736fc MH |
1185 | /* root reference counting symmetric to mem_cgroup_iter_load */ |
1186 | if (last_visited && last_visited != root) | |
519ebea3 JW |
1187 | css_put(&last_visited->css); |
1188 | /* | |
1189 | * We store the sequence count from the time @last_visited was | |
1190 | * loaded successfully instead of rereading it here so that we | |
1191 | * don't lose destruction events in between. We could have | |
1192 | * raced with the destruction of @new_position after all. | |
1193 | */ | |
1194 | iter->last_visited = new_position; | |
1195 | smp_wmb(); | |
1196 | iter->last_dead_count = sequence; | |
1197 | } | |
1198 | ||
5660048c JW |
1199 | /** |
1200 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1201 | * @root: hierarchy root | |
1202 | * @prev: previously returned memcg, NULL on first invocation | |
1203 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1204 | * | |
1205 | * Returns references to children of the hierarchy below @root, or | |
1206 | * @root itself, or %NULL after a full round-trip. | |
1207 | * | |
1208 | * Caller must pass the return value in @prev on subsequent | |
1209 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1210 | * to cancel a hierarchy walk before the round-trip is complete. | |
1211 | * | |
1212 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1213 | * divide up the memcgs in the hierarchy among all concurrent | |
1214 | * reclaimers operating on the same zone and priority. | |
1215 | */ | |
694fbc0f | 1216 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1217 | struct mem_cgroup *prev, |
694fbc0f | 1218 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1219 | { |
9f3a0d09 | 1220 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1221 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1222 | |
694fbc0f AM |
1223 | if (mem_cgroup_disabled()) |
1224 | return NULL; | |
5660048c | 1225 | |
9f3a0d09 JW |
1226 | if (!root) |
1227 | root = root_mem_cgroup; | |
7d74b06f | 1228 | |
9f3a0d09 | 1229 | if (prev && !reclaim) |
542f85f9 | 1230 | last_visited = prev; |
14067bb3 | 1231 | |
9f3a0d09 JW |
1232 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1233 | if (prev) | |
c40046f3 | 1234 | goto out_css_put; |
694fbc0f | 1235 | return root; |
9f3a0d09 | 1236 | } |
14067bb3 | 1237 | |
542f85f9 | 1238 | rcu_read_lock(); |
9f3a0d09 | 1239 | while (!memcg) { |
527a5ec9 | 1240 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1241 | int uninitialized_var(seq); |
711d3d2c | 1242 | |
527a5ec9 JW |
1243 | if (reclaim) { |
1244 | int nid = zone_to_nid(reclaim->zone); | |
1245 | int zid = zone_idx(reclaim->zone); | |
1246 | struct mem_cgroup_per_zone *mz; | |
1247 | ||
1248 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1249 | iter = &mz->reclaim_iter[reclaim->priority]; | |
542f85f9 | 1250 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1251 | iter->last_visited = NULL; |
542f85f9 MH |
1252 | goto out_unlock; |
1253 | } | |
5f578161 | 1254 | |
519ebea3 | 1255 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1256 | } |
7d74b06f | 1257 | |
694fbc0f | 1258 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1259 | |
527a5ec9 | 1260 | if (reclaim) { |
ecc736fc MH |
1261 | mem_cgroup_iter_update(iter, last_visited, memcg, root, |
1262 | seq); | |
542f85f9 | 1263 | |
19f39402 | 1264 | if (!memcg) |
527a5ec9 JW |
1265 | iter->generation++; |
1266 | else if (!prev && memcg) | |
1267 | reclaim->generation = iter->generation; | |
1268 | } | |
9f3a0d09 | 1269 | |
694fbc0f | 1270 | if (prev && !memcg) |
542f85f9 | 1271 | goto out_unlock; |
9f3a0d09 | 1272 | } |
542f85f9 MH |
1273 | out_unlock: |
1274 | rcu_read_unlock(); | |
c40046f3 MH |
1275 | out_css_put: |
1276 | if (prev && prev != root) | |
1277 | css_put(&prev->css); | |
1278 | ||
9f3a0d09 | 1279 | return memcg; |
14067bb3 | 1280 | } |
7d74b06f | 1281 | |
5660048c JW |
1282 | /** |
1283 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1284 | * @root: hierarchy root | |
1285 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1286 | */ | |
1287 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1288 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1289 | { |
1290 | if (!root) | |
1291 | root = root_mem_cgroup; | |
1292 | if (prev && prev != root) | |
1293 | css_put(&prev->css); | |
1294 | } | |
7d74b06f | 1295 | |
9f3a0d09 JW |
1296 | /* |
1297 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1298 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1299 | * be used for reference counting. | |
1300 | */ | |
1301 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1302 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1303 | iter != NULL; \ |
527a5ec9 | 1304 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1305 | |
9f3a0d09 | 1306 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1307 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1308 | iter != NULL; \ |
527a5ec9 | 1309 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1310 | |
68ae564b | 1311 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1312 | { |
c0ff4b85 | 1313 | struct mem_cgroup *memcg; |
456f998e | 1314 | |
456f998e | 1315 | rcu_read_lock(); |
c0ff4b85 R |
1316 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1317 | if (unlikely(!memcg)) | |
456f998e YH |
1318 | goto out; |
1319 | ||
1320 | switch (idx) { | |
456f998e | 1321 | case PGFAULT: |
0e574a93 JW |
1322 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1323 | break; | |
1324 | case PGMAJFAULT: | |
1325 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1326 | break; |
1327 | default: | |
1328 | BUG(); | |
1329 | } | |
1330 | out: | |
1331 | rcu_read_unlock(); | |
1332 | } | |
68ae564b | 1333 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1334 | |
925b7673 JW |
1335 | /** |
1336 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1337 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1338 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1339 | * |
1340 | * Returns the lru list vector holding pages for the given @zone and | |
1341 | * @mem. This can be the global zone lruvec, if the memory controller | |
1342 | * is disabled. | |
1343 | */ | |
1344 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1345 | struct mem_cgroup *memcg) | |
1346 | { | |
1347 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1348 | struct lruvec *lruvec; |
925b7673 | 1349 | |
bea8c150 HD |
1350 | if (mem_cgroup_disabled()) { |
1351 | lruvec = &zone->lruvec; | |
1352 | goto out; | |
1353 | } | |
925b7673 JW |
1354 | |
1355 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1356 | lruvec = &mz->lruvec; |
1357 | out: | |
1358 | /* | |
1359 | * Since a node can be onlined after the mem_cgroup was created, | |
1360 | * we have to be prepared to initialize lruvec->zone here; | |
1361 | * and if offlined then reonlined, we need to reinitialize it. | |
1362 | */ | |
1363 | if (unlikely(lruvec->zone != zone)) | |
1364 | lruvec->zone = zone; | |
1365 | return lruvec; | |
925b7673 JW |
1366 | } |
1367 | ||
08e552c6 KH |
1368 | /* |
1369 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1370 | * Operations are called by routine of global LRU independently from memcg. | |
1371 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1372 | * | |
1373 | * Changes to pc->mem_cgroup happens when | |
1374 | * 1. charge | |
1375 | * 2. moving account | |
1376 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1377 | * It is added to LRU before charge. | |
1378 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1379 | * When moving account, the page is not on LRU. It's isolated. | |
1380 | */ | |
4f98a2fe | 1381 | |
925b7673 | 1382 | /** |
fa9add64 | 1383 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1384 | * @page: the page |
fa9add64 | 1385 | * @zone: zone of the page |
925b7673 | 1386 | */ |
fa9add64 | 1387 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1388 | { |
08e552c6 | 1389 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1390 | struct mem_cgroup *memcg; |
1391 | struct page_cgroup *pc; | |
bea8c150 | 1392 | struct lruvec *lruvec; |
6d12e2d8 | 1393 | |
bea8c150 HD |
1394 | if (mem_cgroup_disabled()) { |
1395 | lruvec = &zone->lruvec; | |
1396 | goto out; | |
1397 | } | |
925b7673 | 1398 | |
08e552c6 | 1399 | pc = lookup_page_cgroup(page); |
38c5d72f | 1400 | memcg = pc->mem_cgroup; |
7512102c HD |
1401 | |
1402 | /* | |
fa9add64 | 1403 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1404 | * an uncharged page off lru does nothing to secure |
1405 | * its former mem_cgroup from sudden removal. | |
1406 | * | |
1407 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1408 | * under page_cgroup lock: between them, they make all uses | |
1409 | * of pc->mem_cgroup safe. | |
1410 | */ | |
fa9add64 | 1411 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1412 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1413 | ||
925b7673 | 1414 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1415 | lruvec = &mz->lruvec; |
1416 | out: | |
1417 | /* | |
1418 | * Since a node can be onlined after the mem_cgroup was created, | |
1419 | * we have to be prepared to initialize lruvec->zone here; | |
1420 | * and if offlined then reonlined, we need to reinitialize it. | |
1421 | */ | |
1422 | if (unlikely(lruvec->zone != zone)) | |
1423 | lruvec->zone = zone; | |
1424 | return lruvec; | |
08e552c6 | 1425 | } |
b69408e8 | 1426 | |
925b7673 | 1427 | /** |
fa9add64 HD |
1428 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1429 | * @lruvec: mem_cgroup per zone lru vector | |
1430 | * @lru: index of lru list the page is sitting on | |
1431 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1432 | * |
fa9add64 HD |
1433 | * This function must be called when a page is added to or removed from an |
1434 | * lru list. | |
3f58a829 | 1435 | */ |
fa9add64 HD |
1436 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1437 | int nr_pages) | |
3f58a829 MK |
1438 | { |
1439 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1440 | unsigned long *lru_size; |
3f58a829 MK |
1441 | |
1442 | if (mem_cgroup_disabled()) | |
1443 | return; | |
1444 | ||
fa9add64 HD |
1445 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1446 | lru_size = mz->lru_size + lru; | |
1447 | *lru_size += nr_pages; | |
1448 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1449 | } |
544122e5 | 1450 | |
3e92041d | 1451 | /* |
c0ff4b85 | 1452 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1453 | * hierarchy subtree |
1454 | */ | |
c3ac9a8a JW |
1455 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1456 | struct mem_cgroup *memcg) | |
3e92041d | 1457 | { |
91c63734 JW |
1458 | if (root_memcg == memcg) |
1459 | return true; | |
3a981f48 | 1460 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1461 | return false; |
b47f77b5 | 1462 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1463 | } |
1464 | ||
1465 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1466 | struct mem_cgroup *memcg) | |
1467 | { | |
1468 | bool ret; | |
1469 | ||
91c63734 | 1470 | rcu_read_lock(); |
c3ac9a8a | 1471 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1472 | rcu_read_unlock(); |
1473 | return ret; | |
3e92041d MH |
1474 | } |
1475 | ||
ffbdccf5 DR |
1476 | bool task_in_mem_cgroup(struct task_struct *task, |
1477 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1478 | { |
0b7f569e | 1479 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1480 | struct task_struct *p; |
ffbdccf5 | 1481 | bool ret; |
4c4a2214 | 1482 | |
158e0a2d | 1483 | p = find_lock_task_mm(task); |
de077d22 DR |
1484 | if (p) { |
1485 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1486 | task_unlock(p); | |
1487 | } else { | |
1488 | /* | |
1489 | * All threads may have already detached their mm's, but the oom | |
1490 | * killer still needs to detect if they have already been oom | |
1491 | * killed to prevent needlessly killing additional tasks. | |
1492 | */ | |
ffbdccf5 | 1493 | rcu_read_lock(); |
de077d22 DR |
1494 | curr = mem_cgroup_from_task(task); |
1495 | if (curr) | |
1496 | css_get(&curr->css); | |
ffbdccf5 | 1497 | rcu_read_unlock(); |
de077d22 | 1498 | } |
0b7f569e | 1499 | if (!curr) |
ffbdccf5 | 1500 | return false; |
d31f56db | 1501 | /* |
c0ff4b85 | 1502 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1503 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1504 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1505 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1506 | */ |
c0ff4b85 | 1507 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1508 | css_put(&curr->css); |
4c4a2214 DR |
1509 | return ret; |
1510 | } | |
1511 | ||
c56d5c7d | 1512 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1513 | { |
9b272977 | 1514 | unsigned long inactive_ratio; |
14797e23 | 1515 | unsigned long inactive; |
9b272977 | 1516 | unsigned long active; |
c772be93 | 1517 | unsigned long gb; |
14797e23 | 1518 | |
4d7dcca2 HD |
1519 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1520 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1521 | |
c772be93 KM |
1522 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1523 | if (gb) | |
1524 | inactive_ratio = int_sqrt(10 * gb); | |
1525 | else | |
1526 | inactive_ratio = 1; | |
1527 | ||
9b272977 | 1528 | return inactive * inactive_ratio < active; |
14797e23 KM |
1529 | } |
1530 | ||
6d61ef40 BS |
1531 | #define mem_cgroup_from_res_counter(counter, member) \ |
1532 | container_of(counter, struct mem_cgroup, member) | |
1533 | ||
19942822 | 1534 | /** |
9d11ea9f | 1535 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1536 | * @memcg: the memory cgroup |
19942822 | 1537 | * |
9d11ea9f | 1538 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1539 | * pages. |
19942822 | 1540 | */ |
c0ff4b85 | 1541 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1542 | { |
9d11ea9f JW |
1543 | unsigned long long margin; |
1544 | ||
c0ff4b85 | 1545 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1546 | if (do_swap_account) |
c0ff4b85 | 1547 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1548 | return margin >> PAGE_SHIFT; |
19942822 JW |
1549 | } |
1550 | ||
1f4c025b | 1551 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1552 | { |
a7885eb8 | 1553 | /* root ? */ |
63876986 | 1554 | if (!css_parent(&memcg->css)) |
a7885eb8 KM |
1555 | return vm_swappiness; |
1556 | ||
bf1ff263 | 1557 | return memcg->swappiness; |
a7885eb8 KM |
1558 | } |
1559 | ||
619d094b KH |
1560 | /* |
1561 | * memcg->moving_account is used for checking possibility that some thread is | |
1562 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1563 | * a memcg, other threads should check memcg->moving_account under | |
1564 | * rcu_read_lock(), like this: | |
1565 | * | |
1566 | * CPU-A CPU-B | |
1567 | * rcu_read_lock() | |
1568 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1569 | * take heavy locks. | |
1570 | * synchronize_rcu() update something. | |
1571 | * rcu_read_unlock() | |
1572 | * start move here. | |
1573 | */ | |
4331f7d3 KH |
1574 | |
1575 | /* for quick checking without looking up memcg */ | |
1576 | atomic_t memcg_moving __read_mostly; | |
1577 | ||
c0ff4b85 | 1578 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1579 | { |
4331f7d3 | 1580 | atomic_inc(&memcg_moving); |
619d094b | 1581 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1582 | synchronize_rcu(); |
1583 | } | |
1584 | ||
c0ff4b85 | 1585 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1586 | { |
619d094b KH |
1587 | /* |
1588 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1589 | * We check NULL in callee rather than caller. | |
1590 | */ | |
4331f7d3 KH |
1591 | if (memcg) { |
1592 | atomic_dec(&memcg_moving); | |
619d094b | 1593 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1594 | } |
32047e2a | 1595 | } |
619d094b | 1596 | |
32047e2a KH |
1597 | /* |
1598 | * 2 routines for checking "mem" is under move_account() or not. | |
1599 | * | |
13fd1dd9 AM |
1600 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1601 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1602 | * pc->mem_cgroup may be overwritten. |
1603 | * | |
1604 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1605 | * under hierarchy of moving cgroups. This is for | |
1606 | * waiting at hith-memory prressure caused by "move". | |
1607 | */ | |
1608 | ||
13fd1dd9 | 1609 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1610 | { |
1611 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1612 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1613 | } |
4b534334 | 1614 | |
c0ff4b85 | 1615 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1616 | { |
2bd9bb20 KH |
1617 | struct mem_cgroup *from; |
1618 | struct mem_cgroup *to; | |
4b534334 | 1619 | bool ret = false; |
2bd9bb20 KH |
1620 | /* |
1621 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1622 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1623 | */ | |
1624 | spin_lock(&mc.lock); | |
1625 | from = mc.from; | |
1626 | to = mc.to; | |
1627 | if (!from) | |
1628 | goto unlock; | |
3e92041d | 1629 | |
c0ff4b85 R |
1630 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1631 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1632 | unlock: |
1633 | spin_unlock(&mc.lock); | |
4b534334 KH |
1634 | return ret; |
1635 | } | |
1636 | ||
c0ff4b85 | 1637 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1638 | { |
1639 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1640 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1641 | DEFINE_WAIT(wait); |
1642 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1643 | /* moving charge context might have finished. */ | |
1644 | if (mc.moving_task) | |
1645 | schedule(); | |
1646 | finish_wait(&mc.waitq, &wait); | |
1647 | return true; | |
1648 | } | |
1649 | } | |
1650 | return false; | |
1651 | } | |
1652 | ||
312734c0 KH |
1653 | /* |
1654 | * Take this lock when | |
1655 | * - a code tries to modify page's memcg while it's USED. | |
1656 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1657 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1658 | */ |
1659 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1660 | unsigned long *flags) | |
1661 | { | |
1662 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1663 | } | |
1664 | ||
1665 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1666 | unsigned long *flags) | |
1667 | { | |
1668 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1669 | } | |
1670 | ||
58cf188e | 1671 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1672 | /** |
58cf188e | 1673 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1674 | * @memcg: The memory cgroup that went over limit |
1675 | * @p: Task that is going to be killed | |
1676 | * | |
1677 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1678 | * enabled | |
1679 | */ | |
1680 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1681 | { | |
e61734c5 | 1682 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1683 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1684 | struct mem_cgroup *iter; |
1685 | unsigned int i; | |
e222432b | 1686 | |
58cf188e | 1687 | if (!p) |
e222432b BS |
1688 | return; |
1689 | ||
08088cb9 | 1690 | mutex_lock(&oom_info_lock); |
e222432b BS |
1691 | rcu_read_lock(); |
1692 | ||
e61734c5 TH |
1693 | pr_info("Task in "); |
1694 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1695 | pr_info(" killed as a result of limit of "); | |
1696 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1697 | pr_info("\n"); | |
e222432b | 1698 | |
e222432b BS |
1699 | rcu_read_unlock(); |
1700 | ||
d045197f | 1701 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1702 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1703 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1704 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1705 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1706 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1707 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1708 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1709 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1710 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1711 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1712 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1713 | |
1714 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1715 | pr_info("Memory cgroup stats for "); |
1716 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1717 | pr_cont(":"); |
1718 | ||
1719 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1720 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1721 | continue; | |
1722 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1723 | K(mem_cgroup_read_stat(iter, i))); | |
1724 | } | |
1725 | ||
1726 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1727 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1728 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1729 | ||
1730 | pr_cont("\n"); | |
1731 | } | |
08088cb9 | 1732 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1733 | } |
1734 | ||
81d39c20 KH |
1735 | /* |
1736 | * This function returns the number of memcg under hierarchy tree. Returns | |
1737 | * 1(self count) if no children. | |
1738 | */ | |
c0ff4b85 | 1739 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1740 | { |
1741 | int num = 0; | |
7d74b06f KH |
1742 | struct mem_cgroup *iter; |
1743 | ||
c0ff4b85 | 1744 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1745 | num++; |
81d39c20 KH |
1746 | return num; |
1747 | } | |
1748 | ||
a63d83f4 DR |
1749 | /* |
1750 | * Return the memory (and swap, if configured) limit for a memcg. | |
1751 | */ | |
9cbb78bb | 1752 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1753 | { |
1754 | u64 limit; | |
a63d83f4 | 1755 | |
f3e8eb70 | 1756 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1757 | |
a63d83f4 | 1758 | /* |
9a5a8f19 | 1759 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1760 | */ |
9a5a8f19 MH |
1761 | if (mem_cgroup_swappiness(memcg)) { |
1762 | u64 memsw; | |
1763 | ||
1764 | limit += total_swap_pages << PAGE_SHIFT; | |
1765 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1766 | ||
1767 | /* | |
1768 | * If memsw is finite and limits the amount of swap space | |
1769 | * available to this memcg, return that limit. | |
1770 | */ | |
1771 | limit = min(limit, memsw); | |
1772 | } | |
1773 | ||
1774 | return limit; | |
a63d83f4 DR |
1775 | } |
1776 | ||
19965460 DR |
1777 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1778 | int order) | |
9cbb78bb DR |
1779 | { |
1780 | struct mem_cgroup *iter; | |
1781 | unsigned long chosen_points = 0; | |
1782 | unsigned long totalpages; | |
1783 | unsigned int points = 0; | |
1784 | struct task_struct *chosen = NULL; | |
1785 | ||
876aafbf | 1786 | /* |
465adcf1 DR |
1787 | * If current has a pending SIGKILL or is exiting, then automatically |
1788 | * select it. The goal is to allow it to allocate so that it may | |
1789 | * quickly exit and free its memory. | |
876aafbf | 1790 | */ |
465adcf1 | 1791 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1792 | set_thread_flag(TIF_MEMDIE); |
1793 | return; | |
1794 | } | |
1795 | ||
1796 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1797 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1798 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1799 | struct css_task_iter it; |
9cbb78bb DR |
1800 | struct task_struct *task; |
1801 | ||
72ec7029 TH |
1802 | css_task_iter_start(&iter->css, &it); |
1803 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1804 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1805 | false)) { | |
1806 | case OOM_SCAN_SELECT: | |
1807 | if (chosen) | |
1808 | put_task_struct(chosen); | |
1809 | chosen = task; | |
1810 | chosen_points = ULONG_MAX; | |
1811 | get_task_struct(chosen); | |
1812 | /* fall through */ | |
1813 | case OOM_SCAN_CONTINUE: | |
1814 | continue; | |
1815 | case OOM_SCAN_ABORT: | |
72ec7029 | 1816 | css_task_iter_end(&it); |
9cbb78bb DR |
1817 | mem_cgroup_iter_break(memcg, iter); |
1818 | if (chosen) | |
1819 | put_task_struct(chosen); | |
1820 | return; | |
1821 | case OOM_SCAN_OK: | |
1822 | break; | |
1823 | }; | |
1824 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1825 | if (!points || points < chosen_points) |
1826 | continue; | |
1827 | /* Prefer thread group leaders for display purposes */ | |
1828 | if (points == chosen_points && | |
1829 | thread_group_leader(chosen)) | |
1830 | continue; | |
1831 | ||
1832 | if (chosen) | |
1833 | put_task_struct(chosen); | |
1834 | chosen = task; | |
1835 | chosen_points = points; | |
1836 | get_task_struct(chosen); | |
9cbb78bb | 1837 | } |
72ec7029 | 1838 | css_task_iter_end(&it); |
9cbb78bb DR |
1839 | } |
1840 | ||
1841 | if (!chosen) | |
1842 | return; | |
1843 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1844 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1845 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1846 | } |
1847 | ||
5660048c JW |
1848 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1849 | gfp_t gfp_mask, | |
1850 | unsigned long flags) | |
1851 | { | |
1852 | unsigned long total = 0; | |
1853 | bool noswap = false; | |
1854 | int loop; | |
1855 | ||
1856 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1857 | noswap = true; | |
1858 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1859 | noswap = true; | |
1860 | ||
1861 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1862 | if (loop) | |
1863 | drain_all_stock_async(memcg); | |
1864 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1865 | /* | |
1866 | * Allow limit shrinkers, which are triggered directly | |
1867 | * by userspace, to catch signals and stop reclaim | |
1868 | * after minimal progress, regardless of the margin. | |
1869 | */ | |
1870 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1871 | break; | |
1872 | if (mem_cgroup_margin(memcg)) | |
1873 | break; | |
1874 | /* | |
1875 | * If nothing was reclaimed after two attempts, there | |
1876 | * may be no reclaimable pages in this hierarchy. | |
1877 | */ | |
1878 | if (loop && !total) | |
1879 | break; | |
1880 | } | |
1881 | return total; | |
1882 | } | |
1883 | ||
4d0c066d KH |
1884 | /** |
1885 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1886 | * @memcg: the target memcg |
4d0c066d KH |
1887 | * @nid: the node ID to be checked. |
1888 | * @noswap : specify true here if the user wants flle only information. | |
1889 | * | |
1890 | * This function returns whether the specified memcg contains any | |
1891 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1892 | * pages in the node. | |
1893 | */ | |
c0ff4b85 | 1894 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1895 | int nid, bool noswap) |
1896 | { | |
c0ff4b85 | 1897 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1898 | return true; |
1899 | if (noswap || !total_swap_pages) | |
1900 | return false; | |
c0ff4b85 | 1901 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1902 | return true; |
1903 | return false; | |
1904 | ||
1905 | } | |
bb4cc1a8 | 1906 | #if MAX_NUMNODES > 1 |
889976db YH |
1907 | |
1908 | /* | |
1909 | * Always updating the nodemask is not very good - even if we have an empty | |
1910 | * list or the wrong list here, we can start from some node and traverse all | |
1911 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1912 | * | |
1913 | */ | |
c0ff4b85 | 1914 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1915 | { |
1916 | int nid; | |
453a9bf3 KH |
1917 | /* |
1918 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1919 | * pagein/pageout changes since the last update. | |
1920 | */ | |
c0ff4b85 | 1921 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1922 | return; |
c0ff4b85 | 1923 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1924 | return; |
1925 | ||
889976db | 1926 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1927 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1928 | |
31aaea4a | 1929 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1930 | |
c0ff4b85 R |
1931 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1932 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1933 | } |
453a9bf3 | 1934 | |
c0ff4b85 R |
1935 | atomic_set(&memcg->numainfo_events, 0); |
1936 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1937 | } |
1938 | ||
1939 | /* | |
1940 | * Selecting a node where we start reclaim from. Because what we need is just | |
1941 | * reducing usage counter, start from anywhere is O,K. Considering | |
1942 | * memory reclaim from current node, there are pros. and cons. | |
1943 | * | |
1944 | * Freeing memory from current node means freeing memory from a node which | |
1945 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1946 | * hit limits, it will see a contention on a node. But freeing from remote | |
1947 | * node means more costs for memory reclaim because of memory latency. | |
1948 | * | |
1949 | * Now, we use round-robin. Better algorithm is welcomed. | |
1950 | */ | |
c0ff4b85 | 1951 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1952 | { |
1953 | int node; | |
1954 | ||
c0ff4b85 R |
1955 | mem_cgroup_may_update_nodemask(memcg); |
1956 | node = memcg->last_scanned_node; | |
889976db | 1957 | |
c0ff4b85 | 1958 | node = next_node(node, memcg->scan_nodes); |
889976db | 1959 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1960 | node = first_node(memcg->scan_nodes); |
889976db YH |
1961 | /* |
1962 | * We call this when we hit limit, not when pages are added to LRU. | |
1963 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1964 | * memcg is too small and all pages are not on LRU. In that case, | |
1965 | * we use curret node. | |
1966 | */ | |
1967 | if (unlikely(node == MAX_NUMNODES)) | |
1968 | node = numa_node_id(); | |
1969 | ||
c0ff4b85 | 1970 | memcg->last_scanned_node = node; |
889976db YH |
1971 | return node; |
1972 | } | |
1973 | ||
bb4cc1a8 AM |
1974 | /* |
1975 | * Check all nodes whether it contains reclaimable pages or not. | |
1976 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1977 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1978 | * enough new information. We need to do double check. | |
1979 | */ | |
1980 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1981 | { | |
1982 | int nid; | |
1983 | ||
1984 | /* | |
1985 | * quick check...making use of scan_node. | |
1986 | * We can skip unused nodes. | |
1987 | */ | |
1988 | if (!nodes_empty(memcg->scan_nodes)) { | |
1989 | for (nid = first_node(memcg->scan_nodes); | |
1990 | nid < MAX_NUMNODES; | |
1991 | nid = next_node(nid, memcg->scan_nodes)) { | |
1992 | ||
1993 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1994 | return true; | |
1995 | } | |
1996 | } | |
1997 | /* | |
1998 | * Check rest of nodes. | |
1999 | */ | |
2000 | for_each_node_state(nid, N_MEMORY) { | |
2001 | if (node_isset(nid, memcg->scan_nodes)) | |
2002 | continue; | |
2003 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
2004 | return true; | |
2005 | } | |
2006 | return false; | |
2007 | } | |
2008 | ||
889976db | 2009 | #else |
c0ff4b85 | 2010 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
2011 | { |
2012 | return 0; | |
2013 | } | |
4d0c066d | 2014 | |
bb4cc1a8 AM |
2015 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
2016 | { | |
2017 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
2018 | } | |
889976db YH |
2019 | #endif |
2020 | ||
0608f43d AM |
2021 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
2022 | struct zone *zone, | |
2023 | gfp_t gfp_mask, | |
2024 | unsigned long *total_scanned) | |
2025 | { | |
2026 | struct mem_cgroup *victim = NULL; | |
2027 | int total = 0; | |
2028 | int loop = 0; | |
2029 | unsigned long excess; | |
2030 | unsigned long nr_scanned; | |
2031 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2032 | .zone = zone, | |
2033 | .priority = 0, | |
2034 | }; | |
2035 | ||
2036 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; | |
2037 | ||
2038 | while (1) { | |
2039 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
2040 | if (!victim) { | |
2041 | loop++; | |
2042 | if (loop >= 2) { | |
2043 | /* | |
2044 | * If we have not been able to reclaim | |
2045 | * anything, it might because there are | |
2046 | * no reclaimable pages under this hierarchy | |
2047 | */ | |
2048 | if (!total) | |
2049 | break; | |
2050 | /* | |
2051 | * We want to do more targeted reclaim. | |
2052 | * excess >> 2 is not to excessive so as to | |
2053 | * reclaim too much, nor too less that we keep | |
2054 | * coming back to reclaim from this cgroup | |
2055 | */ | |
2056 | if (total >= (excess >> 2) || | |
2057 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
2058 | break; | |
2059 | } | |
2060 | continue; | |
2061 | } | |
2062 | if (!mem_cgroup_reclaimable(victim, false)) | |
2063 | continue; | |
2064 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
2065 | zone, &nr_scanned); | |
2066 | *total_scanned += nr_scanned; | |
2067 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
2068 | break; | |
6d61ef40 | 2069 | } |
0608f43d AM |
2070 | mem_cgroup_iter_break(root_memcg, victim); |
2071 | return total; | |
6d61ef40 BS |
2072 | } |
2073 | ||
0056f4e6 JW |
2074 | #ifdef CONFIG_LOCKDEP |
2075 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
2076 | .name = "memcg_oom_lock", | |
2077 | }; | |
2078 | #endif | |
2079 | ||
fb2a6fc5 JW |
2080 | static DEFINE_SPINLOCK(memcg_oom_lock); |
2081 | ||
867578cb KH |
2082 | /* |
2083 | * Check OOM-Killer is already running under our hierarchy. | |
2084 | * If someone is running, return false. | |
2085 | */ | |
fb2a6fc5 | 2086 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2087 | { |
79dfdacc | 2088 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2089 | |
fb2a6fc5 JW |
2090 | spin_lock(&memcg_oom_lock); |
2091 | ||
9f3a0d09 | 2092 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2093 | if (iter->oom_lock) { |
79dfdacc MH |
2094 | /* |
2095 | * this subtree of our hierarchy is already locked | |
2096 | * so we cannot give a lock. | |
2097 | */ | |
79dfdacc | 2098 | failed = iter; |
9f3a0d09 JW |
2099 | mem_cgroup_iter_break(memcg, iter); |
2100 | break; | |
23751be0 JW |
2101 | } else |
2102 | iter->oom_lock = true; | |
7d74b06f | 2103 | } |
867578cb | 2104 | |
fb2a6fc5 JW |
2105 | if (failed) { |
2106 | /* | |
2107 | * OK, we failed to lock the whole subtree so we have | |
2108 | * to clean up what we set up to the failing subtree | |
2109 | */ | |
2110 | for_each_mem_cgroup_tree(iter, memcg) { | |
2111 | if (iter == failed) { | |
2112 | mem_cgroup_iter_break(memcg, iter); | |
2113 | break; | |
2114 | } | |
2115 | iter->oom_lock = false; | |
79dfdacc | 2116 | } |
0056f4e6 JW |
2117 | } else |
2118 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2119 | |
2120 | spin_unlock(&memcg_oom_lock); | |
2121 | ||
2122 | return !failed; | |
a636b327 | 2123 | } |
0b7f569e | 2124 | |
fb2a6fc5 | 2125 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2126 | { |
7d74b06f KH |
2127 | struct mem_cgroup *iter; |
2128 | ||
fb2a6fc5 | 2129 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 2130 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 2131 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2132 | iter->oom_lock = false; |
fb2a6fc5 | 2133 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2134 | } |
2135 | ||
c0ff4b85 | 2136 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2137 | { |
2138 | struct mem_cgroup *iter; | |
2139 | ||
c0ff4b85 | 2140 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2141 | atomic_inc(&iter->under_oom); |
2142 | } | |
2143 | ||
c0ff4b85 | 2144 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2145 | { |
2146 | struct mem_cgroup *iter; | |
2147 | ||
867578cb KH |
2148 | /* |
2149 | * When a new child is created while the hierarchy is under oom, | |
2150 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2151 | * atomic_add_unless() here. | |
2152 | */ | |
c0ff4b85 | 2153 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2154 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2155 | } |
2156 | ||
867578cb KH |
2157 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2158 | ||
dc98df5a | 2159 | struct oom_wait_info { |
d79154bb | 2160 | struct mem_cgroup *memcg; |
dc98df5a KH |
2161 | wait_queue_t wait; |
2162 | }; | |
2163 | ||
2164 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2165 | unsigned mode, int sync, void *arg) | |
2166 | { | |
d79154bb HD |
2167 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2168 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2169 | struct oom_wait_info *oom_wait_info; |
2170 | ||
2171 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2172 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2173 | |
dc98df5a | 2174 | /* |
d79154bb | 2175 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2176 | * Then we can use css_is_ancestor without taking care of RCU. |
2177 | */ | |
c0ff4b85 R |
2178 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2179 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2180 | return 0; |
dc98df5a KH |
2181 | return autoremove_wake_function(wait, mode, sync, arg); |
2182 | } | |
2183 | ||
c0ff4b85 | 2184 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2185 | { |
3812c8c8 | 2186 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2187 | /* for filtering, pass "memcg" as argument. */ |
2188 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2189 | } |
2190 | ||
c0ff4b85 | 2191 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2192 | { |
c0ff4b85 R |
2193 | if (memcg && atomic_read(&memcg->under_oom)) |
2194 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2195 | } |
2196 | ||
3812c8c8 | 2197 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2198 | { |
3812c8c8 JW |
2199 | if (!current->memcg_oom.may_oom) |
2200 | return; | |
867578cb | 2201 | /* |
49426420 JW |
2202 | * We are in the middle of the charge context here, so we |
2203 | * don't want to block when potentially sitting on a callstack | |
2204 | * that holds all kinds of filesystem and mm locks. | |
2205 | * | |
2206 | * Also, the caller may handle a failed allocation gracefully | |
2207 | * (like optional page cache readahead) and so an OOM killer | |
2208 | * invocation might not even be necessary. | |
2209 | * | |
2210 | * That's why we don't do anything here except remember the | |
2211 | * OOM context and then deal with it at the end of the page | |
2212 | * fault when the stack is unwound, the locks are released, | |
2213 | * and when we know whether the fault was overall successful. | |
867578cb | 2214 | */ |
49426420 JW |
2215 | css_get(&memcg->css); |
2216 | current->memcg_oom.memcg = memcg; | |
2217 | current->memcg_oom.gfp_mask = mask; | |
2218 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2219 | } |
2220 | ||
2221 | /** | |
2222 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2223 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2224 | * |
49426420 JW |
2225 | * This has to be called at the end of a page fault if the memcg OOM |
2226 | * handler was enabled. | |
3812c8c8 | 2227 | * |
49426420 | 2228 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2229 | * sleep on a waitqueue until the userspace task resolves the |
2230 | * situation. Sleeping directly in the charge context with all kinds | |
2231 | * of locks held is not a good idea, instead we remember an OOM state | |
2232 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2233 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2234 | * |
2235 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2236 | * completed, %false otherwise. |
3812c8c8 | 2237 | */ |
49426420 | 2238 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2239 | { |
49426420 | 2240 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2241 | struct oom_wait_info owait; |
49426420 | 2242 | bool locked; |
3812c8c8 JW |
2243 | |
2244 | /* OOM is global, do not handle */ | |
3812c8c8 | 2245 | if (!memcg) |
49426420 | 2246 | return false; |
3812c8c8 | 2247 | |
49426420 JW |
2248 | if (!handle) |
2249 | goto cleanup; | |
3812c8c8 JW |
2250 | |
2251 | owait.memcg = memcg; | |
2252 | owait.wait.flags = 0; | |
2253 | owait.wait.func = memcg_oom_wake_function; | |
2254 | owait.wait.private = current; | |
2255 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2256 | |
3812c8c8 | 2257 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2258 | mem_cgroup_mark_under_oom(memcg); |
2259 | ||
2260 | locked = mem_cgroup_oom_trylock(memcg); | |
2261 | ||
2262 | if (locked) | |
2263 | mem_cgroup_oom_notify(memcg); | |
2264 | ||
2265 | if (locked && !memcg->oom_kill_disable) { | |
2266 | mem_cgroup_unmark_under_oom(memcg); | |
2267 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2268 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2269 | current->memcg_oom.order); | |
2270 | } else { | |
3812c8c8 | 2271 | schedule(); |
49426420 JW |
2272 | mem_cgroup_unmark_under_oom(memcg); |
2273 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2274 | } | |
2275 | ||
2276 | if (locked) { | |
fb2a6fc5 JW |
2277 | mem_cgroup_oom_unlock(memcg); |
2278 | /* | |
2279 | * There is no guarantee that an OOM-lock contender | |
2280 | * sees the wakeups triggered by the OOM kill | |
2281 | * uncharges. Wake any sleepers explicitely. | |
2282 | */ | |
2283 | memcg_oom_recover(memcg); | |
2284 | } | |
49426420 JW |
2285 | cleanup: |
2286 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2287 | css_put(&memcg->css); |
867578cb | 2288 | return true; |
0b7f569e KH |
2289 | } |
2290 | ||
d69b042f BS |
2291 | /* |
2292 | * Currently used to update mapped file statistics, but the routine can be | |
2293 | * generalized to update other statistics as well. | |
32047e2a KH |
2294 | * |
2295 | * Notes: Race condition | |
2296 | * | |
2297 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2298 | * it tends to be costly. But considering some conditions, we doesn't need | |
2299 | * to do so _always_. | |
2300 | * | |
2301 | * Considering "charge", lock_page_cgroup() is not required because all | |
2302 | * file-stat operations happen after a page is attached to radix-tree. There | |
2303 | * are no race with "charge". | |
2304 | * | |
2305 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2306 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2307 | * if there are race with "uncharge". Statistics itself is properly handled | |
2308 | * by flags. | |
2309 | * | |
2310 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2311 | * small, we check mm->moving_account and detect there are possibility of race |
2312 | * If there is, we take a lock. | |
d69b042f | 2313 | */ |
26174efd | 2314 | |
89c06bd5 KH |
2315 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2316 | bool *locked, unsigned long *flags) | |
2317 | { | |
2318 | struct mem_cgroup *memcg; | |
2319 | struct page_cgroup *pc; | |
2320 | ||
2321 | pc = lookup_page_cgroup(page); | |
2322 | again: | |
2323 | memcg = pc->mem_cgroup; | |
2324 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2325 | return; | |
2326 | /* | |
2327 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2328 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2329 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2330 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2331 | */ |
13fd1dd9 | 2332 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2333 | return; |
2334 | ||
2335 | move_lock_mem_cgroup(memcg, flags); | |
2336 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2337 | move_unlock_mem_cgroup(memcg, flags); | |
2338 | goto again; | |
2339 | } | |
2340 | *locked = true; | |
2341 | } | |
2342 | ||
2343 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2344 | { | |
2345 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2346 | ||
2347 | /* | |
2348 | * It's guaranteed that pc->mem_cgroup never changes while | |
2349 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2350 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2351 | */ |
2352 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2353 | } | |
2354 | ||
2a7106f2 | 2355 | void mem_cgroup_update_page_stat(struct page *page, |
68b4876d | 2356 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2357 | { |
c0ff4b85 | 2358 | struct mem_cgroup *memcg; |
32047e2a | 2359 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2360 | unsigned long uninitialized_var(flags); |
d69b042f | 2361 | |
cfa44946 | 2362 | if (mem_cgroup_disabled()) |
d69b042f | 2363 | return; |
89c06bd5 | 2364 | |
658b72c5 | 2365 | VM_BUG_ON(!rcu_read_lock_held()); |
c0ff4b85 R |
2366 | memcg = pc->mem_cgroup; |
2367 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2368 | return; |
26174efd | 2369 | |
c0ff4b85 | 2370 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2371 | } |
26174efd | 2372 | |
cdec2e42 KH |
2373 | /* |
2374 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2375 | * TODO: maybe necessary to use big numbers in big irons. | |
2376 | */ | |
7ec99d62 | 2377 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2378 | struct memcg_stock_pcp { |
2379 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2380 | unsigned int nr_pages; |
cdec2e42 | 2381 | struct work_struct work; |
26fe6168 | 2382 | unsigned long flags; |
a0db00fc | 2383 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2384 | }; |
2385 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2386 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2387 | |
a0956d54 SS |
2388 | /** |
2389 | * consume_stock: Try to consume stocked charge on this cpu. | |
2390 | * @memcg: memcg to consume from. | |
2391 | * @nr_pages: how many pages to charge. | |
2392 | * | |
2393 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2394 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2395 | * service an allocation will refill the stock. | |
2396 | * | |
2397 | * returns true if successful, false otherwise. | |
cdec2e42 | 2398 | */ |
a0956d54 | 2399 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2400 | { |
2401 | struct memcg_stock_pcp *stock; | |
2402 | bool ret = true; | |
2403 | ||
a0956d54 SS |
2404 | if (nr_pages > CHARGE_BATCH) |
2405 | return false; | |
2406 | ||
cdec2e42 | 2407 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2408 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2409 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2410 | else /* need to call res_counter_charge */ |
2411 | ret = false; | |
2412 | put_cpu_var(memcg_stock); | |
2413 | return ret; | |
2414 | } | |
2415 | ||
2416 | /* | |
2417 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2418 | */ | |
2419 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2420 | { | |
2421 | struct mem_cgroup *old = stock->cached; | |
2422 | ||
11c9ea4e JW |
2423 | if (stock->nr_pages) { |
2424 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2425 | ||
2426 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2427 | if (do_swap_account) |
11c9ea4e JW |
2428 | res_counter_uncharge(&old->memsw, bytes); |
2429 | stock->nr_pages = 0; | |
cdec2e42 KH |
2430 | } |
2431 | stock->cached = NULL; | |
cdec2e42 KH |
2432 | } |
2433 | ||
2434 | /* | |
2435 | * This must be called under preempt disabled or must be called by | |
2436 | * a thread which is pinned to local cpu. | |
2437 | */ | |
2438 | static void drain_local_stock(struct work_struct *dummy) | |
2439 | { | |
2440 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2441 | drain_stock(stock); | |
26fe6168 | 2442 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2443 | } |
2444 | ||
e4777496 MH |
2445 | static void __init memcg_stock_init(void) |
2446 | { | |
2447 | int cpu; | |
2448 | ||
2449 | for_each_possible_cpu(cpu) { | |
2450 | struct memcg_stock_pcp *stock = | |
2451 | &per_cpu(memcg_stock, cpu); | |
2452 | INIT_WORK(&stock->work, drain_local_stock); | |
2453 | } | |
2454 | } | |
2455 | ||
cdec2e42 KH |
2456 | /* |
2457 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2458 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2459 | */ |
c0ff4b85 | 2460 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2461 | { |
2462 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2463 | ||
c0ff4b85 | 2464 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2465 | drain_stock(stock); |
c0ff4b85 | 2466 | stock->cached = memcg; |
cdec2e42 | 2467 | } |
11c9ea4e | 2468 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2469 | put_cpu_var(memcg_stock); |
2470 | } | |
2471 | ||
2472 | /* | |
c0ff4b85 | 2473 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2474 | * of the hierarchy under it. sync flag says whether we should block |
2475 | * until the work is done. | |
cdec2e42 | 2476 | */ |
c0ff4b85 | 2477 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2478 | { |
26fe6168 | 2479 | int cpu, curcpu; |
d38144b7 | 2480 | |
cdec2e42 | 2481 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2482 | get_online_cpus(); |
5af12d0e | 2483 | curcpu = get_cpu(); |
cdec2e42 KH |
2484 | for_each_online_cpu(cpu) { |
2485 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2486 | struct mem_cgroup *memcg; |
26fe6168 | 2487 | |
c0ff4b85 R |
2488 | memcg = stock->cached; |
2489 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2490 | continue; |
c0ff4b85 | 2491 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2492 | continue; |
d1a05b69 MH |
2493 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2494 | if (cpu == curcpu) | |
2495 | drain_local_stock(&stock->work); | |
2496 | else | |
2497 | schedule_work_on(cpu, &stock->work); | |
2498 | } | |
cdec2e42 | 2499 | } |
5af12d0e | 2500 | put_cpu(); |
d38144b7 MH |
2501 | |
2502 | if (!sync) | |
2503 | goto out; | |
2504 | ||
2505 | for_each_online_cpu(cpu) { | |
2506 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2507 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2508 | flush_work(&stock->work); |
2509 | } | |
2510 | out: | |
f894ffa8 | 2511 | put_online_cpus(); |
d38144b7 MH |
2512 | } |
2513 | ||
2514 | /* | |
2515 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2516 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2517 | * expects some charges will be back to res_counter later but cannot wait for | |
2518 | * it. | |
2519 | */ | |
c0ff4b85 | 2520 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2521 | { |
9f50fad6 MH |
2522 | /* |
2523 | * If someone calls draining, avoid adding more kworker runs. | |
2524 | */ | |
2525 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2526 | return; | |
c0ff4b85 | 2527 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2528 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2529 | } |
2530 | ||
2531 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2532 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2533 | { |
2534 | /* called when force_empty is called */ | |
9f50fad6 | 2535 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2536 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2537 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2538 | } |
2539 | ||
711d3d2c KH |
2540 | /* |
2541 | * This function drains percpu counter value from DEAD cpu and | |
2542 | * move it to local cpu. Note that this function can be preempted. | |
2543 | */ | |
c0ff4b85 | 2544 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2545 | { |
2546 | int i; | |
2547 | ||
c0ff4b85 | 2548 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2549 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2550 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2551 | |
c0ff4b85 R |
2552 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2553 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2554 | } |
e9f8974f | 2555 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2556 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2557 | |
c0ff4b85 R |
2558 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2559 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2560 | } |
c0ff4b85 | 2561 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2562 | } |
2563 | ||
0db0628d | 2564 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2565 | unsigned long action, |
2566 | void *hcpu) | |
2567 | { | |
2568 | int cpu = (unsigned long)hcpu; | |
2569 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2570 | struct mem_cgroup *iter; |
cdec2e42 | 2571 | |
619d094b | 2572 | if (action == CPU_ONLINE) |
1489ebad | 2573 | return NOTIFY_OK; |
1489ebad | 2574 | |
d833049b | 2575 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2576 | return NOTIFY_OK; |
711d3d2c | 2577 | |
9f3a0d09 | 2578 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2579 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2580 | ||
cdec2e42 KH |
2581 | stock = &per_cpu(memcg_stock, cpu); |
2582 | drain_stock(stock); | |
2583 | return NOTIFY_OK; | |
2584 | } | |
2585 | ||
4b534334 KH |
2586 | |
2587 | /* See __mem_cgroup_try_charge() for details */ | |
2588 | enum { | |
2589 | CHARGE_OK, /* success */ | |
2590 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2591 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2592 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
4b534334 KH |
2593 | }; |
2594 | ||
c0ff4b85 | 2595 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 | 2596 | unsigned int nr_pages, unsigned int min_pages, |
3812c8c8 | 2597 | bool invoke_oom) |
4b534334 | 2598 | { |
7ec99d62 | 2599 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2600 | struct mem_cgroup *mem_over_limit; |
2601 | struct res_counter *fail_res; | |
2602 | unsigned long flags = 0; | |
2603 | int ret; | |
2604 | ||
c0ff4b85 | 2605 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2606 | |
2607 | if (likely(!ret)) { | |
2608 | if (!do_swap_account) | |
2609 | return CHARGE_OK; | |
c0ff4b85 | 2610 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2611 | if (likely(!ret)) |
2612 | return CHARGE_OK; | |
2613 | ||
c0ff4b85 | 2614 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2615 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2616 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2617 | } else | |
2618 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2619 | /* |
9221edb7 JW |
2620 | * Never reclaim on behalf of optional batching, retry with a |
2621 | * single page instead. | |
2622 | */ | |
4c9c5359 | 2623 | if (nr_pages > min_pages) |
4b534334 KH |
2624 | return CHARGE_RETRY; |
2625 | ||
2626 | if (!(gfp_mask & __GFP_WAIT)) | |
2627 | return CHARGE_WOULDBLOCK; | |
2628 | ||
4c9c5359 SS |
2629 | if (gfp_mask & __GFP_NORETRY) |
2630 | return CHARGE_NOMEM; | |
2631 | ||
5660048c | 2632 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2633 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2634 | return CHARGE_RETRY; |
4b534334 | 2635 | /* |
19942822 JW |
2636 | * Even though the limit is exceeded at this point, reclaim |
2637 | * may have been able to free some pages. Retry the charge | |
2638 | * before killing the task. | |
2639 | * | |
2640 | * Only for regular pages, though: huge pages are rather | |
2641 | * unlikely to succeed so close to the limit, and we fall back | |
2642 | * to regular pages anyway in case of failure. | |
4b534334 | 2643 | */ |
4c9c5359 | 2644 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2645 | return CHARGE_RETRY; |
2646 | ||
2647 | /* | |
2648 | * At task move, charge accounts can be doubly counted. So, it's | |
2649 | * better to wait until the end of task_move if something is going on. | |
2650 | */ | |
2651 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2652 | return CHARGE_RETRY; | |
2653 | ||
3812c8c8 JW |
2654 | if (invoke_oom) |
2655 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize)); | |
4b534334 | 2656 | |
3812c8c8 | 2657 | return CHARGE_NOMEM; |
4b534334 KH |
2658 | } |
2659 | ||
f817ed48 | 2660 | /* |
38c5d72f KH |
2661 | * __mem_cgroup_try_charge() does |
2662 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2663 | * 2. update res_counter | |
2664 | * 3. call memory reclaim if necessary. | |
2665 | * | |
2666 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2667 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2668 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2669 | * as possible without any hazards. 2: all pages should have a valid | |
2670 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2671 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2672 | * | |
2673 | * So __mem_cgroup_try_charge() will return | |
2674 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2675 | * -ENOMEM ... charge failure because of resource limits. | |
2676 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2677 | * | |
2678 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2679 | * the oom-killer can be invoked. | |
8a9f3ccd | 2680 | */ |
f817ed48 | 2681 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2682 | gfp_t gfp_mask, |
7ec99d62 | 2683 | unsigned int nr_pages, |
c0ff4b85 | 2684 | struct mem_cgroup **ptr, |
7ec99d62 | 2685 | bool oom) |
8a9f3ccd | 2686 | { |
7ec99d62 | 2687 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2688 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2689 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2690 | int ret; |
a636b327 | 2691 | |
867578cb KH |
2692 | /* |
2693 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2694 | * in system level. So, allow to go ahead dying process in addition to | |
2695 | * MEMDIE process. | |
2696 | */ | |
2697 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2698 | || fatal_signal_pending(current))) | |
2699 | goto bypass; | |
a636b327 | 2700 | |
49426420 | 2701 | if (unlikely(task_in_memcg_oom(current))) |
1f14c1ac | 2702 | goto nomem; |
49426420 | 2703 | |
a0d8b00a JW |
2704 | if (gfp_mask & __GFP_NOFAIL) |
2705 | oom = false; | |
2706 | ||
8a9f3ccd | 2707 | /* |
3be91277 HD |
2708 | * We always charge the cgroup the mm_struct belongs to. |
2709 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2710 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2711 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2712 | */ |
c0ff4b85 | 2713 | if (!*ptr && !mm) |
38c5d72f | 2714 | *ptr = root_mem_cgroup; |
f75ca962 | 2715 | again: |
c0ff4b85 R |
2716 | if (*ptr) { /* css should be a valid one */ |
2717 | memcg = *ptr; | |
c0ff4b85 | 2718 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2719 | goto done; |
a0956d54 | 2720 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2721 | goto done; |
c0ff4b85 | 2722 | css_get(&memcg->css); |
4b534334 | 2723 | } else { |
f75ca962 | 2724 | struct task_struct *p; |
54595fe2 | 2725 | |
f75ca962 KH |
2726 | rcu_read_lock(); |
2727 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2728 | /* |
ebb76ce1 | 2729 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2730 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2731 | * race with swapoff. Then, we have small risk of mis-accouning. |
2732 | * But such kind of mis-account by race always happens because | |
2733 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2734 | * small race, here. | |
2735 | * (*) swapoff at el will charge against mm-struct not against | |
2736 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2737 | */ |
c0ff4b85 | 2738 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2739 | if (!memcg) |
2740 | memcg = root_mem_cgroup; | |
2741 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2742 | rcu_read_unlock(); |
2743 | goto done; | |
2744 | } | |
a0956d54 | 2745 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2746 | /* |
2747 | * It seems dagerous to access memcg without css_get(). | |
2748 | * But considering how consume_stok works, it's not | |
2749 | * necessary. If consume_stock success, some charges | |
2750 | * from this memcg are cached on this cpu. So, we | |
2751 | * don't need to call css_get()/css_tryget() before | |
2752 | * calling consume_stock(). | |
2753 | */ | |
2754 | rcu_read_unlock(); | |
2755 | goto done; | |
2756 | } | |
2757 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2758 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2759 | rcu_read_unlock(); |
2760 | goto again; | |
2761 | } | |
2762 | rcu_read_unlock(); | |
2763 | } | |
8a9f3ccd | 2764 | |
4b534334 | 2765 | do { |
3812c8c8 | 2766 | bool invoke_oom = oom && !nr_oom_retries; |
7a81b88c | 2767 | |
4b534334 | 2768 | /* If killed, bypass charge */ |
f75ca962 | 2769 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2770 | css_put(&memcg->css); |
4b534334 | 2771 | goto bypass; |
f75ca962 | 2772 | } |
6d61ef40 | 2773 | |
3812c8c8 JW |
2774 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, |
2775 | nr_pages, invoke_oom); | |
4b534334 KH |
2776 | switch (ret) { |
2777 | case CHARGE_OK: | |
2778 | break; | |
2779 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2780 | batch = nr_pages; |
c0ff4b85 R |
2781 | css_put(&memcg->css); |
2782 | memcg = NULL; | |
f75ca962 | 2783 | goto again; |
4b534334 | 2784 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2785 | css_put(&memcg->css); |
4b534334 KH |
2786 | goto nomem; |
2787 | case CHARGE_NOMEM: /* OOM routine works */ | |
3812c8c8 | 2788 | if (!oom || invoke_oom) { |
c0ff4b85 | 2789 | css_put(&memcg->css); |
867578cb | 2790 | goto nomem; |
f75ca962 | 2791 | } |
4b534334 KH |
2792 | nr_oom_retries--; |
2793 | break; | |
66e1707b | 2794 | } |
4b534334 KH |
2795 | } while (ret != CHARGE_OK); |
2796 | ||
7ec99d62 | 2797 | if (batch > nr_pages) |
c0ff4b85 R |
2798 | refill_stock(memcg, batch - nr_pages); |
2799 | css_put(&memcg->css); | |
0c3e73e8 | 2800 | done: |
c0ff4b85 | 2801 | *ptr = memcg; |
7a81b88c KH |
2802 | return 0; |
2803 | nomem: | |
3168ecbe JW |
2804 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2805 | *ptr = NULL; | |
2806 | return -ENOMEM; | |
2807 | } | |
867578cb | 2808 | bypass: |
38c5d72f KH |
2809 | *ptr = root_mem_cgroup; |
2810 | return -EINTR; | |
7a81b88c | 2811 | } |
8a9f3ccd | 2812 | |
a3032a2c DN |
2813 | /* |
2814 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2815 | * This function is for that and do uncharge, put css's refcnt. | |
2816 | * gotten by try_charge(). | |
2817 | */ | |
c0ff4b85 | 2818 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2819 | unsigned int nr_pages) |
a3032a2c | 2820 | { |
c0ff4b85 | 2821 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2822 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2823 | ||
c0ff4b85 | 2824 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2825 | if (do_swap_account) |
c0ff4b85 | 2826 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2827 | } |
854ffa8d DN |
2828 | } |
2829 | ||
d01dd17f KH |
2830 | /* |
2831 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2832 | * This is useful when moving usage to parent cgroup. | |
2833 | */ | |
2834 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2835 | unsigned int nr_pages) | |
2836 | { | |
2837 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2838 | ||
2839 | if (mem_cgroup_is_root(memcg)) | |
2840 | return; | |
2841 | ||
2842 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2843 | if (do_swap_account) | |
2844 | res_counter_uncharge_until(&memcg->memsw, | |
2845 | memcg->memsw.parent, bytes); | |
2846 | } | |
2847 | ||
a3b2d692 KH |
2848 | /* |
2849 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2850 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2851 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2852 | * called against removed memcg.) | |
a3b2d692 KH |
2853 | */ |
2854 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2855 | { | |
a3b2d692 KH |
2856 | /* ID 0 is unused ID */ |
2857 | if (!id) | |
2858 | return NULL; | |
34c00c31 | 2859 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2860 | } |
2861 | ||
e42d9d5d | 2862 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2863 | { |
c0ff4b85 | 2864 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2865 | struct page_cgroup *pc; |
a3b2d692 | 2866 | unsigned short id; |
b5a84319 KH |
2867 | swp_entry_t ent; |
2868 | ||
309381fe | 2869 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2870 | |
3c776e64 | 2871 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2872 | lock_page_cgroup(pc); |
a3b2d692 | 2873 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2874 | memcg = pc->mem_cgroup; |
2875 | if (memcg && !css_tryget(&memcg->css)) | |
2876 | memcg = NULL; | |
e42d9d5d | 2877 | } else if (PageSwapCache(page)) { |
3c776e64 | 2878 | ent.val = page_private(page); |
9fb4b7cc | 2879 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2880 | rcu_read_lock(); |
c0ff4b85 R |
2881 | memcg = mem_cgroup_lookup(id); |
2882 | if (memcg && !css_tryget(&memcg->css)) | |
2883 | memcg = NULL; | |
a3b2d692 | 2884 | rcu_read_unlock(); |
3c776e64 | 2885 | } |
c0bd3f63 | 2886 | unlock_page_cgroup(pc); |
c0ff4b85 | 2887 | return memcg; |
b5a84319 KH |
2888 | } |
2889 | ||
c0ff4b85 | 2890 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2891 | struct page *page, |
7ec99d62 | 2892 | unsigned int nr_pages, |
9ce70c02 HD |
2893 | enum charge_type ctype, |
2894 | bool lrucare) | |
7a81b88c | 2895 | { |
ce587e65 | 2896 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2897 | struct zone *uninitialized_var(zone); |
fa9add64 | 2898 | struct lruvec *lruvec; |
9ce70c02 | 2899 | bool was_on_lru = false; |
b2402857 | 2900 | bool anon; |
9ce70c02 | 2901 | |
ca3e0214 | 2902 | lock_page_cgroup(pc); |
309381fe | 2903 | VM_BUG_ON_PAGE(PageCgroupUsed(pc), page); |
ca3e0214 KH |
2904 | /* |
2905 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2906 | * accessed by any other context at this point. | |
2907 | */ | |
9ce70c02 HD |
2908 | |
2909 | /* | |
2910 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2911 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2912 | */ | |
2913 | if (lrucare) { | |
2914 | zone = page_zone(page); | |
2915 | spin_lock_irq(&zone->lru_lock); | |
2916 | if (PageLRU(page)) { | |
fa9add64 | 2917 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2918 | ClearPageLRU(page); |
fa9add64 | 2919 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2920 | was_on_lru = true; |
2921 | } | |
2922 | } | |
2923 | ||
c0ff4b85 | 2924 | pc->mem_cgroup = memcg; |
261fb61a KH |
2925 | /* |
2926 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2927 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2928 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2929 | * before USED bit, we need memory barrier here. | |
2930 | * See mem_cgroup_add_lru_list(), etc. | |
f894ffa8 | 2931 | */ |
08e552c6 | 2932 | smp_wmb(); |
b2402857 | 2933 | SetPageCgroupUsed(pc); |
3be91277 | 2934 | |
9ce70c02 HD |
2935 | if (lrucare) { |
2936 | if (was_on_lru) { | |
fa9add64 | 2937 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
309381fe | 2938 | VM_BUG_ON_PAGE(PageLRU(page), page); |
9ce70c02 | 2939 | SetPageLRU(page); |
fa9add64 | 2940 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2941 | } |
2942 | spin_unlock_irq(&zone->lru_lock); | |
2943 | } | |
2944 | ||
41326c17 | 2945 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2946 | anon = true; |
2947 | else | |
2948 | anon = false; | |
2949 | ||
b070e65c | 2950 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
52d4b9ac | 2951 | unlock_page_cgroup(pc); |
9ce70c02 | 2952 | |
430e4863 | 2953 | /* |
bb4cc1a8 AM |
2954 | * "charge_statistics" updated event counter. Then, check it. |
2955 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2956 | * if they exceeds softlimit. | |
430e4863 | 2957 | */ |
c0ff4b85 | 2958 | memcg_check_events(memcg, page); |
7a81b88c | 2959 | } |
66e1707b | 2960 | |
7cf27982 GC |
2961 | static DEFINE_MUTEX(set_limit_mutex); |
2962 | ||
7ae1e1d0 | 2963 | #ifdef CONFIG_MEMCG_KMEM |
d6441637 VD |
2964 | static DEFINE_MUTEX(activate_kmem_mutex); |
2965 | ||
7ae1e1d0 GC |
2966 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) |
2967 | { | |
2968 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
6de64beb | 2969 | memcg_kmem_is_active(memcg); |
7ae1e1d0 GC |
2970 | } |
2971 | ||
1f458cbf GC |
2972 | /* |
2973 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2974 | * in the memcg_cache_params struct. | |
2975 | */ | |
2976 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2977 | { | |
2978 | struct kmem_cache *cachep; | |
2979 | ||
2980 | VM_BUG_ON(p->is_root_cache); | |
2981 | cachep = p->root_cache; | |
7a67d7ab | 2982 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2983 | } |
2984 | ||
749c5415 | 2985 | #ifdef CONFIG_SLABINFO |
2da8ca82 | 2986 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
749c5415 | 2987 | { |
2da8ca82 | 2988 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
749c5415 GC |
2989 | struct memcg_cache_params *params; |
2990 | ||
2991 | if (!memcg_can_account_kmem(memcg)) | |
2992 | return -EIO; | |
2993 | ||
2994 | print_slabinfo_header(m); | |
2995 | ||
2996 | mutex_lock(&memcg->slab_caches_mutex); | |
2997 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
2998 | cache_show(memcg_params_to_cache(params), m); | |
2999 | mutex_unlock(&memcg->slab_caches_mutex); | |
3000 | ||
3001 | return 0; | |
3002 | } | |
3003 | #endif | |
3004 | ||
7ae1e1d0 GC |
3005 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
3006 | { | |
3007 | struct res_counter *fail_res; | |
3008 | struct mem_cgroup *_memcg; | |
3009 | int ret = 0; | |
7ae1e1d0 GC |
3010 | |
3011 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
3012 | if (ret) | |
3013 | return ret; | |
3014 | ||
7ae1e1d0 GC |
3015 | _memcg = memcg; |
3016 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
b9921ecd | 3017 | &_memcg, oom_gfp_allowed(gfp)); |
7ae1e1d0 GC |
3018 | |
3019 | if (ret == -EINTR) { | |
3020 | /* | |
3021 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
3022 | * OOM kill or fatal signal. Since our only options are to | |
3023 | * either fail the allocation or charge it to this cgroup, do | |
3024 | * it as a temporary condition. But we can't fail. From a | |
3025 | * kmem/slab perspective, the cache has already been selected, | |
3026 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
3027 | * our minds. | |
3028 | * | |
3029 | * This condition will only trigger if the task entered | |
3030 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
3031 | * __mem_cgroup_try_charge() above. Tasks that were already | |
3032 | * dying when the allocation triggers should have been already | |
3033 | * directed to the root cgroup in memcontrol.h | |
3034 | */ | |
3035 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
3036 | if (do_swap_account) | |
3037 | res_counter_charge_nofail(&memcg->memsw, size, | |
3038 | &fail_res); | |
3039 | ret = 0; | |
3040 | } else if (ret) | |
3041 | res_counter_uncharge(&memcg->kmem, size); | |
3042 | ||
3043 | return ret; | |
3044 | } | |
3045 | ||
3046 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
3047 | { | |
7ae1e1d0 GC |
3048 | res_counter_uncharge(&memcg->res, size); |
3049 | if (do_swap_account) | |
3050 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
3051 | |
3052 | /* Not down to 0 */ | |
3053 | if (res_counter_uncharge(&memcg->kmem, size)) | |
3054 | return; | |
3055 | ||
10d5ebf4 LZ |
3056 | /* |
3057 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
3058 | * this last uncharge is racing with the offlining code or it is | |
3059 | * outliving the memcg existence. | |
3060 | * | |
3061 | * The memory barrier imposed by test&clear is paired with the | |
3062 | * explicit one in memcg_kmem_mark_dead(). | |
3063 | */ | |
7de37682 | 3064 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 3065 | css_put(&memcg->css); |
7ae1e1d0 GC |
3066 | } |
3067 | ||
2633d7a0 GC |
3068 | /* |
3069 | * helper for acessing a memcg's index. It will be used as an index in the | |
3070 | * child cache array in kmem_cache, and also to derive its name. This function | |
3071 | * will return -1 when this is not a kmem-limited memcg. | |
3072 | */ | |
3073 | int memcg_cache_id(struct mem_cgroup *memcg) | |
3074 | { | |
3075 | return memcg ? memcg->kmemcg_id : -1; | |
3076 | } | |
3077 | ||
55007d84 GC |
3078 | static size_t memcg_caches_array_size(int num_groups) |
3079 | { | |
3080 | ssize_t size; | |
3081 | if (num_groups <= 0) | |
3082 | return 0; | |
3083 | ||
3084 | size = 2 * num_groups; | |
3085 | if (size < MEMCG_CACHES_MIN_SIZE) | |
3086 | size = MEMCG_CACHES_MIN_SIZE; | |
3087 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3088 | size = MEMCG_CACHES_MAX_SIZE; | |
3089 | ||
3090 | return size; | |
3091 | } | |
3092 | ||
3093 | /* | |
3094 | * We should update the current array size iff all caches updates succeed. This | |
3095 | * can only be done from the slab side. The slab mutex needs to be held when | |
3096 | * calling this. | |
3097 | */ | |
3098 | void memcg_update_array_size(int num) | |
3099 | { | |
3100 | if (num > memcg_limited_groups_array_size) | |
3101 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
3102 | } | |
3103 | ||
15cf17d2 KK |
3104 | static void kmem_cache_destroy_work_func(struct work_struct *w); |
3105 | ||
55007d84 GC |
3106 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
3107 | { | |
3108 | struct memcg_cache_params *cur_params = s->memcg_params; | |
3109 | ||
f35c3a8e | 3110 | VM_BUG_ON(!is_root_cache(s)); |
55007d84 GC |
3111 | |
3112 | if (num_groups > memcg_limited_groups_array_size) { | |
3113 | int i; | |
f8570263 | 3114 | struct memcg_cache_params *new_params; |
55007d84 GC |
3115 | ssize_t size = memcg_caches_array_size(num_groups); |
3116 | ||
3117 | size *= sizeof(void *); | |
90c7a79c | 3118 | size += offsetof(struct memcg_cache_params, memcg_caches); |
55007d84 | 3119 | |
f8570263 VD |
3120 | new_params = kzalloc(size, GFP_KERNEL); |
3121 | if (!new_params) | |
55007d84 | 3122 | return -ENOMEM; |
55007d84 | 3123 | |
f8570263 | 3124 | new_params->is_root_cache = true; |
55007d84 GC |
3125 | |
3126 | /* | |
3127 | * There is the chance it will be bigger than | |
3128 | * memcg_limited_groups_array_size, if we failed an allocation | |
3129 | * in a cache, in which case all caches updated before it, will | |
3130 | * have a bigger array. | |
3131 | * | |
3132 | * But if that is the case, the data after | |
3133 | * memcg_limited_groups_array_size is certainly unused | |
3134 | */ | |
3135 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3136 | if (!cur_params->memcg_caches[i]) | |
3137 | continue; | |
f8570263 | 3138 | new_params->memcg_caches[i] = |
55007d84 GC |
3139 | cur_params->memcg_caches[i]; |
3140 | } | |
3141 | ||
3142 | /* | |
3143 | * Ideally, we would wait until all caches succeed, and only | |
3144 | * then free the old one. But this is not worth the extra | |
3145 | * pointer per-cache we'd have to have for this. | |
3146 | * | |
3147 | * It is not a big deal if some caches are left with a size | |
3148 | * bigger than the others. And all updates will reset this | |
3149 | * anyway. | |
3150 | */ | |
f8570263 VD |
3151 | rcu_assign_pointer(s->memcg_params, new_params); |
3152 | if (cur_params) | |
3153 | kfree_rcu(cur_params, rcu_head); | |
55007d84 GC |
3154 | } |
3155 | return 0; | |
3156 | } | |
3157 | ||
363a044f VD |
3158 | int memcg_alloc_cache_params(struct mem_cgroup *memcg, struct kmem_cache *s, |
3159 | struct kmem_cache *root_cache) | |
2633d7a0 | 3160 | { |
90c7a79c | 3161 | size_t size; |
2633d7a0 GC |
3162 | |
3163 | if (!memcg_kmem_enabled()) | |
3164 | return 0; | |
3165 | ||
90c7a79c AV |
3166 | if (!memcg) { |
3167 | size = offsetof(struct memcg_cache_params, memcg_caches); | |
55007d84 | 3168 | size += memcg_limited_groups_array_size * sizeof(void *); |
90c7a79c AV |
3169 | } else |
3170 | size = sizeof(struct memcg_cache_params); | |
55007d84 | 3171 | |
2633d7a0 GC |
3172 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3173 | if (!s->memcg_params) | |
3174 | return -ENOMEM; | |
3175 | ||
943a451a | 3176 | if (memcg) { |
2633d7a0 | 3177 | s->memcg_params->memcg = memcg; |
943a451a | 3178 | s->memcg_params->root_cache = root_cache; |
3e6b11df AV |
3179 | INIT_WORK(&s->memcg_params->destroy, |
3180 | kmem_cache_destroy_work_func); | |
4ba902b5 GC |
3181 | } else |
3182 | s->memcg_params->is_root_cache = true; | |
3183 | ||
2633d7a0 GC |
3184 | return 0; |
3185 | } | |
3186 | ||
363a044f VD |
3187 | void memcg_free_cache_params(struct kmem_cache *s) |
3188 | { | |
3189 | kfree(s->memcg_params); | |
3190 | } | |
3191 | ||
1aa13254 | 3192 | void memcg_register_cache(struct kmem_cache *s) |
2633d7a0 | 3193 | { |
d7f25f8a GC |
3194 | struct kmem_cache *root; |
3195 | struct mem_cgroup *memcg; | |
3196 | int id; | |
3197 | ||
1aa13254 VD |
3198 | if (is_root_cache(s)) |
3199 | return; | |
3200 | ||
2edefe11 VD |
3201 | /* |
3202 | * Holding the slab_mutex assures nobody will touch the memcg_caches | |
3203 | * array while we are modifying it. | |
3204 | */ | |
3205 | lockdep_assert_held(&slab_mutex); | |
3206 | ||
1aa13254 VD |
3207 | root = s->memcg_params->root_cache; |
3208 | memcg = s->memcg_params->memcg; | |
3209 | id = memcg_cache_id(memcg); | |
3210 | ||
3211 | css_get(&memcg->css); | |
3212 | ||
1aa13254 | 3213 | |
d7f25f8a | 3214 | /* |
959c8963 VD |
3215 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
3216 | * barrier here to ensure nobody will see the kmem_cache partially | |
3217 | * initialized. | |
d7f25f8a | 3218 | */ |
959c8963 VD |
3219 | smp_wmb(); |
3220 | ||
96403da2 VD |
3221 | /* |
3222 | * Initialize the pointer to this cache in its parent's memcg_params | |
3223 | * before adding it to the memcg_slab_caches list, otherwise we can | |
3224 | * fail to convert memcg_params_to_cache() while traversing the list. | |
3225 | */ | |
2edefe11 | 3226 | VM_BUG_ON(root->memcg_params->memcg_caches[id]); |
959c8963 | 3227 | root->memcg_params->memcg_caches[id] = s; |
96403da2 VD |
3228 | |
3229 | mutex_lock(&memcg->slab_caches_mutex); | |
3230 | list_add(&s->memcg_params->list, &memcg->memcg_slab_caches); | |
3231 | mutex_unlock(&memcg->slab_caches_mutex); | |
1aa13254 | 3232 | } |
d7f25f8a | 3233 | |
1aa13254 VD |
3234 | void memcg_unregister_cache(struct kmem_cache *s) |
3235 | { | |
3236 | struct kmem_cache *root; | |
3237 | struct mem_cgroup *memcg; | |
3238 | int id; | |
3239 | ||
3240 | if (is_root_cache(s)) | |
3241 | return; | |
d7f25f8a | 3242 | |
2edefe11 VD |
3243 | /* |
3244 | * Holding the slab_mutex assures nobody will touch the memcg_caches | |
3245 | * array while we are modifying it. | |
3246 | */ | |
3247 | lockdep_assert_held(&slab_mutex); | |
3248 | ||
d7f25f8a | 3249 | root = s->memcg_params->root_cache; |
96403da2 VD |
3250 | memcg = s->memcg_params->memcg; |
3251 | id = memcg_cache_id(memcg); | |
d7f25f8a GC |
3252 | |
3253 | mutex_lock(&memcg->slab_caches_mutex); | |
3254 | list_del(&s->memcg_params->list); | |
3255 | mutex_unlock(&memcg->slab_caches_mutex); | |
3256 | ||
96403da2 VD |
3257 | /* |
3258 | * Clear the pointer to this cache in its parent's memcg_params only | |
3259 | * after removing it from the memcg_slab_caches list, otherwise we can | |
3260 | * fail to convert memcg_params_to_cache() while traversing the list. | |
3261 | */ | |
2edefe11 | 3262 | VM_BUG_ON(!root->memcg_params->memcg_caches[id]); |
96403da2 VD |
3263 | root->memcg_params->memcg_caches[id] = NULL; |
3264 | ||
20f05310 | 3265 | css_put(&memcg->css); |
2633d7a0 GC |
3266 | } |
3267 | ||
0e9d92f2 GC |
3268 | /* |
3269 | * During the creation a new cache, we need to disable our accounting mechanism | |
3270 | * altogether. This is true even if we are not creating, but rather just | |
3271 | * enqueing new caches to be created. | |
3272 | * | |
3273 | * This is because that process will trigger allocations; some visible, like | |
3274 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3275 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3276 | * objects during debug. | |
3277 | * | |
3278 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3279 | * to it. This may not be a bounded recursion: since the first cache creation | |
3280 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3281 | * cache again, failing at the same point. | |
3282 | * | |
3283 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3284 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3285 | * inside the following two functions. | |
3286 | */ | |
3287 | static inline void memcg_stop_kmem_account(void) | |
3288 | { | |
3289 | VM_BUG_ON(!current->mm); | |
3290 | current->memcg_kmem_skip_account++; | |
3291 | } | |
3292 | ||
3293 | static inline void memcg_resume_kmem_account(void) | |
3294 | { | |
3295 | VM_BUG_ON(!current->mm); | |
3296 | current->memcg_kmem_skip_account--; | |
3297 | } | |
3298 | ||
1f458cbf GC |
3299 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3300 | { | |
3301 | struct kmem_cache *cachep; | |
3302 | struct memcg_cache_params *p; | |
3303 | ||
3304 | p = container_of(w, struct memcg_cache_params, destroy); | |
3305 | ||
3306 | cachep = memcg_params_to_cache(p); | |
3307 | ||
22933152 GC |
3308 | /* |
3309 | * If we get down to 0 after shrink, we could delete right away. | |
3310 | * However, memcg_release_pages() already puts us back in the workqueue | |
3311 | * in that case. If we proceed deleting, we'll get a dangling | |
3312 | * reference, and removing the object from the workqueue in that case | |
3313 | * is unnecessary complication. We are not a fast path. | |
3314 | * | |
3315 | * Note that this case is fundamentally different from racing with | |
3316 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3317 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3318 | * into the queue, but doing so from inside the worker racing to | |
3319 | * destroy it. | |
3320 | * | |
3321 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3322 | * again | |
3323 | */ | |
0d8a4a37 | 3324 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) |
22933152 | 3325 | kmem_cache_shrink(cachep); |
0d8a4a37 | 3326 | else |
1f458cbf GC |
3327 | kmem_cache_destroy(cachep); |
3328 | } | |
3329 | ||
3330 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3331 | { | |
3332 | if (!cachep->memcg_params->dead) | |
3333 | return; | |
3334 | ||
22933152 GC |
3335 | /* |
3336 | * There are many ways in which we can get here. | |
3337 | * | |
3338 | * We can get to a memory-pressure situation while the delayed work is | |
3339 | * still pending to run. The vmscan shrinkers can then release all | |
3340 | * cache memory and get us to destruction. If this is the case, we'll | |
3341 | * be executed twice, which is a bug (the second time will execute over | |
3342 | * bogus data). In this case, cancelling the work should be fine. | |
3343 | * | |
3344 | * But we can also get here from the worker itself, if | |
3345 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3346 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3347 | * cancel the work (the worker runs with an internal lock held, which | |
3348 | * is the same lock we would hold for cancel_work_sync().) | |
3349 | * | |
3350 | * Since we can't possibly know who got us here, just refrain from | |
3351 | * running if there is already work pending | |
3352 | */ | |
3353 | if (work_pending(&cachep->memcg_params->destroy)) | |
3354 | return; | |
1f458cbf GC |
3355 | /* |
3356 | * We have to defer the actual destroying to a workqueue, because | |
3357 | * we might currently be in a context that cannot sleep. | |
3358 | */ | |
3359 | schedule_work(&cachep->memcg_params->destroy); | |
3360 | } | |
3361 | ||
842e2873 VD |
3362 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3363 | struct kmem_cache *s) | |
d7f25f8a | 3364 | { |
7c094fd6 | 3365 | struct kmem_cache *new = NULL; |
e61734c5 | 3366 | static char *tmp_path = NULL, *tmp_name = NULL; |
842e2873 | 3367 | static DEFINE_MUTEX(mutex); /* protects tmp_name */ |
d7f25f8a | 3368 | |
842e2873 | 3369 | BUG_ON(!memcg_can_account_kmem(memcg)); |
d9c10ddd | 3370 | |
842e2873 | 3371 | mutex_lock(&mutex); |
d9c10ddd MH |
3372 | /* |
3373 | * kmem_cache_create_memcg duplicates the given name and | |
3374 | * cgroup_name for this name requires RCU context. | |
3375 | * This static temporary buffer is used to prevent from | |
3376 | * pointless shortliving allocation. | |
3377 | */ | |
e61734c5 TH |
3378 | if (!tmp_path || !tmp_name) { |
3379 | if (!tmp_path) | |
3380 | tmp_path = kmalloc(PATH_MAX, GFP_KERNEL); | |
d9c10ddd | 3381 | if (!tmp_name) |
e61734c5 TH |
3382 | tmp_name = kmalloc(NAME_MAX + 1, GFP_KERNEL); |
3383 | if (!tmp_path || !tmp_name) | |
7c094fd6 | 3384 | goto out; |
d9c10ddd MH |
3385 | } |
3386 | ||
e61734c5 TH |
3387 | cgroup_name(memcg->css.cgroup, tmp_name, NAME_MAX + 1); |
3388 | snprintf(tmp_path, PATH_MAX, "%s(%d:%s)", s->name, | |
3389 | memcg_cache_id(memcg), tmp_name); | |
d7f25f8a | 3390 | |
e61734c5 | 3391 | new = kmem_cache_create_memcg(memcg, tmp_path, s->object_size, s->align, |
943a451a | 3392 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d79923fa GC |
3393 | if (new) |
3394 | new->allocflags |= __GFP_KMEMCG; | |
842e2873 VD |
3395 | else |
3396 | new = s; | |
7c094fd6 | 3397 | out: |
842e2873 | 3398 | mutex_unlock(&mutex); |
d7f25f8a GC |
3399 | return new; |
3400 | } | |
3401 | ||
7cf27982 GC |
3402 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3403 | { | |
3404 | struct kmem_cache *c; | |
3405 | int i; | |
3406 | ||
3407 | if (!s->memcg_params) | |
3408 | return; | |
3409 | if (!s->memcg_params->is_root_cache) | |
3410 | return; | |
3411 | ||
3412 | /* | |
3413 | * If the cache is being destroyed, we trust that there is no one else | |
3414 | * requesting objects from it. Even if there are, the sanity checks in | |
3415 | * kmem_cache_destroy should caught this ill-case. | |
3416 | * | |
3417 | * Still, we don't want anyone else freeing memcg_caches under our | |
3418 | * noses, which can happen if a new memcg comes to life. As usual, | |
d6441637 VD |
3419 | * we'll take the activate_kmem_mutex to protect ourselves against |
3420 | * this. | |
7cf27982 | 3421 | */ |
d6441637 | 3422 | mutex_lock(&activate_kmem_mutex); |
7a67d7ab QH |
3423 | for_each_memcg_cache_index(i) { |
3424 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
3425 | if (!c) |
3426 | continue; | |
3427 | ||
3428 | /* | |
3429 | * We will now manually delete the caches, so to avoid races | |
3430 | * we need to cancel all pending destruction workers and | |
3431 | * proceed with destruction ourselves. | |
3432 | * | |
3433 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3434 | * and that could spawn the workers again: it is likely that | |
3435 | * the cache still have active pages until this very moment. | |
3436 | * This would lead us back to mem_cgroup_destroy_cache. | |
3437 | * | |
3438 | * But that will not execute at all if the "dead" flag is not | |
3439 | * set, so flip it down to guarantee we are in control. | |
3440 | */ | |
3441 | c->memcg_params->dead = false; | |
22933152 | 3442 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3443 | kmem_cache_destroy(c); |
3444 | } | |
d6441637 | 3445 | mutex_unlock(&activate_kmem_mutex); |
7cf27982 GC |
3446 | } |
3447 | ||
d7f25f8a GC |
3448 | struct create_work { |
3449 | struct mem_cgroup *memcg; | |
3450 | struct kmem_cache *cachep; | |
3451 | struct work_struct work; | |
3452 | }; | |
3453 | ||
1f458cbf GC |
3454 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3455 | { | |
3456 | struct kmem_cache *cachep; | |
3457 | struct memcg_cache_params *params; | |
3458 | ||
3459 | if (!memcg_kmem_is_active(memcg)) | |
3460 | return; | |
3461 | ||
3462 | mutex_lock(&memcg->slab_caches_mutex); | |
3463 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3464 | cachep = memcg_params_to_cache(params); | |
3465 | cachep->memcg_params->dead = true; | |
1f458cbf GC |
3466 | schedule_work(&cachep->memcg_params->destroy); |
3467 | } | |
3468 | mutex_unlock(&memcg->slab_caches_mutex); | |
3469 | } | |
3470 | ||
d7f25f8a GC |
3471 | static void memcg_create_cache_work_func(struct work_struct *w) |
3472 | { | |
3473 | struct create_work *cw; | |
3474 | ||
3475 | cw = container_of(w, struct create_work, work); | |
3476 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
1aa13254 | 3477 | css_put(&cw->memcg->css); |
d7f25f8a GC |
3478 | kfree(cw); |
3479 | } | |
3480 | ||
3481 | /* | |
3482 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3483 | */ |
0e9d92f2 GC |
3484 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3485 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3486 | { |
3487 | struct create_work *cw; | |
3488 | ||
3489 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
ca0dde97 LZ |
3490 | if (cw == NULL) { |
3491 | css_put(&memcg->css); | |
d7f25f8a GC |
3492 | return; |
3493 | } | |
3494 | ||
3495 | cw->memcg = memcg; | |
3496 | cw->cachep = cachep; | |
3497 | ||
3498 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3499 | schedule_work(&cw->work); | |
3500 | } | |
3501 | ||
0e9d92f2 GC |
3502 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3503 | struct kmem_cache *cachep) | |
3504 | { | |
3505 | /* | |
3506 | * We need to stop accounting when we kmalloc, because if the | |
3507 | * corresponding kmalloc cache is not yet created, the first allocation | |
3508 | * in __memcg_create_cache_enqueue will recurse. | |
3509 | * | |
3510 | * However, it is better to enclose the whole function. Depending on | |
3511 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3512 | * trigger an allocation. This too, will make us recurse. Because at | |
3513 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3514 | * the safest choice is to do it like this, wrapping the whole function. | |
3515 | */ | |
3516 | memcg_stop_kmem_account(); | |
3517 | __memcg_create_cache_enqueue(memcg, cachep); | |
3518 | memcg_resume_kmem_account(); | |
3519 | } | |
d7f25f8a GC |
3520 | /* |
3521 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3522 | * We try to use the current memcg's version of the cache. | |
3523 | * | |
3524 | * If the cache does not exist yet, if we are the first user of it, | |
3525 | * we either create it immediately, if possible, or create it asynchronously | |
3526 | * in a workqueue. | |
3527 | * In the latter case, we will let the current allocation go through with | |
3528 | * the original cache. | |
3529 | * | |
3530 | * Can't be called in interrupt context or from kernel threads. | |
3531 | * This function needs to be called with rcu_read_lock() held. | |
3532 | */ | |
3533 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3534 | gfp_t gfp) | |
3535 | { | |
3536 | struct mem_cgroup *memcg; | |
959c8963 | 3537 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
3538 | |
3539 | VM_BUG_ON(!cachep->memcg_params); | |
3540 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3541 | ||
0e9d92f2 GC |
3542 | if (!current->mm || current->memcg_kmem_skip_account) |
3543 | return cachep; | |
3544 | ||
d7f25f8a GC |
3545 | rcu_read_lock(); |
3546 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3547 | |
3548 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3549 | goto out; |
d7f25f8a | 3550 | |
959c8963 VD |
3551 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
3552 | if (likely(memcg_cachep)) { | |
3553 | cachep = memcg_cachep; | |
ca0dde97 | 3554 | goto out; |
d7f25f8a GC |
3555 | } |
3556 | ||
ca0dde97 LZ |
3557 | /* The corresponding put will be done in the workqueue. */ |
3558 | if (!css_tryget(&memcg->css)) | |
3559 | goto out; | |
3560 | rcu_read_unlock(); | |
3561 | ||
3562 | /* | |
3563 | * If we are in a safe context (can wait, and not in interrupt | |
3564 | * context), we could be be predictable and return right away. | |
3565 | * This would guarantee that the allocation being performed | |
3566 | * already belongs in the new cache. | |
3567 | * | |
3568 | * However, there are some clashes that can arrive from locking. | |
3569 | * For instance, because we acquire the slab_mutex while doing | |
3570 | * kmem_cache_dup, this means no further allocation could happen | |
3571 | * with the slab_mutex held. | |
3572 | * | |
3573 | * Also, because cache creation issue get_online_cpus(), this | |
3574 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3575 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3576 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3577 | * better to defer everything. | |
3578 | */ | |
3579 | memcg_create_cache_enqueue(memcg, cachep); | |
3580 | return cachep; | |
3581 | out: | |
3582 | rcu_read_unlock(); | |
3583 | return cachep; | |
d7f25f8a GC |
3584 | } |
3585 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3586 | ||
7ae1e1d0 GC |
3587 | /* |
3588 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3589 | * possible for the given order. But the page is not allocated yet, so we'll | |
3590 | * need a further commit step to do the final arrangements. | |
3591 | * | |
3592 | * It is possible for the task to switch cgroups in this mean time, so at | |
3593 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3594 | * the handle argument to return to the caller which cgroup we should commit | |
3595 | * against. We could also return the memcg directly and avoid the pointer | |
3596 | * passing, but a boolean return value gives better semantics considering | |
3597 | * the compiled-out case as well. | |
3598 | * | |
3599 | * Returning true means the allocation is possible. | |
3600 | */ | |
3601 | bool | |
3602 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3603 | { | |
3604 | struct mem_cgroup *memcg; | |
3605 | int ret; | |
3606 | ||
3607 | *_memcg = NULL; | |
6d42c232 GC |
3608 | |
3609 | /* | |
3610 | * Disabling accounting is only relevant for some specific memcg | |
3611 | * internal allocations. Therefore we would initially not have such | |
3612 | * check here, since direct calls to the page allocator that are marked | |
3613 | * with GFP_KMEMCG only happen outside memcg core. We are mostly | |
3614 | * concerned with cache allocations, and by having this test at | |
3615 | * memcg_kmem_get_cache, we are already able to relay the allocation to | |
3616 | * the root cache and bypass the memcg cache altogether. | |
3617 | * | |
3618 | * There is one exception, though: the SLUB allocator does not create | |
3619 | * large order caches, but rather service large kmallocs directly from | |
3620 | * the page allocator. Therefore, the following sequence when backed by | |
3621 | * the SLUB allocator: | |
3622 | * | |
f894ffa8 AM |
3623 | * memcg_stop_kmem_account(); |
3624 | * kmalloc(<large_number>) | |
3625 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3626 | * |
3627 | * would effectively ignore the fact that we should skip accounting, | |
3628 | * since it will drive us directly to this function without passing | |
3629 | * through the cache selector memcg_kmem_get_cache. Such large | |
3630 | * allocations are extremely rare but can happen, for instance, for the | |
3631 | * cache arrays. We bring this test here. | |
3632 | */ | |
3633 | if (!current->mm || current->memcg_kmem_skip_account) | |
3634 | return true; | |
3635 | ||
7ae1e1d0 GC |
3636 | memcg = try_get_mem_cgroup_from_mm(current->mm); |
3637 | ||
3638 | /* | |
3639 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3640 | * isn't much we can do without complicating this too much, and it would | |
3641 | * be gfp-dependent anyway. Just let it go | |
3642 | */ | |
3643 | if (unlikely(!memcg)) | |
3644 | return true; | |
3645 | ||
3646 | if (!memcg_can_account_kmem(memcg)) { | |
3647 | css_put(&memcg->css); | |
3648 | return true; | |
3649 | } | |
3650 | ||
7ae1e1d0 GC |
3651 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3652 | if (!ret) | |
3653 | *_memcg = memcg; | |
7ae1e1d0 GC |
3654 | |
3655 | css_put(&memcg->css); | |
3656 | return (ret == 0); | |
3657 | } | |
3658 | ||
3659 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3660 | int order) | |
3661 | { | |
3662 | struct page_cgroup *pc; | |
3663 | ||
3664 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3665 | ||
3666 | /* The page allocation failed. Revert */ | |
3667 | if (!page) { | |
3668 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3669 | return; |
3670 | } | |
3671 | ||
3672 | pc = lookup_page_cgroup(page); | |
3673 | lock_page_cgroup(pc); | |
3674 | pc->mem_cgroup = memcg; | |
3675 | SetPageCgroupUsed(pc); | |
3676 | unlock_page_cgroup(pc); | |
3677 | } | |
3678 | ||
3679 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3680 | { | |
3681 | struct mem_cgroup *memcg = NULL; | |
3682 | struct page_cgroup *pc; | |
3683 | ||
3684 | ||
3685 | pc = lookup_page_cgroup(page); | |
3686 | /* | |
3687 | * Fast unlocked return. Theoretically might have changed, have to | |
3688 | * check again after locking. | |
3689 | */ | |
3690 | if (!PageCgroupUsed(pc)) | |
3691 | return; | |
3692 | ||
3693 | lock_page_cgroup(pc); | |
3694 | if (PageCgroupUsed(pc)) { | |
3695 | memcg = pc->mem_cgroup; | |
3696 | ClearPageCgroupUsed(pc); | |
3697 | } | |
3698 | unlock_page_cgroup(pc); | |
3699 | ||
3700 | /* | |
3701 | * We trust that only if there is a memcg associated with the page, it | |
3702 | * is a valid allocation | |
3703 | */ | |
3704 | if (!memcg) | |
3705 | return; | |
3706 | ||
309381fe | 3707 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
7ae1e1d0 | 3708 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); |
7ae1e1d0 | 3709 | } |
1f458cbf GC |
3710 | #else |
3711 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3712 | { | |
3713 | } | |
7ae1e1d0 GC |
3714 | #endif /* CONFIG_MEMCG_KMEM */ |
3715 | ||
ca3e0214 KH |
3716 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3717 | ||
a0db00fc | 3718 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3719 | /* |
3720 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3721 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3722 | * charge/uncharge will be never happen and move_account() is done under | |
3723 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3724 | */ |
e94c8a9c | 3725 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3726 | { |
3727 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3728 | struct page_cgroup *pc; |
b070e65c | 3729 | struct mem_cgroup *memcg; |
e94c8a9c | 3730 | int i; |
ca3e0214 | 3731 | |
3d37c4a9 KH |
3732 | if (mem_cgroup_disabled()) |
3733 | return; | |
b070e65c DR |
3734 | |
3735 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3736 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3737 | pc = head_pc + i; | |
b070e65c | 3738 | pc->mem_cgroup = memcg; |
e94c8a9c | 3739 | smp_wmb();/* see __commit_charge() */ |
e94c8a9c KH |
3740 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3741 | } | |
b070e65c DR |
3742 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3743 | HPAGE_PMD_NR); | |
ca3e0214 | 3744 | } |
12d27107 | 3745 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3746 | |
7af467e8 JW |
3747 | static void mem_cgroup_move_account_page_stat(struct mem_cgroup *from, |
3748 | struct mem_cgroup *to, | |
3749 | unsigned int nr_pages, | |
3750 | enum mem_cgroup_stat_index idx) | |
3ea67d06 SZ |
3751 | { |
3752 | /* Update stat data for mem_cgroup */ | |
5e8cfc3c | 3753 | __this_cpu_sub(from->stat->count[idx], nr_pages); |
3ea67d06 | 3754 | __this_cpu_add(to->stat->count[idx], nr_pages); |
3ea67d06 SZ |
3755 | } |
3756 | ||
f817ed48 | 3757 | /** |
de3638d9 | 3758 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3759 | * @page: the page |
7ec99d62 | 3760 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3761 | * @pc: page_cgroup of the page. |
3762 | * @from: mem_cgroup which the page is moved from. | |
3763 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3764 | * | |
3765 | * The caller must confirm following. | |
08e552c6 | 3766 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3767 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3768 | * |
2f3479b1 KH |
3769 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3770 | * from old cgroup. | |
f817ed48 | 3771 | */ |
7ec99d62 JW |
3772 | static int mem_cgroup_move_account(struct page *page, |
3773 | unsigned int nr_pages, | |
3774 | struct page_cgroup *pc, | |
3775 | struct mem_cgroup *from, | |
2f3479b1 | 3776 | struct mem_cgroup *to) |
f817ed48 | 3777 | { |
de3638d9 JW |
3778 | unsigned long flags; |
3779 | int ret; | |
b2402857 | 3780 | bool anon = PageAnon(page); |
987eba66 | 3781 | |
f817ed48 | 3782 | VM_BUG_ON(from == to); |
309381fe | 3783 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3784 | /* |
3785 | * The page is isolated from LRU. So, collapse function | |
3786 | * will not handle this page. But page splitting can happen. | |
3787 | * Do this check under compound_page_lock(). The caller should | |
3788 | * hold it. | |
3789 | */ | |
3790 | ret = -EBUSY; | |
7ec99d62 | 3791 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3792 | goto out; |
3793 | ||
3794 | lock_page_cgroup(pc); | |
3795 | ||
3796 | ret = -EINVAL; | |
3797 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3798 | goto unlock; | |
3799 | ||
312734c0 | 3800 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3801 | |
3ea67d06 SZ |
3802 | if (!anon && page_mapped(page)) |
3803 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3804 | MEM_CGROUP_STAT_FILE_MAPPED); | |
3805 | ||
3806 | if (PageWriteback(page)) | |
3807 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3808 | MEM_CGROUP_STAT_WRITEBACK); | |
3809 | ||
b070e65c | 3810 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
d69b042f | 3811 | |
854ffa8d | 3812 | /* caller should have done css_get */ |
08e552c6 | 3813 | pc->mem_cgroup = to; |
b070e65c | 3814 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
312734c0 | 3815 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3816 | ret = 0; |
3817 | unlock: | |
57f9fd7d | 3818 | unlock_page_cgroup(pc); |
d2265e6f KH |
3819 | /* |
3820 | * check events | |
3821 | */ | |
5564e88b JW |
3822 | memcg_check_events(to, page); |
3823 | memcg_check_events(from, page); | |
de3638d9 | 3824 | out: |
f817ed48 KH |
3825 | return ret; |
3826 | } | |
3827 | ||
2ef37d3f MH |
3828 | /** |
3829 | * mem_cgroup_move_parent - moves page to the parent group | |
3830 | * @page: the page to move | |
3831 | * @pc: page_cgroup of the page | |
3832 | * @child: page's cgroup | |
3833 | * | |
3834 | * move charges to its parent or the root cgroup if the group has no | |
3835 | * parent (aka use_hierarchy==0). | |
3836 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3837 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3838 | * it signals a race with a page removal/uncharge or migration. In the | |
3839 | * first case the page is on the way out and it will vanish from the LRU | |
3840 | * on the next attempt and the call should be retried later. | |
3841 | * Isolation from the LRU fails only if page has been isolated from | |
3842 | * the LRU since we looked at it and that usually means either global | |
3843 | * reclaim or migration going on. The page will either get back to the | |
3844 | * LRU or vanish. | |
3845 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3846 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3847 | * disappear in the next attempt. | |
f817ed48 | 3848 | */ |
5564e88b JW |
3849 | static int mem_cgroup_move_parent(struct page *page, |
3850 | struct page_cgroup *pc, | |
6068bf01 | 3851 | struct mem_cgroup *child) |
f817ed48 | 3852 | { |
f817ed48 | 3853 | struct mem_cgroup *parent; |
7ec99d62 | 3854 | unsigned int nr_pages; |
4be4489f | 3855 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3856 | int ret; |
3857 | ||
d8423011 | 3858 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3859 | |
57f9fd7d DN |
3860 | ret = -EBUSY; |
3861 | if (!get_page_unless_zero(page)) | |
3862 | goto out; | |
3863 | if (isolate_lru_page(page)) | |
3864 | goto put; | |
52dbb905 | 3865 | |
7ec99d62 | 3866 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3867 | |
cc926f78 KH |
3868 | parent = parent_mem_cgroup(child); |
3869 | /* | |
3870 | * If no parent, move charges to root cgroup. | |
3871 | */ | |
3872 | if (!parent) | |
3873 | parent = root_mem_cgroup; | |
f817ed48 | 3874 | |
2ef37d3f | 3875 | if (nr_pages > 1) { |
309381fe | 3876 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
987eba66 | 3877 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3878 | } |
987eba66 | 3879 | |
cc926f78 | 3880 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3881 | pc, child, parent); |
cc926f78 KH |
3882 | if (!ret) |
3883 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3884 | |
7ec99d62 | 3885 | if (nr_pages > 1) |
987eba66 | 3886 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3887 | putback_lru_page(page); |
57f9fd7d | 3888 | put: |
40d58138 | 3889 | put_page(page); |
57f9fd7d | 3890 | out: |
f817ed48 KH |
3891 | return ret; |
3892 | } | |
3893 | ||
7a81b88c KH |
3894 | /* |
3895 | * Charge the memory controller for page usage. | |
3896 | * Return | |
3897 | * 0 if the charge was successful | |
3898 | * < 0 if the cgroup is over its limit | |
3899 | */ | |
3900 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3901 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3902 | { |
c0ff4b85 | 3903 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3904 | unsigned int nr_pages = 1; |
8493ae43 | 3905 | bool oom = true; |
7a81b88c | 3906 | int ret; |
ec168510 | 3907 | |
37c2ac78 | 3908 | if (PageTransHuge(page)) { |
7ec99d62 | 3909 | nr_pages <<= compound_order(page); |
309381fe | 3910 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
8493ae43 JW |
3911 | /* |
3912 | * Never OOM-kill a process for a huge page. The | |
3913 | * fault handler will fall back to regular pages. | |
3914 | */ | |
3915 | oom = false; | |
37c2ac78 | 3916 | } |
7a81b88c | 3917 | |
c0ff4b85 | 3918 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3919 | if (ret == -ENOMEM) |
7a81b88c | 3920 | return ret; |
ce587e65 | 3921 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3922 | return 0; |
8a9f3ccd BS |
3923 | } |
3924 | ||
7a81b88c KH |
3925 | int mem_cgroup_newpage_charge(struct page *page, |
3926 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3927 | { |
f8d66542 | 3928 | if (mem_cgroup_disabled()) |
cede86ac | 3929 | return 0; |
309381fe SL |
3930 | VM_BUG_ON_PAGE(page_mapped(page), page); |
3931 | VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page); | |
7a0524cf | 3932 | VM_BUG_ON(!mm); |
217bc319 | 3933 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3934 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
3935 | } |
3936 | ||
54595fe2 KH |
3937 | /* |
3938 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3939 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3940 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3941 | * "commit()" or removed by "cancel()" |
3942 | */ | |
0435a2fd JW |
3943 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3944 | struct page *page, | |
3945 | gfp_t mask, | |
3946 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3947 | { |
c0ff4b85 | 3948 | struct mem_cgroup *memcg; |
90deb788 | 3949 | struct page_cgroup *pc; |
54595fe2 | 3950 | int ret; |
8c7c6e34 | 3951 | |
90deb788 JW |
3952 | pc = lookup_page_cgroup(page); |
3953 | /* | |
3954 | * Every swap fault against a single page tries to charge the | |
3955 | * page, bail as early as possible. shmem_unuse() encounters | |
3956 | * already charged pages, too. The USED bit is protected by | |
3957 | * the page lock, which serializes swap cache removal, which | |
3958 | * in turn serializes uncharging. | |
3959 | */ | |
3960 | if (PageCgroupUsed(pc)) | |
3961 | return 0; | |
8c7c6e34 KH |
3962 | if (!do_swap_account) |
3963 | goto charge_cur_mm; | |
c0ff4b85 R |
3964 | memcg = try_get_mem_cgroup_from_page(page); |
3965 | if (!memcg) | |
54595fe2 | 3966 | goto charge_cur_mm; |
72835c86 JW |
3967 | *memcgp = memcg; |
3968 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 3969 | css_put(&memcg->css); |
38c5d72f KH |
3970 | if (ret == -EINTR) |
3971 | ret = 0; | |
54595fe2 | 3972 | return ret; |
8c7c6e34 | 3973 | charge_cur_mm: |
38c5d72f KH |
3974 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
3975 | if (ret == -EINTR) | |
3976 | ret = 0; | |
3977 | return ret; | |
8c7c6e34 KH |
3978 | } |
3979 | ||
0435a2fd JW |
3980 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
3981 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
3982 | { | |
3983 | *memcgp = NULL; | |
3984 | if (mem_cgroup_disabled()) | |
3985 | return 0; | |
bdf4f4d2 JW |
3986 | /* |
3987 | * A racing thread's fault, or swapoff, may have already | |
3988 | * updated the pte, and even removed page from swap cache: in | |
3989 | * those cases unuse_pte()'s pte_same() test will fail; but | |
3990 | * there's also a KSM case which does need to charge the page. | |
3991 | */ | |
3992 | if (!PageSwapCache(page)) { | |
3993 | int ret; | |
3994 | ||
3995 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
3996 | if (ret == -EINTR) | |
3997 | ret = 0; | |
3998 | return ret; | |
3999 | } | |
0435a2fd JW |
4000 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
4001 | } | |
4002 | ||
827a03d2 JW |
4003 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
4004 | { | |
4005 | if (mem_cgroup_disabled()) | |
4006 | return; | |
4007 | if (!memcg) | |
4008 | return; | |
4009 | __mem_cgroup_cancel_charge(memcg, 1); | |
4010 | } | |
4011 | ||
83aae4c7 | 4012 | static void |
72835c86 | 4013 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 4014 | enum charge_type ctype) |
7a81b88c | 4015 | { |
f8d66542 | 4016 | if (mem_cgroup_disabled()) |
7a81b88c | 4017 | return; |
72835c86 | 4018 | if (!memcg) |
7a81b88c | 4019 | return; |
5a6475a4 | 4020 | |
ce587e65 | 4021 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
4022 | /* |
4023 | * Now swap is on-memory. This means this page may be | |
4024 | * counted both as mem and swap....double count. | |
03f3c433 KH |
4025 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
4026 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
4027 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 4028 | */ |
03f3c433 | 4029 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 4030 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 4031 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 4032 | } |
7a81b88c KH |
4033 | } |
4034 | ||
72835c86 JW |
4035 | void mem_cgroup_commit_charge_swapin(struct page *page, |
4036 | struct mem_cgroup *memcg) | |
83aae4c7 | 4037 | { |
72835c86 | 4038 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 4039 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
4040 | } |
4041 | ||
827a03d2 JW |
4042 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
4043 | gfp_t gfp_mask) | |
7a81b88c | 4044 | { |
827a03d2 JW |
4045 | struct mem_cgroup *memcg = NULL; |
4046 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
4047 | int ret; | |
4048 | ||
f8d66542 | 4049 | if (mem_cgroup_disabled()) |
827a03d2 JW |
4050 | return 0; |
4051 | if (PageCompound(page)) | |
4052 | return 0; | |
4053 | ||
827a03d2 JW |
4054 | if (!PageSwapCache(page)) |
4055 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
4056 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
4057 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
4058 | gfp_mask, &memcg); | |
827a03d2 JW |
4059 | if (!ret) |
4060 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
4061 | } | |
4062 | return ret; | |
7a81b88c KH |
4063 | } |
4064 | ||
c0ff4b85 | 4065 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
4066 | unsigned int nr_pages, |
4067 | const enum charge_type ctype) | |
569b846d KH |
4068 | { |
4069 | struct memcg_batch_info *batch = NULL; | |
4070 | bool uncharge_memsw = true; | |
7ec99d62 | 4071 | |
569b846d KH |
4072 | /* If swapout, usage of swap doesn't decrease */ |
4073 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
4074 | uncharge_memsw = false; | |
569b846d KH |
4075 | |
4076 | batch = ¤t->memcg_batch; | |
4077 | /* | |
4078 | * In usual, we do css_get() when we remember memcg pointer. | |
4079 | * But in this case, we keep res->usage until end of a series of | |
4080 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
4081 | */ | |
4082 | if (!batch->memcg) | |
c0ff4b85 | 4083 | batch->memcg = memcg; |
3c11ecf4 KH |
4084 | /* |
4085 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 4086 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
4087 | * the same cgroup and we have chance to coalesce uncharges. |
4088 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
4089 | * because we want to do uncharge as soon as possible. | |
4090 | */ | |
4091 | ||
4092 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
4093 | goto direct_uncharge; | |
4094 | ||
7ec99d62 | 4095 | if (nr_pages > 1) |
ec168510 AA |
4096 | goto direct_uncharge; |
4097 | ||
569b846d KH |
4098 | /* |
4099 | * In typical case, batch->memcg == mem. This means we can | |
4100 | * merge a series of uncharges to an uncharge of res_counter. | |
4101 | * If not, we uncharge res_counter ony by one. | |
4102 | */ | |
c0ff4b85 | 4103 | if (batch->memcg != memcg) |
569b846d KH |
4104 | goto direct_uncharge; |
4105 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 4106 | batch->nr_pages++; |
569b846d | 4107 | if (uncharge_memsw) |
7ffd4ca7 | 4108 | batch->memsw_nr_pages++; |
569b846d KH |
4109 | return; |
4110 | direct_uncharge: | |
c0ff4b85 | 4111 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 4112 | if (uncharge_memsw) |
c0ff4b85 R |
4113 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
4114 | if (unlikely(batch->memcg != memcg)) | |
4115 | memcg_oom_recover(memcg); | |
569b846d | 4116 | } |
7a81b88c | 4117 | |
8a9f3ccd | 4118 | /* |
69029cd5 | 4119 | * uncharge if !page_mapped(page) |
8a9f3ccd | 4120 | */ |
8c7c6e34 | 4121 | static struct mem_cgroup * |
0030f535 JW |
4122 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
4123 | bool end_migration) | |
8a9f3ccd | 4124 | { |
c0ff4b85 | 4125 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
4126 | unsigned int nr_pages = 1; |
4127 | struct page_cgroup *pc; | |
b2402857 | 4128 | bool anon; |
8a9f3ccd | 4129 | |
f8d66542 | 4130 | if (mem_cgroup_disabled()) |
8c7c6e34 | 4131 | return NULL; |
4077960e | 4132 | |
37c2ac78 | 4133 | if (PageTransHuge(page)) { |
7ec99d62 | 4134 | nr_pages <<= compound_order(page); |
309381fe | 4135 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
37c2ac78 | 4136 | } |
8697d331 | 4137 | /* |
3c541e14 | 4138 | * Check if our page_cgroup is valid |
8697d331 | 4139 | */ |
52d4b9ac | 4140 | pc = lookup_page_cgroup(page); |
cfa44946 | 4141 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 4142 | return NULL; |
b9c565d5 | 4143 | |
52d4b9ac | 4144 | lock_page_cgroup(pc); |
d13d1443 | 4145 | |
c0ff4b85 | 4146 | memcg = pc->mem_cgroup; |
8c7c6e34 | 4147 | |
d13d1443 KH |
4148 | if (!PageCgroupUsed(pc)) |
4149 | goto unlock_out; | |
4150 | ||
b2402857 KH |
4151 | anon = PageAnon(page); |
4152 | ||
d13d1443 | 4153 | switch (ctype) { |
41326c17 | 4154 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
4155 | /* |
4156 | * Generally PageAnon tells if it's the anon statistics to be | |
4157 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4158 | * used before page reached the stage of being marked PageAnon. | |
4159 | */ | |
b2402857 KH |
4160 | anon = true; |
4161 | /* fallthrough */ | |
8a9478ca | 4162 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4163 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4164 | if (page_mapped(page)) |
4165 | goto unlock_out; | |
4166 | /* | |
4167 | * Pages under migration may not be uncharged. But | |
4168 | * end_migration() /must/ be the one uncharging the | |
4169 | * unused post-migration page and so it has to call | |
4170 | * here with the migration bit still set. See the | |
4171 | * res_counter handling below. | |
4172 | */ | |
4173 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4174 | goto unlock_out; |
4175 | break; | |
4176 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4177 | if (!PageAnon(page)) { /* Shared memory */ | |
4178 | if (page->mapping && !page_is_file_cache(page)) | |
4179 | goto unlock_out; | |
4180 | } else if (page_mapped(page)) /* Anon */ | |
4181 | goto unlock_out; | |
4182 | break; | |
4183 | default: | |
4184 | break; | |
52d4b9ac | 4185 | } |
d13d1443 | 4186 | |
b070e65c | 4187 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
04046e1a | 4188 | |
52d4b9ac | 4189 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4190 | /* |
4191 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4192 | * freed from LRU. This is safe because uncharged page is expected not | |
4193 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4194 | * special functions. | |
4195 | */ | |
b9c565d5 | 4196 | |
52d4b9ac | 4197 | unlock_page_cgroup(pc); |
f75ca962 | 4198 | /* |
c0ff4b85 | 4199 | * even after unlock, we have memcg->res.usage here and this memcg |
4050377b | 4200 | * will never be freed, so it's safe to call css_get(). |
f75ca962 | 4201 | */ |
c0ff4b85 | 4202 | memcg_check_events(memcg, page); |
f75ca962 | 4203 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 | 4204 | mem_cgroup_swap_statistics(memcg, true); |
4050377b | 4205 | css_get(&memcg->css); |
f75ca962 | 4206 | } |
0030f535 JW |
4207 | /* |
4208 | * Migration does not charge the res_counter for the | |
4209 | * replacement page, so leave it alone when phasing out the | |
4210 | * page that is unused after the migration. | |
4211 | */ | |
4212 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4213 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4214 | |
c0ff4b85 | 4215 | return memcg; |
d13d1443 KH |
4216 | |
4217 | unlock_out: | |
4218 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4219 | return NULL; |
3c541e14 BS |
4220 | } |
4221 | ||
69029cd5 KH |
4222 | void mem_cgroup_uncharge_page(struct page *page) |
4223 | { | |
52d4b9ac KH |
4224 | /* early check. */ |
4225 | if (page_mapped(page)) | |
4226 | return; | |
309381fe | 4227 | VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page); |
28ccddf7 JW |
4228 | /* |
4229 | * If the page is in swap cache, uncharge should be deferred | |
4230 | * to the swap path, which also properly accounts swap usage | |
4231 | * and handles memcg lifetime. | |
4232 | * | |
4233 | * Note that this check is not stable and reclaim may add the | |
4234 | * page to swap cache at any time after this. However, if the | |
4235 | * page is not in swap cache by the time page->mapcount hits | |
4236 | * 0, there won't be any page table references to the swap | |
4237 | * slot, and reclaim will free it and not actually write the | |
4238 | * page to disk. | |
4239 | */ | |
0c59b89c JW |
4240 | if (PageSwapCache(page)) |
4241 | return; | |
0030f535 | 4242 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4243 | } |
4244 | ||
4245 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4246 | { | |
309381fe SL |
4247 | VM_BUG_ON_PAGE(page_mapped(page), page); |
4248 | VM_BUG_ON_PAGE(page->mapping, page); | |
0030f535 | 4249 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4250 | } |
4251 | ||
569b846d KH |
4252 | /* |
4253 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4254 | * In that cases, pages are freed continuously and we can expect pages | |
4255 | * are in the same memcg. All these calls itself limits the number of | |
4256 | * pages freed at once, then uncharge_start/end() is called properly. | |
4257 | * This may be called prural(2) times in a context, | |
4258 | */ | |
4259 | ||
4260 | void mem_cgroup_uncharge_start(void) | |
4261 | { | |
4262 | current->memcg_batch.do_batch++; | |
4263 | /* We can do nest. */ | |
4264 | if (current->memcg_batch.do_batch == 1) { | |
4265 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4266 | current->memcg_batch.nr_pages = 0; |
4267 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4268 | } |
4269 | } | |
4270 | ||
4271 | void mem_cgroup_uncharge_end(void) | |
4272 | { | |
4273 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4274 | ||
4275 | if (!batch->do_batch) | |
4276 | return; | |
4277 | ||
4278 | batch->do_batch--; | |
4279 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4280 | return; | |
4281 | ||
4282 | if (!batch->memcg) | |
4283 | return; | |
4284 | /* | |
4285 | * This "batch->memcg" is valid without any css_get/put etc... | |
4286 | * bacause we hide charges behind us. | |
4287 | */ | |
7ffd4ca7 JW |
4288 | if (batch->nr_pages) |
4289 | res_counter_uncharge(&batch->memcg->res, | |
4290 | batch->nr_pages * PAGE_SIZE); | |
4291 | if (batch->memsw_nr_pages) | |
4292 | res_counter_uncharge(&batch->memcg->memsw, | |
4293 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4294 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4295 | /* forget this pointer (for sanity check) */ |
4296 | batch->memcg = NULL; | |
4297 | } | |
4298 | ||
e767e056 | 4299 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4300 | /* |
e767e056 | 4301 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4302 | * memcg information is recorded to swap_cgroup of "ent" |
4303 | */ | |
8a9478ca KH |
4304 | void |
4305 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4306 | { |
4307 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4308 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4309 | ||
4310 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4311 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4312 | ||
0030f535 | 4313 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4314 | |
f75ca962 KH |
4315 | /* |
4316 | * record memcg information, if swapout && memcg != NULL, | |
4050377b | 4317 | * css_get() was called in uncharge(). |
f75ca962 KH |
4318 | */ |
4319 | if (do_swap_account && swapout && memcg) | |
34c00c31 | 4320 | swap_cgroup_record(ent, mem_cgroup_id(memcg)); |
8c7c6e34 | 4321 | } |
e767e056 | 4322 | #endif |
8c7c6e34 | 4323 | |
c255a458 | 4324 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4325 | /* |
4326 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4327 | * uncharge "memsw" account. | |
4328 | */ | |
4329 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4330 | { |
8c7c6e34 | 4331 | struct mem_cgroup *memcg; |
a3b2d692 | 4332 | unsigned short id; |
8c7c6e34 KH |
4333 | |
4334 | if (!do_swap_account) | |
4335 | return; | |
4336 | ||
a3b2d692 KH |
4337 | id = swap_cgroup_record(ent, 0); |
4338 | rcu_read_lock(); | |
4339 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4340 | if (memcg) { |
a3b2d692 KH |
4341 | /* |
4342 | * We uncharge this because swap is freed. | |
4343 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4344 | */ | |
0c3e73e8 | 4345 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4346 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4347 | mem_cgroup_swap_statistics(memcg, false); |
4050377b | 4348 | css_put(&memcg->css); |
8c7c6e34 | 4349 | } |
a3b2d692 | 4350 | rcu_read_unlock(); |
d13d1443 | 4351 | } |
02491447 DN |
4352 | |
4353 | /** | |
4354 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4355 | * @entry: swap entry to be moved | |
4356 | * @from: mem_cgroup which the entry is moved from | |
4357 | * @to: mem_cgroup which the entry is moved to | |
4358 | * | |
4359 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4360 | * as the mem_cgroup's id of @from. | |
4361 | * | |
4362 | * Returns 0 on success, -EINVAL on failure. | |
4363 | * | |
4364 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4365 | * both res and memsw, and called css_get(). | |
4366 | */ | |
4367 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4368 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4369 | { |
4370 | unsigned short old_id, new_id; | |
4371 | ||
34c00c31 LZ |
4372 | old_id = mem_cgroup_id(from); |
4373 | new_id = mem_cgroup_id(to); | |
02491447 DN |
4374 | |
4375 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4376 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4377 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4378 | /* |
483c30b5 DN |
4379 | * This function is only called from task migration context now. |
4380 | * It postpones res_counter and refcount handling till the end | |
4381 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
4382 | * improvement. But we cannot postpone css_get(to) because if |
4383 | * the process that has been moved to @to does swap-in, the | |
4384 | * refcount of @to might be decreased to 0. | |
4385 | * | |
4386 | * We are in attach() phase, so the cgroup is guaranteed to be | |
4387 | * alive, so we can just call css_get(). | |
02491447 | 4388 | */ |
4050377b | 4389 | css_get(&to->css); |
02491447 DN |
4390 | return 0; |
4391 | } | |
4392 | return -EINVAL; | |
4393 | } | |
4394 | #else | |
4395 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4396 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4397 | { |
4398 | return -EINVAL; | |
4399 | } | |
8c7c6e34 | 4400 | #endif |
d13d1443 | 4401 | |
ae41be37 | 4402 | /* |
01b1ae63 KH |
4403 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4404 | * page belongs to. | |
ae41be37 | 4405 | */ |
0030f535 JW |
4406 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4407 | struct mem_cgroup **memcgp) | |
ae41be37 | 4408 | { |
c0ff4b85 | 4409 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4410 | unsigned int nr_pages = 1; |
7ec99d62 | 4411 | struct page_cgroup *pc; |
ac39cf8c | 4412 | enum charge_type ctype; |
8869b8f6 | 4413 | |
72835c86 | 4414 | *memcgp = NULL; |
56039efa | 4415 | |
f8d66542 | 4416 | if (mem_cgroup_disabled()) |
0030f535 | 4417 | return; |
4077960e | 4418 | |
b32967ff MG |
4419 | if (PageTransHuge(page)) |
4420 | nr_pages <<= compound_order(page); | |
4421 | ||
52d4b9ac KH |
4422 | pc = lookup_page_cgroup(page); |
4423 | lock_page_cgroup(pc); | |
4424 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4425 | memcg = pc->mem_cgroup; |
4426 | css_get(&memcg->css); | |
ac39cf8c AM |
4427 | /* |
4428 | * At migrating an anonymous page, its mapcount goes down | |
4429 | * to 0 and uncharge() will be called. But, even if it's fully | |
4430 | * unmapped, migration may fail and this page has to be | |
4431 | * charged again. We set MIGRATION flag here and delay uncharge | |
4432 | * until end_migration() is called | |
4433 | * | |
4434 | * Corner Case Thinking | |
4435 | * A) | |
4436 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4437 | * while migration was ongoing. | |
4438 | * If unmap finds the old page, uncharge() of it will be delayed | |
4439 | * until end_migration(). If unmap finds a new page, it's | |
4440 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4441 | * finds swap_migration_entry, the new page will not be mapped | |
4442 | * and end_migration() will find it(mapcount==0). | |
4443 | * | |
4444 | * B) | |
4445 | * When the old page was mapped but migraion fails, the kernel | |
4446 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4447 | * if mapcount goes down to 0. We can do remap successfully | |
4448 | * without charging it again. | |
4449 | * | |
4450 | * C) | |
4451 | * The "old" page is under lock_page() until the end of | |
4452 | * migration, so, the old page itself will not be swapped-out. | |
4453 | * If the new page is swapped out before end_migraton, our | |
4454 | * hook to usual swap-out path will catch the event. | |
4455 | */ | |
4456 | if (PageAnon(page)) | |
4457 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4458 | } |
52d4b9ac | 4459 | unlock_page_cgroup(pc); |
ac39cf8c AM |
4460 | /* |
4461 | * If the page is not charged at this point, | |
4462 | * we return here. | |
4463 | */ | |
c0ff4b85 | 4464 | if (!memcg) |
0030f535 | 4465 | return; |
01b1ae63 | 4466 | |
72835c86 | 4467 | *memcgp = memcg; |
ac39cf8c AM |
4468 | /* |
4469 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4470 | * is called before end_migration, we can catch all events on this new | |
4471 | * page. In the case new page is migrated but not remapped, new page's | |
4472 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4473 | */ | |
ac39cf8c | 4474 | if (PageAnon(page)) |
41326c17 | 4475 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4476 | else |
62ba7442 | 4477 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4478 | /* |
4479 | * The page is committed to the memcg, but it's not actually | |
4480 | * charged to the res_counter since we plan on replacing the | |
4481 | * old one and only one page is going to be left afterwards. | |
4482 | */ | |
b32967ff | 4483 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4484 | } |
8869b8f6 | 4485 | |
69029cd5 | 4486 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4487 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4488 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4489 | { |
ac39cf8c | 4490 | struct page *used, *unused; |
01b1ae63 | 4491 | struct page_cgroup *pc; |
b2402857 | 4492 | bool anon; |
01b1ae63 | 4493 | |
c0ff4b85 | 4494 | if (!memcg) |
01b1ae63 | 4495 | return; |
b25ed609 | 4496 | |
50de1dd9 | 4497 | if (!migration_ok) { |
ac39cf8c AM |
4498 | used = oldpage; |
4499 | unused = newpage; | |
01b1ae63 | 4500 | } else { |
ac39cf8c | 4501 | used = newpage; |
01b1ae63 KH |
4502 | unused = oldpage; |
4503 | } | |
0030f535 | 4504 | anon = PageAnon(used); |
7d188958 JW |
4505 | __mem_cgroup_uncharge_common(unused, |
4506 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4507 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4508 | true); | |
0030f535 | 4509 | css_put(&memcg->css); |
69029cd5 | 4510 | /* |
ac39cf8c AM |
4511 | * We disallowed uncharge of pages under migration because mapcount |
4512 | * of the page goes down to zero, temporarly. | |
4513 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4514 | */ |
ac39cf8c AM |
4515 | pc = lookup_page_cgroup(oldpage); |
4516 | lock_page_cgroup(pc); | |
4517 | ClearPageCgroupMigration(pc); | |
4518 | unlock_page_cgroup(pc); | |
ac39cf8c | 4519 | |
01b1ae63 | 4520 | /* |
ac39cf8c AM |
4521 | * If a page is a file cache, radix-tree replacement is very atomic |
4522 | * and we can skip this check. When it was an Anon page, its mapcount | |
4523 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4524 | * uncharged yet. There are several case but page->mapcount check | |
4525 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4526 | * check. (see prepare_charge() also) | |
69029cd5 | 4527 | */ |
b2402857 | 4528 | if (anon) |
ac39cf8c | 4529 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4530 | } |
78fb7466 | 4531 | |
ab936cbc KH |
4532 | /* |
4533 | * At replace page cache, newpage is not under any memcg but it's on | |
4534 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4535 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4536 | */ | |
4537 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4538 | struct page *newpage) | |
4539 | { | |
bde05d1c | 4540 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4541 | struct page_cgroup *pc; |
ab936cbc | 4542 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4543 | |
4544 | if (mem_cgroup_disabled()) | |
4545 | return; | |
4546 | ||
4547 | pc = lookup_page_cgroup(oldpage); | |
4548 | /* fix accounting on old pages */ | |
4549 | lock_page_cgroup(pc); | |
bde05d1c HD |
4550 | if (PageCgroupUsed(pc)) { |
4551 | memcg = pc->mem_cgroup; | |
b070e65c | 4552 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
bde05d1c HD |
4553 | ClearPageCgroupUsed(pc); |
4554 | } | |
ab936cbc KH |
4555 | unlock_page_cgroup(pc); |
4556 | ||
bde05d1c HD |
4557 | /* |
4558 | * When called from shmem_replace_page(), in some cases the | |
4559 | * oldpage has already been charged, and in some cases not. | |
4560 | */ | |
4561 | if (!memcg) | |
4562 | return; | |
ab936cbc KH |
4563 | /* |
4564 | * Even if newpage->mapping was NULL before starting replacement, | |
4565 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4566 | * LRU while we overwrite pc->mem_cgroup. | |
4567 | */ | |
ce587e65 | 4568 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4569 | } |
4570 | ||
f212ad7c DN |
4571 | #ifdef CONFIG_DEBUG_VM |
4572 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4573 | { | |
4574 | struct page_cgroup *pc; | |
4575 | ||
4576 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4577 | /* |
4578 | * Can be NULL while feeding pages into the page allocator for | |
4579 | * the first time, i.e. during boot or memory hotplug; | |
4580 | * or when mem_cgroup_disabled(). | |
4581 | */ | |
f212ad7c DN |
4582 | if (likely(pc) && PageCgroupUsed(pc)) |
4583 | return pc; | |
4584 | return NULL; | |
4585 | } | |
4586 | ||
4587 | bool mem_cgroup_bad_page_check(struct page *page) | |
4588 | { | |
4589 | if (mem_cgroup_disabled()) | |
4590 | return false; | |
4591 | ||
4592 | return lookup_page_cgroup_used(page) != NULL; | |
4593 | } | |
4594 | ||
4595 | void mem_cgroup_print_bad_page(struct page *page) | |
4596 | { | |
4597 | struct page_cgroup *pc; | |
4598 | ||
4599 | pc = lookup_page_cgroup_used(page); | |
4600 | if (pc) { | |
d045197f AM |
4601 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4602 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4603 | } |
4604 | } | |
4605 | #endif | |
4606 | ||
d38d2a75 | 4607 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4608 | unsigned long long val) |
628f4235 | 4609 | { |
81d39c20 | 4610 | int retry_count; |
3c11ecf4 | 4611 | u64 memswlimit, memlimit; |
628f4235 | 4612 | int ret = 0; |
81d39c20 KH |
4613 | int children = mem_cgroup_count_children(memcg); |
4614 | u64 curusage, oldusage; | |
3c11ecf4 | 4615 | int enlarge; |
81d39c20 KH |
4616 | |
4617 | /* | |
4618 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4619 | * is depends on callers. We set our retry-count to be function | |
4620 | * of # of children which we should visit in this loop. | |
4621 | */ | |
4622 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4623 | ||
4624 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4625 | |
3c11ecf4 | 4626 | enlarge = 0; |
8c7c6e34 | 4627 | while (retry_count) { |
628f4235 KH |
4628 | if (signal_pending(current)) { |
4629 | ret = -EINTR; | |
4630 | break; | |
4631 | } | |
8c7c6e34 KH |
4632 | /* |
4633 | * Rather than hide all in some function, I do this in | |
4634 | * open coded manner. You see what this really does. | |
aaad153e | 4635 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4636 | */ |
4637 | mutex_lock(&set_limit_mutex); | |
4638 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4639 | if (memswlimit < val) { | |
4640 | ret = -EINVAL; | |
4641 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4642 | break; |
4643 | } | |
3c11ecf4 KH |
4644 | |
4645 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4646 | if (memlimit < val) | |
4647 | enlarge = 1; | |
4648 | ||
8c7c6e34 | 4649 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4650 | if (!ret) { |
4651 | if (memswlimit == val) | |
4652 | memcg->memsw_is_minimum = true; | |
4653 | else | |
4654 | memcg->memsw_is_minimum = false; | |
4655 | } | |
8c7c6e34 KH |
4656 | mutex_unlock(&set_limit_mutex); |
4657 | ||
4658 | if (!ret) | |
4659 | break; | |
4660 | ||
5660048c JW |
4661 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4662 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4663 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4664 | /* Usage is reduced ? */ | |
f894ffa8 | 4665 | if (curusage >= oldusage) |
81d39c20 KH |
4666 | retry_count--; |
4667 | else | |
4668 | oldusage = curusage; | |
8c7c6e34 | 4669 | } |
3c11ecf4 KH |
4670 | if (!ret && enlarge) |
4671 | memcg_oom_recover(memcg); | |
14797e23 | 4672 | |
8c7c6e34 KH |
4673 | return ret; |
4674 | } | |
4675 | ||
338c8431 LZ |
4676 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4677 | unsigned long long val) | |
8c7c6e34 | 4678 | { |
81d39c20 | 4679 | int retry_count; |
3c11ecf4 | 4680 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4681 | int children = mem_cgroup_count_children(memcg); |
4682 | int ret = -EBUSY; | |
3c11ecf4 | 4683 | int enlarge = 0; |
8c7c6e34 | 4684 | |
81d39c20 | 4685 | /* see mem_cgroup_resize_res_limit */ |
f894ffa8 | 4686 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
81d39c20 | 4687 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
8c7c6e34 KH |
4688 | while (retry_count) { |
4689 | if (signal_pending(current)) { | |
4690 | ret = -EINTR; | |
4691 | break; | |
4692 | } | |
4693 | /* | |
4694 | * Rather than hide all in some function, I do this in | |
4695 | * open coded manner. You see what this really does. | |
aaad153e | 4696 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4697 | */ |
4698 | mutex_lock(&set_limit_mutex); | |
4699 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4700 | if (memlimit > val) { | |
4701 | ret = -EINVAL; | |
4702 | mutex_unlock(&set_limit_mutex); | |
4703 | break; | |
4704 | } | |
3c11ecf4 KH |
4705 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4706 | if (memswlimit < val) | |
4707 | enlarge = 1; | |
8c7c6e34 | 4708 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4709 | if (!ret) { |
4710 | if (memlimit == val) | |
4711 | memcg->memsw_is_minimum = true; | |
4712 | else | |
4713 | memcg->memsw_is_minimum = false; | |
4714 | } | |
8c7c6e34 KH |
4715 | mutex_unlock(&set_limit_mutex); |
4716 | ||
4717 | if (!ret) | |
4718 | break; | |
4719 | ||
5660048c JW |
4720 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4721 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4722 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4723 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4724 | /* Usage is reduced ? */ |
8c7c6e34 | 4725 | if (curusage >= oldusage) |
628f4235 | 4726 | retry_count--; |
81d39c20 KH |
4727 | else |
4728 | oldusage = curusage; | |
628f4235 | 4729 | } |
3c11ecf4 KH |
4730 | if (!ret && enlarge) |
4731 | memcg_oom_recover(memcg); | |
628f4235 KH |
4732 | return ret; |
4733 | } | |
4734 | ||
0608f43d AM |
4735 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
4736 | gfp_t gfp_mask, | |
4737 | unsigned long *total_scanned) | |
4738 | { | |
4739 | unsigned long nr_reclaimed = 0; | |
4740 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
4741 | unsigned long reclaimed; | |
4742 | int loop = 0; | |
4743 | struct mem_cgroup_tree_per_zone *mctz; | |
4744 | unsigned long long excess; | |
4745 | unsigned long nr_scanned; | |
4746 | ||
4747 | if (order > 0) | |
4748 | return 0; | |
4749 | ||
4750 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
4751 | /* | |
4752 | * This loop can run a while, specially if mem_cgroup's continuously | |
4753 | * keep exceeding their soft limit and putting the system under | |
4754 | * pressure | |
4755 | */ | |
4756 | do { | |
4757 | if (next_mz) | |
4758 | mz = next_mz; | |
4759 | else | |
4760 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
4761 | if (!mz) | |
4762 | break; | |
4763 | ||
4764 | nr_scanned = 0; | |
4765 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
4766 | gfp_mask, &nr_scanned); | |
4767 | nr_reclaimed += reclaimed; | |
4768 | *total_scanned += nr_scanned; | |
4769 | spin_lock(&mctz->lock); | |
4770 | ||
4771 | /* | |
4772 | * If we failed to reclaim anything from this memory cgroup | |
4773 | * it is time to move on to the next cgroup | |
4774 | */ | |
4775 | next_mz = NULL; | |
4776 | if (!reclaimed) { | |
4777 | do { | |
4778 | /* | |
4779 | * Loop until we find yet another one. | |
4780 | * | |
4781 | * By the time we get the soft_limit lock | |
4782 | * again, someone might have aded the | |
4783 | * group back on the RB tree. Iterate to | |
4784 | * make sure we get a different mem. | |
4785 | * mem_cgroup_largest_soft_limit_node returns | |
4786 | * NULL if no other cgroup is present on | |
4787 | * the tree | |
4788 | */ | |
4789 | next_mz = | |
4790 | __mem_cgroup_largest_soft_limit_node(mctz); | |
4791 | if (next_mz == mz) | |
4792 | css_put(&next_mz->memcg->css); | |
4793 | else /* next_mz == NULL or other memcg */ | |
4794 | break; | |
4795 | } while (1); | |
4796 | } | |
4797 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); | |
4798 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4799 | /* | |
4800 | * One school of thought says that we should not add | |
4801 | * back the node to the tree if reclaim returns 0. | |
4802 | * But our reclaim could return 0, simply because due | |
4803 | * to priority we are exposing a smaller subset of | |
4804 | * memory to reclaim from. Consider this as a longer | |
4805 | * term TODO. | |
4806 | */ | |
4807 | /* If excess == 0, no tree ops */ | |
4808 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); | |
4809 | spin_unlock(&mctz->lock); | |
4810 | css_put(&mz->memcg->css); | |
4811 | loop++; | |
4812 | /* | |
4813 | * Could not reclaim anything and there are no more | |
4814 | * mem cgroups to try or we seem to be looping without | |
4815 | * reclaiming anything. | |
4816 | */ | |
4817 | if (!nr_reclaimed && | |
4818 | (next_mz == NULL || | |
4819 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4820 | break; | |
4821 | } while (!nr_reclaimed); | |
4822 | if (next_mz) | |
4823 | css_put(&next_mz->memcg->css); | |
4824 | return nr_reclaimed; | |
4825 | } | |
4826 | ||
2ef37d3f MH |
4827 | /** |
4828 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4829 | * @memcg: group to clear | |
4830 | * @node: NUMA node | |
4831 | * @zid: zone id | |
4832 | * @lru: lru to to clear | |
4833 | * | |
3c935d18 | 4834 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4835 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4836 | * group. | |
cc847582 | 4837 | */ |
2ef37d3f | 4838 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4839 | int node, int zid, enum lru_list lru) |
cc847582 | 4840 | { |
bea8c150 | 4841 | struct lruvec *lruvec; |
2ef37d3f | 4842 | unsigned long flags; |
072c56c1 | 4843 | struct list_head *list; |
925b7673 JW |
4844 | struct page *busy; |
4845 | struct zone *zone; | |
072c56c1 | 4846 | |
08e552c6 | 4847 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4848 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4849 | list = &lruvec->lists[lru]; | |
cc847582 | 4850 | |
f817ed48 | 4851 | busy = NULL; |
2ef37d3f | 4852 | do { |
925b7673 | 4853 | struct page_cgroup *pc; |
5564e88b JW |
4854 | struct page *page; |
4855 | ||
08e552c6 | 4856 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4857 | if (list_empty(list)) { |
08e552c6 | 4858 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4859 | break; |
f817ed48 | 4860 | } |
925b7673 JW |
4861 | page = list_entry(list->prev, struct page, lru); |
4862 | if (busy == page) { | |
4863 | list_move(&page->lru, list); | |
648bcc77 | 4864 | busy = NULL; |
08e552c6 | 4865 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4866 | continue; |
4867 | } | |
08e552c6 | 4868 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4869 | |
925b7673 | 4870 | pc = lookup_page_cgroup(page); |
5564e88b | 4871 | |
3c935d18 | 4872 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4873 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4874 | busy = page; |
f817ed48 KH |
4875 | cond_resched(); |
4876 | } else | |
4877 | busy = NULL; | |
2ef37d3f | 4878 | } while (!list_empty(list)); |
cc847582 KH |
4879 | } |
4880 | ||
4881 | /* | |
c26251f9 MH |
4882 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4883 | * all the charges and pages to the parent. | |
cc847582 | 4884 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4885 | * |
4886 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4887 | */ |
ab5196c2 | 4888 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4889 | { |
c26251f9 | 4890 | int node, zid; |
bea207c8 | 4891 | u64 usage; |
f817ed48 | 4892 | |
fce66477 | 4893 | do { |
52d4b9ac KH |
4894 | /* This is for making all *used* pages to be on LRU. */ |
4895 | lru_add_drain_all(); | |
c0ff4b85 | 4896 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4897 | mem_cgroup_start_move(memcg); |
31aaea4a | 4898 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4899 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4900 | enum lru_list lru; |
4901 | for_each_lru(lru) { | |
2ef37d3f | 4902 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4903 | node, zid, lru); |
f817ed48 | 4904 | } |
1ecaab2b | 4905 | } |
f817ed48 | 4906 | } |
c0ff4b85 R |
4907 | mem_cgroup_end_move(memcg); |
4908 | memcg_oom_recover(memcg); | |
52d4b9ac | 4909 | cond_resched(); |
f817ed48 | 4910 | |
2ef37d3f | 4911 | /* |
bea207c8 GC |
4912 | * Kernel memory may not necessarily be trackable to a specific |
4913 | * process. So they are not migrated, and therefore we can't | |
4914 | * expect their value to drop to 0 here. | |
4915 | * Having res filled up with kmem only is enough. | |
4916 | * | |
2ef37d3f MH |
4917 | * This is a safety check because mem_cgroup_force_empty_list |
4918 | * could have raced with mem_cgroup_replace_page_cache callers | |
4919 | * so the lru seemed empty but the page could have been added | |
4920 | * right after the check. RES_USAGE should be safe as we always | |
4921 | * charge before adding to the LRU. | |
4922 | */ | |
bea207c8 GC |
4923 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4924 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4925 | } while (usage > 0); | |
c26251f9 MH |
4926 | } |
4927 | ||
b5f99b53 GC |
4928 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
4929 | { | |
696ac172 JW |
4930 | lockdep_assert_held(&memcg_create_mutex); |
4931 | /* | |
4932 | * The lock does not prevent addition or deletion to the list | |
4933 | * of children, but it prevents a new child from being | |
4934 | * initialized based on this parent in css_online(), so it's | |
4935 | * enough to decide whether hierarchically inherited | |
4936 | * attributes can still be changed or not. | |
4937 | */ | |
4938 | return memcg->use_hierarchy && | |
4939 | !list_empty(&memcg->css.cgroup->children); | |
b5f99b53 GC |
4940 | } |
4941 | ||
c26251f9 MH |
4942 | /* |
4943 | * Reclaims as many pages from the given memcg as possible and moves | |
4944 | * the rest to the parent. | |
4945 | * | |
4946 | * Caller is responsible for holding css reference for memcg. | |
4947 | */ | |
4948 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4949 | { | |
4950 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
4951 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 4952 | |
c1e862c1 | 4953 | /* returns EBUSY if there is a task or if we come here twice. */ |
07bc356e | 4954 | if (cgroup_has_tasks(cgrp) || !list_empty(&cgrp->children)) |
c26251f9 MH |
4955 | return -EBUSY; |
4956 | ||
c1e862c1 KH |
4957 | /* we call try-to-free pages for make this cgroup empty */ |
4958 | lru_add_drain_all(); | |
f817ed48 | 4959 | /* try to free all pages in this cgroup */ |
569530fb | 4960 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4961 | int progress; |
c1e862c1 | 4962 | |
c26251f9 MH |
4963 | if (signal_pending(current)) |
4964 | return -EINTR; | |
4965 | ||
c0ff4b85 | 4966 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4967 | false); |
c1e862c1 | 4968 | if (!progress) { |
f817ed48 | 4969 | nr_retries--; |
c1e862c1 | 4970 | /* maybe some writeback is necessary */ |
8aa7e847 | 4971 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4972 | } |
f817ed48 KH |
4973 | |
4974 | } | |
08e552c6 | 4975 | lru_add_drain(); |
ab5196c2 MH |
4976 | mem_cgroup_reparent_charges(memcg); |
4977 | ||
4978 | return 0; | |
cc847582 KH |
4979 | } |
4980 | ||
182446d0 TH |
4981 | static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css, |
4982 | unsigned int event) | |
c1e862c1 | 4983 | { |
182446d0 | 4984 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c26251f9 | 4985 | |
d8423011 MH |
4986 | if (mem_cgroup_is_root(memcg)) |
4987 | return -EINVAL; | |
c33bd835 | 4988 | return mem_cgroup_force_empty(memcg); |
c1e862c1 KH |
4989 | } |
4990 | ||
182446d0 TH |
4991 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
4992 | struct cftype *cft) | |
18f59ea7 | 4993 | { |
182446d0 | 4994 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
4995 | } |
4996 | ||
182446d0 TH |
4997 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
4998 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
4999 | { |
5000 | int retval = 0; | |
182446d0 | 5001 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5002 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css)); |
18f59ea7 | 5003 | |
0999821b | 5004 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
5005 | |
5006 | if (memcg->use_hierarchy == val) | |
5007 | goto out; | |
5008 | ||
18f59ea7 | 5009 | /* |
af901ca1 | 5010 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
5011 | * in the child subtrees. If it is unset, then the change can |
5012 | * occur, provided the current cgroup has no children. | |
5013 | * | |
5014 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
5015 | * set if there are no children. | |
5016 | */ | |
c0ff4b85 | 5017 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 5018 | (val == 1 || val == 0)) { |
696ac172 | 5019 | if (list_empty(&memcg->css.cgroup->children)) |
c0ff4b85 | 5020 | memcg->use_hierarchy = val; |
18f59ea7 BS |
5021 | else |
5022 | retval = -EBUSY; | |
5023 | } else | |
5024 | retval = -EINVAL; | |
567fb435 GC |
5025 | |
5026 | out: | |
0999821b | 5027 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
5028 | |
5029 | return retval; | |
5030 | } | |
5031 | ||
0c3e73e8 | 5032 | |
c0ff4b85 | 5033 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 5034 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 5035 | { |
7d74b06f | 5036 | struct mem_cgroup *iter; |
7a159cc9 | 5037 | long val = 0; |
0c3e73e8 | 5038 | |
7a159cc9 | 5039 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 5040 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
5041 | val += mem_cgroup_read_stat(iter, idx); |
5042 | ||
5043 | if (val < 0) /* race ? */ | |
5044 | val = 0; | |
5045 | return val; | |
0c3e73e8 BS |
5046 | } |
5047 | ||
c0ff4b85 | 5048 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 5049 | { |
7d74b06f | 5050 | u64 val; |
104f3928 | 5051 | |
c0ff4b85 | 5052 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 5053 | if (!swap) |
65c64ce8 | 5054 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 5055 | else |
65c64ce8 | 5056 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
5057 | } |
5058 | ||
b070e65c DR |
5059 | /* |
5060 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
5061 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
5062 | */ | |
c0ff4b85 R |
5063 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
5064 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 5065 | |
7d74b06f | 5066 | if (swap) |
bff6bb83 | 5067 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
5068 | |
5069 | return val << PAGE_SHIFT; | |
5070 | } | |
5071 | ||
791badbd TH |
5072 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
5073 | struct cftype *cft) | |
8cdea7c0 | 5074 | { |
182446d0 | 5075 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
104f3928 | 5076 | u64 val; |
791badbd | 5077 | int name; |
86ae53e1 | 5078 | enum res_type type; |
8c7c6e34 KH |
5079 | |
5080 | type = MEMFILE_TYPE(cft->private); | |
5081 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5082 | |
8c7c6e34 KH |
5083 | switch (type) { |
5084 | case _MEM: | |
104f3928 | 5085 | if (name == RES_USAGE) |
c0ff4b85 | 5086 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 5087 | else |
c0ff4b85 | 5088 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
5089 | break; |
5090 | case _MEMSWAP: | |
104f3928 | 5091 | if (name == RES_USAGE) |
c0ff4b85 | 5092 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 5093 | else |
c0ff4b85 | 5094 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 5095 | break; |
510fc4e1 GC |
5096 | case _KMEM: |
5097 | val = res_counter_read_u64(&memcg->kmem, name); | |
5098 | break; | |
8c7c6e34 KH |
5099 | default: |
5100 | BUG(); | |
8c7c6e34 | 5101 | } |
af36f906 | 5102 | |
791badbd | 5103 | return val; |
8cdea7c0 | 5104 | } |
510fc4e1 | 5105 | |
510fc4e1 | 5106 | #ifdef CONFIG_MEMCG_KMEM |
d6441637 VD |
5107 | /* should be called with activate_kmem_mutex held */ |
5108 | static int __memcg_activate_kmem(struct mem_cgroup *memcg, | |
5109 | unsigned long long limit) | |
5110 | { | |
5111 | int err = 0; | |
5112 | int memcg_id; | |
5113 | ||
5114 | if (memcg_kmem_is_active(memcg)) | |
5115 | return 0; | |
5116 | ||
5117 | /* | |
5118 | * We are going to allocate memory for data shared by all memory | |
5119 | * cgroups so let's stop accounting here. | |
5120 | */ | |
5121 | memcg_stop_kmem_account(); | |
5122 | ||
510fc4e1 GC |
5123 | /* |
5124 | * For simplicity, we won't allow this to be disabled. It also can't | |
5125 | * be changed if the cgroup has children already, or if tasks had | |
5126 | * already joined. | |
5127 | * | |
5128 | * If tasks join before we set the limit, a person looking at | |
5129 | * kmem.usage_in_bytes will have no way to determine when it took | |
5130 | * place, which makes the value quite meaningless. | |
5131 | * | |
5132 | * After it first became limited, changes in the value of the limit are | |
5133 | * of course permitted. | |
510fc4e1 | 5134 | */ |
0999821b | 5135 | mutex_lock(&memcg_create_mutex); |
07bc356e | 5136 | if (cgroup_has_tasks(memcg->css.cgroup) || memcg_has_children(memcg)) |
d6441637 VD |
5137 | err = -EBUSY; |
5138 | mutex_unlock(&memcg_create_mutex); | |
5139 | if (err) | |
5140 | goto out; | |
510fc4e1 | 5141 | |
d6441637 VD |
5142 | memcg_id = ida_simple_get(&kmem_limited_groups, |
5143 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
5144 | if (memcg_id < 0) { | |
5145 | err = memcg_id; | |
5146 | goto out; | |
5147 | } | |
5148 | ||
5149 | /* | |
5150 | * Make sure we have enough space for this cgroup in each root cache's | |
5151 | * memcg_params. | |
5152 | */ | |
5153 | err = memcg_update_all_caches(memcg_id + 1); | |
5154 | if (err) | |
5155 | goto out_rmid; | |
5156 | ||
5157 | memcg->kmemcg_id = memcg_id; | |
5158 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
5159 | mutex_init(&memcg->slab_caches_mutex); | |
5160 | ||
5161 | /* | |
5162 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
5163 | * active bit set yet, so this should succeed. | |
5164 | */ | |
5165 | err = res_counter_set_limit(&memcg->kmem, limit); | |
5166 | VM_BUG_ON(err); | |
5167 | ||
5168 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
5169 | /* | |
5170 | * Setting the active bit after enabling static branching will | |
5171 | * guarantee no one starts accounting before all call sites are | |
5172 | * patched. | |
5173 | */ | |
5174 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 5175 | out: |
d6441637 VD |
5176 | memcg_resume_kmem_account(); |
5177 | return err; | |
5178 | ||
5179 | out_rmid: | |
5180 | ida_simple_remove(&kmem_limited_groups, memcg_id); | |
5181 | goto out; | |
5182 | } | |
5183 | ||
5184 | static int memcg_activate_kmem(struct mem_cgroup *memcg, | |
5185 | unsigned long long limit) | |
5186 | { | |
5187 | int ret; | |
5188 | ||
5189 | mutex_lock(&activate_kmem_mutex); | |
5190 | ret = __memcg_activate_kmem(memcg, limit); | |
5191 | mutex_unlock(&activate_kmem_mutex); | |
5192 | return ret; | |
5193 | } | |
5194 | ||
5195 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
5196 | unsigned long long val) | |
5197 | { | |
5198 | int ret; | |
5199 | ||
5200 | if (!memcg_kmem_is_active(memcg)) | |
5201 | ret = memcg_activate_kmem(memcg, val); | |
5202 | else | |
5203 | ret = res_counter_set_limit(&memcg->kmem, val); | |
510fc4e1 GC |
5204 | return ret; |
5205 | } | |
5206 | ||
55007d84 | 5207 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 5208 | { |
55007d84 | 5209 | int ret = 0; |
510fc4e1 | 5210 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 5211 | |
d6441637 VD |
5212 | if (!parent) |
5213 | return 0; | |
55007d84 | 5214 | |
d6441637 | 5215 | mutex_lock(&activate_kmem_mutex); |
55007d84 | 5216 | /* |
d6441637 VD |
5217 | * If the parent cgroup is not kmem-active now, it cannot be activated |
5218 | * after this point, because it has at least one child already. | |
55007d84 | 5219 | */ |
d6441637 VD |
5220 | if (memcg_kmem_is_active(parent)) |
5221 | ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX); | |
5222 | mutex_unlock(&activate_kmem_mutex); | |
55007d84 | 5223 | return ret; |
510fc4e1 | 5224 | } |
d6441637 VD |
5225 | #else |
5226 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
5227 | unsigned long long val) | |
5228 | { | |
5229 | return -EINVAL; | |
5230 | } | |
6d043990 | 5231 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 5232 | |
628f4235 KH |
5233 | /* |
5234 | * The user of this function is... | |
5235 | * RES_LIMIT. | |
5236 | */ | |
182446d0 | 5237 | static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft, |
4d3bb511 | 5238 | char *buffer) |
8cdea7c0 | 5239 | { |
182446d0 | 5240 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5241 | enum res_type type; |
5242 | int name; | |
628f4235 KH |
5243 | unsigned long long val; |
5244 | int ret; | |
5245 | ||
8c7c6e34 KH |
5246 | type = MEMFILE_TYPE(cft->private); |
5247 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5248 | |
8c7c6e34 | 5249 | switch (name) { |
628f4235 | 5250 | case RES_LIMIT: |
4b3bde4c BS |
5251 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
5252 | ret = -EINVAL; | |
5253 | break; | |
5254 | } | |
628f4235 KH |
5255 | /* This function does all necessary parse...reuse it */ |
5256 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
5257 | if (ret) |
5258 | break; | |
5259 | if (type == _MEM) | |
628f4235 | 5260 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 5261 | else if (type == _MEMSWAP) |
8c7c6e34 | 5262 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 5263 | else if (type == _KMEM) |
d6441637 | 5264 | ret = memcg_update_kmem_limit(memcg, val); |
510fc4e1 GC |
5265 | else |
5266 | return -EINVAL; | |
628f4235 | 5267 | break; |
296c81d8 BS |
5268 | case RES_SOFT_LIMIT: |
5269 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
5270 | if (ret) | |
5271 | break; | |
5272 | /* | |
5273 | * For memsw, soft limits are hard to implement in terms | |
5274 | * of semantics, for now, we support soft limits for | |
5275 | * control without swap | |
5276 | */ | |
5277 | if (type == _MEM) | |
5278 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5279 | else | |
5280 | ret = -EINVAL; | |
5281 | break; | |
628f4235 KH |
5282 | default: |
5283 | ret = -EINVAL; /* should be BUG() ? */ | |
5284 | break; | |
5285 | } | |
5286 | return ret; | |
8cdea7c0 BS |
5287 | } |
5288 | ||
fee7b548 KH |
5289 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5290 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5291 | { | |
fee7b548 KH |
5292 | unsigned long long min_limit, min_memsw_limit, tmp; |
5293 | ||
5294 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5295 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
5296 | if (!memcg->use_hierarchy) |
5297 | goto out; | |
5298 | ||
63876986 TH |
5299 | while (css_parent(&memcg->css)) { |
5300 | memcg = mem_cgroup_from_css(css_parent(&memcg->css)); | |
fee7b548 KH |
5301 | if (!memcg->use_hierarchy) |
5302 | break; | |
5303 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5304 | min_limit = min(min_limit, tmp); | |
5305 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5306 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5307 | } | |
5308 | out: | |
5309 | *mem_limit = min_limit; | |
5310 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5311 | } |
5312 | ||
182446d0 | 5313 | static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event) |
c84872e1 | 5314 | { |
182446d0 | 5315 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5316 | int name; |
5317 | enum res_type type; | |
c84872e1 | 5318 | |
8c7c6e34 KH |
5319 | type = MEMFILE_TYPE(event); |
5320 | name = MEMFILE_ATTR(event); | |
af36f906 | 5321 | |
8c7c6e34 | 5322 | switch (name) { |
29f2a4da | 5323 | case RES_MAX_USAGE: |
8c7c6e34 | 5324 | if (type == _MEM) |
c0ff4b85 | 5325 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5326 | else if (type == _MEMSWAP) |
c0ff4b85 | 5327 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5328 | else if (type == _KMEM) |
5329 | res_counter_reset_max(&memcg->kmem); | |
5330 | else | |
5331 | return -EINVAL; | |
29f2a4da PE |
5332 | break; |
5333 | case RES_FAILCNT: | |
8c7c6e34 | 5334 | if (type == _MEM) |
c0ff4b85 | 5335 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5336 | else if (type == _MEMSWAP) |
c0ff4b85 | 5337 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5338 | else if (type == _KMEM) |
5339 | res_counter_reset_failcnt(&memcg->kmem); | |
5340 | else | |
5341 | return -EINVAL; | |
29f2a4da PE |
5342 | break; |
5343 | } | |
f64c3f54 | 5344 | |
85cc59db | 5345 | return 0; |
c84872e1 PE |
5346 | } |
5347 | ||
182446d0 | 5348 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5349 | struct cftype *cft) |
5350 | { | |
182446d0 | 5351 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
5352 | } |
5353 | ||
02491447 | 5354 | #ifdef CONFIG_MMU |
182446d0 | 5355 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5356 | struct cftype *cft, u64 val) |
5357 | { | |
182446d0 | 5358 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
5359 | |
5360 | if (val >= (1 << NR_MOVE_TYPE)) | |
5361 | return -EINVAL; | |
ee5e8472 | 5362 | |
7dc74be0 | 5363 | /* |
ee5e8472 GC |
5364 | * No kind of locking is needed in here, because ->can_attach() will |
5365 | * check this value once in the beginning of the process, and then carry | |
5366 | * on with stale data. This means that changes to this value will only | |
5367 | * affect task migrations starting after the change. | |
7dc74be0 | 5368 | */ |
c0ff4b85 | 5369 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5370 | return 0; |
5371 | } | |
02491447 | 5372 | #else |
182446d0 | 5373 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
5374 | struct cftype *cft, u64 val) |
5375 | { | |
5376 | return -ENOSYS; | |
5377 | } | |
5378 | #endif | |
7dc74be0 | 5379 | |
406eb0c9 | 5380 | #ifdef CONFIG_NUMA |
2da8ca82 | 5381 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 5382 | { |
25485de6 GT |
5383 | struct numa_stat { |
5384 | const char *name; | |
5385 | unsigned int lru_mask; | |
5386 | }; | |
5387 | ||
5388 | static const struct numa_stat stats[] = { | |
5389 | { "total", LRU_ALL }, | |
5390 | { "file", LRU_ALL_FILE }, | |
5391 | { "anon", LRU_ALL_ANON }, | |
5392 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
5393 | }; | |
5394 | const struct numa_stat *stat; | |
406eb0c9 | 5395 | int nid; |
25485de6 | 5396 | unsigned long nr; |
2da8ca82 | 5397 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 5398 | |
25485de6 GT |
5399 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
5400 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
5401 | seq_printf(m, "%s=%lu", stat->name, nr); | |
5402 | for_each_node_state(nid, N_MEMORY) { | |
5403 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
5404 | stat->lru_mask); | |
5405 | seq_printf(m, " N%d=%lu", nid, nr); | |
5406 | } | |
5407 | seq_putc(m, '\n'); | |
406eb0c9 | 5408 | } |
406eb0c9 | 5409 | |
071aee13 YH |
5410 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
5411 | struct mem_cgroup *iter; | |
5412 | ||
5413 | nr = 0; | |
5414 | for_each_mem_cgroup_tree(iter, memcg) | |
5415 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
5416 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
5417 | for_each_node_state(nid, N_MEMORY) { | |
5418 | nr = 0; | |
5419 | for_each_mem_cgroup_tree(iter, memcg) | |
5420 | nr += mem_cgroup_node_nr_lru_pages( | |
5421 | iter, nid, stat->lru_mask); | |
5422 | seq_printf(m, " N%d=%lu", nid, nr); | |
5423 | } | |
5424 | seq_putc(m, '\n'); | |
406eb0c9 | 5425 | } |
406eb0c9 | 5426 | |
406eb0c9 YH |
5427 | return 0; |
5428 | } | |
5429 | #endif /* CONFIG_NUMA */ | |
5430 | ||
af7c4b0e JW |
5431 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5432 | { | |
5433 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5434 | } | |
5435 | ||
2da8ca82 | 5436 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 5437 | { |
2da8ca82 | 5438 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
af7c4b0e JW |
5439 | struct mem_cgroup *mi; |
5440 | unsigned int i; | |
406eb0c9 | 5441 | |
af7c4b0e | 5442 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5443 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5444 | continue; |
af7c4b0e JW |
5445 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5446 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5447 | } |
7b854121 | 5448 | |
af7c4b0e JW |
5449 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5450 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5451 | mem_cgroup_read_events(memcg, i)); | |
5452 | ||
5453 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5454 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5455 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5456 | ||
14067bb3 | 5457 | /* Hierarchical information */ |
fee7b548 KH |
5458 | { |
5459 | unsigned long long limit, memsw_limit; | |
d79154bb | 5460 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5461 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5462 | if (do_swap_account) |
78ccf5b5 JW |
5463 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5464 | memsw_limit); | |
fee7b548 | 5465 | } |
7f016ee8 | 5466 | |
af7c4b0e JW |
5467 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5468 | long long val = 0; | |
5469 | ||
bff6bb83 | 5470 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5471 | continue; |
af7c4b0e JW |
5472 | for_each_mem_cgroup_tree(mi, memcg) |
5473 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5474 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5475 | } | |
5476 | ||
5477 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5478 | unsigned long long val = 0; | |
5479 | ||
5480 | for_each_mem_cgroup_tree(mi, memcg) | |
5481 | val += mem_cgroup_read_events(mi, i); | |
5482 | seq_printf(m, "total_%s %llu\n", | |
5483 | mem_cgroup_events_names[i], val); | |
5484 | } | |
5485 | ||
5486 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5487 | unsigned long long val = 0; | |
5488 | ||
5489 | for_each_mem_cgroup_tree(mi, memcg) | |
5490 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5491 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5492 | } |
14067bb3 | 5493 | |
7f016ee8 | 5494 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5495 | { |
5496 | int nid, zid; | |
5497 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5498 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5499 | unsigned long recent_rotated[2] = {0, 0}; |
5500 | unsigned long recent_scanned[2] = {0, 0}; | |
5501 | ||
5502 | for_each_online_node(nid) | |
5503 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5504 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5505 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5506 | |
89abfab1 HD |
5507 | recent_rotated[0] += rstat->recent_rotated[0]; |
5508 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5509 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5510 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5511 | } |
78ccf5b5 JW |
5512 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5513 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5514 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5515 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5516 | } |
5517 | #endif | |
5518 | ||
d2ceb9b7 KH |
5519 | return 0; |
5520 | } | |
5521 | ||
182446d0 TH |
5522 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
5523 | struct cftype *cft) | |
a7885eb8 | 5524 | { |
182446d0 | 5525 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 5526 | |
1f4c025b | 5527 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5528 | } |
5529 | ||
182446d0 TH |
5530 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
5531 | struct cftype *cft, u64 val) | |
a7885eb8 | 5532 | { |
182446d0 | 5533 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5534 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
a7885eb8 | 5535 | |
63876986 | 5536 | if (val > 100 || !parent) |
a7885eb8 KM |
5537 | return -EINVAL; |
5538 | ||
0999821b | 5539 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5540 | |
a7885eb8 | 5541 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5542 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5543 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5544 | return -EINVAL; |
068b38c1 | 5545 | } |
a7885eb8 | 5546 | |
a7885eb8 | 5547 | memcg->swappiness = val; |
a7885eb8 | 5548 | |
0999821b | 5549 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5550 | |
a7885eb8 KM |
5551 | return 0; |
5552 | } | |
5553 | ||
2e72b634 KS |
5554 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5555 | { | |
5556 | struct mem_cgroup_threshold_ary *t; | |
5557 | u64 usage; | |
5558 | int i; | |
5559 | ||
5560 | rcu_read_lock(); | |
5561 | if (!swap) | |
2c488db2 | 5562 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5563 | else |
2c488db2 | 5564 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5565 | |
5566 | if (!t) | |
5567 | goto unlock; | |
5568 | ||
5569 | usage = mem_cgroup_usage(memcg, swap); | |
5570 | ||
5571 | /* | |
748dad36 | 5572 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5573 | * If it's not true, a threshold was crossed after last |
5574 | * call of __mem_cgroup_threshold(). | |
5575 | */ | |
5407a562 | 5576 | i = t->current_threshold; |
2e72b634 KS |
5577 | |
5578 | /* | |
5579 | * Iterate backward over array of thresholds starting from | |
5580 | * current_threshold and check if a threshold is crossed. | |
5581 | * If none of thresholds below usage is crossed, we read | |
5582 | * only one element of the array here. | |
5583 | */ | |
5584 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5585 | eventfd_signal(t->entries[i].eventfd, 1); | |
5586 | ||
5587 | /* i = current_threshold + 1 */ | |
5588 | i++; | |
5589 | ||
5590 | /* | |
5591 | * Iterate forward over array of thresholds starting from | |
5592 | * current_threshold+1 and check if a threshold is crossed. | |
5593 | * If none of thresholds above usage is crossed, we read | |
5594 | * only one element of the array here. | |
5595 | */ | |
5596 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5597 | eventfd_signal(t->entries[i].eventfd, 1); | |
5598 | ||
5599 | /* Update current_threshold */ | |
5407a562 | 5600 | t->current_threshold = i - 1; |
2e72b634 KS |
5601 | unlock: |
5602 | rcu_read_unlock(); | |
5603 | } | |
5604 | ||
5605 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5606 | { | |
ad4ca5f4 KS |
5607 | while (memcg) { |
5608 | __mem_cgroup_threshold(memcg, false); | |
5609 | if (do_swap_account) | |
5610 | __mem_cgroup_threshold(memcg, true); | |
5611 | ||
5612 | memcg = parent_mem_cgroup(memcg); | |
5613 | } | |
2e72b634 KS |
5614 | } |
5615 | ||
5616 | static int compare_thresholds(const void *a, const void *b) | |
5617 | { | |
5618 | const struct mem_cgroup_threshold *_a = a; | |
5619 | const struct mem_cgroup_threshold *_b = b; | |
5620 | ||
2bff24a3 GT |
5621 | if (_a->threshold > _b->threshold) |
5622 | return 1; | |
5623 | ||
5624 | if (_a->threshold < _b->threshold) | |
5625 | return -1; | |
5626 | ||
5627 | return 0; | |
2e72b634 KS |
5628 | } |
5629 | ||
c0ff4b85 | 5630 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5631 | { |
5632 | struct mem_cgroup_eventfd_list *ev; | |
5633 | ||
c0ff4b85 | 5634 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5635 | eventfd_signal(ev->eventfd, 1); |
5636 | return 0; | |
5637 | } | |
5638 | ||
c0ff4b85 | 5639 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5640 | { |
7d74b06f KH |
5641 | struct mem_cgroup *iter; |
5642 | ||
c0ff4b85 | 5643 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5644 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5645 | } |
5646 | ||
59b6f873 | 5647 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 5648 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 5649 | { |
2c488db2 KS |
5650 | struct mem_cgroup_thresholds *thresholds; |
5651 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 5652 | u64 threshold, usage; |
2c488db2 | 5653 | int i, size, ret; |
2e72b634 KS |
5654 | |
5655 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5656 | if (ret) | |
5657 | return ret; | |
5658 | ||
5659 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5660 | |
2e72b634 | 5661 | if (type == _MEM) |
2c488db2 | 5662 | thresholds = &memcg->thresholds; |
2e72b634 | 5663 | else if (type == _MEMSWAP) |
2c488db2 | 5664 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5665 | else |
5666 | BUG(); | |
5667 | ||
5668 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5669 | ||
5670 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5671 | if (thresholds->primary) |
2e72b634 KS |
5672 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5673 | ||
2c488db2 | 5674 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5675 | |
5676 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5677 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5678 | GFP_KERNEL); |
2c488db2 | 5679 | if (!new) { |
2e72b634 KS |
5680 | ret = -ENOMEM; |
5681 | goto unlock; | |
5682 | } | |
2c488db2 | 5683 | new->size = size; |
2e72b634 KS |
5684 | |
5685 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5686 | if (thresholds->primary) { |
5687 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5688 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5689 | } |
5690 | ||
2e72b634 | 5691 | /* Add new threshold */ |
2c488db2 KS |
5692 | new->entries[size - 1].eventfd = eventfd; |
5693 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5694 | |
5695 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5696 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5697 | compare_thresholds, NULL); |
5698 | ||
5699 | /* Find current threshold */ | |
2c488db2 | 5700 | new->current_threshold = -1; |
2e72b634 | 5701 | for (i = 0; i < size; i++) { |
748dad36 | 5702 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5703 | /* |
2c488db2 KS |
5704 | * new->current_threshold will not be used until |
5705 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5706 | * it here. |
5707 | */ | |
2c488db2 | 5708 | ++new->current_threshold; |
748dad36 SZ |
5709 | } else |
5710 | break; | |
2e72b634 KS |
5711 | } |
5712 | ||
2c488db2 KS |
5713 | /* Free old spare buffer and save old primary buffer as spare */ |
5714 | kfree(thresholds->spare); | |
5715 | thresholds->spare = thresholds->primary; | |
5716 | ||
5717 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5718 | |
907860ed | 5719 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5720 | synchronize_rcu(); |
5721 | ||
2e72b634 KS |
5722 | unlock: |
5723 | mutex_unlock(&memcg->thresholds_lock); | |
5724 | ||
5725 | return ret; | |
5726 | } | |
5727 | ||
59b6f873 | 5728 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5729 | struct eventfd_ctx *eventfd, const char *args) |
5730 | { | |
59b6f873 | 5731 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
5732 | } |
5733 | ||
59b6f873 | 5734 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5735 | struct eventfd_ctx *eventfd, const char *args) |
5736 | { | |
59b6f873 | 5737 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
5738 | } |
5739 | ||
59b6f873 | 5740 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 5741 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 5742 | { |
2c488db2 KS |
5743 | struct mem_cgroup_thresholds *thresholds; |
5744 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 5745 | u64 usage; |
2c488db2 | 5746 | int i, j, size; |
2e72b634 KS |
5747 | |
5748 | mutex_lock(&memcg->thresholds_lock); | |
5749 | if (type == _MEM) | |
2c488db2 | 5750 | thresholds = &memcg->thresholds; |
2e72b634 | 5751 | else if (type == _MEMSWAP) |
2c488db2 | 5752 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5753 | else |
5754 | BUG(); | |
5755 | ||
371528ca AV |
5756 | if (!thresholds->primary) |
5757 | goto unlock; | |
5758 | ||
2e72b634 KS |
5759 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5760 | ||
5761 | /* Check if a threshold crossed before removing */ | |
5762 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5763 | ||
5764 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5765 | size = 0; |
5766 | for (i = 0; i < thresholds->primary->size; i++) { | |
5767 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5768 | size++; |
5769 | } | |
5770 | ||
2c488db2 | 5771 | new = thresholds->spare; |
907860ed | 5772 | |
2e72b634 KS |
5773 | /* Set thresholds array to NULL if we don't have thresholds */ |
5774 | if (!size) { | |
2c488db2 KS |
5775 | kfree(new); |
5776 | new = NULL; | |
907860ed | 5777 | goto swap_buffers; |
2e72b634 KS |
5778 | } |
5779 | ||
2c488db2 | 5780 | new->size = size; |
2e72b634 KS |
5781 | |
5782 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5783 | new->current_threshold = -1; |
5784 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5785 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5786 | continue; |
5787 | ||
2c488db2 | 5788 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5789 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5790 | /* |
2c488db2 | 5791 | * new->current_threshold will not be used |
2e72b634 KS |
5792 | * until rcu_assign_pointer(), so it's safe to increment |
5793 | * it here. | |
5794 | */ | |
2c488db2 | 5795 | ++new->current_threshold; |
2e72b634 KS |
5796 | } |
5797 | j++; | |
5798 | } | |
5799 | ||
907860ed | 5800 | swap_buffers: |
2c488db2 KS |
5801 | /* Swap primary and spare array */ |
5802 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5803 | /* If all events are unregistered, free the spare array */ |
5804 | if (!new) { | |
5805 | kfree(thresholds->spare); | |
5806 | thresholds->spare = NULL; | |
5807 | } | |
5808 | ||
2c488db2 | 5809 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5810 | |
907860ed | 5811 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5812 | synchronize_rcu(); |
371528ca | 5813 | unlock: |
2e72b634 | 5814 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5815 | } |
c1e862c1 | 5816 | |
59b6f873 | 5817 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5818 | struct eventfd_ctx *eventfd) |
5819 | { | |
59b6f873 | 5820 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
5821 | } |
5822 | ||
59b6f873 | 5823 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5824 | struct eventfd_ctx *eventfd) |
5825 | { | |
59b6f873 | 5826 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
5827 | } |
5828 | ||
59b6f873 | 5829 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 5830 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 5831 | { |
9490ff27 | 5832 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 5833 | |
9490ff27 KH |
5834 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
5835 | if (!event) | |
5836 | return -ENOMEM; | |
5837 | ||
1af8efe9 | 5838 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5839 | |
5840 | event->eventfd = eventfd; | |
5841 | list_add(&event->list, &memcg->oom_notify); | |
5842 | ||
5843 | /* already in OOM ? */ | |
79dfdacc | 5844 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5845 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5846 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5847 | |
5848 | return 0; | |
5849 | } | |
5850 | ||
59b6f873 | 5851 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 5852 | struct eventfd_ctx *eventfd) |
9490ff27 | 5853 | { |
9490ff27 | 5854 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 5855 | |
1af8efe9 | 5856 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5857 | |
c0ff4b85 | 5858 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5859 | if (ev->eventfd == eventfd) { |
5860 | list_del(&ev->list); | |
5861 | kfree(ev); | |
5862 | } | |
5863 | } | |
5864 | ||
1af8efe9 | 5865 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5866 | } |
5867 | ||
2da8ca82 | 5868 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 5869 | { |
2da8ca82 | 5870 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 5871 | |
791badbd TH |
5872 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
5873 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
5874 | return 0; |
5875 | } | |
5876 | ||
182446d0 | 5877 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5878 | struct cftype *cft, u64 val) |
5879 | { | |
182446d0 | 5880 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5881 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
3c11ecf4 KH |
5882 | |
5883 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
63876986 | 5884 | if (!parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
5885 | return -EINVAL; |
5886 | ||
0999821b | 5887 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5888 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5889 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5890 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5891 | return -EINVAL; |
5892 | } | |
c0ff4b85 | 5893 | memcg->oom_kill_disable = val; |
4d845ebf | 5894 | if (!val) |
c0ff4b85 | 5895 | memcg_oom_recover(memcg); |
0999821b | 5896 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5897 | return 0; |
5898 | } | |
5899 | ||
c255a458 | 5900 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5901 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5902 | { |
55007d84 GC |
5903 | int ret; |
5904 | ||
2633d7a0 | 5905 | memcg->kmemcg_id = -1; |
55007d84 GC |
5906 | ret = memcg_propagate_kmem(memcg); |
5907 | if (ret) | |
5908 | return ret; | |
2633d7a0 | 5909 | |
1d62e436 | 5910 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5911 | } |
e5671dfa | 5912 | |
10d5ebf4 | 5913 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 5914 | { |
1d62e436 | 5915 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
5916 | } |
5917 | ||
5918 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
5919 | { | |
5920 | if (!memcg_kmem_is_active(memcg)) | |
5921 | return; | |
5922 | ||
5923 | /* | |
5924 | * kmem charges can outlive the cgroup. In the case of slab | |
5925 | * pages, for instance, a page contain objects from various | |
5926 | * processes. As we prevent from taking a reference for every | |
5927 | * such allocation we have to be careful when doing uncharge | |
5928 | * (see memcg_uncharge_kmem) and here during offlining. | |
5929 | * | |
5930 | * The idea is that that only the _last_ uncharge which sees | |
5931 | * the dead memcg will drop the last reference. An additional | |
5932 | * reference is taken here before the group is marked dead | |
5933 | * which is then paired with css_put during uncharge resp. here. | |
5934 | * | |
5935 | * Although this might sound strange as this path is called from | |
5936 | * css_offline() when the referencemight have dropped down to 0 | |
5937 | * and shouldn't be incremented anymore (css_tryget would fail) | |
5938 | * we do not have other options because of the kmem allocations | |
5939 | * lifetime. | |
5940 | */ | |
5941 | css_get(&memcg->css); | |
7de37682 GC |
5942 | |
5943 | memcg_kmem_mark_dead(memcg); | |
5944 | ||
5945 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5946 | return; | |
5947 | ||
7de37682 | 5948 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5949 | css_put(&memcg->css); |
d1a4c0b3 | 5950 | } |
e5671dfa | 5951 | #else |
cbe128e3 | 5952 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5953 | { |
5954 | return 0; | |
5955 | } | |
d1a4c0b3 | 5956 | |
10d5ebf4 LZ |
5957 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
5958 | { | |
5959 | } | |
5960 | ||
5961 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
5962 | { |
5963 | } | |
e5671dfa GC |
5964 | #endif |
5965 | ||
3bc942f3 TH |
5966 | /* |
5967 | * DO NOT USE IN NEW FILES. | |
5968 | * | |
5969 | * "cgroup.event_control" implementation. | |
5970 | * | |
5971 | * This is way over-engineered. It tries to support fully configurable | |
5972 | * events for each user. Such level of flexibility is completely | |
5973 | * unnecessary especially in the light of the planned unified hierarchy. | |
5974 | * | |
5975 | * Please deprecate this and replace with something simpler if at all | |
5976 | * possible. | |
5977 | */ | |
5978 | ||
79bd9814 TH |
5979 | /* |
5980 | * Unregister event and free resources. | |
5981 | * | |
5982 | * Gets called from workqueue. | |
5983 | */ | |
3bc942f3 | 5984 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 5985 | { |
3bc942f3 TH |
5986 | struct mem_cgroup_event *event = |
5987 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 5988 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
5989 | |
5990 | remove_wait_queue(event->wqh, &event->wait); | |
5991 | ||
59b6f873 | 5992 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
5993 | |
5994 | /* Notify userspace the event is going away. */ | |
5995 | eventfd_signal(event->eventfd, 1); | |
5996 | ||
5997 | eventfd_ctx_put(event->eventfd); | |
5998 | kfree(event); | |
59b6f873 | 5999 | css_put(&memcg->css); |
79bd9814 TH |
6000 | } |
6001 | ||
6002 | /* | |
6003 | * Gets called on POLLHUP on eventfd when user closes it. | |
6004 | * | |
6005 | * Called with wqh->lock held and interrupts disabled. | |
6006 | */ | |
3bc942f3 TH |
6007 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
6008 | int sync, void *key) | |
79bd9814 | 6009 | { |
3bc942f3 TH |
6010 | struct mem_cgroup_event *event = |
6011 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 6012 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
6013 | unsigned long flags = (unsigned long)key; |
6014 | ||
6015 | if (flags & POLLHUP) { | |
6016 | /* | |
6017 | * If the event has been detached at cgroup removal, we | |
6018 | * can simply return knowing the other side will cleanup | |
6019 | * for us. | |
6020 | * | |
6021 | * We can't race against event freeing since the other | |
6022 | * side will require wqh->lock via remove_wait_queue(), | |
6023 | * which we hold. | |
6024 | */ | |
fba94807 | 6025 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
6026 | if (!list_empty(&event->list)) { |
6027 | list_del_init(&event->list); | |
6028 | /* | |
6029 | * We are in atomic context, but cgroup_event_remove() | |
6030 | * may sleep, so we have to call it in workqueue. | |
6031 | */ | |
6032 | schedule_work(&event->remove); | |
6033 | } | |
fba94807 | 6034 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
6035 | } |
6036 | ||
6037 | return 0; | |
6038 | } | |
6039 | ||
3bc942f3 | 6040 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
6041 | wait_queue_head_t *wqh, poll_table *pt) |
6042 | { | |
3bc942f3 TH |
6043 | struct mem_cgroup_event *event = |
6044 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
6045 | |
6046 | event->wqh = wqh; | |
6047 | add_wait_queue(wqh, &event->wait); | |
6048 | } | |
6049 | ||
6050 | /* | |
3bc942f3 TH |
6051 | * DO NOT USE IN NEW FILES. |
6052 | * | |
79bd9814 TH |
6053 | * Parse input and register new cgroup event handler. |
6054 | * | |
6055 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
6056 | * Interpretation of args is defined by control file implementation. | |
6057 | */ | |
3bc942f3 | 6058 | static int memcg_write_event_control(struct cgroup_subsys_state *css, |
4d3bb511 | 6059 | struct cftype *cft, char *buffer) |
79bd9814 | 6060 | { |
fba94807 | 6061 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 6062 | struct mem_cgroup_event *event; |
79bd9814 TH |
6063 | struct cgroup_subsys_state *cfile_css; |
6064 | unsigned int efd, cfd; | |
6065 | struct fd efile; | |
6066 | struct fd cfile; | |
fba94807 | 6067 | const char *name; |
79bd9814 TH |
6068 | char *endp; |
6069 | int ret; | |
6070 | ||
6071 | efd = simple_strtoul(buffer, &endp, 10); | |
6072 | if (*endp != ' ') | |
6073 | return -EINVAL; | |
6074 | buffer = endp + 1; | |
6075 | ||
6076 | cfd = simple_strtoul(buffer, &endp, 10); | |
6077 | if ((*endp != ' ') && (*endp != '\0')) | |
6078 | return -EINVAL; | |
6079 | buffer = endp + 1; | |
6080 | ||
6081 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
6082 | if (!event) | |
6083 | return -ENOMEM; | |
6084 | ||
59b6f873 | 6085 | event->memcg = memcg; |
79bd9814 | 6086 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
6087 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
6088 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
6089 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
6090 | |
6091 | efile = fdget(efd); | |
6092 | if (!efile.file) { | |
6093 | ret = -EBADF; | |
6094 | goto out_kfree; | |
6095 | } | |
6096 | ||
6097 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
6098 | if (IS_ERR(event->eventfd)) { | |
6099 | ret = PTR_ERR(event->eventfd); | |
6100 | goto out_put_efile; | |
6101 | } | |
6102 | ||
6103 | cfile = fdget(cfd); | |
6104 | if (!cfile.file) { | |
6105 | ret = -EBADF; | |
6106 | goto out_put_eventfd; | |
6107 | } | |
6108 | ||
6109 | /* the process need read permission on control file */ | |
6110 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
6111 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
6112 | if (ret < 0) | |
6113 | goto out_put_cfile; | |
6114 | ||
fba94807 TH |
6115 | /* |
6116 | * Determine the event callbacks and set them in @event. This used | |
6117 | * to be done via struct cftype but cgroup core no longer knows | |
6118 | * about these events. The following is crude but the whole thing | |
6119 | * is for compatibility anyway. | |
3bc942f3 TH |
6120 | * |
6121 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
6122 | */ |
6123 | name = cfile.file->f_dentry->d_name.name; | |
6124 | ||
6125 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
6126 | event->register_event = mem_cgroup_usage_register_event; | |
6127 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
6128 | } else if (!strcmp(name, "memory.oom_control")) { | |
6129 | event->register_event = mem_cgroup_oom_register_event; | |
6130 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
6131 | } else if (!strcmp(name, "memory.pressure_level")) { | |
6132 | event->register_event = vmpressure_register_event; | |
6133 | event->unregister_event = vmpressure_unregister_event; | |
6134 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
6135 | event->register_event = memsw_cgroup_usage_register_event; |
6136 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
6137 | } else { |
6138 | ret = -EINVAL; | |
6139 | goto out_put_cfile; | |
6140 | } | |
6141 | ||
79bd9814 | 6142 | /* |
b5557c4c TH |
6143 | * Verify @cfile should belong to @css. Also, remaining events are |
6144 | * automatically removed on cgroup destruction but the removal is | |
6145 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 6146 | */ |
5a17f543 TH |
6147 | cfile_css = css_tryget_from_dir(cfile.file->f_dentry->d_parent, |
6148 | &memory_cgrp_subsys); | |
79bd9814 | 6149 | ret = -EINVAL; |
5a17f543 | 6150 | if (IS_ERR(cfile_css)) |
79bd9814 | 6151 | goto out_put_cfile; |
5a17f543 TH |
6152 | if (cfile_css != css) { |
6153 | css_put(cfile_css); | |
79bd9814 | 6154 | goto out_put_cfile; |
5a17f543 | 6155 | } |
79bd9814 | 6156 | |
59b6f873 | 6157 | ret = event->register_event(memcg, event->eventfd, buffer); |
79bd9814 TH |
6158 | if (ret) |
6159 | goto out_put_css; | |
6160 | ||
6161 | efile.file->f_op->poll(efile.file, &event->pt); | |
6162 | ||
fba94807 TH |
6163 | spin_lock(&memcg->event_list_lock); |
6164 | list_add(&event->list, &memcg->event_list); | |
6165 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
6166 | |
6167 | fdput(cfile); | |
6168 | fdput(efile); | |
6169 | ||
6170 | return 0; | |
6171 | ||
6172 | out_put_css: | |
b5557c4c | 6173 | css_put(css); |
79bd9814 TH |
6174 | out_put_cfile: |
6175 | fdput(cfile); | |
6176 | out_put_eventfd: | |
6177 | eventfd_ctx_put(event->eventfd); | |
6178 | out_put_efile: | |
6179 | fdput(efile); | |
6180 | out_kfree: | |
6181 | kfree(event); | |
6182 | ||
6183 | return ret; | |
6184 | } | |
6185 | ||
8cdea7c0 BS |
6186 | static struct cftype mem_cgroup_files[] = { |
6187 | { | |
0eea1030 | 6188 | .name = "usage_in_bytes", |
8c7c6e34 | 6189 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 6190 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 6191 | }, |
c84872e1 PE |
6192 | { |
6193 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 6194 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 6195 | .trigger = mem_cgroup_reset, |
791badbd | 6196 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 6197 | }, |
8cdea7c0 | 6198 | { |
0eea1030 | 6199 | .name = "limit_in_bytes", |
8c7c6e34 | 6200 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 6201 | .write_string = mem_cgroup_write, |
791badbd | 6202 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 6203 | }, |
296c81d8 BS |
6204 | { |
6205 | .name = "soft_limit_in_bytes", | |
6206 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
6207 | .write_string = mem_cgroup_write, | |
791badbd | 6208 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 6209 | }, |
8cdea7c0 BS |
6210 | { |
6211 | .name = "failcnt", | |
8c7c6e34 | 6212 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 6213 | .trigger = mem_cgroup_reset, |
791badbd | 6214 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 6215 | }, |
d2ceb9b7 KH |
6216 | { |
6217 | .name = "stat", | |
2da8ca82 | 6218 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 6219 | }, |
c1e862c1 KH |
6220 | { |
6221 | .name = "force_empty", | |
6222 | .trigger = mem_cgroup_force_empty_write, | |
6223 | }, | |
18f59ea7 BS |
6224 | { |
6225 | .name = "use_hierarchy", | |
f00baae7 | 6226 | .flags = CFTYPE_INSANE, |
18f59ea7 BS |
6227 | .write_u64 = mem_cgroup_hierarchy_write, |
6228 | .read_u64 = mem_cgroup_hierarchy_read, | |
6229 | }, | |
79bd9814 | 6230 | { |
3bc942f3 TH |
6231 | .name = "cgroup.event_control", /* XXX: for compat */ |
6232 | .write_string = memcg_write_event_control, | |
79bd9814 TH |
6233 | .flags = CFTYPE_NO_PREFIX, |
6234 | .mode = S_IWUGO, | |
6235 | }, | |
a7885eb8 KM |
6236 | { |
6237 | .name = "swappiness", | |
6238 | .read_u64 = mem_cgroup_swappiness_read, | |
6239 | .write_u64 = mem_cgroup_swappiness_write, | |
6240 | }, | |
7dc74be0 DN |
6241 | { |
6242 | .name = "move_charge_at_immigrate", | |
6243 | .read_u64 = mem_cgroup_move_charge_read, | |
6244 | .write_u64 = mem_cgroup_move_charge_write, | |
6245 | }, | |
9490ff27 KH |
6246 | { |
6247 | .name = "oom_control", | |
2da8ca82 | 6248 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 6249 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
6250 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
6251 | }, | |
70ddf637 AV |
6252 | { |
6253 | .name = "pressure_level", | |
70ddf637 | 6254 | }, |
406eb0c9 YH |
6255 | #ifdef CONFIG_NUMA |
6256 | { | |
6257 | .name = "numa_stat", | |
2da8ca82 | 6258 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
6259 | }, |
6260 | #endif | |
510fc4e1 GC |
6261 | #ifdef CONFIG_MEMCG_KMEM |
6262 | { | |
6263 | .name = "kmem.limit_in_bytes", | |
6264 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
6265 | .write_string = mem_cgroup_write, | |
791badbd | 6266 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
6267 | }, |
6268 | { | |
6269 | .name = "kmem.usage_in_bytes", | |
6270 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 6271 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
6272 | }, |
6273 | { | |
6274 | .name = "kmem.failcnt", | |
6275 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6276 | .trigger = mem_cgroup_reset, | |
791badbd | 6277 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
6278 | }, |
6279 | { | |
6280 | .name = "kmem.max_usage_in_bytes", | |
6281 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6282 | .trigger = mem_cgroup_reset, | |
791badbd | 6283 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 6284 | }, |
749c5415 GC |
6285 | #ifdef CONFIG_SLABINFO |
6286 | { | |
6287 | .name = "kmem.slabinfo", | |
2da8ca82 | 6288 | .seq_show = mem_cgroup_slabinfo_read, |
749c5415 GC |
6289 | }, |
6290 | #endif | |
8c7c6e34 | 6291 | #endif |
6bc10349 | 6292 | { }, /* terminate */ |
af36f906 | 6293 | }; |
8c7c6e34 | 6294 | |
2d11085e MH |
6295 | #ifdef CONFIG_MEMCG_SWAP |
6296 | static struct cftype memsw_cgroup_files[] = { | |
6297 | { | |
6298 | .name = "memsw.usage_in_bytes", | |
6299 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 6300 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
6301 | }, |
6302 | { | |
6303 | .name = "memsw.max_usage_in_bytes", | |
6304 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6305 | .trigger = mem_cgroup_reset, | |
791badbd | 6306 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
6307 | }, |
6308 | { | |
6309 | .name = "memsw.limit_in_bytes", | |
6310 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6311 | .write_string = mem_cgroup_write, | |
791badbd | 6312 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
6313 | }, |
6314 | { | |
6315 | .name = "memsw.failcnt", | |
6316 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6317 | .trigger = mem_cgroup_reset, | |
791badbd | 6318 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
6319 | }, |
6320 | { }, /* terminate */ | |
6321 | }; | |
6322 | #endif | |
c0ff4b85 | 6323 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
6324 | { |
6325 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 6326 | struct mem_cgroup_per_zone *mz; |
41e3355d | 6327 | int zone, tmp = node; |
1ecaab2b KH |
6328 | /* |
6329 | * This routine is called against possible nodes. | |
6330 | * But it's BUG to call kmalloc() against offline node. | |
6331 | * | |
6332 | * TODO: this routine can waste much memory for nodes which will | |
6333 | * never be onlined. It's better to use memory hotplug callback | |
6334 | * function. | |
6335 | */ | |
41e3355d KH |
6336 | if (!node_state(node, N_NORMAL_MEMORY)) |
6337 | tmp = -1; | |
17295c88 | 6338 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
6339 | if (!pn) |
6340 | return 1; | |
1ecaab2b | 6341 | |
1ecaab2b KH |
6342 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
6343 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 6344 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
6345 | mz->usage_in_excess = 0; |
6346 | mz->on_tree = false; | |
d79154bb | 6347 | mz->memcg = memcg; |
1ecaab2b | 6348 | } |
54f72fe0 | 6349 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
6350 | return 0; |
6351 | } | |
6352 | ||
c0ff4b85 | 6353 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 6354 | { |
54f72fe0 | 6355 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
6356 | } |
6357 | ||
33327948 KH |
6358 | static struct mem_cgroup *mem_cgroup_alloc(void) |
6359 | { | |
d79154bb | 6360 | struct mem_cgroup *memcg; |
8ff69e2c | 6361 | size_t size; |
33327948 | 6362 | |
8ff69e2c VD |
6363 | size = sizeof(struct mem_cgroup); |
6364 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 6365 | |
8ff69e2c | 6366 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 6367 | if (!memcg) |
e7bbcdf3 DC |
6368 | return NULL; |
6369 | ||
d79154bb HD |
6370 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
6371 | if (!memcg->stat) | |
d2e61b8d | 6372 | goto out_free; |
d79154bb HD |
6373 | spin_lock_init(&memcg->pcp_counter_lock); |
6374 | return memcg; | |
d2e61b8d DC |
6375 | |
6376 | out_free: | |
8ff69e2c | 6377 | kfree(memcg); |
d2e61b8d | 6378 | return NULL; |
33327948 KH |
6379 | } |
6380 | ||
59927fb9 | 6381 | /* |
c8b2a36f GC |
6382 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
6383 | * (scanning all at force_empty is too costly...) | |
6384 | * | |
6385 | * Instead of clearing all references at force_empty, we remember | |
6386 | * the number of reference from swap_cgroup and free mem_cgroup when | |
6387 | * it goes down to 0. | |
6388 | * | |
6389 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 6390 | */ |
c8b2a36f GC |
6391 | |
6392 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 6393 | { |
c8b2a36f | 6394 | int node; |
59927fb9 | 6395 | |
bb4cc1a8 | 6396 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
6397 | |
6398 | for_each_node(node) | |
6399 | free_mem_cgroup_per_zone_info(memcg, node); | |
6400 | ||
6401 | free_percpu(memcg->stat); | |
6402 | ||
3f134619 GC |
6403 | /* |
6404 | * We need to make sure that (at least for now), the jump label | |
6405 | * destruction code runs outside of the cgroup lock. This is because | |
6406 | * get_online_cpus(), which is called from the static_branch update, | |
6407 | * can't be called inside the cgroup_lock. cpusets are the ones | |
6408 | * enforcing this dependency, so if they ever change, we might as well. | |
6409 | * | |
6410 | * schedule_work() will guarantee this happens. Be careful if you need | |
6411 | * to move this code around, and make sure it is outside | |
6412 | * the cgroup_lock. | |
6413 | */ | |
a8964b9b | 6414 | disarm_static_keys(memcg); |
8ff69e2c | 6415 | kfree(memcg); |
59927fb9 | 6416 | } |
3afe36b1 | 6417 | |
7bcc1bb1 DN |
6418 | /* |
6419 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6420 | */ | |
e1aab161 | 6421 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6422 | { |
c0ff4b85 | 6423 | if (!memcg->res.parent) |
7bcc1bb1 | 6424 | return NULL; |
c0ff4b85 | 6425 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6426 | } |
e1aab161 | 6427 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6428 | |
bb4cc1a8 AM |
6429 | static void __init mem_cgroup_soft_limit_tree_init(void) |
6430 | { | |
6431 | struct mem_cgroup_tree_per_node *rtpn; | |
6432 | struct mem_cgroup_tree_per_zone *rtpz; | |
6433 | int tmp, node, zone; | |
6434 | ||
6435 | for_each_node(node) { | |
6436 | tmp = node; | |
6437 | if (!node_state(node, N_NORMAL_MEMORY)) | |
6438 | tmp = -1; | |
6439 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
6440 | BUG_ON(!rtpn); | |
6441 | ||
6442 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
6443 | ||
6444 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
6445 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
6446 | rtpz->rb_root = RB_ROOT; | |
6447 | spin_lock_init(&rtpz->lock); | |
6448 | } | |
6449 | } | |
6450 | } | |
6451 | ||
0eb253e2 | 6452 | static struct cgroup_subsys_state * __ref |
eb95419b | 6453 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 6454 | { |
d142e3e6 | 6455 | struct mem_cgroup *memcg; |
04046e1a | 6456 | long error = -ENOMEM; |
6d12e2d8 | 6457 | int node; |
8cdea7c0 | 6458 | |
c0ff4b85 R |
6459 | memcg = mem_cgroup_alloc(); |
6460 | if (!memcg) | |
04046e1a | 6461 | return ERR_PTR(error); |
78fb7466 | 6462 | |
3ed28fa1 | 6463 | for_each_node(node) |
c0ff4b85 | 6464 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6465 | goto free_out; |
f64c3f54 | 6466 | |
c077719b | 6467 | /* root ? */ |
eb95419b | 6468 | if (parent_css == NULL) { |
a41c58a6 | 6469 | root_mem_cgroup = memcg; |
d142e3e6 GC |
6470 | res_counter_init(&memcg->res, NULL); |
6471 | res_counter_init(&memcg->memsw, NULL); | |
6472 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6473 | } |
28dbc4b6 | 6474 | |
d142e3e6 GC |
6475 | memcg->last_scanned_node = MAX_NUMNODES; |
6476 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
6477 | memcg->move_charge_at_immigrate = 0; |
6478 | mutex_init(&memcg->thresholds_lock); | |
6479 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 6480 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
6481 | INIT_LIST_HEAD(&memcg->event_list); |
6482 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
6483 | |
6484 | return &memcg->css; | |
6485 | ||
6486 | free_out: | |
6487 | __mem_cgroup_free(memcg); | |
6488 | return ERR_PTR(error); | |
6489 | } | |
6490 | ||
6491 | static int | |
eb95419b | 6492 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 6493 | { |
eb95419b TH |
6494 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6495 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css)); | |
d142e3e6 | 6496 | |
4219b2da LZ |
6497 | if (css->cgroup->id > MEM_CGROUP_ID_MAX) |
6498 | return -ENOSPC; | |
6499 | ||
63876986 | 6500 | if (!parent) |
d142e3e6 GC |
6501 | return 0; |
6502 | ||
0999821b | 6503 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6504 | |
6505 | memcg->use_hierarchy = parent->use_hierarchy; | |
6506 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6507 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6508 | ||
6509 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6510 | res_counter_init(&memcg->res, &parent->res); |
6511 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6512 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6513 | |
7bcc1bb1 | 6514 | /* |
8d76a979 LZ |
6515 | * No need to take a reference to the parent because cgroup |
6516 | * core guarantees its existence. | |
7bcc1bb1 | 6517 | */ |
18f59ea7 | 6518 | } else { |
c0ff4b85 R |
6519 | res_counter_init(&memcg->res, NULL); |
6520 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 6521 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
6522 | /* |
6523 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6524 | * much sense so let cgroup subsystem know about this | |
6525 | * unfortunate state in our controller. | |
6526 | */ | |
d142e3e6 | 6527 | if (parent != root_mem_cgroup) |
073219e9 | 6528 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 6529 | } |
0999821b | 6530 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 6531 | |
073219e9 | 6532 | return memcg_init_kmem(memcg, &memory_cgrp_subsys); |
8cdea7c0 BS |
6533 | } |
6534 | ||
5f578161 MH |
6535 | /* |
6536 | * Announce all parents that a group from their hierarchy is gone. | |
6537 | */ | |
6538 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6539 | { | |
6540 | struct mem_cgroup *parent = memcg; | |
6541 | ||
6542 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 6543 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
6544 | |
6545 | /* | |
6546 | * if the root memcg is not hierarchical we have to check it | |
6547 | * explicitely. | |
6548 | */ | |
6549 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 6550 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
6551 | } |
6552 | ||
eb95419b | 6553 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 6554 | { |
eb95419b | 6555 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 6556 | struct mem_cgroup_event *event, *tmp; |
4fb1a86f | 6557 | struct cgroup_subsys_state *iter; |
79bd9814 TH |
6558 | |
6559 | /* | |
6560 | * Unregister events and notify userspace. | |
6561 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
6562 | * directory to avoid race between userspace and kernelspace. | |
6563 | */ | |
fba94807 TH |
6564 | spin_lock(&memcg->event_list_lock); |
6565 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
6566 | list_del_init(&event->list); |
6567 | schedule_work(&event->remove); | |
6568 | } | |
fba94807 | 6569 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 6570 | |
10d5ebf4 LZ |
6571 | kmem_cgroup_css_offline(memcg); |
6572 | ||
5f578161 | 6573 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
4fb1a86f FB |
6574 | |
6575 | /* | |
6576 | * This requires that offlining is serialized. Right now that is | |
6577 | * guaranteed because css_killed_work_fn() holds the cgroup_mutex. | |
6578 | */ | |
6579 | css_for_each_descendant_post(iter, css) | |
6580 | mem_cgroup_reparent_charges(mem_cgroup_from_css(iter)); | |
6581 | ||
1f458cbf | 6582 | mem_cgroup_destroy_all_caches(memcg); |
33cb876e | 6583 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
6584 | } |
6585 | ||
eb95419b | 6586 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 6587 | { |
eb95419b | 6588 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
96f1c58d JW |
6589 | /* |
6590 | * XXX: css_offline() would be where we should reparent all | |
6591 | * memory to prepare the cgroup for destruction. However, | |
6592 | * memcg does not do css_tryget() and res_counter charging | |
6593 | * under the same RCU lock region, which means that charging | |
6594 | * could race with offlining. Offlining only happens to | |
6595 | * cgroups with no tasks in them but charges can show up | |
6596 | * without any tasks from the swapin path when the target | |
6597 | * memcg is looked up from the swapout record and not from the | |
6598 | * current task as it usually is. A race like this can leak | |
6599 | * charges and put pages with stale cgroup pointers into | |
6600 | * circulation: | |
6601 | * | |
6602 | * #0 #1 | |
6603 | * lookup_swap_cgroup_id() | |
6604 | * rcu_read_lock() | |
6605 | * mem_cgroup_lookup() | |
6606 | * css_tryget() | |
6607 | * rcu_read_unlock() | |
6608 | * disable css_tryget() | |
6609 | * call_rcu() | |
6610 | * offline_css() | |
6611 | * reparent_charges() | |
6612 | * res_counter_charge() | |
6613 | * css_put() | |
6614 | * css_free() | |
6615 | * pc->mem_cgroup = dead memcg | |
6616 | * add page to lru | |
6617 | * | |
6618 | * The bulk of the charges are still moved in offline_css() to | |
6619 | * avoid pinning a lot of pages in case a long-term reference | |
6620 | * like a swapout record is deferring the css_free() to long | |
6621 | * after offlining. But this makes sure we catch any charges | |
6622 | * made after offlining: | |
6623 | */ | |
6624 | mem_cgroup_reparent_charges(memcg); | |
c268e994 | 6625 | |
10d5ebf4 | 6626 | memcg_destroy_kmem(memcg); |
465939a1 | 6627 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
6628 | } |
6629 | ||
02491447 | 6630 | #ifdef CONFIG_MMU |
7dc74be0 | 6631 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6632 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6633 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6634 | { |
854ffa8d DN |
6635 | int ret = 0; |
6636 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6637 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6638 | |
c0ff4b85 | 6639 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6640 | mc.precharge += count; |
6641 | /* we don't need css_get for root */ | |
6642 | return ret; | |
6643 | } | |
6644 | /* try to charge at once */ | |
6645 | if (count > 1) { | |
6646 | struct res_counter *dummy; | |
6647 | /* | |
c0ff4b85 | 6648 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6649 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6650 | * are still under the same cgroup_mutex. So we can postpone | |
6651 | * css_get(). | |
6652 | */ | |
c0ff4b85 | 6653 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6654 | goto one_by_one; |
c0ff4b85 | 6655 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6656 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6657 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6658 | goto one_by_one; |
6659 | } | |
6660 | mc.precharge += count; | |
854ffa8d DN |
6661 | return ret; |
6662 | } | |
6663 | one_by_one: | |
6664 | /* fall back to one by one charge */ | |
6665 | while (count--) { | |
6666 | if (signal_pending(current)) { | |
6667 | ret = -EINTR; | |
6668 | break; | |
6669 | } | |
6670 | if (!batch_count--) { | |
6671 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6672 | cond_resched(); | |
6673 | } | |
c0ff4b85 R |
6674 | ret = __mem_cgroup_try_charge(NULL, |
6675 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6676 | if (ret) |
854ffa8d | 6677 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6678 | return ret; |
854ffa8d DN |
6679 | mc.precharge++; |
6680 | } | |
4ffef5fe DN |
6681 | return ret; |
6682 | } | |
6683 | ||
6684 | /** | |
8d32ff84 | 6685 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6686 | * @vma: the vma the pte to be checked belongs |
6687 | * @addr: the address corresponding to the pte to be checked | |
6688 | * @ptent: the pte to be checked | |
02491447 | 6689 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6690 | * |
6691 | * Returns | |
6692 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6693 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6694 | * move charge. if @target is not NULL, the page is stored in target->page | |
6695 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6696 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6697 | * target for charge migration. if @target is not NULL, the entry is stored | |
6698 | * in target->ent. | |
4ffef5fe DN |
6699 | * |
6700 | * Called with pte lock held. | |
6701 | */ | |
4ffef5fe DN |
6702 | union mc_target { |
6703 | struct page *page; | |
02491447 | 6704 | swp_entry_t ent; |
4ffef5fe DN |
6705 | }; |
6706 | ||
4ffef5fe | 6707 | enum mc_target_type { |
8d32ff84 | 6708 | MC_TARGET_NONE = 0, |
4ffef5fe | 6709 | MC_TARGET_PAGE, |
02491447 | 6710 | MC_TARGET_SWAP, |
4ffef5fe DN |
6711 | }; |
6712 | ||
90254a65 DN |
6713 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6714 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6715 | { |
90254a65 | 6716 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6717 | |
90254a65 DN |
6718 | if (!page || !page_mapped(page)) |
6719 | return NULL; | |
6720 | if (PageAnon(page)) { | |
6721 | /* we don't move shared anon */ | |
4b91355e | 6722 | if (!move_anon()) |
90254a65 | 6723 | return NULL; |
87946a72 DN |
6724 | } else if (!move_file()) |
6725 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6726 | return NULL; |
6727 | if (!get_page_unless_zero(page)) | |
6728 | return NULL; | |
6729 | ||
6730 | return page; | |
6731 | } | |
6732 | ||
4b91355e | 6733 | #ifdef CONFIG_SWAP |
90254a65 DN |
6734 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6735 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6736 | { | |
90254a65 DN |
6737 | struct page *page = NULL; |
6738 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6739 | ||
6740 | if (!move_anon() || non_swap_entry(ent)) | |
6741 | return NULL; | |
4b91355e KH |
6742 | /* |
6743 | * Because lookup_swap_cache() updates some statistics counter, | |
6744 | * we call find_get_page() with swapper_space directly. | |
6745 | */ | |
33806f06 | 6746 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6747 | if (do_swap_account) |
6748 | entry->val = ent.val; | |
6749 | ||
6750 | return page; | |
6751 | } | |
4b91355e KH |
6752 | #else |
6753 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6754 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6755 | { | |
6756 | return NULL; | |
6757 | } | |
6758 | #endif | |
90254a65 | 6759 | |
87946a72 DN |
6760 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6761 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6762 | { | |
6763 | struct page *page = NULL; | |
87946a72 DN |
6764 | struct address_space *mapping; |
6765 | pgoff_t pgoff; | |
6766 | ||
6767 | if (!vma->vm_file) /* anonymous vma */ | |
6768 | return NULL; | |
6769 | if (!move_file()) | |
6770 | return NULL; | |
6771 | ||
87946a72 DN |
6772 | mapping = vma->vm_file->f_mapping; |
6773 | if (pte_none(ptent)) | |
6774 | pgoff = linear_page_index(vma, addr); | |
6775 | else /* pte_file(ptent) is true */ | |
6776 | pgoff = pte_to_pgoff(ptent); | |
6777 | ||
6778 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6779 | page = find_get_page(mapping, pgoff); |
6780 | ||
6781 | #ifdef CONFIG_SWAP | |
6782 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6783 | if (radix_tree_exceptional_entry(page)) { | |
6784 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6785 | if (do_swap_account) |
aa3b1895 | 6786 | *entry = swap; |
33806f06 | 6787 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6788 | } |
aa3b1895 | 6789 | #endif |
87946a72 DN |
6790 | return page; |
6791 | } | |
6792 | ||
8d32ff84 | 6793 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6794 | unsigned long addr, pte_t ptent, union mc_target *target) |
6795 | { | |
6796 | struct page *page = NULL; | |
6797 | struct page_cgroup *pc; | |
8d32ff84 | 6798 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6799 | swp_entry_t ent = { .val = 0 }; |
6800 | ||
6801 | if (pte_present(ptent)) | |
6802 | page = mc_handle_present_pte(vma, addr, ptent); | |
6803 | else if (is_swap_pte(ptent)) | |
6804 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6805 | else if (pte_none(ptent) || pte_file(ptent)) |
6806 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6807 | |
6808 | if (!page && !ent.val) | |
8d32ff84 | 6809 | return ret; |
02491447 DN |
6810 | if (page) { |
6811 | pc = lookup_page_cgroup(page); | |
6812 | /* | |
6813 | * Do only loose check w/o page_cgroup lock. | |
6814 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6815 | * the lock. | |
6816 | */ | |
6817 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6818 | ret = MC_TARGET_PAGE; | |
6819 | if (target) | |
6820 | target->page = page; | |
6821 | } | |
6822 | if (!ret || !target) | |
6823 | put_page(page); | |
6824 | } | |
90254a65 DN |
6825 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6826 | if (ent.val && !ret && | |
34c00c31 | 6827 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6828 | ret = MC_TARGET_SWAP; |
6829 | if (target) | |
6830 | target->ent = ent; | |
4ffef5fe | 6831 | } |
4ffef5fe DN |
6832 | return ret; |
6833 | } | |
6834 | ||
12724850 NH |
6835 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6836 | /* | |
6837 | * We don't consider swapping or file mapped pages because THP does not | |
6838 | * support them for now. | |
6839 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6840 | */ | |
6841 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6842 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6843 | { | |
6844 | struct page *page = NULL; | |
6845 | struct page_cgroup *pc; | |
6846 | enum mc_target_type ret = MC_TARGET_NONE; | |
6847 | ||
6848 | page = pmd_page(pmd); | |
309381fe | 6849 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
6850 | if (!move_anon()) |
6851 | return ret; | |
6852 | pc = lookup_page_cgroup(page); | |
6853 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6854 | ret = MC_TARGET_PAGE; | |
6855 | if (target) { | |
6856 | get_page(page); | |
6857 | target->page = page; | |
6858 | } | |
6859 | } | |
6860 | return ret; | |
6861 | } | |
6862 | #else | |
6863 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6864 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6865 | { | |
6866 | return MC_TARGET_NONE; | |
6867 | } | |
6868 | #endif | |
6869 | ||
4ffef5fe DN |
6870 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6871 | unsigned long addr, unsigned long end, | |
6872 | struct mm_walk *walk) | |
6873 | { | |
6874 | struct vm_area_struct *vma = walk->private; | |
6875 | pte_t *pte; | |
6876 | spinlock_t *ptl; | |
6877 | ||
bf929152 | 6878 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
6879 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
6880 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 6881 | spin_unlock(ptl); |
1a5a9906 | 6882 | return 0; |
12724850 | 6883 | } |
03319327 | 6884 | |
45f83cef AA |
6885 | if (pmd_trans_unstable(pmd)) |
6886 | return 0; | |
4ffef5fe DN |
6887 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6888 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6889 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6890 | mc.precharge++; /* increment precharge temporarily */ |
6891 | pte_unmap_unlock(pte - 1, ptl); | |
6892 | cond_resched(); | |
6893 | ||
7dc74be0 DN |
6894 | return 0; |
6895 | } | |
6896 | ||
4ffef5fe DN |
6897 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6898 | { | |
6899 | unsigned long precharge; | |
6900 | struct vm_area_struct *vma; | |
6901 | ||
dfe076b0 | 6902 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6903 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6904 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6905 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6906 | .mm = mm, | |
6907 | .private = vma, | |
6908 | }; | |
6909 | if (is_vm_hugetlb_page(vma)) | |
6910 | continue; | |
4ffef5fe DN |
6911 | walk_page_range(vma->vm_start, vma->vm_end, |
6912 | &mem_cgroup_count_precharge_walk); | |
6913 | } | |
dfe076b0 | 6914 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6915 | |
6916 | precharge = mc.precharge; | |
6917 | mc.precharge = 0; | |
6918 | ||
6919 | return precharge; | |
6920 | } | |
6921 | ||
4ffef5fe DN |
6922 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6923 | { | |
dfe076b0 DN |
6924 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6925 | ||
6926 | VM_BUG_ON(mc.moving_task); | |
6927 | mc.moving_task = current; | |
6928 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6929 | } |
6930 | ||
dfe076b0 DN |
6931 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6932 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6933 | { |
2bd9bb20 KH |
6934 | struct mem_cgroup *from = mc.from; |
6935 | struct mem_cgroup *to = mc.to; | |
4050377b | 6936 | int i; |
2bd9bb20 | 6937 | |
4ffef5fe | 6938 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6939 | if (mc.precharge) { |
6940 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6941 | mc.precharge = 0; | |
6942 | } | |
6943 | /* | |
6944 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6945 | * we must uncharge here. | |
6946 | */ | |
6947 | if (mc.moved_charge) { | |
6948 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6949 | mc.moved_charge = 0; | |
4ffef5fe | 6950 | } |
483c30b5 DN |
6951 | /* we must fixup refcnts and charges */ |
6952 | if (mc.moved_swap) { | |
483c30b5 DN |
6953 | /* uncharge swap account from the old cgroup */ |
6954 | if (!mem_cgroup_is_root(mc.from)) | |
6955 | res_counter_uncharge(&mc.from->memsw, | |
6956 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6957 | |
6958 | for (i = 0; i < mc.moved_swap; i++) | |
6959 | css_put(&mc.from->css); | |
483c30b5 DN |
6960 | |
6961 | if (!mem_cgroup_is_root(mc.to)) { | |
6962 | /* | |
6963 | * we charged both to->res and to->memsw, so we should | |
6964 | * uncharge to->res. | |
6965 | */ | |
6966 | res_counter_uncharge(&mc.to->res, | |
6967 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 | 6968 | } |
4050377b | 6969 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6970 | mc.moved_swap = 0; |
6971 | } | |
dfe076b0 DN |
6972 | memcg_oom_recover(from); |
6973 | memcg_oom_recover(to); | |
6974 | wake_up_all(&mc.waitq); | |
6975 | } | |
6976 | ||
6977 | static void mem_cgroup_clear_mc(void) | |
6978 | { | |
6979 | struct mem_cgroup *from = mc.from; | |
6980 | ||
6981 | /* | |
6982 | * we must clear moving_task before waking up waiters at the end of | |
6983 | * task migration. | |
6984 | */ | |
6985 | mc.moving_task = NULL; | |
6986 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6987 | spin_lock(&mc.lock); |
4ffef5fe DN |
6988 | mc.from = NULL; |
6989 | mc.to = NULL; | |
2bd9bb20 | 6990 | spin_unlock(&mc.lock); |
32047e2a | 6991 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6992 | } |
6993 | ||
eb95419b | 6994 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6995 | struct cgroup_taskset *tset) |
7dc74be0 | 6996 | { |
2f7ee569 | 6997 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6998 | int ret = 0; |
eb95419b | 6999 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 7000 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 7001 | |
ee5e8472 GC |
7002 | /* |
7003 | * We are now commited to this value whatever it is. Changes in this | |
7004 | * tunable will only affect upcoming migrations, not the current one. | |
7005 | * So we need to save it, and keep it going. | |
7006 | */ | |
7007 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
7008 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
7009 | struct mm_struct *mm; |
7010 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
7011 | ||
c0ff4b85 | 7012 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
7013 | |
7014 | mm = get_task_mm(p); | |
7015 | if (!mm) | |
7016 | return 0; | |
7dc74be0 | 7017 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
7018 | if (mm->owner == p) { |
7019 | VM_BUG_ON(mc.from); | |
7020 | VM_BUG_ON(mc.to); | |
7021 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 7022 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 7023 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 7024 | mem_cgroup_start_move(from); |
2bd9bb20 | 7025 | spin_lock(&mc.lock); |
4ffef5fe | 7026 | mc.from = from; |
c0ff4b85 | 7027 | mc.to = memcg; |
ee5e8472 | 7028 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 7029 | spin_unlock(&mc.lock); |
dfe076b0 | 7030 | /* We set mc.moving_task later */ |
4ffef5fe DN |
7031 | |
7032 | ret = mem_cgroup_precharge_mc(mm); | |
7033 | if (ret) | |
7034 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
7035 | } |
7036 | mmput(mm); | |
7dc74be0 DN |
7037 | } |
7038 | return ret; | |
7039 | } | |
7040 | ||
eb95419b | 7041 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7042 | struct cgroup_taskset *tset) |
7dc74be0 | 7043 | { |
4ffef5fe | 7044 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
7045 | } |
7046 | ||
4ffef5fe DN |
7047 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
7048 | unsigned long addr, unsigned long end, | |
7049 | struct mm_walk *walk) | |
7dc74be0 | 7050 | { |
4ffef5fe DN |
7051 | int ret = 0; |
7052 | struct vm_area_struct *vma = walk->private; | |
7053 | pte_t *pte; | |
7054 | spinlock_t *ptl; | |
12724850 NH |
7055 | enum mc_target_type target_type; |
7056 | union mc_target target; | |
7057 | struct page *page; | |
7058 | struct page_cgroup *pc; | |
4ffef5fe | 7059 | |
12724850 NH |
7060 | /* |
7061 | * We don't take compound_lock() here but no race with splitting thp | |
7062 | * happens because: | |
7063 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
7064 | * under splitting, which means there's no concurrent thp split, | |
7065 | * - if another thread runs into split_huge_page() just after we | |
7066 | * entered this if-block, the thread must wait for page table lock | |
7067 | * to be unlocked in __split_huge_page_splitting(), where the main | |
7068 | * part of thp split is not executed yet. | |
7069 | */ | |
bf929152 | 7070 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 7071 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 7072 | spin_unlock(ptl); |
12724850 NH |
7073 | return 0; |
7074 | } | |
7075 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
7076 | if (target_type == MC_TARGET_PAGE) { | |
7077 | page = target.page; | |
7078 | if (!isolate_lru_page(page)) { | |
7079 | pc = lookup_page_cgroup(page); | |
7080 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 7081 | pc, mc.from, mc.to)) { |
12724850 NH |
7082 | mc.precharge -= HPAGE_PMD_NR; |
7083 | mc.moved_charge += HPAGE_PMD_NR; | |
7084 | } | |
7085 | putback_lru_page(page); | |
7086 | } | |
7087 | put_page(page); | |
7088 | } | |
bf929152 | 7089 | spin_unlock(ptl); |
1a5a9906 | 7090 | return 0; |
12724850 NH |
7091 | } |
7092 | ||
45f83cef AA |
7093 | if (pmd_trans_unstable(pmd)) |
7094 | return 0; | |
4ffef5fe DN |
7095 | retry: |
7096 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
7097 | for (; addr != end; addr += PAGE_SIZE) { | |
7098 | pte_t ptent = *(pte++); | |
02491447 | 7099 | swp_entry_t ent; |
4ffef5fe DN |
7100 | |
7101 | if (!mc.precharge) | |
7102 | break; | |
7103 | ||
8d32ff84 | 7104 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
7105 | case MC_TARGET_PAGE: |
7106 | page = target.page; | |
7107 | if (isolate_lru_page(page)) | |
7108 | goto put; | |
7109 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 7110 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 7111 | mc.from, mc.to)) { |
4ffef5fe | 7112 | mc.precharge--; |
854ffa8d DN |
7113 | /* we uncharge from mc.from later. */ |
7114 | mc.moved_charge++; | |
4ffef5fe DN |
7115 | } |
7116 | putback_lru_page(page); | |
8d32ff84 | 7117 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
7118 | put_page(page); |
7119 | break; | |
02491447 DN |
7120 | case MC_TARGET_SWAP: |
7121 | ent = target.ent; | |
e91cbb42 | 7122 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 7123 | mc.precharge--; |
483c30b5 DN |
7124 | /* we fixup refcnts and charges later. */ |
7125 | mc.moved_swap++; | |
7126 | } | |
02491447 | 7127 | break; |
4ffef5fe DN |
7128 | default: |
7129 | break; | |
7130 | } | |
7131 | } | |
7132 | pte_unmap_unlock(pte - 1, ptl); | |
7133 | cond_resched(); | |
7134 | ||
7135 | if (addr != end) { | |
7136 | /* | |
7137 | * We have consumed all precharges we got in can_attach(). | |
7138 | * We try charge one by one, but don't do any additional | |
7139 | * charges to mc.to if we have failed in charge once in attach() | |
7140 | * phase. | |
7141 | */ | |
854ffa8d | 7142 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
7143 | if (!ret) |
7144 | goto retry; | |
7145 | } | |
7146 | ||
7147 | return ret; | |
7148 | } | |
7149 | ||
7150 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
7151 | { | |
7152 | struct vm_area_struct *vma; | |
7153 | ||
7154 | lru_add_drain_all(); | |
dfe076b0 DN |
7155 | retry: |
7156 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
7157 | /* | |
7158 | * Someone who are holding the mmap_sem might be waiting in | |
7159 | * waitq. So we cancel all extra charges, wake up all waiters, | |
7160 | * and retry. Because we cancel precharges, we might not be able | |
7161 | * to move enough charges, but moving charge is a best-effort | |
7162 | * feature anyway, so it wouldn't be a big problem. | |
7163 | */ | |
7164 | __mem_cgroup_clear_mc(); | |
7165 | cond_resched(); | |
7166 | goto retry; | |
7167 | } | |
4ffef5fe DN |
7168 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
7169 | int ret; | |
7170 | struct mm_walk mem_cgroup_move_charge_walk = { | |
7171 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
7172 | .mm = mm, | |
7173 | .private = vma, | |
7174 | }; | |
7175 | if (is_vm_hugetlb_page(vma)) | |
7176 | continue; | |
4ffef5fe DN |
7177 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
7178 | &mem_cgroup_move_charge_walk); | |
7179 | if (ret) | |
7180 | /* | |
7181 | * means we have consumed all precharges and failed in | |
7182 | * doing additional charge. Just abandon here. | |
7183 | */ | |
7184 | break; | |
7185 | } | |
dfe076b0 | 7186 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
7187 | } |
7188 | ||
eb95419b | 7189 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 7190 | struct cgroup_taskset *tset) |
67e465a7 | 7191 | { |
2f7ee569 | 7192 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 7193 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 7194 | |
dfe076b0 | 7195 | if (mm) { |
a433658c KM |
7196 | if (mc.to) |
7197 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
7198 | mmput(mm); |
7199 | } | |
a433658c KM |
7200 | if (mc.to) |
7201 | mem_cgroup_clear_mc(); | |
67e465a7 | 7202 | } |
5cfb80a7 | 7203 | #else /* !CONFIG_MMU */ |
eb95419b | 7204 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7205 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7206 | { |
7207 | return 0; | |
7208 | } | |
eb95419b | 7209 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7210 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7211 | { |
7212 | } | |
eb95419b | 7213 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 7214 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7215 | { |
7216 | } | |
7217 | #endif | |
67e465a7 | 7218 | |
f00baae7 TH |
7219 | /* |
7220 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
7221 | * to verify sane_behavior flag on each mount attempt. | |
7222 | */ | |
eb95419b | 7223 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
7224 | { |
7225 | /* | |
7226 | * use_hierarchy is forced with sane_behavior. cgroup core | |
7227 | * guarantees that @root doesn't have any children, so turning it | |
7228 | * on for the root memcg is enough. | |
7229 | */ | |
eb95419b TH |
7230 | if (cgroup_sane_behavior(root_css->cgroup)) |
7231 | mem_cgroup_from_css(root_css)->use_hierarchy = true; | |
f00baae7 TH |
7232 | } |
7233 | ||
073219e9 | 7234 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 7235 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 7236 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
7237 | .css_offline = mem_cgroup_css_offline, |
7238 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
7239 | .can_attach = mem_cgroup_can_attach, |
7240 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 7241 | .attach = mem_cgroup_move_task, |
f00baae7 | 7242 | .bind = mem_cgroup_bind, |
6bc10349 | 7243 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 7244 | .early_init = 0, |
8cdea7c0 | 7245 | }; |
c077719b | 7246 | |
c255a458 | 7247 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
7248 | static int __init enable_swap_account(char *s) |
7249 | { | |
a2c8990a | 7250 | if (!strcmp(s, "1")) |
a42c390c | 7251 | really_do_swap_account = 1; |
a2c8990a | 7252 | else if (!strcmp(s, "0")) |
a42c390c MH |
7253 | really_do_swap_account = 0; |
7254 | return 1; | |
7255 | } | |
a2c8990a | 7256 | __setup("swapaccount=", enable_swap_account); |
c077719b | 7257 | |
2d11085e MH |
7258 | static void __init memsw_file_init(void) |
7259 | { | |
073219e9 | 7260 | WARN_ON(cgroup_add_cftypes(&memory_cgrp_subsys, memsw_cgroup_files)); |
6acc8b02 MH |
7261 | } |
7262 | ||
7263 | static void __init enable_swap_cgroup(void) | |
7264 | { | |
7265 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
7266 | do_swap_account = 1; | |
7267 | memsw_file_init(); | |
7268 | } | |
2d11085e | 7269 | } |
6acc8b02 | 7270 | |
2d11085e | 7271 | #else |
6acc8b02 | 7272 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
7273 | { |
7274 | } | |
c077719b | 7275 | #endif |
2d11085e MH |
7276 | |
7277 | /* | |
1081312f MH |
7278 | * subsys_initcall() for memory controller. |
7279 | * | |
7280 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
7281 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
7282 | * everything that doesn't depend on a specific mem_cgroup structure should | |
7283 | * be initialized from here. | |
2d11085e MH |
7284 | */ |
7285 | static int __init mem_cgroup_init(void) | |
7286 | { | |
7287 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 7288 | enable_swap_cgroup(); |
bb4cc1a8 | 7289 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 7290 | memcg_stock_init(); |
2d11085e MH |
7291 | return 0; |
7292 | } | |
7293 | subsys_initcall(mem_cgroup_init); |