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 | { |
c62b1a3b KH |
924 | preempt_disable(); |
925 | ||
b2402857 KH |
926 | /* |
927 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
928 | * counted as CACHE even if it's on ANON LRU. | |
929 | */ | |
930 | if (anon) | |
931 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 932 | nr_pages); |
d52aa412 | 933 | else |
b2402857 | 934 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 935 | nr_pages); |
55e462b0 | 936 | |
b070e65c DR |
937 | if (PageTransHuge(page)) |
938 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
939 | nr_pages); | |
940 | ||
e401f176 KH |
941 | /* pagein of a big page is an event. So, ignore page size */ |
942 | if (nr_pages > 0) | |
c0ff4b85 | 943 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 944 | else { |
c0ff4b85 | 945 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
946 | nr_pages = -nr_pages; /* for event */ |
947 | } | |
e401f176 | 948 | |
13114716 | 949 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 950 | |
c62b1a3b | 951 | preempt_enable(); |
6d12e2d8 KH |
952 | } |
953 | ||
bb2a0de9 | 954 | unsigned long |
4d7dcca2 | 955 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
956 | { |
957 | struct mem_cgroup_per_zone *mz; | |
958 | ||
959 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
960 | return mz->lru_size[lru]; | |
961 | } | |
962 | ||
963 | static unsigned long | |
c0ff4b85 | 964 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 965 | unsigned int lru_mask) |
889976db YH |
966 | { |
967 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 968 | enum lru_list lru; |
bb2a0de9 KH |
969 | unsigned long ret = 0; |
970 | ||
c0ff4b85 | 971 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 972 | |
f156ab93 HD |
973 | for_each_lru(lru) { |
974 | if (BIT(lru) & lru_mask) | |
975 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
976 | } |
977 | return ret; | |
978 | } | |
979 | ||
980 | static unsigned long | |
c0ff4b85 | 981 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
982 | int nid, unsigned int lru_mask) |
983 | { | |
889976db YH |
984 | u64 total = 0; |
985 | int zid; | |
986 | ||
bb2a0de9 | 987 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
988 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
989 | nid, zid, lru_mask); | |
bb2a0de9 | 990 | |
889976db YH |
991 | return total; |
992 | } | |
bb2a0de9 | 993 | |
c0ff4b85 | 994 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 995 | unsigned int lru_mask) |
6d12e2d8 | 996 | { |
889976db | 997 | int nid; |
6d12e2d8 KH |
998 | u64 total = 0; |
999 | ||
31aaea4a | 1000 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 1001 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 1002 | return total; |
d52aa412 KH |
1003 | } |
1004 | ||
f53d7ce3 JW |
1005 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
1006 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
1007 | { |
1008 | unsigned long val, next; | |
1009 | ||
13114716 | 1010 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 1011 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 1012 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
1013 | if ((long)next - (long)val < 0) { |
1014 | switch (target) { | |
1015 | case MEM_CGROUP_TARGET_THRESH: | |
1016 | next = val + THRESHOLDS_EVENTS_TARGET; | |
1017 | break; | |
bb4cc1a8 AM |
1018 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
1019 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
1020 | break; | |
f53d7ce3 JW |
1021 | case MEM_CGROUP_TARGET_NUMAINFO: |
1022 | next = val + NUMAINFO_EVENTS_TARGET; | |
1023 | break; | |
1024 | default: | |
1025 | break; | |
1026 | } | |
1027 | __this_cpu_write(memcg->stat->targets[target], next); | |
1028 | return true; | |
7a159cc9 | 1029 | } |
f53d7ce3 | 1030 | return false; |
d2265e6f KH |
1031 | } |
1032 | ||
1033 | /* | |
1034 | * Check events in order. | |
1035 | * | |
1036 | */ | |
c0ff4b85 | 1037 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 1038 | { |
4799401f | 1039 | preempt_disable(); |
d2265e6f | 1040 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1041 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1042 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1043 | bool do_softlimit; |
82b3f2a7 | 1044 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1045 | |
bb4cc1a8 AM |
1046 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1047 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1048 | #if MAX_NUMNODES > 1 |
1049 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1050 | MEM_CGROUP_TARGET_NUMAINFO); | |
1051 | #endif | |
1052 | preempt_enable(); | |
1053 | ||
c0ff4b85 | 1054 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1055 | if (unlikely(do_softlimit)) |
1056 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1057 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1058 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1059 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1060 | #endif |
f53d7ce3 JW |
1061 | } else |
1062 | preempt_enable(); | |
d2265e6f KH |
1063 | } |
1064 | ||
cf475ad2 | 1065 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1066 | { |
31a78f23 BS |
1067 | /* |
1068 | * mm_update_next_owner() may clear mm->owner to NULL | |
1069 | * if it races with swapoff, page migration, etc. | |
1070 | * So this can be called with p == NULL. | |
1071 | */ | |
1072 | if (unlikely(!p)) | |
1073 | return NULL; | |
1074 | ||
073219e9 | 1075 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1076 | } |
1077 | ||
a433658c | 1078 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1079 | { |
c0ff4b85 | 1080 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
1081 | |
1082 | if (!mm) | |
1083 | return NULL; | |
54595fe2 KH |
1084 | /* |
1085 | * Because we have no locks, mm->owner's may be being moved to other | |
1086 | * cgroup. We use css_tryget() here even if this looks | |
1087 | * pessimistic (rather than adding locks here). | |
1088 | */ | |
1089 | rcu_read_lock(); | |
1090 | do { | |
c0ff4b85 R |
1091 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1092 | if (unlikely(!memcg)) | |
54595fe2 | 1093 | break; |
c0ff4b85 | 1094 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1095 | rcu_read_unlock(); |
c0ff4b85 | 1096 | return memcg; |
54595fe2 KH |
1097 | } |
1098 | ||
16248d8f MH |
1099 | /* |
1100 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1101 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1102 | * | |
1103 | * helper function to be used by mem_cgroup_iter | |
1104 | */ | |
1105 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
694fbc0f | 1106 | struct mem_cgroup *last_visited) |
16248d8f | 1107 | { |
492eb21b | 1108 | struct cgroup_subsys_state *prev_css, *next_css; |
16248d8f | 1109 | |
bd8815a6 | 1110 | prev_css = last_visited ? &last_visited->css : NULL; |
16248d8f | 1111 | skip_node: |
492eb21b | 1112 | next_css = css_next_descendant_pre(prev_css, &root->css); |
16248d8f MH |
1113 | |
1114 | /* | |
1115 | * Even if we found a group we have to make sure it is | |
1116 | * alive. css && !memcg means that the groups should be | |
1117 | * skipped and we should continue the tree walk. | |
1118 | * last_visited css is safe to use because it is | |
1119 | * protected by css_get and the tree walk is rcu safe. | |
0eef6156 MH |
1120 | * |
1121 | * We do not take a reference on the root of the tree walk | |
1122 | * because we might race with the root removal when it would | |
1123 | * be the only node in the iterated hierarchy and mem_cgroup_iter | |
1124 | * would end up in an endless loop because it expects that at | |
1125 | * least one valid node will be returned. Root cannot disappear | |
1126 | * because caller of the iterator should hold it already so | |
1127 | * skipping css reference should be safe. | |
16248d8f | 1128 | */ |
492eb21b | 1129 | if (next_css) { |
ce48225f HD |
1130 | if ((next_css == &root->css) || |
1131 | ((next_css->flags & CSS_ONLINE) && css_tryget(next_css))) | |
d8ad3055 | 1132 | return mem_cgroup_from_css(next_css); |
0eef6156 MH |
1133 | |
1134 | prev_css = next_css; | |
1135 | goto skip_node; | |
16248d8f MH |
1136 | } |
1137 | ||
1138 | return NULL; | |
1139 | } | |
1140 | ||
519ebea3 JW |
1141 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
1142 | { | |
1143 | /* | |
1144 | * When a group in the hierarchy below root is destroyed, the | |
1145 | * hierarchy iterator can no longer be trusted since it might | |
1146 | * have pointed to the destroyed group. Invalidate it. | |
1147 | */ | |
1148 | atomic_inc(&root->dead_count); | |
1149 | } | |
1150 | ||
1151 | static struct mem_cgroup * | |
1152 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, | |
1153 | struct mem_cgroup *root, | |
1154 | int *sequence) | |
1155 | { | |
1156 | struct mem_cgroup *position = NULL; | |
1157 | /* | |
1158 | * A cgroup destruction happens in two stages: offlining and | |
1159 | * release. They are separated by a RCU grace period. | |
1160 | * | |
1161 | * If the iterator is valid, we may still race with an | |
1162 | * offlining. The RCU lock ensures the object won't be | |
1163 | * released, tryget will fail if we lost the race. | |
1164 | */ | |
1165 | *sequence = atomic_read(&root->dead_count); | |
1166 | if (iter->last_dead_count == *sequence) { | |
1167 | smp_rmb(); | |
1168 | position = iter->last_visited; | |
ecc736fc MH |
1169 | |
1170 | /* | |
1171 | * We cannot take a reference to root because we might race | |
1172 | * with root removal and returning NULL would end up in | |
1173 | * an endless loop on the iterator user level when root | |
1174 | * would be returned all the time. | |
1175 | */ | |
1176 | if (position && position != root && | |
1177 | !css_tryget(&position->css)) | |
519ebea3 JW |
1178 | position = NULL; |
1179 | } | |
1180 | return position; | |
1181 | } | |
1182 | ||
1183 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, | |
1184 | struct mem_cgroup *last_visited, | |
1185 | struct mem_cgroup *new_position, | |
ecc736fc | 1186 | struct mem_cgroup *root, |
519ebea3 JW |
1187 | int sequence) |
1188 | { | |
ecc736fc MH |
1189 | /* root reference counting symmetric to mem_cgroup_iter_load */ |
1190 | if (last_visited && last_visited != root) | |
519ebea3 JW |
1191 | css_put(&last_visited->css); |
1192 | /* | |
1193 | * We store the sequence count from the time @last_visited was | |
1194 | * loaded successfully instead of rereading it here so that we | |
1195 | * don't lose destruction events in between. We could have | |
1196 | * raced with the destruction of @new_position after all. | |
1197 | */ | |
1198 | iter->last_visited = new_position; | |
1199 | smp_wmb(); | |
1200 | iter->last_dead_count = sequence; | |
1201 | } | |
1202 | ||
5660048c JW |
1203 | /** |
1204 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1205 | * @root: hierarchy root | |
1206 | * @prev: previously returned memcg, NULL on first invocation | |
1207 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1208 | * | |
1209 | * Returns references to children of the hierarchy below @root, or | |
1210 | * @root itself, or %NULL after a full round-trip. | |
1211 | * | |
1212 | * Caller must pass the return value in @prev on subsequent | |
1213 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1214 | * to cancel a hierarchy walk before the round-trip is complete. | |
1215 | * | |
1216 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1217 | * divide up the memcgs in the hierarchy among all concurrent | |
1218 | * reclaimers operating on the same zone and priority. | |
1219 | */ | |
694fbc0f | 1220 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1221 | struct mem_cgroup *prev, |
694fbc0f | 1222 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1223 | { |
9f3a0d09 | 1224 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1225 | struct mem_cgroup *last_visited = NULL; |
711d3d2c | 1226 | |
694fbc0f AM |
1227 | if (mem_cgroup_disabled()) |
1228 | return NULL; | |
5660048c | 1229 | |
9f3a0d09 JW |
1230 | if (!root) |
1231 | root = root_mem_cgroup; | |
7d74b06f | 1232 | |
9f3a0d09 | 1233 | if (prev && !reclaim) |
542f85f9 | 1234 | last_visited = prev; |
14067bb3 | 1235 | |
9f3a0d09 JW |
1236 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1237 | if (prev) | |
c40046f3 | 1238 | goto out_css_put; |
694fbc0f | 1239 | return root; |
9f3a0d09 | 1240 | } |
14067bb3 | 1241 | |
542f85f9 | 1242 | rcu_read_lock(); |
9f3a0d09 | 1243 | while (!memcg) { |
527a5ec9 | 1244 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
519ebea3 | 1245 | int uninitialized_var(seq); |
711d3d2c | 1246 | |
527a5ec9 JW |
1247 | if (reclaim) { |
1248 | int nid = zone_to_nid(reclaim->zone); | |
1249 | int zid = zone_idx(reclaim->zone); | |
1250 | struct mem_cgroup_per_zone *mz; | |
1251 | ||
1252 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1253 | iter = &mz->reclaim_iter[reclaim->priority]; | |
542f85f9 | 1254 | if (prev && reclaim->generation != iter->generation) { |
5f578161 | 1255 | iter->last_visited = NULL; |
542f85f9 MH |
1256 | goto out_unlock; |
1257 | } | |
5f578161 | 1258 | |
519ebea3 | 1259 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
527a5ec9 | 1260 | } |
7d74b06f | 1261 | |
694fbc0f | 1262 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1263 | |
527a5ec9 | 1264 | if (reclaim) { |
ecc736fc MH |
1265 | mem_cgroup_iter_update(iter, last_visited, memcg, root, |
1266 | seq); | |
542f85f9 | 1267 | |
19f39402 | 1268 | if (!memcg) |
527a5ec9 JW |
1269 | iter->generation++; |
1270 | else if (!prev && memcg) | |
1271 | reclaim->generation = iter->generation; | |
1272 | } | |
9f3a0d09 | 1273 | |
694fbc0f | 1274 | if (prev && !memcg) |
542f85f9 | 1275 | goto out_unlock; |
9f3a0d09 | 1276 | } |
542f85f9 MH |
1277 | out_unlock: |
1278 | rcu_read_unlock(); | |
c40046f3 MH |
1279 | out_css_put: |
1280 | if (prev && prev != root) | |
1281 | css_put(&prev->css); | |
1282 | ||
9f3a0d09 | 1283 | return memcg; |
14067bb3 | 1284 | } |
7d74b06f | 1285 | |
5660048c JW |
1286 | /** |
1287 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1288 | * @root: hierarchy root | |
1289 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1290 | */ | |
1291 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1292 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1293 | { |
1294 | if (!root) | |
1295 | root = root_mem_cgroup; | |
1296 | if (prev && prev != root) | |
1297 | css_put(&prev->css); | |
1298 | } | |
7d74b06f | 1299 | |
9f3a0d09 JW |
1300 | /* |
1301 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1302 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1303 | * be used for reference counting. | |
1304 | */ | |
1305 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1306 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1307 | iter != NULL; \ |
527a5ec9 | 1308 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1309 | |
9f3a0d09 | 1310 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1311 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1312 | iter != NULL; \ |
527a5ec9 | 1313 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1314 | |
68ae564b | 1315 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1316 | { |
c0ff4b85 | 1317 | struct mem_cgroup *memcg; |
456f998e | 1318 | |
456f998e | 1319 | rcu_read_lock(); |
c0ff4b85 R |
1320 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1321 | if (unlikely(!memcg)) | |
456f998e YH |
1322 | goto out; |
1323 | ||
1324 | switch (idx) { | |
456f998e | 1325 | case PGFAULT: |
0e574a93 JW |
1326 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1327 | break; | |
1328 | case PGMAJFAULT: | |
1329 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1330 | break; |
1331 | default: | |
1332 | BUG(); | |
1333 | } | |
1334 | out: | |
1335 | rcu_read_unlock(); | |
1336 | } | |
68ae564b | 1337 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1338 | |
925b7673 JW |
1339 | /** |
1340 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1341 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1342 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1343 | * |
1344 | * Returns the lru list vector holding pages for the given @zone and | |
1345 | * @mem. This can be the global zone lruvec, if the memory controller | |
1346 | * is disabled. | |
1347 | */ | |
1348 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1349 | struct mem_cgroup *memcg) | |
1350 | { | |
1351 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1352 | struct lruvec *lruvec; |
925b7673 | 1353 | |
bea8c150 HD |
1354 | if (mem_cgroup_disabled()) { |
1355 | lruvec = &zone->lruvec; | |
1356 | goto out; | |
1357 | } | |
925b7673 JW |
1358 | |
1359 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1360 | lruvec = &mz->lruvec; |
1361 | out: | |
1362 | /* | |
1363 | * Since a node can be onlined after the mem_cgroup was created, | |
1364 | * we have to be prepared to initialize lruvec->zone here; | |
1365 | * and if offlined then reonlined, we need to reinitialize it. | |
1366 | */ | |
1367 | if (unlikely(lruvec->zone != zone)) | |
1368 | lruvec->zone = zone; | |
1369 | return lruvec; | |
925b7673 JW |
1370 | } |
1371 | ||
08e552c6 KH |
1372 | /* |
1373 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1374 | * Operations are called by routine of global LRU independently from memcg. | |
1375 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1376 | * | |
1377 | * Changes to pc->mem_cgroup happens when | |
1378 | * 1. charge | |
1379 | * 2. moving account | |
1380 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1381 | * It is added to LRU before charge. | |
1382 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1383 | * When moving account, the page is not on LRU. It's isolated. | |
1384 | */ | |
4f98a2fe | 1385 | |
925b7673 | 1386 | /** |
fa9add64 | 1387 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1388 | * @page: the page |
fa9add64 | 1389 | * @zone: zone of the page |
925b7673 | 1390 | */ |
fa9add64 | 1391 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1392 | { |
08e552c6 | 1393 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1394 | struct mem_cgroup *memcg; |
1395 | struct page_cgroup *pc; | |
bea8c150 | 1396 | struct lruvec *lruvec; |
6d12e2d8 | 1397 | |
bea8c150 HD |
1398 | if (mem_cgroup_disabled()) { |
1399 | lruvec = &zone->lruvec; | |
1400 | goto out; | |
1401 | } | |
925b7673 | 1402 | |
08e552c6 | 1403 | pc = lookup_page_cgroup(page); |
38c5d72f | 1404 | memcg = pc->mem_cgroup; |
7512102c HD |
1405 | |
1406 | /* | |
fa9add64 | 1407 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1408 | * an uncharged page off lru does nothing to secure |
1409 | * its former mem_cgroup from sudden removal. | |
1410 | * | |
1411 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1412 | * under page_cgroup lock: between them, they make all uses | |
1413 | * of pc->mem_cgroup safe. | |
1414 | */ | |
fa9add64 | 1415 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1416 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1417 | ||
925b7673 | 1418 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1419 | lruvec = &mz->lruvec; |
1420 | out: | |
1421 | /* | |
1422 | * Since a node can be onlined after the mem_cgroup was created, | |
1423 | * we have to be prepared to initialize lruvec->zone here; | |
1424 | * and if offlined then reonlined, we need to reinitialize it. | |
1425 | */ | |
1426 | if (unlikely(lruvec->zone != zone)) | |
1427 | lruvec->zone = zone; | |
1428 | return lruvec; | |
08e552c6 | 1429 | } |
b69408e8 | 1430 | |
925b7673 | 1431 | /** |
fa9add64 HD |
1432 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1433 | * @lruvec: mem_cgroup per zone lru vector | |
1434 | * @lru: index of lru list the page is sitting on | |
1435 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1436 | * |
fa9add64 HD |
1437 | * This function must be called when a page is added to or removed from an |
1438 | * lru list. | |
3f58a829 | 1439 | */ |
fa9add64 HD |
1440 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1441 | int nr_pages) | |
3f58a829 MK |
1442 | { |
1443 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1444 | unsigned long *lru_size; |
3f58a829 MK |
1445 | |
1446 | if (mem_cgroup_disabled()) | |
1447 | return; | |
1448 | ||
fa9add64 HD |
1449 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1450 | lru_size = mz->lru_size + lru; | |
1451 | *lru_size += nr_pages; | |
1452 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1453 | } |
544122e5 | 1454 | |
3e92041d | 1455 | /* |
c0ff4b85 | 1456 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1457 | * hierarchy subtree |
1458 | */ | |
c3ac9a8a JW |
1459 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1460 | struct mem_cgroup *memcg) | |
3e92041d | 1461 | { |
91c63734 JW |
1462 | if (root_memcg == memcg) |
1463 | return true; | |
3a981f48 | 1464 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1465 | return false; |
b47f77b5 | 1466 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1467 | } |
1468 | ||
1469 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1470 | struct mem_cgroup *memcg) | |
1471 | { | |
1472 | bool ret; | |
1473 | ||
91c63734 | 1474 | rcu_read_lock(); |
c3ac9a8a | 1475 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1476 | rcu_read_unlock(); |
1477 | return ret; | |
3e92041d MH |
1478 | } |
1479 | ||
ffbdccf5 DR |
1480 | bool task_in_mem_cgroup(struct task_struct *task, |
1481 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1482 | { |
0b7f569e | 1483 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1484 | struct task_struct *p; |
ffbdccf5 | 1485 | bool ret; |
4c4a2214 | 1486 | |
158e0a2d | 1487 | p = find_lock_task_mm(task); |
de077d22 DR |
1488 | if (p) { |
1489 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1490 | task_unlock(p); | |
1491 | } else { | |
1492 | /* | |
1493 | * All threads may have already detached their mm's, but the oom | |
1494 | * killer still needs to detect if they have already been oom | |
1495 | * killed to prevent needlessly killing additional tasks. | |
1496 | */ | |
ffbdccf5 | 1497 | rcu_read_lock(); |
de077d22 DR |
1498 | curr = mem_cgroup_from_task(task); |
1499 | if (curr) | |
1500 | css_get(&curr->css); | |
ffbdccf5 | 1501 | rcu_read_unlock(); |
de077d22 | 1502 | } |
0b7f569e | 1503 | if (!curr) |
ffbdccf5 | 1504 | return false; |
d31f56db | 1505 | /* |
c0ff4b85 | 1506 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1507 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1508 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1509 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1510 | */ |
c0ff4b85 | 1511 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1512 | css_put(&curr->css); |
4c4a2214 DR |
1513 | return ret; |
1514 | } | |
1515 | ||
c56d5c7d | 1516 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1517 | { |
9b272977 | 1518 | unsigned long inactive_ratio; |
14797e23 | 1519 | unsigned long inactive; |
9b272977 | 1520 | unsigned long active; |
c772be93 | 1521 | unsigned long gb; |
14797e23 | 1522 | |
4d7dcca2 HD |
1523 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1524 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1525 | |
c772be93 KM |
1526 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1527 | if (gb) | |
1528 | inactive_ratio = int_sqrt(10 * gb); | |
1529 | else | |
1530 | inactive_ratio = 1; | |
1531 | ||
9b272977 | 1532 | return inactive * inactive_ratio < active; |
14797e23 KM |
1533 | } |
1534 | ||
6d61ef40 BS |
1535 | #define mem_cgroup_from_res_counter(counter, member) \ |
1536 | container_of(counter, struct mem_cgroup, member) | |
1537 | ||
19942822 | 1538 | /** |
9d11ea9f | 1539 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1540 | * @memcg: the memory cgroup |
19942822 | 1541 | * |
9d11ea9f | 1542 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1543 | * pages. |
19942822 | 1544 | */ |
c0ff4b85 | 1545 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1546 | { |
9d11ea9f JW |
1547 | unsigned long long margin; |
1548 | ||
c0ff4b85 | 1549 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1550 | if (do_swap_account) |
c0ff4b85 | 1551 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1552 | return margin >> PAGE_SHIFT; |
19942822 JW |
1553 | } |
1554 | ||
1f4c025b | 1555 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1556 | { |
a7885eb8 | 1557 | /* root ? */ |
63876986 | 1558 | if (!css_parent(&memcg->css)) |
a7885eb8 KM |
1559 | return vm_swappiness; |
1560 | ||
bf1ff263 | 1561 | return memcg->swappiness; |
a7885eb8 KM |
1562 | } |
1563 | ||
619d094b KH |
1564 | /* |
1565 | * memcg->moving_account is used for checking possibility that some thread is | |
1566 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1567 | * a memcg, other threads should check memcg->moving_account under | |
1568 | * rcu_read_lock(), like this: | |
1569 | * | |
1570 | * CPU-A CPU-B | |
1571 | * rcu_read_lock() | |
1572 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1573 | * take heavy locks. | |
1574 | * synchronize_rcu() update something. | |
1575 | * rcu_read_unlock() | |
1576 | * start move here. | |
1577 | */ | |
4331f7d3 KH |
1578 | |
1579 | /* for quick checking without looking up memcg */ | |
1580 | atomic_t memcg_moving __read_mostly; | |
1581 | ||
c0ff4b85 | 1582 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1583 | { |
4331f7d3 | 1584 | atomic_inc(&memcg_moving); |
619d094b | 1585 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1586 | synchronize_rcu(); |
1587 | } | |
1588 | ||
c0ff4b85 | 1589 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1590 | { |
619d094b KH |
1591 | /* |
1592 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1593 | * We check NULL in callee rather than caller. | |
1594 | */ | |
4331f7d3 KH |
1595 | if (memcg) { |
1596 | atomic_dec(&memcg_moving); | |
619d094b | 1597 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1598 | } |
32047e2a | 1599 | } |
619d094b | 1600 | |
32047e2a KH |
1601 | /* |
1602 | * 2 routines for checking "mem" is under move_account() or not. | |
1603 | * | |
13fd1dd9 AM |
1604 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1605 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1606 | * pc->mem_cgroup may be overwritten. |
1607 | * | |
1608 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1609 | * under hierarchy of moving cgroups. This is for | |
1610 | * waiting at hith-memory prressure caused by "move". | |
1611 | */ | |
1612 | ||
13fd1dd9 | 1613 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1614 | { |
1615 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1616 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1617 | } |
4b534334 | 1618 | |
c0ff4b85 | 1619 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1620 | { |
2bd9bb20 KH |
1621 | struct mem_cgroup *from; |
1622 | struct mem_cgroup *to; | |
4b534334 | 1623 | bool ret = false; |
2bd9bb20 KH |
1624 | /* |
1625 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1626 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1627 | */ | |
1628 | spin_lock(&mc.lock); | |
1629 | from = mc.from; | |
1630 | to = mc.to; | |
1631 | if (!from) | |
1632 | goto unlock; | |
3e92041d | 1633 | |
c0ff4b85 R |
1634 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1635 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1636 | unlock: |
1637 | spin_unlock(&mc.lock); | |
4b534334 KH |
1638 | return ret; |
1639 | } | |
1640 | ||
c0ff4b85 | 1641 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1642 | { |
1643 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1644 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1645 | DEFINE_WAIT(wait); |
1646 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1647 | /* moving charge context might have finished. */ | |
1648 | if (mc.moving_task) | |
1649 | schedule(); | |
1650 | finish_wait(&mc.waitq, &wait); | |
1651 | return true; | |
1652 | } | |
1653 | } | |
1654 | return false; | |
1655 | } | |
1656 | ||
312734c0 KH |
1657 | /* |
1658 | * Take this lock when | |
1659 | * - a code tries to modify page's memcg while it's USED. | |
1660 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1661 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1662 | */ |
1663 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1664 | unsigned long *flags) | |
1665 | { | |
1666 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1667 | } | |
1668 | ||
1669 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1670 | unsigned long *flags) | |
1671 | { | |
1672 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1673 | } | |
1674 | ||
58cf188e | 1675 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1676 | /** |
58cf188e | 1677 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1678 | * @memcg: The memory cgroup that went over limit |
1679 | * @p: Task that is going to be killed | |
1680 | * | |
1681 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1682 | * enabled | |
1683 | */ | |
1684 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1685 | { | |
e61734c5 | 1686 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1687 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1688 | struct mem_cgroup *iter; |
1689 | unsigned int i; | |
e222432b | 1690 | |
58cf188e | 1691 | if (!p) |
e222432b BS |
1692 | return; |
1693 | ||
08088cb9 | 1694 | mutex_lock(&oom_info_lock); |
e222432b BS |
1695 | rcu_read_lock(); |
1696 | ||
e61734c5 TH |
1697 | pr_info("Task in "); |
1698 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1699 | pr_info(" killed as a result of limit of "); | |
1700 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1701 | pr_info("\n"); | |
e222432b | 1702 | |
e222432b BS |
1703 | rcu_read_unlock(); |
1704 | ||
d045197f | 1705 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1706 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1707 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1708 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1709 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1710 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1711 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1712 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1713 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1714 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1715 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1716 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1717 | |
1718 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1719 | pr_info("Memory cgroup stats for "); |
1720 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1721 | pr_cont(":"); |
1722 | ||
1723 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1724 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1725 | continue; | |
1726 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1727 | K(mem_cgroup_read_stat(iter, i))); | |
1728 | } | |
1729 | ||
1730 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1731 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1732 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1733 | ||
1734 | pr_cont("\n"); | |
1735 | } | |
08088cb9 | 1736 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1737 | } |
1738 | ||
81d39c20 KH |
1739 | /* |
1740 | * This function returns the number of memcg under hierarchy tree. Returns | |
1741 | * 1(self count) if no children. | |
1742 | */ | |
c0ff4b85 | 1743 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1744 | { |
1745 | int num = 0; | |
7d74b06f KH |
1746 | struct mem_cgroup *iter; |
1747 | ||
c0ff4b85 | 1748 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1749 | num++; |
81d39c20 KH |
1750 | return num; |
1751 | } | |
1752 | ||
a63d83f4 DR |
1753 | /* |
1754 | * Return the memory (and swap, if configured) limit for a memcg. | |
1755 | */ | |
9cbb78bb | 1756 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1757 | { |
1758 | u64 limit; | |
a63d83f4 | 1759 | |
f3e8eb70 | 1760 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1761 | |
a63d83f4 | 1762 | /* |
9a5a8f19 | 1763 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1764 | */ |
9a5a8f19 MH |
1765 | if (mem_cgroup_swappiness(memcg)) { |
1766 | u64 memsw; | |
1767 | ||
1768 | limit += total_swap_pages << PAGE_SHIFT; | |
1769 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1770 | ||
1771 | /* | |
1772 | * If memsw is finite and limits the amount of swap space | |
1773 | * available to this memcg, return that limit. | |
1774 | */ | |
1775 | limit = min(limit, memsw); | |
1776 | } | |
1777 | ||
1778 | return limit; | |
a63d83f4 DR |
1779 | } |
1780 | ||
19965460 DR |
1781 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1782 | int order) | |
9cbb78bb DR |
1783 | { |
1784 | struct mem_cgroup *iter; | |
1785 | unsigned long chosen_points = 0; | |
1786 | unsigned long totalpages; | |
1787 | unsigned int points = 0; | |
1788 | struct task_struct *chosen = NULL; | |
1789 | ||
876aafbf | 1790 | /* |
465adcf1 DR |
1791 | * If current has a pending SIGKILL or is exiting, then automatically |
1792 | * select it. The goal is to allow it to allocate so that it may | |
1793 | * quickly exit and free its memory. | |
876aafbf | 1794 | */ |
465adcf1 | 1795 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1796 | set_thread_flag(TIF_MEMDIE); |
1797 | return; | |
1798 | } | |
1799 | ||
1800 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1801 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1802 | for_each_mem_cgroup_tree(iter, memcg) { | |
72ec7029 | 1803 | struct css_task_iter it; |
9cbb78bb DR |
1804 | struct task_struct *task; |
1805 | ||
72ec7029 TH |
1806 | css_task_iter_start(&iter->css, &it); |
1807 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1808 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1809 | false)) { | |
1810 | case OOM_SCAN_SELECT: | |
1811 | if (chosen) | |
1812 | put_task_struct(chosen); | |
1813 | chosen = task; | |
1814 | chosen_points = ULONG_MAX; | |
1815 | get_task_struct(chosen); | |
1816 | /* fall through */ | |
1817 | case OOM_SCAN_CONTINUE: | |
1818 | continue; | |
1819 | case OOM_SCAN_ABORT: | |
72ec7029 | 1820 | css_task_iter_end(&it); |
9cbb78bb DR |
1821 | mem_cgroup_iter_break(memcg, iter); |
1822 | if (chosen) | |
1823 | put_task_struct(chosen); | |
1824 | return; | |
1825 | case OOM_SCAN_OK: | |
1826 | break; | |
1827 | }; | |
1828 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1829 | if (!points || points < chosen_points) |
1830 | continue; | |
1831 | /* Prefer thread group leaders for display purposes */ | |
1832 | if (points == chosen_points && | |
1833 | thread_group_leader(chosen)) | |
1834 | continue; | |
1835 | ||
1836 | if (chosen) | |
1837 | put_task_struct(chosen); | |
1838 | chosen = task; | |
1839 | chosen_points = points; | |
1840 | get_task_struct(chosen); | |
9cbb78bb | 1841 | } |
72ec7029 | 1842 | css_task_iter_end(&it); |
9cbb78bb DR |
1843 | } |
1844 | ||
1845 | if (!chosen) | |
1846 | return; | |
1847 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1848 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1849 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1850 | } |
1851 | ||
5660048c JW |
1852 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1853 | gfp_t gfp_mask, | |
1854 | unsigned long flags) | |
1855 | { | |
1856 | unsigned long total = 0; | |
1857 | bool noswap = false; | |
1858 | int loop; | |
1859 | ||
1860 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1861 | noswap = true; | |
1862 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1863 | noswap = true; | |
1864 | ||
1865 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1866 | if (loop) | |
1867 | drain_all_stock_async(memcg); | |
1868 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1869 | /* | |
1870 | * Allow limit shrinkers, which are triggered directly | |
1871 | * by userspace, to catch signals and stop reclaim | |
1872 | * after minimal progress, regardless of the margin. | |
1873 | */ | |
1874 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1875 | break; | |
1876 | if (mem_cgroup_margin(memcg)) | |
1877 | break; | |
1878 | /* | |
1879 | * If nothing was reclaimed after two attempts, there | |
1880 | * may be no reclaimable pages in this hierarchy. | |
1881 | */ | |
1882 | if (loop && !total) | |
1883 | break; | |
1884 | } | |
1885 | return total; | |
1886 | } | |
1887 | ||
4d0c066d KH |
1888 | /** |
1889 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1890 | * @memcg: the target memcg |
4d0c066d KH |
1891 | * @nid: the node ID to be checked. |
1892 | * @noswap : specify true here if the user wants flle only information. | |
1893 | * | |
1894 | * This function returns whether the specified memcg contains any | |
1895 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1896 | * pages in the node. | |
1897 | */ | |
c0ff4b85 | 1898 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1899 | int nid, bool noswap) |
1900 | { | |
c0ff4b85 | 1901 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1902 | return true; |
1903 | if (noswap || !total_swap_pages) | |
1904 | return false; | |
c0ff4b85 | 1905 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1906 | return true; |
1907 | return false; | |
1908 | ||
1909 | } | |
bb4cc1a8 | 1910 | #if MAX_NUMNODES > 1 |
889976db YH |
1911 | |
1912 | /* | |
1913 | * Always updating the nodemask is not very good - even if we have an empty | |
1914 | * list or the wrong list here, we can start from some node and traverse all | |
1915 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1916 | * | |
1917 | */ | |
c0ff4b85 | 1918 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1919 | { |
1920 | int nid; | |
453a9bf3 KH |
1921 | /* |
1922 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1923 | * pagein/pageout changes since the last update. | |
1924 | */ | |
c0ff4b85 | 1925 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1926 | return; |
c0ff4b85 | 1927 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1928 | return; |
1929 | ||
889976db | 1930 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1931 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1932 | |
31aaea4a | 1933 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1934 | |
c0ff4b85 R |
1935 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1936 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1937 | } |
453a9bf3 | 1938 | |
c0ff4b85 R |
1939 | atomic_set(&memcg->numainfo_events, 0); |
1940 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1941 | } |
1942 | ||
1943 | /* | |
1944 | * Selecting a node where we start reclaim from. Because what we need is just | |
1945 | * reducing usage counter, start from anywhere is O,K. Considering | |
1946 | * memory reclaim from current node, there are pros. and cons. | |
1947 | * | |
1948 | * Freeing memory from current node means freeing memory from a node which | |
1949 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1950 | * hit limits, it will see a contention on a node. But freeing from remote | |
1951 | * node means more costs for memory reclaim because of memory latency. | |
1952 | * | |
1953 | * Now, we use round-robin. Better algorithm is welcomed. | |
1954 | */ | |
c0ff4b85 | 1955 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1956 | { |
1957 | int node; | |
1958 | ||
c0ff4b85 R |
1959 | mem_cgroup_may_update_nodemask(memcg); |
1960 | node = memcg->last_scanned_node; | |
889976db | 1961 | |
c0ff4b85 | 1962 | node = next_node(node, memcg->scan_nodes); |
889976db | 1963 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1964 | node = first_node(memcg->scan_nodes); |
889976db YH |
1965 | /* |
1966 | * We call this when we hit limit, not when pages are added to LRU. | |
1967 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1968 | * memcg is too small and all pages are not on LRU. In that case, | |
1969 | * we use curret node. | |
1970 | */ | |
1971 | if (unlikely(node == MAX_NUMNODES)) | |
1972 | node = numa_node_id(); | |
1973 | ||
c0ff4b85 | 1974 | memcg->last_scanned_node = node; |
889976db YH |
1975 | return node; |
1976 | } | |
1977 | ||
bb4cc1a8 AM |
1978 | /* |
1979 | * Check all nodes whether it contains reclaimable pages or not. | |
1980 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1981 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1982 | * enough new information. We need to do double check. | |
1983 | */ | |
1984 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1985 | { | |
1986 | int nid; | |
1987 | ||
1988 | /* | |
1989 | * quick check...making use of scan_node. | |
1990 | * We can skip unused nodes. | |
1991 | */ | |
1992 | if (!nodes_empty(memcg->scan_nodes)) { | |
1993 | for (nid = first_node(memcg->scan_nodes); | |
1994 | nid < MAX_NUMNODES; | |
1995 | nid = next_node(nid, memcg->scan_nodes)) { | |
1996 | ||
1997 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1998 | return true; | |
1999 | } | |
2000 | } | |
2001 | /* | |
2002 | * Check rest of nodes. | |
2003 | */ | |
2004 | for_each_node_state(nid, N_MEMORY) { | |
2005 | if (node_isset(nid, memcg->scan_nodes)) | |
2006 | continue; | |
2007 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
2008 | return true; | |
2009 | } | |
2010 | return false; | |
2011 | } | |
2012 | ||
889976db | 2013 | #else |
c0ff4b85 | 2014 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
2015 | { |
2016 | return 0; | |
2017 | } | |
4d0c066d | 2018 | |
bb4cc1a8 AM |
2019 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
2020 | { | |
2021 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
2022 | } | |
889976db YH |
2023 | #endif |
2024 | ||
0608f43d AM |
2025 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
2026 | struct zone *zone, | |
2027 | gfp_t gfp_mask, | |
2028 | unsigned long *total_scanned) | |
2029 | { | |
2030 | struct mem_cgroup *victim = NULL; | |
2031 | int total = 0; | |
2032 | int loop = 0; | |
2033 | unsigned long excess; | |
2034 | unsigned long nr_scanned; | |
2035 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2036 | .zone = zone, | |
2037 | .priority = 0, | |
2038 | }; | |
2039 | ||
2040 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; | |
2041 | ||
2042 | while (1) { | |
2043 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
2044 | if (!victim) { | |
2045 | loop++; | |
2046 | if (loop >= 2) { | |
2047 | /* | |
2048 | * If we have not been able to reclaim | |
2049 | * anything, it might because there are | |
2050 | * no reclaimable pages under this hierarchy | |
2051 | */ | |
2052 | if (!total) | |
2053 | break; | |
2054 | /* | |
2055 | * We want to do more targeted reclaim. | |
2056 | * excess >> 2 is not to excessive so as to | |
2057 | * reclaim too much, nor too less that we keep | |
2058 | * coming back to reclaim from this cgroup | |
2059 | */ | |
2060 | if (total >= (excess >> 2) || | |
2061 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
2062 | break; | |
2063 | } | |
2064 | continue; | |
2065 | } | |
2066 | if (!mem_cgroup_reclaimable(victim, false)) | |
2067 | continue; | |
2068 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
2069 | zone, &nr_scanned); | |
2070 | *total_scanned += nr_scanned; | |
2071 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
2072 | break; | |
6d61ef40 | 2073 | } |
0608f43d AM |
2074 | mem_cgroup_iter_break(root_memcg, victim); |
2075 | return total; | |
6d61ef40 BS |
2076 | } |
2077 | ||
0056f4e6 JW |
2078 | #ifdef CONFIG_LOCKDEP |
2079 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
2080 | .name = "memcg_oom_lock", | |
2081 | }; | |
2082 | #endif | |
2083 | ||
fb2a6fc5 JW |
2084 | static DEFINE_SPINLOCK(memcg_oom_lock); |
2085 | ||
867578cb KH |
2086 | /* |
2087 | * Check OOM-Killer is already running under our hierarchy. | |
2088 | * If someone is running, return false. | |
2089 | */ | |
fb2a6fc5 | 2090 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 2091 | { |
79dfdacc | 2092 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2093 | |
fb2a6fc5 JW |
2094 | spin_lock(&memcg_oom_lock); |
2095 | ||
9f3a0d09 | 2096 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2097 | if (iter->oom_lock) { |
79dfdacc MH |
2098 | /* |
2099 | * this subtree of our hierarchy is already locked | |
2100 | * so we cannot give a lock. | |
2101 | */ | |
79dfdacc | 2102 | failed = iter; |
9f3a0d09 JW |
2103 | mem_cgroup_iter_break(memcg, iter); |
2104 | break; | |
23751be0 JW |
2105 | } else |
2106 | iter->oom_lock = true; | |
7d74b06f | 2107 | } |
867578cb | 2108 | |
fb2a6fc5 JW |
2109 | if (failed) { |
2110 | /* | |
2111 | * OK, we failed to lock the whole subtree so we have | |
2112 | * to clean up what we set up to the failing subtree | |
2113 | */ | |
2114 | for_each_mem_cgroup_tree(iter, memcg) { | |
2115 | if (iter == failed) { | |
2116 | mem_cgroup_iter_break(memcg, iter); | |
2117 | break; | |
2118 | } | |
2119 | iter->oom_lock = false; | |
79dfdacc | 2120 | } |
0056f4e6 JW |
2121 | } else |
2122 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
2123 | |
2124 | spin_unlock(&memcg_oom_lock); | |
2125 | ||
2126 | return !failed; | |
a636b327 | 2127 | } |
0b7f569e | 2128 | |
fb2a6fc5 | 2129 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2130 | { |
7d74b06f KH |
2131 | struct mem_cgroup *iter; |
2132 | ||
fb2a6fc5 | 2133 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 2134 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 2135 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2136 | iter->oom_lock = false; |
fb2a6fc5 | 2137 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
2138 | } |
2139 | ||
c0ff4b85 | 2140 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2141 | { |
2142 | struct mem_cgroup *iter; | |
2143 | ||
c0ff4b85 | 2144 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2145 | atomic_inc(&iter->under_oom); |
2146 | } | |
2147 | ||
c0ff4b85 | 2148 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2149 | { |
2150 | struct mem_cgroup *iter; | |
2151 | ||
867578cb KH |
2152 | /* |
2153 | * When a new child is created while the hierarchy is under oom, | |
2154 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2155 | * atomic_add_unless() here. | |
2156 | */ | |
c0ff4b85 | 2157 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2158 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2159 | } |
2160 | ||
867578cb KH |
2161 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2162 | ||
dc98df5a | 2163 | struct oom_wait_info { |
d79154bb | 2164 | struct mem_cgroup *memcg; |
dc98df5a KH |
2165 | wait_queue_t wait; |
2166 | }; | |
2167 | ||
2168 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2169 | unsigned mode, int sync, void *arg) | |
2170 | { | |
d79154bb HD |
2171 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2172 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2173 | struct oom_wait_info *oom_wait_info; |
2174 | ||
2175 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2176 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2177 | |
dc98df5a | 2178 | /* |
d79154bb | 2179 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2180 | * Then we can use css_is_ancestor without taking care of RCU. |
2181 | */ | |
c0ff4b85 R |
2182 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2183 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2184 | return 0; |
dc98df5a KH |
2185 | return autoremove_wake_function(wait, mode, sync, arg); |
2186 | } | |
2187 | ||
c0ff4b85 | 2188 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2189 | { |
3812c8c8 | 2190 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2191 | /* for filtering, pass "memcg" as argument. */ |
2192 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2193 | } |
2194 | ||
c0ff4b85 | 2195 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2196 | { |
c0ff4b85 R |
2197 | if (memcg && atomic_read(&memcg->under_oom)) |
2198 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2199 | } |
2200 | ||
3812c8c8 | 2201 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2202 | { |
3812c8c8 JW |
2203 | if (!current->memcg_oom.may_oom) |
2204 | return; | |
867578cb | 2205 | /* |
49426420 JW |
2206 | * We are in the middle of the charge context here, so we |
2207 | * don't want to block when potentially sitting on a callstack | |
2208 | * that holds all kinds of filesystem and mm locks. | |
2209 | * | |
2210 | * Also, the caller may handle a failed allocation gracefully | |
2211 | * (like optional page cache readahead) and so an OOM killer | |
2212 | * invocation might not even be necessary. | |
2213 | * | |
2214 | * That's why we don't do anything here except remember the | |
2215 | * OOM context and then deal with it at the end of the page | |
2216 | * fault when the stack is unwound, the locks are released, | |
2217 | * and when we know whether the fault was overall successful. | |
867578cb | 2218 | */ |
49426420 JW |
2219 | css_get(&memcg->css); |
2220 | current->memcg_oom.memcg = memcg; | |
2221 | current->memcg_oom.gfp_mask = mask; | |
2222 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2223 | } |
2224 | ||
2225 | /** | |
2226 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2227 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2228 | * |
49426420 JW |
2229 | * This has to be called at the end of a page fault if the memcg OOM |
2230 | * handler was enabled. | |
3812c8c8 | 2231 | * |
49426420 | 2232 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2233 | * sleep on a waitqueue until the userspace task resolves the |
2234 | * situation. Sleeping directly in the charge context with all kinds | |
2235 | * of locks held is not a good idea, instead we remember an OOM state | |
2236 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2237 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2238 | * |
2239 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2240 | * completed, %false otherwise. |
3812c8c8 | 2241 | */ |
49426420 | 2242 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2243 | { |
49426420 | 2244 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2245 | struct oom_wait_info owait; |
49426420 | 2246 | bool locked; |
3812c8c8 JW |
2247 | |
2248 | /* OOM is global, do not handle */ | |
3812c8c8 | 2249 | if (!memcg) |
49426420 | 2250 | return false; |
3812c8c8 | 2251 | |
49426420 JW |
2252 | if (!handle) |
2253 | goto cleanup; | |
3812c8c8 JW |
2254 | |
2255 | owait.memcg = memcg; | |
2256 | owait.wait.flags = 0; | |
2257 | owait.wait.func = memcg_oom_wake_function; | |
2258 | owait.wait.private = current; | |
2259 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2260 | |
3812c8c8 | 2261 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2262 | mem_cgroup_mark_under_oom(memcg); |
2263 | ||
2264 | locked = mem_cgroup_oom_trylock(memcg); | |
2265 | ||
2266 | if (locked) | |
2267 | mem_cgroup_oom_notify(memcg); | |
2268 | ||
2269 | if (locked && !memcg->oom_kill_disable) { | |
2270 | mem_cgroup_unmark_under_oom(memcg); | |
2271 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2272 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2273 | current->memcg_oom.order); | |
2274 | } else { | |
3812c8c8 | 2275 | schedule(); |
49426420 JW |
2276 | mem_cgroup_unmark_under_oom(memcg); |
2277 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2278 | } | |
2279 | ||
2280 | if (locked) { | |
fb2a6fc5 JW |
2281 | mem_cgroup_oom_unlock(memcg); |
2282 | /* | |
2283 | * There is no guarantee that an OOM-lock contender | |
2284 | * sees the wakeups triggered by the OOM kill | |
2285 | * uncharges. Wake any sleepers explicitely. | |
2286 | */ | |
2287 | memcg_oom_recover(memcg); | |
2288 | } | |
49426420 JW |
2289 | cleanup: |
2290 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2291 | css_put(&memcg->css); |
867578cb | 2292 | return true; |
0b7f569e KH |
2293 | } |
2294 | ||
d69b042f BS |
2295 | /* |
2296 | * Currently used to update mapped file statistics, but the routine can be | |
2297 | * generalized to update other statistics as well. | |
32047e2a KH |
2298 | * |
2299 | * Notes: Race condition | |
2300 | * | |
2301 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2302 | * it tends to be costly. But considering some conditions, we doesn't need | |
2303 | * to do so _always_. | |
2304 | * | |
2305 | * Considering "charge", lock_page_cgroup() is not required because all | |
2306 | * file-stat operations happen after a page is attached to radix-tree. There | |
2307 | * are no race with "charge". | |
2308 | * | |
2309 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2310 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2311 | * if there are race with "uncharge". Statistics itself is properly handled | |
2312 | * by flags. | |
2313 | * | |
2314 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2315 | * small, we check mm->moving_account and detect there are possibility of race |
2316 | * If there is, we take a lock. | |
d69b042f | 2317 | */ |
26174efd | 2318 | |
89c06bd5 KH |
2319 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2320 | bool *locked, unsigned long *flags) | |
2321 | { | |
2322 | struct mem_cgroup *memcg; | |
2323 | struct page_cgroup *pc; | |
2324 | ||
2325 | pc = lookup_page_cgroup(page); | |
2326 | again: | |
2327 | memcg = pc->mem_cgroup; | |
2328 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2329 | return; | |
2330 | /* | |
2331 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2332 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2333 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2334 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2335 | */ |
13fd1dd9 | 2336 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2337 | return; |
2338 | ||
2339 | move_lock_mem_cgroup(memcg, flags); | |
2340 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2341 | move_unlock_mem_cgroup(memcg, flags); | |
2342 | goto again; | |
2343 | } | |
2344 | *locked = true; | |
2345 | } | |
2346 | ||
2347 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2348 | { | |
2349 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2350 | ||
2351 | /* | |
2352 | * It's guaranteed that pc->mem_cgroup never changes while | |
2353 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2354 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2355 | */ |
2356 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2357 | } | |
2358 | ||
2a7106f2 | 2359 | void mem_cgroup_update_page_stat(struct page *page, |
68b4876d | 2360 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2361 | { |
c0ff4b85 | 2362 | struct mem_cgroup *memcg; |
32047e2a | 2363 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2364 | unsigned long uninitialized_var(flags); |
d69b042f | 2365 | |
cfa44946 | 2366 | if (mem_cgroup_disabled()) |
d69b042f | 2367 | return; |
89c06bd5 | 2368 | |
658b72c5 | 2369 | VM_BUG_ON(!rcu_read_lock_held()); |
c0ff4b85 R |
2370 | memcg = pc->mem_cgroup; |
2371 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2372 | return; |
26174efd | 2373 | |
c0ff4b85 | 2374 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2375 | } |
26174efd | 2376 | |
cdec2e42 KH |
2377 | /* |
2378 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2379 | * TODO: maybe necessary to use big numbers in big irons. | |
2380 | */ | |
7ec99d62 | 2381 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2382 | struct memcg_stock_pcp { |
2383 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2384 | unsigned int nr_pages; |
cdec2e42 | 2385 | struct work_struct work; |
26fe6168 | 2386 | unsigned long flags; |
a0db00fc | 2387 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2388 | }; |
2389 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2390 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2391 | |
a0956d54 SS |
2392 | /** |
2393 | * consume_stock: Try to consume stocked charge on this cpu. | |
2394 | * @memcg: memcg to consume from. | |
2395 | * @nr_pages: how many pages to charge. | |
2396 | * | |
2397 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2398 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2399 | * service an allocation will refill the stock. | |
2400 | * | |
2401 | * returns true if successful, false otherwise. | |
cdec2e42 | 2402 | */ |
a0956d54 | 2403 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2404 | { |
2405 | struct memcg_stock_pcp *stock; | |
2406 | bool ret = true; | |
2407 | ||
a0956d54 SS |
2408 | if (nr_pages > CHARGE_BATCH) |
2409 | return false; | |
2410 | ||
cdec2e42 | 2411 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2412 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2413 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2414 | else /* need to call res_counter_charge */ |
2415 | ret = false; | |
2416 | put_cpu_var(memcg_stock); | |
2417 | return ret; | |
2418 | } | |
2419 | ||
2420 | /* | |
2421 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2422 | */ | |
2423 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2424 | { | |
2425 | struct mem_cgroup *old = stock->cached; | |
2426 | ||
11c9ea4e JW |
2427 | if (stock->nr_pages) { |
2428 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2429 | ||
2430 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2431 | if (do_swap_account) |
11c9ea4e JW |
2432 | res_counter_uncharge(&old->memsw, bytes); |
2433 | stock->nr_pages = 0; | |
cdec2e42 KH |
2434 | } |
2435 | stock->cached = NULL; | |
cdec2e42 KH |
2436 | } |
2437 | ||
2438 | /* | |
2439 | * This must be called under preempt disabled or must be called by | |
2440 | * a thread which is pinned to local cpu. | |
2441 | */ | |
2442 | static void drain_local_stock(struct work_struct *dummy) | |
2443 | { | |
2444 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2445 | drain_stock(stock); | |
26fe6168 | 2446 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2447 | } |
2448 | ||
e4777496 MH |
2449 | static void __init memcg_stock_init(void) |
2450 | { | |
2451 | int cpu; | |
2452 | ||
2453 | for_each_possible_cpu(cpu) { | |
2454 | struct memcg_stock_pcp *stock = | |
2455 | &per_cpu(memcg_stock, cpu); | |
2456 | INIT_WORK(&stock->work, drain_local_stock); | |
2457 | } | |
2458 | } | |
2459 | ||
cdec2e42 KH |
2460 | /* |
2461 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2462 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2463 | */ |
c0ff4b85 | 2464 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2465 | { |
2466 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2467 | ||
c0ff4b85 | 2468 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2469 | drain_stock(stock); |
c0ff4b85 | 2470 | stock->cached = memcg; |
cdec2e42 | 2471 | } |
11c9ea4e | 2472 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2473 | put_cpu_var(memcg_stock); |
2474 | } | |
2475 | ||
2476 | /* | |
c0ff4b85 | 2477 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2478 | * of the hierarchy under it. sync flag says whether we should block |
2479 | * until the work is done. | |
cdec2e42 | 2480 | */ |
c0ff4b85 | 2481 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2482 | { |
26fe6168 | 2483 | int cpu, curcpu; |
d38144b7 | 2484 | |
cdec2e42 | 2485 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2486 | get_online_cpus(); |
5af12d0e | 2487 | curcpu = get_cpu(); |
cdec2e42 KH |
2488 | for_each_online_cpu(cpu) { |
2489 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2490 | struct mem_cgroup *memcg; |
26fe6168 | 2491 | |
c0ff4b85 R |
2492 | memcg = stock->cached; |
2493 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2494 | continue; |
c0ff4b85 | 2495 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2496 | continue; |
d1a05b69 MH |
2497 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2498 | if (cpu == curcpu) | |
2499 | drain_local_stock(&stock->work); | |
2500 | else | |
2501 | schedule_work_on(cpu, &stock->work); | |
2502 | } | |
cdec2e42 | 2503 | } |
5af12d0e | 2504 | put_cpu(); |
d38144b7 MH |
2505 | |
2506 | if (!sync) | |
2507 | goto out; | |
2508 | ||
2509 | for_each_online_cpu(cpu) { | |
2510 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2511 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2512 | flush_work(&stock->work); |
2513 | } | |
2514 | out: | |
f894ffa8 | 2515 | put_online_cpus(); |
d38144b7 MH |
2516 | } |
2517 | ||
2518 | /* | |
2519 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2520 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2521 | * expects some charges will be back to res_counter later but cannot wait for | |
2522 | * it. | |
2523 | */ | |
c0ff4b85 | 2524 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2525 | { |
9f50fad6 MH |
2526 | /* |
2527 | * If someone calls draining, avoid adding more kworker runs. | |
2528 | */ | |
2529 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2530 | return; | |
c0ff4b85 | 2531 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2532 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2533 | } |
2534 | ||
2535 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2536 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2537 | { |
2538 | /* called when force_empty is called */ | |
9f50fad6 | 2539 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2540 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2541 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2542 | } |
2543 | ||
711d3d2c KH |
2544 | /* |
2545 | * This function drains percpu counter value from DEAD cpu and | |
2546 | * move it to local cpu. Note that this function can be preempted. | |
2547 | */ | |
c0ff4b85 | 2548 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2549 | { |
2550 | int i; | |
2551 | ||
c0ff4b85 | 2552 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2553 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2554 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2555 | |
c0ff4b85 R |
2556 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2557 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2558 | } |
e9f8974f | 2559 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2560 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2561 | |
c0ff4b85 R |
2562 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2563 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2564 | } |
c0ff4b85 | 2565 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2566 | } |
2567 | ||
0db0628d | 2568 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2569 | unsigned long action, |
2570 | void *hcpu) | |
2571 | { | |
2572 | int cpu = (unsigned long)hcpu; | |
2573 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2574 | struct mem_cgroup *iter; |
cdec2e42 | 2575 | |
619d094b | 2576 | if (action == CPU_ONLINE) |
1489ebad | 2577 | return NOTIFY_OK; |
1489ebad | 2578 | |
d833049b | 2579 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2580 | return NOTIFY_OK; |
711d3d2c | 2581 | |
9f3a0d09 | 2582 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2583 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2584 | ||
cdec2e42 KH |
2585 | stock = &per_cpu(memcg_stock, cpu); |
2586 | drain_stock(stock); | |
2587 | return NOTIFY_OK; | |
2588 | } | |
2589 | ||
4b534334 KH |
2590 | |
2591 | /* See __mem_cgroup_try_charge() for details */ | |
2592 | enum { | |
2593 | CHARGE_OK, /* success */ | |
2594 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2595 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2596 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
4b534334 KH |
2597 | }; |
2598 | ||
c0ff4b85 | 2599 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 | 2600 | unsigned int nr_pages, unsigned int min_pages, |
3812c8c8 | 2601 | bool invoke_oom) |
4b534334 | 2602 | { |
7ec99d62 | 2603 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2604 | struct mem_cgroup *mem_over_limit; |
2605 | struct res_counter *fail_res; | |
2606 | unsigned long flags = 0; | |
2607 | int ret; | |
2608 | ||
c0ff4b85 | 2609 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2610 | |
2611 | if (likely(!ret)) { | |
2612 | if (!do_swap_account) | |
2613 | return CHARGE_OK; | |
c0ff4b85 | 2614 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2615 | if (likely(!ret)) |
2616 | return CHARGE_OK; | |
2617 | ||
c0ff4b85 | 2618 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2619 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2620 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2621 | } else | |
2622 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2623 | /* |
9221edb7 JW |
2624 | * Never reclaim on behalf of optional batching, retry with a |
2625 | * single page instead. | |
2626 | */ | |
4c9c5359 | 2627 | if (nr_pages > min_pages) |
4b534334 KH |
2628 | return CHARGE_RETRY; |
2629 | ||
2630 | if (!(gfp_mask & __GFP_WAIT)) | |
2631 | return CHARGE_WOULDBLOCK; | |
2632 | ||
4c9c5359 SS |
2633 | if (gfp_mask & __GFP_NORETRY) |
2634 | return CHARGE_NOMEM; | |
2635 | ||
5660048c | 2636 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2637 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2638 | return CHARGE_RETRY; |
4b534334 | 2639 | /* |
19942822 JW |
2640 | * Even though the limit is exceeded at this point, reclaim |
2641 | * may have been able to free some pages. Retry the charge | |
2642 | * before killing the task. | |
2643 | * | |
2644 | * Only for regular pages, though: huge pages are rather | |
2645 | * unlikely to succeed so close to the limit, and we fall back | |
2646 | * to regular pages anyway in case of failure. | |
4b534334 | 2647 | */ |
4c9c5359 | 2648 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2649 | return CHARGE_RETRY; |
2650 | ||
2651 | /* | |
2652 | * At task move, charge accounts can be doubly counted. So, it's | |
2653 | * better to wait until the end of task_move if something is going on. | |
2654 | */ | |
2655 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2656 | return CHARGE_RETRY; | |
2657 | ||
3812c8c8 JW |
2658 | if (invoke_oom) |
2659 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize)); | |
4b534334 | 2660 | |
3812c8c8 | 2661 | return CHARGE_NOMEM; |
4b534334 KH |
2662 | } |
2663 | ||
f817ed48 | 2664 | /* |
38c5d72f KH |
2665 | * __mem_cgroup_try_charge() does |
2666 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2667 | * 2. update res_counter | |
2668 | * 3. call memory reclaim if necessary. | |
2669 | * | |
2670 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2671 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2672 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2673 | * as possible without any hazards. 2: all pages should have a valid | |
2674 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2675 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2676 | * | |
2677 | * So __mem_cgroup_try_charge() will return | |
2678 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2679 | * -ENOMEM ... charge failure because of resource limits. | |
2680 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2681 | * | |
2682 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2683 | * the oom-killer can be invoked. | |
8a9f3ccd | 2684 | */ |
f817ed48 | 2685 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2686 | gfp_t gfp_mask, |
7ec99d62 | 2687 | unsigned int nr_pages, |
c0ff4b85 | 2688 | struct mem_cgroup **ptr, |
7ec99d62 | 2689 | bool oom) |
8a9f3ccd | 2690 | { |
7ec99d62 | 2691 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2692 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2693 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2694 | int ret; |
a636b327 | 2695 | |
867578cb KH |
2696 | /* |
2697 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2698 | * in system level. So, allow to go ahead dying process in addition to | |
2699 | * MEMDIE process. | |
2700 | */ | |
2701 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2702 | || fatal_signal_pending(current))) | |
2703 | goto bypass; | |
a636b327 | 2704 | |
49426420 | 2705 | if (unlikely(task_in_memcg_oom(current))) |
1f14c1ac | 2706 | goto nomem; |
49426420 | 2707 | |
a0d8b00a JW |
2708 | if (gfp_mask & __GFP_NOFAIL) |
2709 | oom = false; | |
2710 | ||
8a9f3ccd | 2711 | /* |
3be91277 HD |
2712 | * We always charge the cgroup the mm_struct belongs to. |
2713 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2714 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2715 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2716 | */ |
c0ff4b85 | 2717 | if (!*ptr && !mm) |
38c5d72f | 2718 | *ptr = root_mem_cgroup; |
f75ca962 | 2719 | again: |
c0ff4b85 R |
2720 | if (*ptr) { /* css should be a valid one */ |
2721 | memcg = *ptr; | |
c0ff4b85 | 2722 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2723 | goto done; |
a0956d54 | 2724 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2725 | goto done; |
c0ff4b85 | 2726 | css_get(&memcg->css); |
4b534334 | 2727 | } else { |
f75ca962 | 2728 | struct task_struct *p; |
54595fe2 | 2729 | |
f75ca962 KH |
2730 | rcu_read_lock(); |
2731 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2732 | /* |
ebb76ce1 | 2733 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2734 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2735 | * race with swapoff. Then, we have small risk of mis-accouning. |
2736 | * But such kind of mis-account by race always happens because | |
2737 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2738 | * small race, here. | |
2739 | * (*) swapoff at el will charge against mm-struct not against | |
2740 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2741 | */ |
c0ff4b85 | 2742 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2743 | if (!memcg) |
2744 | memcg = root_mem_cgroup; | |
2745 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2746 | rcu_read_unlock(); |
2747 | goto done; | |
2748 | } | |
a0956d54 | 2749 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2750 | /* |
2751 | * It seems dagerous to access memcg without css_get(). | |
2752 | * But considering how consume_stok works, it's not | |
2753 | * necessary. If consume_stock success, some charges | |
2754 | * from this memcg are cached on this cpu. So, we | |
2755 | * don't need to call css_get()/css_tryget() before | |
2756 | * calling consume_stock(). | |
2757 | */ | |
2758 | rcu_read_unlock(); | |
2759 | goto done; | |
2760 | } | |
2761 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2762 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2763 | rcu_read_unlock(); |
2764 | goto again; | |
2765 | } | |
2766 | rcu_read_unlock(); | |
2767 | } | |
8a9f3ccd | 2768 | |
4b534334 | 2769 | do { |
3812c8c8 | 2770 | bool invoke_oom = oom && !nr_oom_retries; |
7a81b88c | 2771 | |
4b534334 | 2772 | /* If killed, bypass charge */ |
f75ca962 | 2773 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2774 | css_put(&memcg->css); |
4b534334 | 2775 | goto bypass; |
f75ca962 | 2776 | } |
6d61ef40 | 2777 | |
3812c8c8 JW |
2778 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, |
2779 | nr_pages, invoke_oom); | |
4b534334 KH |
2780 | switch (ret) { |
2781 | case CHARGE_OK: | |
2782 | break; | |
2783 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2784 | batch = nr_pages; |
c0ff4b85 R |
2785 | css_put(&memcg->css); |
2786 | memcg = NULL; | |
f75ca962 | 2787 | goto again; |
4b534334 | 2788 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2789 | css_put(&memcg->css); |
4b534334 KH |
2790 | goto nomem; |
2791 | case CHARGE_NOMEM: /* OOM routine works */ | |
3812c8c8 | 2792 | if (!oom || invoke_oom) { |
c0ff4b85 | 2793 | css_put(&memcg->css); |
867578cb | 2794 | goto nomem; |
f75ca962 | 2795 | } |
4b534334 KH |
2796 | nr_oom_retries--; |
2797 | break; | |
66e1707b | 2798 | } |
4b534334 KH |
2799 | } while (ret != CHARGE_OK); |
2800 | ||
7ec99d62 | 2801 | if (batch > nr_pages) |
c0ff4b85 R |
2802 | refill_stock(memcg, batch - nr_pages); |
2803 | css_put(&memcg->css); | |
0c3e73e8 | 2804 | done: |
c0ff4b85 | 2805 | *ptr = memcg; |
7a81b88c KH |
2806 | return 0; |
2807 | nomem: | |
3168ecbe JW |
2808 | if (!(gfp_mask & __GFP_NOFAIL)) { |
2809 | *ptr = NULL; | |
2810 | return -ENOMEM; | |
2811 | } | |
867578cb | 2812 | bypass: |
38c5d72f KH |
2813 | *ptr = root_mem_cgroup; |
2814 | return -EINTR; | |
7a81b88c | 2815 | } |
8a9f3ccd | 2816 | |
a3032a2c DN |
2817 | /* |
2818 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2819 | * This function is for that and do uncharge, put css's refcnt. | |
2820 | * gotten by try_charge(). | |
2821 | */ | |
c0ff4b85 | 2822 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2823 | unsigned int nr_pages) |
a3032a2c | 2824 | { |
c0ff4b85 | 2825 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2826 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2827 | ||
c0ff4b85 | 2828 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2829 | if (do_swap_account) |
c0ff4b85 | 2830 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2831 | } |
854ffa8d DN |
2832 | } |
2833 | ||
d01dd17f KH |
2834 | /* |
2835 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2836 | * This is useful when moving usage to parent cgroup. | |
2837 | */ | |
2838 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2839 | unsigned int nr_pages) | |
2840 | { | |
2841 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2842 | ||
2843 | if (mem_cgroup_is_root(memcg)) | |
2844 | return; | |
2845 | ||
2846 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2847 | if (do_swap_account) | |
2848 | res_counter_uncharge_until(&memcg->memsw, | |
2849 | memcg->memsw.parent, bytes); | |
2850 | } | |
2851 | ||
a3b2d692 KH |
2852 | /* |
2853 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2854 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2855 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2856 | * called against removed memcg.) | |
a3b2d692 KH |
2857 | */ |
2858 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2859 | { | |
a3b2d692 KH |
2860 | /* ID 0 is unused ID */ |
2861 | if (!id) | |
2862 | return NULL; | |
34c00c31 | 2863 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2864 | } |
2865 | ||
e42d9d5d | 2866 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2867 | { |
c0ff4b85 | 2868 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2869 | struct page_cgroup *pc; |
a3b2d692 | 2870 | unsigned short id; |
b5a84319 KH |
2871 | swp_entry_t ent; |
2872 | ||
309381fe | 2873 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2874 | |
3c776e64 | 2875 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2876 | lock_page_cgroup(pc); |
a3b2d692 | 2877 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2878 | memcg = pc->mem_cgroup; |
2879 | if (memcg && !css_tryget(&memcg->css)) | |
2880 | memcg = NULL; | |
e42d9d5d | 2881 | } else if (PageSwapCache(page)) { |
3c776e64 | 2882 | ent.val = page_private(page); |
9fb4b7cc | 2883 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2884 | rcu_read_lock(); |
c0ff4b85 R |
2885 | memcg = mem_cgroup_lookup(id); |
2886 | if (memcg && !css_tryget(&memcg->css)) | |
2887 | memcg = NULL; | |
a3b2d692 | 2888 | rcu_read_unlock(); |
3c776e64 | 2889 | } |
c0bd3f63 | 2890 | unlock_page_cgroup(pc); |
c0ff4b85 | 2891 | return memcg; |
b5a84319 KH |
2892 | } |
2893 | ||
c0ff4b85 | 2894 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2895 | struct page *page, |
7ec99d62 | 2896 | unsigned int nr_pages, |
9ce70c02 HD |
2897 | enum charge_type ctype, |
2898 | bool lrucare) | |
7a81b88c | 2899 | { |
ce587e65 | 2900 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2901 | struct zone *uninitialized_var(zone); |
fa9add64 | 2902 | struct lruvec *lruvec; |
9ce70c02 | 2903 | bool was_on_lru = false; |
b2402857 | 2904 | bool anon; |
9ce70c02 | 2905 | |
ca3e0214 | 2906 | lock_page_cgroup(pc); |
309381fe | 2907 | VM_BUG_ON_PAGE(PageCgroupUsed(pc), page); |
ca3e0214 KH |
2908 | /* |
2909 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2910 | * accessed by any other context at this point. | |
2911 | */ | |
9ce70c02 HD |
2912 | |
2913 | /* | |
2914 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2915 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2916 | */ | |
2917 | if (lrucare) { | |
2918 | zone = page_zone(page); | |
2919 | spin_lock_irq(&zone->lru_lock); | |
2920 | if (PageLRU(page)) { | |
fa9add64 | 2921 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2922 | ClearPageLRU(page); |
fa9add64 | 2923 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2924 | was_on_lru = true; |
2925 | } | |
2926 | } | |
2927 | ||
c0ff4b85 | 2928 | pc->mem_cgroup = memcg; |
261fb61a KH |
2929 | /* |
2930 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2931 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2932 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2933 | * before USED bit, we need memory barrier here. | |
2934 | * See mem_cgroup_add_lru_list(), etc. | |
f894ffa8 | 2935 | */ |
08e552c6 | 2936 | smp_wmb(); |
b2402857 | 2937 | SetPageCgroupUsed(pc); |
3be91277 | 2938 | |
9ce70c02 HD |
2939 | if (lrucare) { |
2940 | if (was_on_lru) { | |
fa9add64 | 2941 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
309381fe | 2942 | VM_BUG_ON_PAGE(PageLRU(page), page); |
9ce70c02 | 2943 | SetPageLRU(page); |
fa9add64 | 2944 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2945 | } |
2946 | spin_unlock_irq(&zone->lru_lock); | |
2947 | } | |
2948 | ||
41326c17 | 2949 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2950 | anon = true; |
2951 | else | |
2952 | anon = false; | |
2953 | ||
b070e65c | 2954 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
52d4b9ac | 2955 | unlock_page_cgroup(pc); |
9ce70c02 | 2956 | |
430e4863 | 2957 | /* |
bb4cc1a8 AM |
2958 | * "charge_statistics" updated event counter. Then, check it. |
2959 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2960 | * if they exceeds softlimit. | |
430e4863 | 2961 | */ |
c0ff4b85 | 2962 | memcg_check_events(memcg, page); |
7a81b88c | 2963 | } |
66e1707b | 2964 | |
7cf27982 GC |
2965 | static DEFINE_MUTEX(set_limit_mutex); |
2966 | ||
7ae1e1d0 | 2967 | #ifdef CONFIG_MEMCG_KMEM |
d6441637 VD |
2968 | static DEFINE_MUTEX(activate_kmem_mutex); |
2969 | ||
7ae1e1d0 GC |
2970 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) |
2971 | { | |
2972 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
6de64beb | 2973 | memcg_kmem_is_active(memcg); |
7ae1e1d0 GC |
2974 | } |
2975 | ||
1f458cbf GC |
2976 | /* |
2977 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2978 | * in the memcg_cache_params struct. | |
2979 | */ | |
2980 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2981 | { | |
2982 | struct kmem_cache *cachep; | |
2983 | ||
2984 | VM_BUG_ON(p->is_root_cache); | |
2985 | cachep = p->root_cache; | |
7a67d7ab | 2986 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2987 | } |
2988 | ||
749c5415 | 2989 | #ifdef CONFIG_SLABINFO |
2da8ca82 | 2990 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
749c5415 | 2991 | { |
2da8ca82 | 2992 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
749c5415 GC |
2993 | struct memcg_cache_params *params; |
2994 | ||
2995 | if (!memcg_can_account_kmem(memcg)) | |
2996 | return -EIO; | |
2997 | ||
2998 | print_slabinfo_header(m); | |
2999 | ||
3000 | mutex_lock(&memcg->slab_caches_mutex); | |
3001 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
3002 | cache_show(memcg_params_to_cache(params), m); | |
3003 | mutex_unlock(&memcg->slab_caches_mutex); | |
3004 | ||
3005 | return 0; | |
3006 | } | |
3007 | #endif | |
3008 | ||
7ae1e1d0 GC |
3009 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
3010 | { | |
3011 | struct res_counter *fail_res; | |
3012 | struct mem_cgroup *_memcg; | |
3013 | int ret = 0; | |
7ae1e1d0 GC |
3014 | |
3015 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
3016 | if (ret) | |
3017 | return ret; | |
3018 | ||
7ae1e1d0 GC |
3019 | _memcg = memcg; |
3020 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
b9921ecd | 3021 | &_memcg, oom_gfp_allowed(gfp)); |
7ae1e1d0 GC |
3022 | |
3023 | if (ret == -EINTR) { | |
3024 | /* | |
3025 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
3026 | * OOM kill or fatal signal. Since our only options are to | |
3027 | * either fail the allocation or charge it to this cgroup, do | |
3028 | * it as a temporary condition. But we can't fail. From a | |
3029 | * kmem/slab perspective, the cache has already been selected, | |
3030 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
3031 | * our minds. | |
3032 | * | |
3033 | * This condition will only trigger if the task entered | |
3034 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
3035 | * __mem_cgroup_try_charge() above. Tasks that were already | |
3036 | * dying when the allocation triggers should have been already | |
3037 | * directed to the root cgroup in memcontrol.h | |
3038 | */ | |
3039 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
3040 | if (do_swap_account) | |
3041 | res_counter_charge_nofail(&memcg->memsw, size, | |
3042 | &fail_res); | |
3043 | ret = 0; | |
3044 | } else if (ret) | |
3045 | res_counter_uncharge(&memcg->kmem, size); | |
3046 | ||
3047 | return ret; | |
3048 | } | |
3049 | ||
3050 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
3051 | { | |
7ae1e1d0 GC |
3052 | res_counter_uncharge(&memcg->res, size); |
3053 | if (do_swap_account) | |
3054 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
3055 | |
3056 | /* Not down to 0 */ | |
3057 | if (res_counter_uncharge(&memcg->kmem, size)) | |
3058 | return; | |
3059 | ||
10d5ebf4 LZ |
3060 | /* |
3061 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
3062 | * this last uncharge is racing with the offlining code or it is | |
3063 | * outliving the memcg existence. | |
3064 | * | |
3065 | * The memory barrier imposed by test&clear is paired with the | |
3066 | * explicit one in memcg_kmem_mark_dead(). | |
3067 | */ | |
7de37682 | 3068 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 3069 | css_put(&memcg->css); |
7ae1e1d0 GC |
3070 | } |
3071 | ||
2633d7a0 GC |
3072 | /* |
3073 | * helper for acessing a memcg's index. It will be used as an index in the | |
3074 | * child cache array in kmem_cache, and also to derive its name. This function | |
3075 | * will return -1 when this is not a kmem-limited memcg. | |
3076 | */ | |
3077 | int memcg_cache_id(struct mem_cgroup *memcg) | |
3078 | { | |
3079 | return memcg ? memcg->kmemcg_id : -1; | |
3080 | } | |
3081 | ||
55007d84 GC |
3082 | static size_t memcg_caches_array_size(int num_groups) |
3083 | { | |
3084 | ssize_t size; | |
3085 | if (num_groups <= 0) | |
3086 | return 0; | |
3087 | ||
3088 | size = 2 * num_groups; | |
3089 | if (size < MEMCG_CACHES_MIN_SIZE) | |
3090 | size = MEMCG_CACHES_MIN_SIZE; | |
3091 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3092 | size = MEMCG_CACHES_MAX_SIZE; | |
3093 | ||
3094 | return size; | |
3095 | } | |
3096 | ||
3097 | /* | |
3098 | * We should update the current array size iff all caches updates succeed. This | |
3099 | * can only be done from the slab side. The slab mutex needs to be held when | |
3100 | * calling this. | |
3101 | */ | |
3102 | void memcg_update_array_size(int num) | |
3103 | { | |
3104 | if (num > memcg_limited_groups_array_size) | |
3105 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
3106 | } | |
3107 | ||
15cf17d2 KK |
3108 | static void kmem_cache_destroy_work_func(struct work_struct *w); |
3109 | ||
55007d84 GC |
3110 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
3111 | { | |
3112 | struct memcg_cache_params *cur_params = s->memcg_params; | |
3113 | ||
f35c3a8e | 3114 | VM_BUG_ON(!is_root_cache(s)); |
55007d84 GC |
3115 | |
3116 | if (num_groups > memcg_limited_groups_array_size) { | |
3117 | int i; | |
f8570263 | 3118 | struct memcg_cache_params *new_params; |
55007d84 GC |
3119 | ssize_t size = memcg_caches_array_size(num_groups); |
3120 | ||
3121 | size *= sizeof(void *); | |
90c7a79c | 3122 | size += offsetof(struct memcg_cache_params, memcg_caches); |
55007d84 | 3123 | |
f8570263 VD |
3124 | new_params = kzalloc(size, GFP_KERNEL); |
3125 | if (!new_params) | |
55007d84 | 3126 | return -ENOMEM; |
55007d84 | 3127 | |
f8570263 | 3128 | new_params->is_root_cache = true; |
55007d84 GC |
3129 | |
3130 | /* | |
3131 | * There is the chance it will be bigger than | |
3132 | * memcg_limited_groups_array_size, if we failed an allocation | |
3133 | * in a cache, in which case all caches updated before it, will | |
3134 | * have a bigger array. | |
3135 | * | |
3136 | * But if that is the case, the data after | |
3137 | * memcg_limited_groups_array_size is certainly unused | |
3138 | */ | |
3139 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3140 | if (!cur_params->memcg_caches[i]) | |
3141 | continue; | |
f8570263 | 3142 | new_params->memcg_caches[i] = |
55007d84 GC |
3143 | cur_params->memcg_caches[i]; |
3144 | } | |
3145 | ||
3146 | /* | |
3147 | * Ideally, we would wait until all caches succeed, and only | |
3148 | * then free the old one. But this is not worth the extra | |
3149 | * pointer per-cache we'd have to have for this. | |
3150 | * | |
3151 | * It is not a big deal if some caches are left with a size | |
3152 | * bigger than the others. And all updates will reset this | |
3153 | * anyway. | |
3154 | */ | |
f8570263 VD |
3155 | rcu_assign_pointer(s->memcg_params, new_params); |
3156 | if (cur_params) | |
3157 | kfree_rcu(cur_params, rcu_head); | |
55007d84 GC |
3158 | } |
3159 | return 0; | |
3160 | } | |
3161 | ||
363a044f VD |
3162 | int memcg_alloc_cache_params(struct mem_cgroup *memcg, struct kmem_cache *s, |
3163 | struct kmem_cache *root_cache) | |
2633d7a0 | 3164 | { |
90c7a79c | 3165 | size_t size; |
2633d7a0 GC |
3166 | |
3167 | if (!memcg_kmem_enabled()) | |
3168 | return 0; | |
3169 | ||
90c7a79c AV |
3170 | if (!memcg) { |
3171 | size = offsetof(struct memcg_cache_params, memcg_caches); | |
55007d84 | 3172 | size += memcg_limited_groups_array_size * sizeof(void *); |
90c7a79c AV |
3173 | } else |
3174 | size = sizeof(struct memcg_cache_params); | |
55007d84 | 3175 | |
2633d7a0 GC |
3176 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3177 | if (!s->memcg_params) | |
3178 | return -ENOMEM; | |
3179 | ||
943a451a | 3180 | if (memcg) { |
2633d7a0 | 3181 | s->memcg_params->memcg = memcg; |
943a451a | 3182 | s->memcg_params->root_cache = root_cache; |
3e6b11df AV |
3183 | INIT_WORK(&s->memcg_params->destroy, |
3184 | kmem_cache_destroy_work_func); | |
4ba902b5 GC |
3185 | } else |
3186 | s->memcg_params->is_root_cache = true; | |
3187 | ||
2633d7a0 GC |
3188 | return 0; |
3189 | } | |
3190 | ||
363a044f VD |
3191 | void memcg_free_cache_params(struct kmem_cache *s) |
3192 | { | |
3193 | kfree(s->memcg_params); | |
3194 | } | |
3195 | ||
1aa13254 | 3196 | void memcg_register_cache(struct kmem_cache *s) |
2633d7a0 | 3197 | { |
d7f25f8a GC |
3198 | struct kmem_cache *root; |
3199 | struct mem_cgroup *memcg; | |
3200 | int id; | |
3201 | ||
1aa13254 VD |
3202 | if (is_root_cache(s)) |
3203 | return; | |
3204 | ||
2edefe11 VD |
3205 | /* |
3206 | * Holding the slab_mutex assures nobody will touch the memcg_caches | |
3207 | * array while we are modifying it. | |
3208 | */ | |
3209 | lockdep_assert_held(&slab_mutex); | |
3210 | ||
1aa13254 VD |
3211 | root = s->memcg_params->root_cache; |
3212 | memcg = s->memcg_params->memcg; | |
3213 | id = memcg_cache_id(memcg); | |
3214 | ||
3215 | css_get(&memcg->css); | |
3216 | ||
1aa13254 | 3217 | |
d7f25f8a | 3218 | /* |
959c8963 VD |
3219 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
3220 | * barrier here to ensure nobody will see the kmem_cache partially | |
3221 | * initialized. | |
d7f25f8a | 3222 | */ |
959c8963 VD |
3223 | smp_wmb(); |
3224 | ||
96403da2 VD |
3225 | /* |
3226 | * Initialize the pointer to this cache in its parent's memcg_params | |
3227 | * before adding it to the memcg_slab_caches list, otherwise we can | |
3228 | * fail to convert memcg_params_to_cache() while traversing the list. | |
3229 | */ | |
2edefe11 | 3230 | VM_BUG_ON(root->memcg_params->memcg_caches[id]); |
959c8963 | 3231 | root->memcg_params->memcg_caches[id] = s; |
96403da2 VD |
3232 | |
3233 | mutex_lock(&memcg->slab_caches_mutex); | |
3234 | list_add(&s->memcg_params->list, &memcg->memcg_slab_caches); | |
3235 | mutex_unlock(&memcg->slab_caches_mutex); | |
1aa13254 | 3236 | } |
d7f25f8a | 3237 | |
1aa13254 VD |
3238 | void memcg_unregister_cache(struct kmem_cache *s) |
3239 | { | |
3240 | struct kmem_cache *root; | |
3241 | struct mem_cgroup *memcg; | |
3242 | int id; | |
3243 | ||
3244 | if (is_root_cache(s)) | |
3245 | return; | |
d7f25f8a | 3246 | |
2edefe11 VD |
3247 | /* |
3248 | * Holding the slab_mutex assures nobody will touch the memcg_caches | |
3249 | * array while we are modifying it. | |
3250 | */ | |
3251 | lockdep_assert_held(&slab_mutex); | |
3252 | ||
d7f25f8a | 3253 | root = s->memcg_params->root_cache; |
96403da2 VD |
3254 | memcg = s->memcg_params->memcg; |
3255 | id = memcg_cache_id(memcg); | |
d7f25f8a GC |
3256 | |
3257 | mutex_lock(&memcg->slab_caches_mutex); | |
3258 | list_del(&s->memcg_params->list); | |
3259 | mutex_unlock(&memcg->slab_caches_mutex); | |
3260 | ||
96403da2 VD |
3261 | /* |
3262 | * Clear the pointer to this cache in its parent's memcg_params only | |
3263 | * after removing it from the memcg_slab_caches list, otherwise we can | |
3264 | * fail to convert memcg_params_to_cache() while traversing the list. | |
3265 | */ | |
2edefe11 | 3266 | VM_BUG_ON(!root->memcg_params->memcg_caches[id]); |
96403da2 VD |
3267 | root->memcg_params->memcg_caches[id] = NULL; |
3268 | ||
20f05310 | 3269 | css_put(&memcg->css); |
2633d7a0 GC |
3270 | } |
3271 | ||
0e9d92f2 GC |
3272 | /* |
3273 | * During the creation a new cache, we need to disable our accounting mechanism | |
3274 | * altogether. This is true even if we are not creating, but rather just | |
3275 | * enqueing new caches to be created. | |
3276 | * | |
3277 | * This is because that process will trigger allocations; some visible, like | |
3278 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3279 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3280 | * objects during debug. | |
3281 | * | |
3282 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3283 | * to it. This may not be a bounded recursion: since the first cache creation | |
3284 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3285 | * cache again, failing at the same point. | |
3286 | * | |
3287 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3288 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3289 | * inside the following two functions. | |
3290 | */ | |
3291 | static inline void memcg_stop_kmem_account(void) | |
3292 | { | |
3293 | VM_BUG_ON(!current->mm); | |
3294 | current->memcg_kmem_skip_account++; | |
3295 | } | |
3296 | ||
3297 | static inline void memcg_resume_kmem_account(void) | |
3298 | { | |
3299 | VM_BUG_ON(!current->mm); | |
3300 | current->memcg_kmem_skip_account--; | |
3301 | } | |
3302 | ||
1f458cbf GC |
3303 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3304 | { | |
3305 | struct kmem_cache *cachep; | |
3306 | struct memcg_cache_params *p; | |
3307 | ||
3308 | p = container_of(w, struct memcg_cache_params, destroy); | |
3309 | ||
3310 | cachep = memcg_params_to_cache(p); | |
3311 | ||
22933152 GC |
3312 | /* |
3313 | * If we get down to 0 after shrink, we could delete right away. | |
3314 | * However, memcg_release_pages() already puts us back in the workqueue | |
3315 | * in that case. If we proceed deleting, we'll get a dangling | |
3316 | * reference, and removing the object from the workqueue in that case | |
3317 | * is unnecessary complication. We are not a fast path. | |
3318 | * | |
3319 | * Note that this case is fundamentally different from racing with | |
3320 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3321 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3322 | * into the queue, but doing so from inside the worker racing to | |
3323 | * destroy it. | |
3324 | * | |
3325 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3326 | * again | |
3327 | */ | |
0d8a4a37 | 3328 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) |
22933152 | 3329 | kmem_cache_shrink(cachep); |
0d8a4a37 | 3330 | else |
1f458cbf GC |
3331 | kmem_cache_destroy(cachep); |
3332 | } | |
3333 | ||
3334 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3335 | { | |
3336 | if (!cachep->memcg_params->dead) | |
3337 | return; | |
3338 | ||
22933152 GC |
3339 | /* |
3340 | * There are many ways in which we can get here. | |
3341 | * | |
3342 | * We can get to a memory-pressure situation while the delayed work is | |
3343 | * still pending to run. The vmscan shrinkers can then release all | |
3344 | * cache memory and get us to destruction. If this is the case, we'll | |
3345 | * be executed twice, which is a bug (the second time will execute over | |
3346 | * bogus data). In this case, cancelling the work should be fine. | |
3347 | * | |
3348 | * But we can also get here from the worker itself, if | |
3349 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3350 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3351 | * cancel the work (the worker runs with an internal lock held, which | |
3352 | * is the same lock we would hold for cancel_work_sync().) | |
3353 | * | |
3354 | * Since we can't possibly know who got us here, just refrain from | |
3355 | * running if there is already work pending | |
3356 | */ | |
3357 | if (work_pending(&cachep->memcg_params->destroy)) | |
3358 | return; | |
1f458cbf GC |
3359 | /* |
3360 | * We have to defer the actual destroying to a workqueue, because | |
3361 | * we might currently be in a context that cannot sleep. | |
3362 | */ | |
3363 | schedule_work(&cachep->memcg_params->destroy); | |
3364 | } | |
3365 | ||
842e2873 VD |
3366 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3367 | struct kmem_cache *s) | |
d7f25f8a | 3368 | { |
7c094fd6 | 3369 | struct kmem_cache *new = NULL; |
e61734c5 | 3370 | static char *tmp_path = NULL, *tmp_name = NULL; |
842e2873 | 3371 | static DEFINE_MUTEX(mutex); /* protects tmp_name */ |
d7f25f8a | 3372 | |
842e2873 | 3373 | BUG_ON(!memcg_can_account_kmem(memcg)); |
d9c10ddd | 3374 | |
842e2873 | 3375 | mutex_lock(&mutex); |
d9c10ddd MH |
3376 | /* |
3377 | * kmem_cache_create_memcg duplicates the given name and | |
3378 | * cgroup_name for this name requires RCU context. | |
3379 | * This static temporary buffer is used to prevent from | |
3380 | * pointless shortliving allocation. | |
3381 | */ | |
e61734c5 TH |
3382 | if (!tmp_path || !tmp_name) { |
3383 | if (!tmp_path) | |
3384 | tmp_path = kmalloc(PATH_MAX, GFP_KERNEL); | |
d9c10ddd | 3385 | if (!tmp_name) |
e61734c5 TH |
3386 | tmp_name = kmalloc(NAME_MAX + 1, GFP_KERNEL); |
3387 | if (!tmp_path || !tmp_name) | |
7c094fd6 | 3388 | goto out; |
d9c10ddd MH |
3389 | } |
3390 | ||
e61734c5 TH |
3391 | cgroup_name(memcg->css.cgroup, tmp_name, NAME_MAX + 1); |
3392 | snprintf(tmp_path, PATH_MAX, "%s(%d:%s)", s->name, | |
3393 | memcg_cache_id(memcg), tmp_name); | |
d7f25f8a | 3394 | |
e61734c5 | 3395 | new = kmem_cache_create_memcg(memcg, tmp_path, s->object_size, s->align, |
943a451a | 3396 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d79923fa GC |
3397 | if (new) |
3398 | new->allocflags |= __GFP_KMEMCG; | |
842e2873 VD |
3399 | else |
3400 | new = s; | |
7c094fd6 | 3401 | out: |
842e2873 | 3402 | mutex_unlock(&mutex); |
d7f25f8a GC |
3403 | return new; |
3404 | } | |
3405 | ||
7cf27982 GC |
3406 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3407 | { | |
3408 | struct kmem_cache *c; | |
3409 | int i; | |
3410 | ||
3411 | if (!s->memcg_params) | |
3412 | return; | |
3413 | if (!s->memcg_params->is_root_cache) | |
3414 | return; | |
3415 | ||
3416 | /* | |
3417 | * If the cache is being destroyed, we trust that there is no one else | |
3418 | * requesting objects from it. Even if there are, the sanity checks in | |
3419 | * kmem_cache_destroy should caught this ill-case. | |
3420 | * | |
3421 | * Still, we don't want anyone else freeing memcg_caches under our | |
3422 | * noses, which can happen if a new memcg comes to life. As usual, | |
d6441637 VD |
3423 | * we'll take the activate_kmem_mutex to protect ourselves against |
3424 | * this. | |
7cf27982 | 3425 | */ |
d6441637 | 3426 | mutex_lock(&activate_kmem_mutex); |
7a67d7ab QH |
3427 | for_each_memcg_cache_index(i) { |
3428 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
3429 | if (!c) |
3430 | continue; | |
3431 | ||
3432 | /* | |
3433 | * We will now manually delete the caches, so to avoid races | |
3434 | * we need to cancel all pending destruction workers and | |
3435 | * proceed with destruction ourselves. | |
3436 | * | |
3437 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3438 | * and that could spawn the workers again: it is likely that | |
3439 | * the cache still have active pages until this very moment. | |
3440 | * This would lead us back to mem_cgroup_destroy_cache. | |
3441 | * | |
3442 | * But that will not execute at all if the "dead" flag is not | |
3443 | * set, so flip it down to guarantee we are in control. | |
3444 | */ | |
3445 | c->memcg_params->dead = false; | |
22933152 | 3446 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3447 | kmem_cache_destroy(c); |
3448 | } | |
d6441637 | 3449 | mutex_unlock(&activate_kmem_mutex); |
7cf27982 GC |
3450 | } |
3451 | ||
d7f25f8a GC |
3452 | struct create_work { |
3453 | struct mem_cgroup *memcg; | |
3454 | struct kmem_cache *cachep; | |
3455 | struct work_struct work; | |
3456 | }; | |
3457 | ||
1f458cbf GC |
3458 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3459 | { | |
3460 | struct kmem_cache *cachep; | |
3461 | struct memcg_cache_params *params; | |
3462 | ||
3463 | if (!memcg_kmem_is_active(memcg)) | |
3464 | return; | |
3465 | ||
3466 | mutex_lock(&memcg->slab_caches_mutex); | |
3467 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3468 | cachep = memcg_params_to_cache(params); | |
3469 | cachep->memcg_params->dead = true; | |
1f458cbf GC |
3470 | schedule_work(&cachep->memcg_params->destroy); |
3471 | } | |
3472 | mutex_unlock(&memcg->slab_caches_mutex); | |
3473 | } | |
3474 | ||
d7f25f8a GC |
3475 | static void memcg_create_cache_work_func(struct work_struct *w) |
3476 | { | |
3477 | struct create_work *cw; | |
3478 | ||
3479 | cw = container_of(w, struct create_work, work); | |
3480 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
1aa13254 | 3481 | css_put(&cw->memcg->css); |
d7f25f8a GC |
3482 | kfree(cw); |
3483 | } | |
3484 | ||
3485 | /* | |
3486 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3487 | */ |
0e9d92f2 GC |
3488 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3489 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3490 | { |
3491 | struct create_work *cw; | |
3492 | ||
3493 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
ca0dde97 LZ |
3494 | if (cw == NULL) { |
3495 | css_put(&memcg->css); | |
d7f25f8a GC |
3496 | return; |
3497 | } | |
3498 | ||
3499 | cw->memcg = memcg; | |
3500 | cw->cachep = cachep; | |
3501 | ||
3502 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3503 | schedule_work(&cw->work); | |
3504 | } | |
3505 | ||
0e9d92f2 GC |
3506 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3507 | struct kmem_cache *cachep) | |
3508 | { | |
3509 | /* | |
3510 | * We need to stop accounting when we kmalloc, because if the | |
3511 | * corresponding kmalloc cache is not yet created, the first allocation | |
3512 | * in __memcg_create_cache_enqueue will recurse. | |
3513 | * | |
3514 | * However, it is better to enclose the whole function. Depending on | |
3515 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3516 | * trigger an allocation. This too, will make us recurse. Because at | |
3517 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3518 | * the safest choice is to do it like this, wrapping the whole function. | |
3519 | */ | |
3520 | memcg_stop_kmem_account(); | |
3521 | __memcg_create_cache_enqueue(memcg, cachep); | |
3522 | memcg_resume_kmem_account(); | |
3523 | } | |
d7f25f8a GC |
3524 | /* |
3525 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3526 | * We try to use the current memcg's version of the cache. | |
3527 | * | |
3528 | * If the cache does not exist yet, if we are the first user of it, | |
3529 | * we either create it immediately, if possible, or create it asynchronously | |
3530 | * in a workqueue. | |
3531 | * In the latter case, we will let the current allocation go through with | |
3532 | * the original cache. | |
3533 | * | |
3534 | * Can't be called in interrupt context or from kernel threads. | |
3535 | * This function needs to be called with rcu_read_lock() held. | |
3536 | */ | |
3537 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3538 | gfp_t gfp) | |
3539 | { | |
3540 | struct mem_cgroup *memcg; | |
959c8963 | 3541 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
3542 | |
3543 | VM_BUG_ON(!cachep->memcg_params); | |
3544 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3545 | ||
0e9d92f2 GC |
3546 | if (!current->mm || current->memcg_kmem_skip_account) |
3547 | return cachep; | |
3548 | ||
d7f25f8a GC |
3549 | rcu_read_lock(); |
3550 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3551 | |
3552 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3553 | goto out; |
d7f25f8a | 3554 | |
959c8963 VD |
3555 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
3556 | if (likely(memcg_cachep)) { | |
3557 | cachep = memcg_cachep; | |
ca0dde97 | 3558 | goto out; |
d7f25f8a GC |
3559 | } |
3560 | ||
ca0dde97 LZ |
3561 | /* The corresponding put will be done in the workqueue. */ |
3562 | if (!css_tryget(&memcg->css)) | |
3563 | goto out; | |
3564 | rcu_read_unlock(); | |
3565 | ||
3566 | /* | |
3567 | * If we are in a safe context (can wait, and not in interrupt | |
3568 | * context), we could be be predictable and return right away. | |
3569 | * This would guarantee that the allocation being performed | |
3570 | * already belongs in the new cache. | |
3571 | * | |
3572 | * However, there are some clashes that can arrive from locking. | |
3573 | * For instance, because we acquire the slab_mutex while doing | |
3574 | * kmem_cache_dup, this means no further allocation could happen | |
3575 | * with the slab_mutex held. | |
3576 | * | |
3577 | * Also, because cache creation issue get_online_cpus(), this | |
3578 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3579 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3580 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3581 | * better to defer everything. | |
3582 | */ | |
3583 | memcg_create_cache_enqueue(memcg, cachep); | |
3584 | return cachep; | |
3585 | out: | |
3586 | rcu_read_unlock(); | |
3587 | return cachep; | |
d7f25f8a GC |
3588 | } |
3589 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3590 | ||
7ae1e1d0 GC |
3591 | /* |
3592 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3593 | * possible for the given order. But the page is not allocated yet, so we'll | |
3594 | * need a further commit step to do the final arrangements. | |
3595 | * | |
3596 | * It is possible for the task to switch cgroups in this mean time, so at | |
3597 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3598 | * the handle argument to return to the caller which cgroup we should commit | |
3599 | * against. We could also return the memcg directly and avoid the pointer | |
3600 | * passing, but a boolean return value gives better semantics considering | |
3601 | * the compiled-out case as well. | |
3602 | * | |
3603 | * Returning true means the allocation is possible. | |
3604 | */ | |
3605 | bool | |
3606 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3607 | { | |
3608 | struct mem_cgroup *memcg; | |
3609 | int ret; | |
3610 | ||
3611 | *_memcg = NULL; | |
6d42c232 GC |
3612 | |
3613 | /* | |
3614 | * Disabling accounting is only relevant for some specific memcg | |
3615 | * internal allocations. Therefore we would initially not have such | |
3616 | * check here, since direct calls to the page allocator that are marked | |
3617 | * with GFP_KMEMCG only happen outside memcg core. We are mostly | |
3618 | * concerned with cache allocations, and by having this test at | |
3619 | * memcg_kmem_get_cache, we are already able to relay the allocation to | |
3620 | * the root cache and bypass the memcg cache altogether. | |
3621 | * | |
3622 | * There is one exception, though: the SLUB allocator does not create | |
3623 | * large order caches, but rather service large kmallocs directly from | |
3624 | * the page allocator. Therefore, the following sequence when backed by | |
3625 | * the SLUB allocator: | |
3626 | * | |
f894ffa8 AM |
3627 | * memcg_stop_kmem_account(); |
3628 | * kmalloc(<large_number>) | |
3629 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3630 | * |
3631 | * would effectively ignore the fact that we should skip accounting, | |
3632 | * since it will drive us directly to this function without passing | |
3633 | * through the cache selector memcg_kmem_get_cache. Such large | |
3634 | * allocations are extremely rare but can happen, for instance, for the | |
3635 | * cache arrays. We bring this test here. | |
3636 | */ | |
3637 | if (!current->mm || current->memcg_kmem_skip_account) | |
3638 | return true; | |
3639 | ||
7ae1e1d0 GC |
3640 | memcg = try_get_mem_cgroup_from_mm(current->mm); |
3641 | ||
3642 | /* | |
3643 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3644 | * isn't much we can do without complicating this too much, and it would | |
3645 | * be gfp-dependent anyway. Just let it go | |
3646 | */ | |
3647 | if (unlikely(!memcg)) | |
3648 | return true; | |
3649 | ||
3650 | if (!memcg_can_account_kmem(memcg)) { | |
3651 | css_put(&memcg->css); | |
3652 | return true; | |
3653 | } | |
3654 | ||
7ae1e1d0 GC |
3655 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3656 | if (!ret) | |
3657 | *_memcg = memcg; | |
7ae1e1d0 GC |
3658 | |
3659 | css_put(&memcg->css); | |
3660 | return (ret == 0); | |
3661 | } | |
3662 | ||
3663 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3664 | int order) | |
3665 | { | |
3666 | struct page_cgroup *pc; | |
3667 | ||
3668 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3669 | ||
3670 | /* The page allocation failed. Revert */ | |
3671 | if (!page) { | |
3672 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3673 | return; |
3674 | } | |
3675 | ||
3676 | pc = lookup_page_cgroup(page); | |
3677 | lock_page_cgroup(pc); | |
3678 | pc->mem_cgroup = memcg; | |
3679 | SetPageCgroupUsed(pc); | |
3680 | unlock_page_cgroup(pc); | |
3681 | } | |
3682 | ||
3683 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3684 | { | |
3685 | struct mem_cgroup *memcg = NULL; | |
3686 | struct page_cgroup *pc; | |
3687 | ||
3688 | ||
3689 | pc = lookup_page_cgroup(page); | |
3690 | /* | |
3691 | * Fast unlocked return. Theoretically might have changed, have to | |
3692 | * check again after locking. | |
3693 | */ | |
3694 | if (!PageCgroupUsed(pc)) | |
3695 | return; | |
3696 | ||
3697 | lock_page_cgroup(pc); | |
3698 | if (PageCgroupUsed(pc)) { | |
3699 | memcg = pc->mem_cgroup; | |
3700 | ClearPageCgroupUsed(pc); | |
3701 | } | |
3702 | unlock_page_cgroup(pc); | |
3703 | ||
3704 | /* | |
3705 | * We trust that only if there is a memcg associated with the page, it | |
3706 | * is a valid allocation | |
3707 | */ | |
3708 | if (!memcg) | |
3709 | return; | |
3710 | ||
309381fe | 3711 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
7ae1e1d0 | 3712 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); |
7ae1e1d0 | 3713 | } |
1f458cbf GC |
3714 | #else |
3715 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3716 | { | |
3717 | } | |
7ae1e1d0 GC |
3718 | #endif /* CONFIG_MEMCG_KMEM */ |
3719 | ||
ca3e0214 KH |
3720 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3721 | ||
a0db00fc | 3722 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3723 | /* |
3724 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3725 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3726 | * charge/uncharge will be never happen and move_account() is done under | |
3727 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3728 | */ |
e94c8a9c | 3729 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3730 | { |
3731 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3732 | struct page_cgroup *pc; |
b070e65c | 3733 | struct mem_cgroup *memcg; |
e94c8a9c | 3734 | int i; |
ca3e0214 | 3735 | |
3d37c4a9 KH |
3736 | if (mem_cgroup_disabled()) |
3737 | return; | |
b070e65c DR |
3738 | |
3739 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3740 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3741 | pc = head_pc + i; | |
b070e65c | 3742 | pc->mem_cgroup = memcg; |
e94c8a9c | 3743 | smp_wmb();/* see __commit_charge() */ |
e94c8a9c KH |
3744 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3745 | } | |
b070e65c DR |
3746 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3747 | HPAGE_PMD_NR); | |
ca3e0214 | 3748 | } |
12d27107 | 3749 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3750 | |
3ea67d06 SZ |
3751 | static inline |
3752 | void mem_cgroup_move_account_page_stat(struct mem_cgroup *from, | |
3753 | struct mem_cgroup *to, | |
3754 | unsigned int nr_pages, | |
3755 | enum mem_cgroup_stat_index idx) | |
3756 | { | |
3757 | /* Update stat data for mem_cgroup */ | |
3758 | preempt_disable(); | |
5e8cfc3c | 3759 | __this_cpu_sub(from->stat->count[idx], nr_pages); |
3ea67d06 SZ |
3760 | __this_cpu_add(to->stat->count[idx], nr_pages); |
3761 | preempt_enable(); | |
3762 | } | |
3763 | ||
f817ed48 | 3764 | /** |
de3638d9 | 3765 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3766 | * @page: the page |
7ec99d62 | 3767 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3768 | * @pc: page_cgroup of the page. |
3769 | * @from: mem_cgroup which the page is moved from. | |
3770 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3771 | * | |
3772 | * The caller must confirm following. | |
08e552c6 | 3773 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3774 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3775 | * |
2f3479b1 KH |
3776 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3777 | * from old cgroup. | |
f817ed48 | 3778 | */ |
7ec99d62 JW |
3779 | static int mem_cgroup_move_account(struct page *page, |
3780 | unsigned int nr_pages, | |
3781 | struct page_cgroup *pc, | |
3782 | struct mem_cgroup *from, | |
2f3479b1 | 3783 | struct mem_cgroup *to) |
f817ed48 | 3784 | { |
de3638d9 JW |
3785 | unsigned long flags; |
3786 | int ret; | |
b2402857 | 3787 | bool anon = PageAnon(page); |
987eba66 | 3788 | |
f817ed48 | 3789 | VM_BUG_ON(from == to); |
309381fe | 3790 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3791 | /* |
3792 | * The page is isolated from LRU. So, collapse function | |
3793 | * will not handle this page. But page splitting can happen. | |
3794 | * Do this check under compound_page_lock(). The caller should | |
3795 | * hold it. | |
3796 | */ | |
3797 | ret = -EBUSY; | |
7ec99d62 | 3798 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3799 | goto out; |
3800 | ||
3801 | lock_page_cgroup(pc); | |
3802 | ||
3803 | ret = -EINVAL; | |
3804 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3805 | goto unlock; | |
3806 | ||
312734c0 | 3807 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3808 | |
3ea67d06 SZ |
3809 | if (!anon && page_mapped(page)) |
3810 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3811 | MEM_CGROUP_STAT_FILE_MAPPED); | |
3812 | ||
3813 | if (PageWriteback(page)) | |
3814 | mem_cgroup_move_account_page_stat(from, to, nr_pages, | |
3815 | MEM_CGROUP_STAT_WRITEBACK); | |
3816 | ||
b070e65c | 3817 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
d69b042f | 3818 | |
854ffa8d | 3819 | /* caller should have done css_get */ |
08e552c6 | 3820 | pc->mem_cgroup = to; |
b070e65c | 3821 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
312734c0 | 3822 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3823 | ret = 0; |
3824 | unlock: | |
57f9fd7d | 3825 | unlock_page_cgroup(pc); |
d2265e6f KH |
3826 | /* |
3827 | * check events | |
3828 | */ | |
5564e88b JW |
3829 | memcg_check_events(to, page); |
3830 | memcg_check_events(from, page); | |
de3638d9 | 3831 | out: |
f817ed48 KH |
3832 | return ret; |
3833 | } | |
3834 | ||
2ef37d3f MH |
3835 | /** |
3836 | * mem_cgroup_move_parent - moves page to the parent group | |
3837 | * @page: the page to move | |
3838 | * @pc: page_cgroup of the page | |
3839 | * @child: page's cgroup | |
3840 | * | |
3841 | * move charges to its parent or the root cgroup if the group has no | |
3842 | * parent (aka use_hierarchy==0). | |
3843 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3844 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3845 | * it signals a race with a page removal/uncharge or migration. In the | |
3846 | * first case the page is on the way out and it will vanish from the LRU | |
3847 | * on the next attempt and the call should be retried later. | |
3848 | * Isolation from the LRU fails only if page has been isolated from | |
3849 | * the LRU since we looked at it and that usually means either global | |
3850 | * reclaim or migration going on. The page will either get back to the | |
3851 | * LRU or vanish. | |
3852 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3853 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3854 | * disappear in the next attempt. | |
f817ed48 | 3855 | */ |
5564e88b JW |
3856 | static int mem_cgroup_move_parent(struct page *page, |
3857 | struct page_cgroup *pc, | |
6068bf01 | 3858 | struct mem_cgroup *child) |
f817ed48 | 3859 | { |
f817ed48 | 3860 | struct mem_cgroup *parent; |
7ec99d62 | 3861 | unsigned int nr_pages; |
4be4489f | 3862 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3863 | int ret; |
3864 | ||
d8423011 | 3865 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3866 | |
57f9fd7d DN |
3867 | ret = -EBUSY; |
3868 | if (!get_page_unless_zero(page)) | |
3869 | goto out; | |
3870 | if (isolate_lru_page(page)) | |
3871 | goto put; | |
52dbb905 | 3872 | |
7ec99d62 | 3873 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3874 | |
cc926f78 KH |
3875 | parent = parent_mem_cgroup(child); |
3876 | /* | |
3877 | * If no parent, move charges to root cgroup. | |
3878 | */ | |
3879 | if (!parent) | |
3880 | parent = root_mem_cgroup; | |
f817ed48 | 3881 | |
2ef37d3f | 3882 | if (nr_pages > 1) { |
309381fe | 3883 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
987eba66 | 3884 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3885 | } |
987eba66 | 3886 | |
cc926f78 | 3887 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3888 | pc, child, parent); |
cc926f78 KH |
3889 | if (!ret) |
3890 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3891 | |
7ec99d62 | 3892 | if (nr_pages > 1) |
987eba66 | 3893 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3894 | putback_lru_page(page); |
57f9fd7d | 3895 | put: |
40d58138 | 3896 | put_page(page); |
57f9fd7d | 3897 | out: |
f817ed48 KH |
3898 | return ret; |
3899 | } | |
3900 | ||
7a81b88c KH |
3901 | /* |
3902 | * Charge the memory controller for page usage. | |
3903 | * Return | |
3904 | * 0 if the charge was successful | |
3905 | * < 0 if the cgroup is over its limit | |
3906 | */ | |
3907 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3908 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3909 | { |
c0ff4b85 | 3910 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3911 | unsigned int nr_pages = 1; |
8493ae43 | 3912 | bool oom = true; |
7a81b88c | 3913 | int ret; |
ec168510 | 3914 | |
37c2ac78 | 3915 | if (PageTransHuge(page)) { |
7ec99d62 | 3916 | nr_pages <<= compound_order(page); |
309381fe | 3917 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
8493ae43 JW |
3918 | /* |
3919 | * Never OOM-kill a process for a huge page. The | |
3920 | * fault handler will fall back to regular pages. | |
3921 | */ | |
3922 | oom = false; | |
37c2ac78 | 3923 | } |
7a81b88c | 3924 | |
c0ff4b85 | 3925 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3926 | if (ret == -ENOMEM) |
7a81b88c | 3927 | return ret; |
ce587e65 | 3928 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3929 | return 0; |
8a9f3ccd BS |
3930 | } |
3931 | ||
7a81b88c KH |
3932 | int mem_cgroup_newpage_charge(struct page *page, |
3933 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3934 | { |
f8d66542 | 3935 | if (mem_cgroup_disabled()) |
cede86ac | 3936 | return 0; |
309381fe SL |
3937 | VM_BUG_ON_PAGE(page_mapped(page), page); |
3938 | VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page); | |
7a0524cf | 3939 | VM_BUG_ON(!mm); |
217bc319 | 3940 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3941 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
3942 | } |
3943 | ||
54595fe2 KH |
3944 | /* |
3945 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3946 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3947 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3948 | * "commit()" or removed by "cancel()" |
3949 | */ | |
0435a2fd JW |
3950 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3951 | struct page *page, | |
3952 | gfp_t mask, | |
3953 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3954 | { |
c0ff4b85 | 3955 | struct mem_cgroup *memcg; |
90deb788 | 3956 | struct page_cgroup *pc; |
54595fe2 | 3957 | int ret; |
8c7c6e34 | 3958 | |
90deb788 JW |
3959 | pc = lookup_page_cgroup(page); |
3960 | /* | |
3961 | * Every swap fault against a single page tries to charge the | |
3962 | * page, bail as early as possible. shmem_unuse() encounters | |
3963 | * already charged pages, too. The USED bit is protected by | |
3964 | * the page lock, which serializes swap cache removal, which | |
3965 | * in turn serializes uncharging. | |
3966 | */ | |
3967 | if (PageCgroupUsed(pc)) | |
3968 | return 0; | |
8c7c6e34 KH |
3969 | if (!do_swap_account) |
3970 | goto charge_cur_mm; | |
c0ff4b85 R |
3971 | memcg = try_get_mem_cgroup_from_page(page); |
3972 | if (!memcg) | |
54595fe2 | 3973 | goto charge_cur_mm; |
72835c86 JW |
3974 | *memcgp = memcg; |
3975 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 3976 | css_put(&memcg->css); |
38c5d72f KH |
3977 | if (ret == -EINTR) |
3978 | ret = 0; | |
54595fe2 | 3979 | return ret; |
8c7c6e34 | 3980 | charge_cur_mm: |
38c5d72f KH |
3981 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
3982 | if (ret == -EINTR) | |
3983 | ret = 0; | |
3984 | return ret; | |
8c7c6e34 KH |
3985 | } |
3986 | ||
0435a2fd JW |
3987 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
3988 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
3989 | { | |
3990 | *memcgp = NULL; | |
3991 | if (mem_cgroup_disabled()) | |
3992 | return 0; | |
bdf4f4d2 JW |
3993 | /* |
3994 | * A racing thread's fault, or swapoff, may have already | |
3995 | * updated the pte, and even removed page from swap cache: in | |
3996 | * those cases unuse_pte()'s pte_same() test will fail; but | |
3997 | * there's also a KSM case which does need to charge the page. | |
3998 | */ | |
3999 | if (!PageSwapCache(page)) { | |
4000 | int ret; | |
4001 | ||
4002 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
4003 | if (ret == -EINTR) | |
4004 | ret = 0; | |
4005 | return ret; | |
4006 | } | |
0435a2fd JW |
4007 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
4008 | } | |
4009 | ||
827a03d2 JW |
4010 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
4011 | { | |
4012 | if (mem_cgroup_disabled()) | |
4013 | return; | |
4014 | if (!memcg) | |
4015 | return; | |
4016 | __mem_cgroup_cancel_charge(memcg, 1); | |
4017 | } | |
4018 | ||
83aae4c7 | 4019 | static void |
72835c86 | 4020 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 4021 | enum charge_type ctype) |
7a81b88c | 4022 | { |
f8d66542 | 4023 | if (mem_cgroup_disabled()) |
7a81b88c | 4024 | return; |
72835c86 | 4025 | if (!memcg) |
7a81b88c | 4026 | return; |
5a6475a4 | 4027 | |
ce587e65 | 4028 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
4029 | /* |
4030 | * Now swap is on-memory. This means this page may be | |
4031 | * counted both as mem and swap....double count. | |
03f3c433 KH |
4032 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
4033 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
4034 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 4035 | */ |
03f3c433 | 4036 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 4037 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 4038 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 4039 | } |
7a81b88c KH |
4040 | } |
4041 | ||
72835c86 JW |
4042 | void mem_cgroup_commit_charge_swapin(struct page *page, |
4043 | struct mem_cgroup *memcg) | |
83aae4c7 | 4044 | { |
72835c86 | 4045 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 4046 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
4047 | } |
4048 | ||
827a03d2 JW |
4049 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
4050 | gfp_t gfp_mask) | |
7a81b88c | 4051 | { |
827a03d2 JW |
4052 | struct mem_cgroup *memcg = NULL; |
4053 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
4054 | int ret; | |
4055 | ||
f8d66542 | 4056 | if (mem_cgroup_disabled()) |
827a03d2 JW |
4057 | return 0; |
4058 | if (PageCompound(page)) | |
4059 | return 0; | |
4060 | ||
827a03d2 JW |
4061 | if (!PageSwapCache(page)) |
4062 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
4063 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
4064 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
4065 | gfp_mask, &memcg); | |
827a03d2 JW |
4066 | if (!ret) |
4067 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
4068 | } | |
4069 | return ret; | |
7a81b88c KH |
4070 | } |
4071 | ||
c0ff4b85 | 4072 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
4073 | unsigned int nr_pages, |
4074 | const enum charge_type ctype) | |
569b846d KH |
4075 | { |
4076 | struct memcg_batch_info *batch = NULL; | |
4077 | bool uncharge_memsw = true; | |
7ec99d62 | 4078 | |
569b846d KH |
4079 | /* If swapout, usage of swap doesn't decrease */ |
4080 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
4081 | uncharge_memsw = false; | |
569b846d KH |
4082 | |
4083 | batch = ¤t->memcg_batch; | |
4084 | /* | |
4085 | * In usual, we do css_get() when we remember memcg pointer. | |
4086 | * But in this case, we keep res->usage until end of a series of | |
4087 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
4088 | */ | |
4089 | if (!batch->memcg) | |
c0ff4b85 | 4090 | batch->memcg = memcg; |
3c11ecf4 KH |
4091 | /* |
4092 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 4093 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
4094 | * the same cgroup and we have chance to coalesce uncharges. |
4095 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
4096 | * because we want to do uncharge as soon as possible. | |
4097 | */ | |
4098 | ||
4099 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
4100 | goto direct_uncharge; | |
4101 | ||
7ec99d62 | 4102 | if (nr_pages > 1) |
ec168510 AA |
4103 | goto direct_uncharge; |
4104 | ||
569b846d KH |
4105 | /* |
4106 | * In typical case, batch->memcg == mem. This means we can | |
4107 | * merge a series of uncharges to an uncharge of res_counter. | |
4108 | * If not, we uncharge res_counter ony by one. | |
4109 | */ | |
c0ff4b85 | 4110 | if (batch->memcg != memcg) |
569b846d KH |
4111 | goto direct_uncharge; |
4112 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 4113 | batch->nr_pages++; |
569b846d | 4114 | if (uncharge_memsw) |
7ffd4ca7 | 4115 | batch->memsw_nr_pages++; |
569b846d KH |
4116 | return; |
4117 | direct_uncharge: | |
c0ff4b85 | 4118 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 4119 | if (uncharge_memsw) |
c0ff4b85 R |
4120 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
4121 | if (unlikely(batch->memcg != memcg)) | |
4122 | memcg_oom_recover(memcg); | |
569b846d | 4123 | } |
7a81b88c | 4124 | |
8a9f3ccd | 4125 | /* |
69029cd5 | 4126 | * uncharge if !page_mapped(page) |
8a9f3ccd | 4127 | */ |
8c7c6e34 | 4128 | static struct mem_cgroup * |
0030f535 JW |
4129 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
4130 | bool end_migration) | |
8a9f3ccd | 4131 | { |
c0ff4b85 | 4132 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
4133 | unsigned int nr_pages = 1; |
4134 | struct page_cgroup *pc; | |
b2402857 | 4135 | bool anon; |
8a9f3ccd | 4136 | |
f8d66542 | 4137 | if (mem_cgroup_disabled()) |
8c7c6e34 | 4138 | return NULL; |
4077960e | 4139 | |
37c2ac78 | 4140 | if (PageTransHuge(page)) { |
7ec99d62 | 4141 | nr_pages <<= compound_order(page); |
309381fe | 4142 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
37c2ac78 | 4143 | } |
8697d331 | 4144 | /* |
3c541e14 | 4145 | * Check if our page_cgroup is valid |
8697d331 | 4146 | */ |
52d4b9ac | 4147 | pc = lookup_page_cgroup(page); |
cfa44946 | 4148 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 4149 | return NULL; |
b9c565d5 | 4150 | |
52d4b9ac | 4151 | lock_page_cgroup(pc); |
d13d1443 | 4152 | |
c0ff4b85 | 4153 | memcg = pc->mem_cgroup; |
8c7c6e34 | 4154 | |
d13d1443 KH |
4155 | if (!PageCgroupUsed(pc)) |
4156 | goto unlock_out; | |
4157 | ||
b2402857 KH |
4158 | anon = PageAnon(page); |
4159 | ||
d13d1443 | 4160 | switch (ctype) { |
41326c17 | 4161 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
4162 | /* |
4163 | * Generally PageAnon tells if it's the anon statistics to be | |
4164 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4165 | * used before page reached the stage of being marked PageAnon. | |
4166 | */ | |
b2402857 KH |
4167 | anon = true; |
4168 | /* fallthrough */ | |
8a9478ca | 4169 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4170 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4171 | if (page_mapped(page)) |
4172 | goto unlock_out; | |
4173 | /* | |
4174 | * Pages under migration may not be uncharged. But | |
4175 | * end_migration() /must/ be the one uncharging the | |
4176 | * unused post-migration page and so it has to call | |
4177 | * here with the migration bit still set. See the | |
4178 | * res_counter handling below. | |
4179 | */ | |
4180 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4181 | goto unlock_out; |
4182 | break; | |
4183 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4184 | if (!PageAnon(page)) { /* Shared memory */ | |
4185 | if (page->mapping && !page_is_file_cache(page)) | |
4186 | goto unlock_out; | |
4187 | } else if (page_mapped(page)) /* Anon */ | |
4188 | goto unlock_out; | |
4189 | break; | |
4190 | default: | |
4191 | break; | |
52d4b9ac | 4192 | } |
d13d1443 | 4193 | |
b070e65c | 4194 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
04046e1a | 4195 | |
52d4b9ac | 4196 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4197 | /* |
4198 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4199 | * freed from LRU. This is safe because uncharged page is expected not | |
4200 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4201 | * special functions. | |
4202 | */ | |
b9c565d5 | 4203 | |
52d4b9ac | 4204 | unlock_page_cgroup(pc); |
f75ca962 | 4205 | /* |
c0ff4b85 | 4206 | * even after unlock, we have memcg->res.usage here and this memcg |
4050377b | 4207 | * will never be freed, so it's safe to call css_get(). |
f75ca962 | 4208 | */ |
c0ff4b85 | 4209 | memcg_check_events(memcg, page); |
f75ca962 | 4210 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 | 4211 | mem_cgroup_swap_statistics(memcg, true); |
4050377b | 4212 | css_get(&memcg->css); |
f75ca962 | 4213 | } |
0030f535 JW |
4214 | /* |
4215 | * Migration does not charge the res_counter for the | |
4216 | * replacement page, so leave it alone when phasing out the | |
4217 | * page that is unused after the migration. | |
4218 | */ | |
4219 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4220 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4221 | |
c0ff4b85 | 4222 | return memcg; |
d13d1443 KH |
4223 | |
4224 | unlock_out: | |
4225 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4226 | return NULL; |
3c541e14 BS |
4227 | } |
4228 | ||
69029cd5 KH |
4229 | void mem_cgroup_uncharge_page(struct page *page) |
4230 | { | |
52d4b9ac KH |
4231 | /* early check. */ |
4232 | if (page_mapped(page)) | |
4233 | return; | |
309381fe | 4234 | VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page); |
28ccddf7 JW |
4235 | /* |
4236 | * If the page is in swap cache, uncharge should be deferred | |
4237 | * to the swap path, which also properly accounts swap usage | |
4238 | * and handles memcg lifetime. | |
4239 | * | |
4240 | * Note that this check is not stable and reclaim may add the | |
4241 | * page to swap cache at any time after this. However, if the | |
4242 | * page is not in swap cache by the time page->mapcount hits | |
4243 | * 0, there won't be any page table references to the swap | |
4244 | * slot, and reclaim will free it and not actually write the | |
4245 | * page to disk. | |
4246 | */ | |
0c59b89c JW |
4247 | if (PageSwapCache(page)) |
4248 | return; | |
0030f535 | 4249 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4250 | } |
4251 | ||
4252 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4253 | { | |
309381fe SL |
4254 | VM_BUG_ON_PAGE(page_mapped(page), page); |
4255 | VM_BUG_ON_PAGE(page->mapping, page); | |
0030f535 | 4256 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4257 | } |
4258 | ||
569b846d KH |
4259 | /* |
4260 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4261 | * In that cases, pages are freed continuously and we can expect pages | |
4262 | * are in the same memcg. All these calls itself limits the number of | |
4263 | * pages freed at once, then uncharge_start/end() is called properly. | |
4264 | * This may be called prural(2) times in a context, | |
4265 | */ | |
4266 | ||
4267 | void mem_cgroup_uncharge_start(void) | |
4268 | { | |
4269 | current->memcg_batch.do_batch++; | |
4270 | /* We can do nest. */ | |
4271 | if (current->memcg_batch.do_batch == 1) { | |
4272 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4273 | current->memcg_batch.nr_pages = 0; |
4274 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4275 | } |
4276 | } | |
4277 | ||
4278 | void mem_cgroup_uncharge_end(void) | |
4279 | { | |
4280 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4281 | ||
4282 | if (!batch->do_batch) | |
4283 | return; | |
4284 | ||
4285 | batch->do_batch--; | |
4286 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4287 | return; | |
4288 | ||
4289 | if (!batch->memcg) | |
4290 | return; | |
4291 | /* | |
4292 | * This "batch->memcg" is valid without any css_get/put etc... | |
4293 | * bacause we hide charges behind us. | |
4294 | */ | |
7ffd4ca7 JW |
4295 | if (batch->nr_pages) |
4296 | res_counter_uncharge(&batch->memcg->res, | |
4297 | batch->nr_pages * PAGE_SIZE); | |
4298 | if (batch->memsw_nr_pages) | |
4299 | res_counter_uncharge(&batch->memcg->memsw, | |
4300 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4301 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4302 | /* forget this pointer (for sanity check) */ |
4303 | batch->memcg = NULL; | |
4304 | } | |
4305 | ||
e767e056 | 4306 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4307 | /* |
e767e056 | 4308 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4309 | * memcg information is recorded to swap_cgroup of "ent" |
4310 | */ | |
8a9478ca KH |
4311 | void |
4312 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4313 | { |
4314 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4315 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4316 | ||
4317 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4318 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4319 | ||
0030f535 | 4320 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4321 | |
f75ca962 KH |
4322 | /* |
4323 | * record memcg information, if swapout && memcg != NULL, | |
4050377b | 4324 | * css_get() was called in uncharge(). |
f75ca962 KH |
4325 | */ |
4326 | if (do_swap_account && swapout && memcg) | |
34c00c31 | 4327 | swap_cgroup_record(ent, mem_cgroup_id(memcg)); |
8c7c6e34 | 4328 | } |
e767e056 | 4329 | #endif |
8c7c6e34 | 4330 | |
c255a458 | 4331 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4332 | /* |
4333 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4334 | * uncharge "memsw" account. | |
4335 | */ | |
4336 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4337 | { |
8c7c6e34 | 4338 | struct mem_cgroup *memcg; |
a3b2d692 | 4339 | unsigned short id; |
8c7c6e34 KH |
4340 | |
4341 | if (!do_swap_account) | |
4342 | return; | |
4343 | ||
a3b2d692 KH |
4344 | id = swap_cgroup_record(ent, 0); |
4345 | rcu_read_lock(); | |
4346 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4347 | if (memcg) { |
a3b2d692 KH |
4348 | /* |
4349 | * We uncharge this because swap is freed. | |
4350 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4351 | */ | |
0c3e73e8 | 4352 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4353 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4354 | mem_cgroup_swap_statistics(memcg, false); |
4050377b | 4355 | css_put(&memcg->css); |
8c7c6e34 | 4356 | } |
a3b2d692 | 4357 | rcu_read_unlock(); |
d13d1443 | 4358 | } |
02491447 DN |
4359 | |
4360 | /** | |
4361 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4362 | * @entry: swap entry to be moved | |
4363 | * @from: mem_cgroup which the entry is moved from | |
4364 | * @to: mem_cgroup which the entry is moved to | |
4365 | * | |
4366 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4367 | * as the mem_cgroup's id of @from. | |
4368 | * | |
4369 | * Returns 0 on success, -EINVAL on failure. | |
4370 | * | |
4371 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4372 | * both res and memsw, and called css_get(). | |
4373 | */ | |
4374 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4375 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4376 | { |
4377 | unsigned short old_id, new_id; | |
4378 | ||
34c00c31 LZ |
4379 | old_id = mem_cgroup_id(from); |
4380 | new_id = mem_cgroup_id(to); | |
02491447 DN |
4381 | |
4382 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4383 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4384 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4385 | /* |
483c30b5 DN |
4386 | * This function is only called from task migration context now. |
4387 | * It postpones res_counter and refcount handling till the end | |
4388 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4050377b LZ |
4389 | * improvement. But we cannot postpone css_get(to) because if |
4390 | * the process that has been moved to @to does swap-in, the | |
4391 | * refcount of @to might be decreased to 0. | |
4392 | * | |
4393 | * We are in attach() phase, so the cgroup is guaranteed to be | |
4394 | * alive, so we can just call css_get(). | |
02491447 | 4395 | */ |
4050377b | 4396 | css_get(&to->css); |
02491447 DN |
4397 | return 0; |
4398 | } | |
4399 | return -EINVAL; | |
4400 | } | |
4401 | #else | |
4402 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4403 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4404 | { |
4405 | return -EINVAL; | |
4406 | } | |
8c7c6e34 | 4407 | #endif |
d13d1443 | 4408 | |
ae41be37 | 4409 | /* |
01b1ae63 KH |
4410 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4411 | * page belongs to. | |
ae41be37 | 4412 | */ |
0030f535 JW |
4413 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4414 | struct mem_cgroup **memcgp) | |
ae41be37 | 4415 | { |
c0ff4b85 | 4416 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4417 | unsigned int nr_pages = 1; |
7ec99d62 | 4418 | struct page_cgroup *pc; |
ac39cf8c | 4419 | enum charge_type ctype; |
8869b8f6 | 4420 | |
72835c86 | 4421 | *memcgp = NULL; |
56039efa | 4422 | |
f8d66542 | 4423 | if (mem_cgroup_disabled()) |
0030f535 | 4424 | return; |
4077960e | 4425 | |
b32967ff MG |
4426 | if (PageTransHuge(page)) |
4427 | nr_pages <<= compound_order(page); | |
4428 | ||
52d4b9ac KH |
4429 | pc = lookup_page_cgroup(page); |
4430 | lock_page_cgroup(pc); | |
4431 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4432 | memcg = pc->mem_cgroup; |
4433 | css_get(&memcg->css); | |
ac39cf8c | 4434 | /* |
4435 | * At migrating an anonymous page, its mapcount goes down | |
4436 | * to 0 and uncharge() will be called. But, even if it's fully | |
4437 | * unmapped, migration may fail and this page has to be | |
4438 | * charged again. We set MIGRATION flag here and delay uncharge | |
4439 | * until end_migration() is called | |
4440 | * | |
4441 | * Corner Case Thinking | |
4442 | * A) | |
4443 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4444 | * while migration was ongoing. | |
4445 | * If unmap finds the old page, uncharge() of it will be delayed | |
4446 | * until end_migration(). If unmap finds a new page, it's | |
4447 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4448 | * finds swap_migration_entry, the new page will not be mapped | |
4449 | * and end_migration() will find it(mapcount==0). | |
4450 | * | |
4451 | * B) | |
4452 | * When the old page was mapped but migraion fails, the kernel | |
4453 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4454 | * if mapcount goes down to 0. We can do remap successfully | |
4455 | * without charging it again. | |
4456 | * | |
4457 | * C) | |
4458 | * The "old" page is under lock_page() until the end of | |
4459 | * migration, so, the old page itself will not be swapped-out. | |
4460 | * If the new page is swapped out before end_migraton, our | |
4461 | * hook to usual swap-out path will catch the event. | |
4462 | */ | |
4463 | if (PageAnon(page)) | |
4464 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4465 | } |
52d4b9ac | 4466 | unlock_page_cgroup(pc); |
ac39cf8c | 4467 | /* |
4468 | * If the page is not charged at this point, | |
4469 | * we return here. | |
4470 | */ | |
c0ff4b85 | 4471 | if (!memcg) |
0030f535 | 4472 | return; |
01b1ae63 | 4473 | |
72835c86 | 4474 | *memcgp = memcg; |
ac39cf8c | 4475 | /* |
4476 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4477 | * is called before end_migration, we can catch all events on this new | |
4478 | * page. In the case new page is migrated but not remapped, new page's | |
4479 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4480 | */ | |
ac39cf8c | 4481 | if (PageAnon(page)) |
41326c17 | 4482 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4483 | else |
62ba7442 | 4484 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4485 | /* |
4486 | * The page is committed to the memcg, but it's not actually | |
4487 | * charged to the res_counter since we plan on replacing the | |
4488 | * old one and only one page is going to be left afterwards. | |
4489 | */ | |
b32967ff | 4490 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4491 | } |
8869b8f6 | 4492 | |
69029cd5 | 4493 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4494 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4495 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4496 | { |
ac39cf8c | 4497 | struct page *used, *unused; |
01b1ae63 | 4498 | struct page_cgroup *pc; |
b2402857 | 4499 | bool anon; |
01b1ae63 | 4500 | |
c0ff4b85 | 4501 | if (!memcg) |
01b1ae63 | 4502 | return; |
b25ed609 | 4503 | |
50de1dd9 | 4504 | if (!migration_ok) { |
ac39cf8c | 4505 | used = oldpage; |
4506 | unused = newpage; | |
01b1ae63 | 4507 | } else { |
ac39cf8c | 4508 | used = newpage; |
01b1ae63 KH |
4509 | unused = oldpage; |
4510 | } | |
0030f535 | 4511 | anon = PageAnon(used); |
7d188958 JW |
4512 | __mem_cgroup_uncharge_common(unused, |
4513 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4514 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4515 | true); | |
0030f535 | 4516 | css_put(&memcg->css); |
69029cd5 | 4517 | /* |
ac39cf8c | 4518 | * We disallowed uncharge of pages under migration because mapcount |
4519 | * of the page goes down to zero, temporarly. | |
4520 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4521 | */ |
ac39cf8c | 4522 | pc = lookup_page_cgroup(oldpage); |
4523 | lock_page_cgroup(pc); | |
4524 | ClearPageCgroupMigration(pc); | |
4525 | unlock_page_cgroup(pc); | |
ac39cf8c | 4526 | |
01b1ae63 | 4527 | /* |
ac39cf8c | 4528 | * If a page is a file cache, radix-tree replacement is very atomic |
4529 | * and we can skip this check. When it was an Anon page, its mapcount | |
4530 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4531 | * uncharged yet. There are several case but page->mapcount check | |
4532 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4533 | * check. (see prepare_charge() also) | |
69029cd5 | 4534 | */ |
b2402857 | 4535 | if (anon) |
ac39cf8c | 4536 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4537 | } |
78fb7466 | 4538 | |
ab936cbc KH |
4539 | /* |
4540 | * At replace page cache, newpage is not under any memcg but it's on | |
4541 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4542 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4543 | */ | |
4544 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4545 | struct page *newpage) | |
4546 | { | |
bde05d1c | 4547 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4548 | struct page_cgroup *pc; |
ab936cbc | 4549 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4550 | |
4551 | if (mem_cgroup_disabled()) | |
4552 | return; | |
4553 | ||
4554 | pc = lookup_page_cgroup(oldpage); | |
4555 | /* fix accounting on old pages */ | |
4556 | lock_page_cgroup(pc); | |
bde05d1c HD |
4557 | if (PageCgroupUsed(pc)) { |
4558 | memcg = pc->mem_cgroup; | |
b070e65c | 4559 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
bde05d1c HD |
4560 | ClearPageCgroupUsed(pc); |
4561 | } | |
ab936cbc KH |
4562 | unlock_page_cgroup(pc); |
4563 | ||
bde05d1c HD |
4564 | /* |
4565 | * When called from shmem_replace_page(), in some cases the | |
4566 | * oldpage has already been charged, and in some cases not. | |
4567 | */ | |
4568 | if (!memcg) | |
4569 | return; | |
ab936cbc KH |
4570 | /* |
4571 | * Even if newpage->mapping was NULL before starting replacement, | |
4572 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4573 | * LRU while we overwrite pc->mem_cgroup. | |
4574 | */ | |
ce587e65 | 4575 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4576 | } |
4577 | ||
f212ad7c DN |
4578 | #ifdef CONFIG_DEBUG_VM |
4579 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4580 | { | |
4581 | struct page_cgroup *pc; | |
4582 | ||
4583 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4584 | /* |
4585 | * Can be NULL while feeding pages into the page allocator for | |
4586 | * the first time, i.e. during boot or memory hotplug; | |
4587 | * or when mem_cgroup_disabled(). | |
4588 | */ | |
f212ad7c DN |
4589 | if (likely(pc) && PageCgroupUsed(pc)) |
4590 | return pc; | |
4591 | return NULL; | |
4592 | } | |
4593 | ||
4594 | bool mem_cgroup_bad_page_check(struct page *page) | |
4595 | { | |
4596 | if (mem_cgroup_disabled()) | |
4597 | return false; | |
4598 | ||
4599 | return lookup_page_cgroup_used(page) != NULL; | |
4600 | } | |
4601 | ||
4602 | void mem_cgroup_print_bad_page(struct page *page) | |
4603 | { | |
4604 | struct page_cgroup *pc; | |
4605 | ||
4606 | pc = lookup_page_cgroup_used(page); | |
4607 | if (pc) { | |
d045197f AM |
4608 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4609 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4610 | } |
4611 | } | |
4612 | #endif | |
4613 | ||
d38d2a75 | 4614 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4615 | unsigned long long val) |
628f4235 | 4616 | { |
81d39c20 | 4617 | int retry_count; |
3c11ecf4 | 4618 | u64 memswlimit, memlimit; |
628f4235 | 4619 | int ret = 0; |
81d39c20 KH |
4620 | int children = mem_cgroup_count_children(memcg); |
4621 | u64 curusage, oldusage; | |
3c11ecf4 | 4622 | int enlarge; |
81d39c20 KH |
4623 | |
4624 | /* | |
4625 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4626 | * is depends on callers. We set our retry-count to be function | |
4627 | * of # of children which we should visit in this loop. | |
4628 | */ | |
4629 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4630 | ||
4631 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4632 | |
3c11ecf4 | 4633 | enlarge = 0; |
8c7c6e34 | 4634 | while (retry_count) { |
628f4235 KH |
4635 | if (signal_pending(current)) { |
4636 | ret = -EINTR; | |
4637 | break; | |
4638 | } | |
8c7c6e34 KH |
4639 | /* |
4640 | * Rather than hide all in some function, I do this in | |
4641 | * open coded manner. You see what this really does. | |
aaad153e | 4642 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4643 | */ |
4644 | mutex_lock(&set_limit_mutex); | |
4645 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4646 | if (memswlimit < val) { | |
4647 | ret = -EINVAL; | |
4648 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4649 | break; |
4650 | } | |
3c11ecf4 KH |
4651 | |
4652 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4653 | if (memlimit < val) | |
4654 | enlarge = 1; | |
4655 | ||
8c7c6e34 | 4656 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4657 | if (!ret) { |
4658 | if (memswlimit == val) | |
4659 | memcg->memsw_is_minimum = true; | |
4660 | else | |
4661 | memcg->memsw_is_minimum = false; | |
4662 | } | |
8c7c6e34 KH |
4663 | mutex_unlock(&set_limit_mutex); |
4664 | ||
4665 | if (!ret) | |
4666 | break; | |
4667 | ||
5660048c JW |
4668 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4669 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4670 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4671 | /* Usage is reduced ? */ | |
f894ffa8 | 4672 | if (curusage >= oldusage) |
81d39c20 KH |
4673 | retry_count--; |
4674 | else | |
4675 | oldusage = curusage; | |
8c7c6e34 | 4676 | } |
3c11ecf4 KH |
4677 | if (!ret && enlarge) |
4678 | memcg_oom_recover(memcg); | |
14797e23 | 4679 | |
8c7c6e34 KH |
4680 | return ret; |
4681 | } | |
4682 | ||
338c8431 LZ |
4683 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4684 | unsigned long long val) | |
8c7c6e34 | 4685 | { |
81d39c20 | 4686 | int retry_count; |
3c11ecf4 | 4687 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4688 | int children = mem_cgroup_count_children(memcg); |
4689 | int ret = -EBUSY; | |
3c11ecf4 | 4690 | int enlarge = 0; |
8c7c6e34 | 4691 | |
81d39c20 | 4692 | /* see mem_cgroup_resize_res_limit */ |
f894ffa8 | 4693 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
81d39c20 | 4694 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
8c7c6e34 KH |
4695 | while (retry_count) { |
4696 | if (signal_pending(current)) { | |
4697 | ret = -EINTR; | |
4698 | break; | |
4699 | } | |
4700 | /* | |
4701 | * Rather than hide all in some function, I do this in | |
4702 | * open coded manner. You see what this really does. | |
aaad153e | 4703 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4704 | */ |
4705 | mutex_lock(&set_limit_mutex); | |
4706 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4707 | if (memlimit > val) { | |
4708 | ret = -EINVAL; | |
4709 | mutex_unlock(&set_limit_mutex); | |
4710 | break; | |
4711 | } | |
3c11ecf4 KH |
4712 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4713 | if (memswlimit < val) | |
4714 | enlarge = 1; | |
8c7c6e34 | 4715 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4716 | if (!ret) { |
4717 | if (memlimit == val) | |
4718 | memcg->memsw_is_minimum = true; | |
4719 | else | |
4720 | memcg->memsw_is_minimum = false; | |
4721 | } | |
8c7c6e34 KH |
4722 | mutex_unlock(&set_limit_mutex); |
4723 | ||
4724 | if (!ret) | |
4725 | break; | |
4726 | ||
5660048c JW |
4727 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4728 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4729 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4730 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4731 | /* Usage is reduced ? */ |
8c7c6e34 | 4732 | if (curusage >= oldusage) |
628f4235 | 4733 | retry_count--; |
81d39c20 KH |
4734 | else |
4735 | oldusage = curusage; | |
628f4235 | 4736 | } |
3c11ecf4 KH |
4737 | if (!ret && enlarge) |
4738 | memcg_oom_recover(memcg); | |
628f4235 KH |
4739 | return ret; |
4740 | } | |
4741 | ||
0608f43d AM |
4742 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
4743 | gfp_t gfp_mask, | |
4744 | unsigned long *total_scanned) | |
4745 | { | |
4746 | unsigned long nr_reclaimed = 0; | |
4747 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
4748 | unsigned long reclaimed; | |
4749 | int loop = 0; | |
4750 | struct mem_cgroup_tree_per_zone *mctz; | |
4751 | unsigned long long excess; | |
4752 | unsigned long nr_scanned; | |
4753 | ||
4754 | if (order > 0) | |
4755 | return 0; | |
4756 | ||
4757 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
4758 | /* | |
4759 | * This loop can run a while, specially if mem_cgroup's continuously | |
4760 | * keep exceeding their soft limit and putting the system under | |
4761 | * pressure | |
4762 | */ | |
4763 | do { | |
4764 | if (next_mz) | |
4765 | mz = next_mz; | |
4766 | else | |
4767 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
4768 | if (!mz) | |
4769 | break; | |
4770 | ||
4771 | nr_scanned = 0; | |
4772 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
4773 | gfp_mask, &nr_scanned); | |
4774 | nr_reclaimed += reclaimed; | |
4775 | *total_scanned += nr_scanned; | |
4776 | spin_lock(&mctz->lock); | |
4777 | ||
4778 | /* | |
4779 | * If we failed to reclaim anything from this memory cgroup | |
4780 | * it is time to move on to the next cgroup | |
4781 | */ | |
4782 | next_mz = NULL; | |
4783 | if (!reclaimed) { | |
4784 | do { | |
4785 | /* | |
4786 | * Loop until we find yet another one. | |
4787 | * | |
4788 | * By the time we get the soft_limit lock | |
4789 | * again, someone might have aded the | |
4790 | * group back on the RB tree. Iterate to | |
4791 | * make sure we get a different mem. | |
4792 | * mem_cgroup_largest_soft_limit_node returns | |
4793 | * NULL if no other cgroup is present on | |
4794 | * the tree | |
4795 | */ | |
4796 | next_mz = | |
4797 | __mem_cgroup_largest_soft_limit_node(mctz); | |
4798 | if (next_mz == mz) | |
4799 | css_put(&next_mz->memcg->css); | |
4800 | else /* next_mz == NULL or other memcg */ | |
4801 | break; | |
4802 | } while (1); | |
4803 | } | |
4804 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); | |
4805 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4806 | /* | |
4807 | * One school of thought says that we should not add | |
4808 | * back the node to the tree if reclaim returns 0. | |
4809 | * But our reclaim could return 0, simply because due | |
4810 | * to priority we are exposing a smaller subset of | |
4811 | * memory to reclaim from. Consider this as a longer | |
4812 | * term TODO. | |
4813 | */ | |
4814 | /* If excess == 0, no tree ops */ | |
4815 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); | |
4816 | spin_unlock(&mctz->lock); | |
4817 | css_put(&mz->memcg->css); | |
4818 | loop++; | |
4819 | /* | |
4820 | * Could not reclaim anything and there are no more | |
4821 | * mem cgroups to try or we seem to be looping without | |
4822 | * reclaiming anything. | |
4823 | */ | |
4824 | if (!nr_reclaimed && | |
4825 | (next_mz == NULL || | |
4826 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4827 | break; | |
4828 | } while (!nr_reclaimed); | |
4829 | if (next_mz) | |
4830 | css_put(&next_mz->memcg->css); | |
4831 | return nr_reclaimed; | |
4832 | } | |
4833 | ||
2ef37d3f MH |
4834 | /** |
4835 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4836 | * @memcg: group to clear | |
4837 | * @node: NUMA node | |
4838 | * @zid: zone id | |
4839 | * @lru: lru to to clear | |
4840 | * | |
3c935d18 | 4841 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4842 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4843 | * group. | |
cc847582 | 4844 | */ |
2ef37d3f | 4845 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4846 | int node, int zid, enum lru_list lru) |
cc847582 | 4847 | { |
bea8c150 | 4848 | struct lruvec *lruvec; |
2ef37d3f | 4849 | unsigned long flags; |
072c56c1 | 4850 | struct list_head *list; |
925b7673 JW |
4851 | struct page *busy; |
4852 | struct zone *zone; | |
072c56c1 | 4853 | |
08e552c6 | 4854 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4855 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4856 | list = &lruvec->lists[lru]; | |
cc847582 | 4857 | |
f817ed48 | 4858 | busy = NULL; |
2ef37d3f | 4859 | do { |
925b7673 | 4860 | struct page_cgroup *pc; |
5564e88b JW |
4861 | struct page *page; |
4862 | ||
08e552c6 | 4863 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4864 | if (list_empty(list)) { |
08e552c6 | 4865 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4866 | break; |
f817ed48 | 4867 | } |
925b7673 JW |
4868 | page = list_entry(list->prev, struct page, lru); |
4869 | if (busy == page) { | |
4870 | list_move(&page->lru, list); | |
648bcc77 | 4871 | busy = NULL; |
08e552c6 | 4872 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4873 | continue; |
4874 | } | |
08e552c6 | 4875 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4876 | |
925b7673 | 4877 | pc = lookup_page_cgroup(page); |
5564e88b | 4878 | |
3c935d18 | 4879 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4880 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4881 | busy = page; |
f817ed48 KH |
4882 | cond_resched(); |
4883 | } else | |
4884 | busy = NULL; | |
2ef37d3f | 4885 | } while (!list_empty(list)); |
cc847582 KH |
4886 | } |
4887 | ||
4888 | /* | |
c26251f9 MH |
4889 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4890 | * all the charges and pages to the parent. | |
cc847582 | 4891 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4892 | * |
4893 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4894 | */ |
ab5196c2 | 4895 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4896 | { |
c26251f9 | 4897 | int node, zid; |
bea207c8 | 4898 | u64 usage; |
f817ed48 | 4899 | |
fce66477 | 4900 | do { |
52d4b9ac KH |
4901 | /* This is for making all *used* pages to be on LRU. */ |
4902 | lru_add_drain_all(); | |
c0ff4b85 | 4903 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4904 | mem_cgroup_start_move(memcg); |
31aaea4a | 4905 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4906 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4907 | enum lru_list lru; |
4908 | for_each_lru(lru) { | |
2ef37d3f | 4909 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4910 | node, zid, lru); |
f817ed48 | 4911 | } |
1ecaab2b | 4912 | } |
f817ed48 | 4913 | } |
c0ff4b85 R |
4914 | mem_cgroup_end_move(memcg); |
4915 | memcg_oom_recover(memcg); | |
52d4b9ac | 4916 | cond_resched(); |
f817ed48 | 4917 | |
2ef37d3f | 4918 | /* |
bea207c8 GC |
4919 | * Kernel memory may not necessarily be trackable to a specific |
4920 | * process. So they are not migrated, and therefore we can't | |
4921 | * expect their value to drop to 0 here. | |
4922 | * Having res filled up with kmem only is enough. | |
4923 | * | |
2ef37d3f MH |
4924 | * This is a safety check because mem_cgroup_force_empty_list |
4925 | * could have raced with mem_cgroup_replace_page_cache callers | |
4926 | * so the lru seemed empty but the page could have been added | |
4927 | * right after the check. RES_USAGE should be safe as we always | |
4928 | * charge before adding to the LRU. | |
4929 | */ | |
bea207c8 GC |
4930 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4931 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4932 | } while (usage > 0); | |
c26251f9 MH |
4933 | } |
4934 | ||
b5f99b53 GC |
4935 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
4936 | { | |
696ac172 JW |
4937 | lockdep_assert_held(&memcg_create_mutex); |
4938 | /* | |
4939 | * The lock does not prevent addition or deletion to the list | |
4940 | * of children, but it prevents a new child from being | |
4941 | * initialized based on this parent in css_online(), so it's | |
4942 | * enough to decide whether hierarchically inherited | |
4943 | * attributes can still be changed or not. | |
4944 | */ | |
4945 | return memcg->use_hierarchy && | |
4946 | !list_empty(&memcg->css.cgroup->children); | |
b5f99b53 GC |
4947 | } |
4948 | ||
c26251f9 MH |
4949 | /* |
4950 | * Reclaims as many pages from the given memcg as possible and moves | |
4951 | * the rest to the parent. | |
4952 | * | |
4953 | * Caller is responsible for holding css reference for memcg. | |
4954 | */ | |
4955 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4956 | { | |
4957 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
4958 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 4959 | |
c1e862c1 | 4960 | /* returns EBUSY if there is a task or if we come here twice. */ |
07bc356e | 4961 | if (cgroup_has_tasks(cgrp) || !list_empty(&cgrp->children)) |
c26251f9 MH |
4962 | return -EBUSY; |
4963 | ||
c1e862c1 KH |
4964 | /* we call try-to-free pages for make this cgroup empty */ |
4965 | lru_add_drain_all(); | |
f817ed48 | 4966 | /* try to free all pages in this cgroup */ |
569530fb | 4967 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4968 | int progress; |
c1e862c1 | 4969 | |
c26251f9 MH |
4970 | if (signal_pending(current)) |
4971 | return -EINTR; | |
4972 | ||
c0ff4b85 | 4973 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4974 | false); |
c1e862c1 | 4975 | if (!progress) { |
f817ed48 | 4976 | nr_retries--; |
c1e862c1 | 4977 | /* maybe some writeback is necessary */ |
8aa7e847 | 4978 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4979 | } |
f817ed48 KH |
4980 | |
4981 | } | |
08e552c6 | 4982 | lru_add_drain(); |
ab5196c2 MH |
4983 | mem_cgroup_reparent_charges(memcg); |
4984 | ||
4985 | return 0; | |
cc847582 KH |
4986 | } |
4987 | ||
182446d0 TH |
4988 | static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css, |
4989 | unsigned int event) | |
c1e862c1 | 4990 | { |
182446d0 | 4991 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c26251f9 | 4992 | |
d8423011 MH |
4993 | if (mem_cgroup_is_root(memcg)) |
4994 | return -EINVAL; | |
c33bd835 | 4995 | return mem_cgroup_force_empty(memcg); |
c1e862c1 KH |
4996 | } |
4997 | ||
182446d0 TH |
4998 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
4999 | struct cftype *cft) | |
18f59ea7 | 5000 | { |
182446d0 | 5001 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
5002 | } |
5003 | ||
182446d0 TH |
5004 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
5005 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
5006 | { |
5007 | int retval = 0; | |
182446d0 | 5008 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5009 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css)); |
18f59ea7 | 5010 | |
0999821b | 5011 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
5012 | |
5013 | if (memcg->use_hierarchy == val) | |
5014 | goto out; | |
5015 | ||
18f59ea7 | 5016 | /* |
af901ca1 | 5017 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
5018 | * in the child subtrees. If it is unset, then the change can |
5019 | * occur, provided the current cgroup has no children. | |
5020 | * | |
5021 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
5022 | * set if there are no children. | |
5023 | */ | |
c0ff4b85 | 5024 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 5025 | (val == 1 || val == 0)) { |
696ac172 | 5026 | if (list_empty(&memcg->css.cgroup->children)) |
c0ff4b85 | 5027 | memcg->use_hierarchy = val; |
18f59ea7 BS |
5028 | else |
5029 | retval = -EBUSY; | |
5030 | } else | |
5031 | retval = -EINVAL; | |
567fb435 GC |
5032 | |
5033 | out: | |
0999821b | 5034 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
5035 | |
5036 | return retval; | |
5037 | } | |
5038 | ||
0c3e73e8 | 5039 | |
c0ff4b85 | 5040 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 5041 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 5042 | { |
7d74b06f | 5043 | struct mem_cgroup *iter; |
7a159cc9 | 5044 | long val = 0; |
0c3e73e8 | 5045 | |
7a159cc9 | 5046 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 5047 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
5048 | val += mem_cgroup_read_stat(iter, idx); |
5049 | ||
5050 | if (val < 0) /* race ? */ | |
5051 | val = 0; | |
5052 | return val; | |
0c3e73e8 BS |
5053 | } |
5054 | ||
c0ff4b85 | 5055 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 5056 | { |
7d74b06f | 5057 | u64 val; |
104f3928 | 5058 | |
c0ff4b85 | 5059 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 5060 | if (!swap) |
65c64ce8 | 5061 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 5062 | else |
65c64ce8 | 5063 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
5064 | } |
5065 | ||
b070e65c DR |
5066 | /* |
5067 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS | |
5068 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. | |
5069 | */ | |
c0ff4b85 R |
5070 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
5071 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 5072 | |
7d74b06f | 5073 | if (swap) |
bff6bb83 | 5074 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
5075 | |
5076 | return val << PAGE_SHIFT; | |
5077 | } | |
5078 | ||
791badbd TH |
5079 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
5080 | struct cftype *cft) | |
8cdea7c0 | 5081 | { |
182446d0 | 5082 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
104f3928 | 5083 | u64 val; |
791badbd | 5084 | int name; |
86ae53e1 | 5085 | enum res_type type; |
8c7c6e34 KH |
5086 | |
5087 | type = MEMFILE_TYPE(cft->private); | |
5088 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5089 | |
8c7c6e34 KH |
5090 | switch (type) { |
5091 | case _MEM: | |
104f3928 | 5092 | if (name == RES_USAGE) |
c0ff4b85 | 5093 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 5094 | else |
c0ff4b85 | 5095 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
5096 | break; |
5097 | case _MEMSWAP: | |
104f3928 | 5098 | if (name == RES_USAGE) |
c0ff4b85 | 5099 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 5100 | else |
c0ff4b85 | 5101 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 5102 | break; |
510fc4e1 GC |
5103 | case _KMEM: |
5104 | val = res_counter_read_u64(&memcg->kmem, name); | |
5105 | break; | |
8c7c6e34 KH |
5106 | default: |
5107 | BUG(); | |
8c7c6e34 | 5108 | } |
af36f906 | 5109 | |
791badbd | 5110 | return val; |
8cdea7c0 | 5111 | } |
510fc4e1 | 5112 | |
510fc4e1 | 5113 | #ifdef CONFIG_MEMCG_KMEM |
d6441637 VD |
5114 | /* should be called with activate_kmem_mutex held */ |
5115 | static int __memcg_activate_kmem(struct mem_cgroup *memcg, | |
5116 | unsigned long long limit) | |
5117 | { | |
5118 | int err = 0; | |
5119 | int memcg_id; | |
5120 | ||
5121 | if (memcg_kmem_is_active(memcg)) | |
5122 | return 0; | |
5123 | ||
5124 | /* | |
5125 | * We are going to allocate memory for data shared by all memory | |
5126 | * cgroups so let's stop accounting here. | |
5127 | */ | |
5128 | memcg_stop_kmem_account(); | |
5129 | ||
510fc4e1 GC |
5130 | /* |
5131 | * For simplicity, we won't allow this to be disabled. It also can't | |
5132 | * be changed if the cgroup has children already, or if tasks had | |
5133 | * already joined. | |
5134 | * | |
5135 | * If tasks join before we set the limit, a person looking at | |
5136 | * kmem.usage_in_bytes will have no way to determine when it took | |
5137 | * place, which makes the value quite meaningless. | |
5138 | * | |
5139 | * After it first became limited, changes in the value of the limit are | |
5140 | * of course permitted. | |
510fc4e1 | 5141 | */ |
0999821b | 5142 | mutex_lock(&memcg_create_mutex); |
07bc356e | 5143 | if (cgroup_has_tasks(memcg->css.cgroup) || memcg_has_children(memcg)) |
d6441637 VD |
5144 | err = -EBUSY; |
5145 | mutex_unlock(&memcg_create_mutex); | |
5146 | if (err) | |
5147 | goto out; | |
510fc4e1 | 5148 | |
d6441637 VD |
5149 | memcg_id = ida_simple_get(&kmem_limited_groups, |
5150 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
5151 | if (memcg_id < 0) { | |
5152 | err = memcg_id; | |
5153 | goto out; | |
5154 | } | |
5155 | ||
5156 | /* | |
5157 | * Make sure we have enough space for this cgroup in each root cache's | |
5158 | * memcg_params. | |
5159 | */ | |
5160 | err = memcg_update_all_caches(memcg_id + 1); | |
5161 | if (err) | |
5162 | goto out_rmid; | |
5163 | ||
5164 | memcg->kmemcg_id = memcg_id; | |
5165 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
5166 | mutex_init(&memcg->slab_caches_mutex); | |
5167 | ||
5168 | /* | |
5169 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
5170 | * active bit set yet, so this should succeed. | |
5171 | */ | |
5172 | err = res_counter_set_limit(&memcg->kmem, limit); | |
5173 | VM_BUG_ON(err); | |
5174 | ||
5175 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
5176 | /* | |
5177 | * Setting the active bit after enabling static branching will | |
5178 | * guarantee no one starts accounting before all call sites are | |
5179 | * patched. | |
5180 | */ | |
5181 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 5182 | out: |
d6441637 VD |
5183 | memcg_resume_kmem_account(); |
5184 | return err; | |
5185 | ||
5186 | out_rmid: | |
5187 | ida_simple_remove(&kmem_limited_groups, memcg_id); | |
5188 | goto out; | |
5189 | } | |
5190 | ||
5191 | static int memcg_activate_kmem(struct mem_cgroup *memcg, | |
5192 | unsigned long long limit) | |
5193 | { | |
5194 | int ret; | |
5195 | ||
5196 | mutex_lock(&activate_kmem_mutex); | |
5197 | ret = __memcg_activate_kmem(memcg, limit); | |
5198 | mutex_unlock(&activate_kmem_mutex); | |
5199 | return ret; | |
5200 | } | |
5201 | ||
5202 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
5203 | unsigned long long val) | |
5204 | { | |
5205 | int ret; | |
5206 | ||
5207 | if (!memcg_kmem_is_active(memcg)) | |
5208 | ret = memcg_activate_kmem(memcg, val); | |
5209 | else | |
5210 | ret = res_counter_set_limit(&memcg->kmem, val); | |
510fc4e1 GC |
5211 | return ret; |
5212 | } | |
5213 | ||
55007d84 | 5214 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 5215 | { |
55007d84 | 5216 | int ret = 0; |
510fc4e1 | 5217 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 5218 | |
d6441637 VD |
5219 | if (!parent) |
5220 | return 0; | |
55007d84 | 5221 | |
d6441637 | 5222 | mutex_lock(&activate_kmem_mutex); |
55007d84 | 5223 | /* |
d6441637 VD |
5224 | * If the parent cgroup is not kmem-active now, it cannot be activated |
5225 | * after this point, because it has at least one child already. | |
55007d84 | 5226 | */ |
d6441637 VD |
5227 | if (memcg_kmem_is_active(parent)) |
5228 | ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX); | |
5229 | mutex_unlock(&activate_kmem_mutex); | |
55007d84 | 5230 | return ret; |
510fc4e1 | 5231 | } |
d6441637 VD |
5232 | #else |
5233 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
5234 | unsigned long long val) | |
5235 | { | |
5236 | return -EINVAL; | |
5237 | } | |
6d043990 | 5238 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 5239 | |
628f4235 KH |
5240 | /* |
5241 | * The user of this function is... | |
5242 | * RES_LIMIT. | |
5243 | */ | |
182446d0 | 5244 | static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft, |
4d3bb511 | 5245 | char *buffer) |
8cdea7c0 | 5246 | { |
182446d0 | 5247 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5248 | enum res_type type; |
5249 | int name; | |
628f4235 KH |
5250 | unsigned long long val; |
5251 | int ret; | |
5252 | ||
8c7c6e34 KH |
5253 | type = MEMFILE_TYPE(cft->private); |
5254 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5255 | |
8c7c6e34 | 5256 | switch (name) { |
628f4235 | 5257 | case RES_LIMIT: |
4b3bde4c BS |
5258 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
5259 | ret = -EINVAL; | |
5260 | break; | |
5261 | } | |
628f4235 KH |
5262 | /* This function does all necessary parse...reuse it */ |
5263 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
5264 | if (ret) |
5265 | break; | |
5266 | if (type == _MEM) | |
628f4235 | 5267 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 5268 | else if (type == _MEMSWAP) |
8c7c6e34 | 5269 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 | 5270 | else if (type == _KMEM) |
d6441637 | 5271 | ret = memcg_update_kmem_limit(memcg, val); |
510fc4e1 GC |
5272 | else |
5273 | return -EINVAL; | |
628f4235 | 5274 | break; |
296c81d8 BS |
5275 | case RES_SOFT_LIMIT: |
5276 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
5277 | if (ret) | |
5278 | break; | |
5279 | /* | |
5280 | * For memsw, soft limits are hard to implement in terms | |
5281 | * of semantics, for now, we support soft limits for | |
5282 | * control without swap | |
5283 | */ | |
5284 | if (type == _MEM) | |
5285 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5286 | else | |
5287 | ret = -EINVAL; | |
5288 | break; | |
628f4235 KH |
5289 | default: |
5290 | ret = -EINVAL; /* should be BUG() ? */ | |
5291 | break; | |
5292 | } | |
5293 | return ret; | |
8cdea7c0 BS |
5294 | } |
5295 | ||
fee7b548 KH |
5296 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5297 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5298 | { | |
fee7b548 KH |
5299 | unsigned long long min_limit, min_memsw_limit, tmp; |
5300 | ||
5301 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5302 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
fee7b548 KH |
5303 | if (!memcg->use_hierarchy) |
5304 | goto out; | |
5305 | ||
63876986 TH |
5306 | while (css_parent(&memcg->css)) { |
5307 | memcg = mem_cgroup_from_css(css_parent(&memcg->css)); | |
fee7b548 KH |
5308 | if (!memcg->use_hierarchy) |
5309 | break; | |
5310 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5311 | min_limit = min(min_limit, tmp); | |
5312 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5313 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5314 | } | |
5315 | out: | |
5316 | *mem_limit = min_limit; | |
5317 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5318 | } |
5319 | ||
182446d0 | 5320 | static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event) |
c84872e1 | 5321 | { |
182446d0 | 5322 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
86ae53e1 GC |
5323 | int name; |
5324 | enum res_type type; | |
c84872e1 | 5325 | |
8c7c6e34 KH |
5326 | type = MEMFILE_TYPE(event); |
5327 | name = MEMFILE_ATTR(event); | |
af36f906 | 5328 | |
8c7c6e34 | 5329 | switch (name) { |
29f2a4da | 5330 | case RES_MAX_USAGE: |
8c7c6e34 | 5331 | if (type == _MEM) |
c0ff4b85 | 5332 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5333 | else if (type == _MEMSWAP) |
c0ff4b85 | 5334 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5335 | else if (type == _KMEM) |
5336 | res_counter_reset_max(&memcg->kmem); | |
5337 | else | |
5338 | return -EINVAL; | |
29f2a4da PE |
5339 | break; |
5340 | case RES_FAILCNT: | |
8c7c6e34 | 5341 | if (type == _MEM) |
c0ff4b85 | 5342 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5343 | else if (type == _MEMSWAP) |
c0ff4b85 | 5344 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5345 | else if (type == _KMEM) |
5346 | res_counter_reset_failcnt(&memcg->kmem); | |
5347 | else | |
5348 | return -EINVAL; | |
29f2a4da PE |
5349 | break; |
5350 | } | |
f64c3f54 | 5351 | |
85cc59db | 5352 | return 0; |
c84872e1 PE |
5353 | } |
5354 | ||
182446d0 | 5355 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5356 | struct cftype *cft) |
5357 | { | |
182446d0 | 5358 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
5359 | } |
5360 | ||
02491447 | 5361 | #ifdef CONFIG_MMU |
182446d0 | 5362 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
5363 | struct cftype *cft, u64 val) |
5364 | { | |
182446d0 | 5365 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
5366 | |
5367 | if (val >= (1 << NR_MOVE_TYPE)) | |
5368 | return -EINVAL; | |
ee5e8472 | 5369 | |
7dc74be0 | 5370 | /* |
ee5e8472 GC |
5371 | * No kind of locking is needed in here, because ->can_attach() will |
5372 | * check this value once in the beginning of the process, and then carry | |
5373 | * on with stale data. This means that changes to this value will only | |
5374 | * affect task migrations starting after the change. | |
7dc74be0 | 5375 | */ |
c0ff4b85 | 5376 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5377 | return 0; |
5378 | } | |
02491447 | 5379 | #else |
182446d0 | 5380 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
5381 | struct cftype *cft, u64 val) |
5382 | { | |
5383 | return -ENOSYS; | |
5384 | } | |
5385 | #endif | |
7dc74be0 | 5386 | |
406eb0c9 | 5387 | #ifdef CONFIG_NUMA |
2da8ca82 | 5388 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 5389 | { |
25485de6 GT |
5390 | struct numa_stat { |
5391 | const char *name; | |
5392 | unsigned int lru_mask; | |
5393 | }; | |
5394 | ||
5395 | static const struct numa_stat stats[] = { | |
5396 | { "total", LRU_ALL }, | |
5397 | { "file", LRU_ALL_FILE }, | |
5398 | { "anon", LRU_ALL_ANON }, | |
5399 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
5400 | }; | |
5401 | const struct numa_stat *stat; | |
406eb0c9 | 5402 | int nid; |
25485de6 | 5403 | unsigned long nr; |
2da8ca82 | 5404 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 5405 | |
25485de6 GT |
5406 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
5407 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
5408 | seq_printf(m, "%s=%lu", stat->name, nr); | |
5409 | for_each_node_state(nid, N_MEMORY) { | |
5410 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
5411 | stat->lru_mask); | |
5412 | seq_printf(m, " N%d=%lu", nid, nr); | |
5413 | } | |
5414 | seq_putc(m, '\n'); | |
406eb0c9 | 5415 | } |
406eb0c9 | 5416 | |
071aee13 YH |
5417 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
5418 | struct mem_cgroup *iter; | |
5419 | ||
5420 | nr = 0; | |
5421 | for_each_mem_cgroup_tree(iter, memcg) | |
5422 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
5423 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
5424 | for_each_node_state(nid, N_MEMORY) { | |
5425 | nr = 0; | |
5426 | for_each_mem_cgroup_tree(iter, memcg) | |
5427 | nr += mem_cgroup_node_nr_lru_pages( | |
5428 | iter, nid, stat->lru_mask); | |
5429 | seq_printf(m, " N%d=%lu", nid, nr); | |
5430 | } | |
5431 | seq_putc(m, '\n'); | |
406eb0c9 | 5432 | } |
406eb0c9 | 5433 | |
406eb0c9 YH |
5434 | return 0; |
5435 | } | |
5436 | #endif /* CONFIG_NUMA */ | |
5437 | ||
af7c4b0e JW |
5438 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5439 | { | |
5440 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5441 | } | |
5442 | ||
2da8ca82 | 5443 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 5444 | { |
2da8ca82 | 5445 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
af7c4b0e JW |
5446 | struct mem_cgroup *mi; |
5447 | unsigned int i; | |
406eb0c9 | 5448 | |
af7c4b0e | 5449 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5450 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5451 | continue; |
af7c4b0e JW |
5452 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5453 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5454 | } |
7b854121 | 5455 | |
af7c4b0e JW |
5456 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5457 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5458 | mem_cgroup_read_events(memcg, i)); | |
5459 | ||
5460 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5461 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5462 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5463 | ||
14067bb3 | 5464 | /* Hierarchical information */ |
fee7b548 KH |
5465 | { |
5466 | unsigned long long limit, memsw_limit; | |
d79154bb | 5467 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5468 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5469 | if (do_swap_account) |
78ccf5b5 JW |
5470 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5471 | memsw_limit); | |
fee7b548 | 5472 | } |
7f016ee8 | 5473 | |
af7c4b0e JW |
5474 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5475 | long long val = 0; | |
5476 | ||
bff6bb83 | 5477 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5478 | continue; |
af7c4b0e JW |
5479 | for_each_mem_cgroup_tree(mi, memcg) |
5480 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5481 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5482 | } | |
5483 | ||
5484 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5485 | unsigned long long val = 0; | |
5486 | ||
5487 | for_each_mem_cgroup_tree(mi, memcg) | |
5488 | val += mem_cgroup_read_events(mi, i); | |
5489 | seq_printf(m, "total_%s %llu\n", | |
5490 | mem_cgroup_events_names[i], val); | |
5491 | } | |
5492 | ||
5493 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5494 | unsigned long long val = 0; | |
5495 | ||
5496 | for_each_mem_cgroup_tree(mi, memcg) | |
5497 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5498 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5499 | } |
14067bb3 | 5500 | |
7f016ee8 | 5501 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5502 | { |
5503 | int nid, zid; | |
5504 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5505 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5506 | unsigned long recent_rotated[2] = {0, 0}; |
5507 | unsigned long recent_scanned[2] = {0, 0}; | |
5508 | ||
5509 | for_each_online_node(nid) | |
5510 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5511 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5512 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5513 | |
89abfab1 HD |
5514 | recent_rotated[0] += rstat->recent_rotated[0]; |
5515 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5516 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5517 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5518 | } |
78ccf5b5 JW |
5519 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5520 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5521 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5522 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5523 | } |
5524 | #endif | |
5525 | ||
d2ceb9b7 KH |
5526 | return 0; |
5527 | } | |
5528 | ||
182446d0 TH |
5529 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
5530 | struct cftype *cft) | |
a7885eb8 | 5531 | { |
182446d0 | 5532 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 5533 | |
1f4c025b | 5534 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5535 | } |
5536 | ||
182446d0 TH |
5537 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
5538 | struct cftype *cft, u64 val) | |
a7885eb8 | 5539 | { |
182446d0 | 5540 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5541 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
a7885eb8 | 5542 | |
63876986 | 5543 | if (val > 100 || !parent) |
a7885eb8 KM |
5544 | return -EINVAL; |
5545 | ||
0999821b | 5546 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5547 | |
a7885eb8 | 5548 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5549 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5550 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5551 | return -EINVAL; |
068b38c1 | 5552 | } |
a7885eb8 | 5553 | |
a7885eb8 | 5554 | memcg->swappiness = val; |
a7885eb8 | 5555 | |
0999821b | 5556 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5557 | |
a7885eb8 KM |
5558 | return 0; |
5559 | } | |
5560 | ||
2e72b634 KS |
5561 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5562 | { | |
5563 | struct mem_cgroup_threshold_ary *t; | |
5564 | u64 usage; | |
5565 | int i; | |
5566 | ||
5567 | rcu_read_lock(); | |
5568 | if (!swap) | |
2c488db2 | 5569 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5570 | else |
2c488db2 | 5571 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5572 | |
5573 | if (!t) | |
5574 | goto unlock; | |
5575 | ||
5576 | usage = mem_cgroup_usage(memcg, swap); | |
5577 | ||
5578 | /* | |
748dad36 | 5579 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5580 | * If it's not true, a threshold was crossed after last |
5581 | * call of __mem_cgroup_threshold(). | |
5582 | */ | |
5407a562 | 5583 | i = t->current_threshold; |
2e72b634 KS |
5584 | |
5585 | /* | |
5586 | * Iterate backward over array of thresholds starting from | |
5587 | * current_threshold and check if a threshold is crossed. | |
5588 | * If none of thresholds below usage is crossed, we read | |
5589 | * only one element of the array here. | |
5590 | */ | |
5591 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5592 | eventfd_signal(t->entries[i].eventfd, 1); | |
5593 | ||
5594 | /* i = current_threshold + 1 */ | |
5595 | i++; | |
5596 | ||
5597 | /* | |
5598 | * Iterate forward over array of thresholds starting from | |
5599 | * current_threshold+1 and check if a threshold is crossed. | |
5600 | * If none of thresholds above usage is crossed, we read | |
5601 | * only one element of the array here. | |
5602 | */ | |
5603 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5604 | eventfd_signal(t->entries[i].eventfd, 1); | |
5605 | ||
5606 | /* Update current_threshold */ | |
5407a562 | 5607 | t->current_threshold = i - 1; |
2e72b634 KS |
5608 | unlock: |
5609 | rcu_read_unlock(); | |
5610 | } | |
5611 | ||
5612 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5613 | { | |
ad4ca5f4 KS |
5614 | while (memcg) { |
5615 | __mem_cgroup_threshold(memcg, false); | |
5616 | if (do_swap_account) | |
5617 | __mem_cgroup_threshold(memcg, true); | |
5618 | ||
5619 | memcg = parent_mem_cgroup(memcg); | |
5620 | } | |
2e72b634 KS |
5621 | } |
5622 | ||
5623 | static int compare_thresholds(const void *a, const void *b) | |
5624 | { | |
5625 | const struct mem_cgroup_threshold *_a = a; | |
5626 | const struct mem_cgroup_threshold *_b = b; | |
5627 | ||
2bff24a3 GT |
5628 | if (_a->threshold > _b->threshold) |
5629 | return 1; | |
5630 | ||
5631 | if (_a->threshold < _b->threshold) | |
5632 | return -1; | |
5633 | ||
5634 | return 0; | |
2e72b634 KS |
5635 | } |
5636 | ||
c0ff4b85 | 5637 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5638 | { |
5639 | struct mem_cgroup_eventfd_list *ev; | |
5640 | ||
c0ff4b85 | 5641 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5642 | eventfd_signal(ev->eventfd, 1); |
5643 | return 0; | |
5644 | } | |
5645 | ||
c0ff4b85 | 5646 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5647 | { |
7d74b06f KH |
5648 | struct mem_cgroup *iter; |
5649 | ||
c0ff4b85 | 5650 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5651 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5652 | } |
5653 | ||
59b6f873 | 5654 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 5655 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 5656 | { |
2c488db2 KS |
5657 | struct mem_cgroup_thresholds *thresholds; |
5658 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 5659 | u64 threshold, usage; |
2c488db2 | 5660 | int i, size, ret; |
2e72b634 KS |
5661 | |
5662 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5663 | if (ret) | |
5664 | return ret; | |
5665 | ||
5666 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5667 | |
2e72b634 | 5668 | if (type == _MEM) |
2c488db2 | 5669 | thresholds = &memcg->thresholds; |
2e72b634 | 5670 | else if (type == _MEMSWAP) |
2c488db2 | 5671 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5672 | else |
5673 | BUG(); | |
5674 | ||
5675 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5676 | ||
5677 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5678 | if (thresholds->primary) |
2e72b634 KS |
5679 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5680 | ||
2c488db2 | 5681 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5682 | |
5683 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5684 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5685 | GFP_KERNEL); |
2c488db2 | 5686 | if (!new) { |
2e72b634 KS |
5687 | ret = -ENOMEM; |
5688 | goto unlock; | |
5689 | } | |
2c488db2 | 5690 | new->size = size; |
2e72b634 KS |
5691 | |
5692 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5693 | if (thresholds->primary) { |
5694 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5695 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5696 | } |
5697 | ||
2e72b634 | 5698 | /* Add new threshold */ |
2c488db2 KS |
5699 | new->entries[size - 1].eventfd = eventfd; |
5700 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5701 | |
5702 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5703 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5704 | compare_thresholds, NULL); |
5705 | ||
5706 | /* Find current threshold */ | |
2c488db2 | 5707 | new->current_threshold = -1; |
2e72b634 | 5708 | for (i = 0; i < size; i++) { |
748dad36 | 5709 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5710 | /* |
2c488db2 KS |
5711 | * new->current_threshold will not be used until |
5712 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5713 | * it here. |
5714 | */ | |
2c488db2 | 5715 | ++new->current_threshold; |
748dad36 SZ |
5716 | } else |
5717 | break; | |
2e72b634 KS |
5718 | } |
5719 | ||
2c488db2 KS |
5720 | /* Free old spare buffer and save old primary buffer as spare */ |
5721 | kfree(thresholds->spare); | |
5722 | thresholds->spare = thresholds->primary; | |
5723 | ||
5724 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5725 | |
907860ed | 5726 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5727 | synchronize_rcu(); |
5728 | ||
2e72b634 KS |
5729 | unlock: |
5730 | mutex_unlock(&memcg->thresholds_lock); | |
5731 | ||
5732 | return ret; | |
5733 | } | |
5734 | ||
59b6f873 | 5735 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5736 | struct eventfd_ctx *eventfd, const char *args) |
5737 | { | |
59b6f873 | 5738 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
5739 | } |
5740 | ||
59b6f873 | 5741 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5742 | struct eventfd_ctx *eventfd, const char *args) |
5743 | { | |
59b6f873 | 5744 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
5745 | } |
5746 | ||
59b6f873 | 5747 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 5748 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 5749 | { |
2c488db2 KS |
5750 | struct mem_cgroup_thresholds *thresholds; |
5751 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 | 5752 | u64 usage; |
2c488db2 | 5753 | int i, j, size; |
2e72b634 KS |
5754 | |
5755 | mutex_lock(&memcg->thresholds_lock); | |
5756 | if (type == _MEM) | |
2c488db2 | 5757 | thresholds = &memcg->thresholds; |
2e72b634 | 5758 | else if (type == _MEMSWAP) |
2c488db2 | 5759 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5760 | else |
5761 | BUG(); | |
5762 | ||
371528ca AV |
5763 | if (!thresholds->primary) |
5764 | goto unlock; | |
5765 | ||
2e72b634 KS |
5766 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5767 | ||
5768 | /* Check if a threshold crossed before removing */ | |
5769 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5770 | ||
5771 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5772 | size = 0; |
5773 | for (i = 0; i < thresholds->primary->size; i++) { | |
5774 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5775 | size++; |
5776 | } | |
5777 | ||
2c488db2 | 5778 | new = thresholds->spare; |
907860ed | 5779 | |
2e72b634 KS |
5780 | /* Set thresholds array to NULL if we don't have thresholds */ |
5781 | if (!size) { | |
2c488db2 KS |
5782 | kfree(new); |
5783 | new = NULL; | |
907860ed | 5784 | goto swap_buffers; |
2e72b634 KS |
5785 | } |
5786 | ||
2c488db2 | 5787 | new->size = size; |
2e72b634 KS |
5788 | |
5789 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5790 | new->current_threshold = -1; |
5791 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5792 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5793 | continue; |
5794 | ||
2c488db2 | 5795 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5796 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5797 | /* |
2c488db2 | 5798 | * new->current_threshold will not be used |
2e72b634 KS |
5799 | * until rcu_assign_pointer(), so it's safe to increment |
5800 | * it here. | |
5801 | */ | |
2c488db2 | 5802 | ++new->current_threshold; |
2e72b634 KS |
5803 | } |
5804 | j++; | |
5805 | } | |
5806 | ||
907860ed | 5807 | swap_buffers: |
2c488db2 KS |
5808 | /* Swap primary and spare array */ |
5809 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5810 | /* If all events are unregistered, free the spare array */ |
5811 | if (!new) { | |
5812 | kfree(thresholds->spare); | |
5813 | thresholds->spare = NULL; | |
5814 | } | |
5815 | ||
2c488db2 | 5816 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5817 | |
907860ed | 5818 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5819 | synchronize_rcu(); |
371528ca | 5820 | unlock: |
2e72b634 | 5821 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5822 | } |
c1e862c1 | 5823 | |
59b6f873 | 5824 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5825 | struct eventfd_ctx *eventfd) |
5826 | { | |
59b6f873 | 5827 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
5828 | } |
5829 | ||
59b6f873 | 5830 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
5831 | struct eventfd_ctx *eventfd) |
5832 | { | |
59b6f873 | 5833 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
5834 | } |
5835 | ||
59b6f873 | 5836 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 5837 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 5838 | { |
9490ff27 | 5839 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 5840 | |
9490ff27 KH |
5841 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
5842 | if (!event) | |
5843 | return -ENOMEM; | |
5844 | ||
1af8efe9 | 5845 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5846 | |
5847 | event->eventfd = eventfd; | |
5848 | list_add(&event->list, &memcg->oom_notify); | |
5849 | ||
5850 | /* already in OOM ? */ | |
79dfdacc | 5851 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5852 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5853 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5854 | |
5855 | return 0; | |
5856 | } | |
5857 | ||
59b6f873 | 5858 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 5859 | struct eventfd_ctx *eventfd) |
9490ff27 | 5860 | { |
9490ff27 | 5861 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 5862 | |
1af8efe9 | 5863 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5864 | |
c0ff4b85 | 5865 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5866 | if (ev->eventfd == eventfd) { |
5867 | list_del(&ev->list); | |
5868 | kfree(ev); | |
5869 | } | |
5870 | } | |
5871 | ||
1af8efe9 | 5872 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5873 | } |
5874 | ||
2da8ca82 | 5875 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 5876 | { |
2da8ca82 | 5877 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 5878 | |
791badbd TH |
5879 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
5880 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
5881 | return 0; |
5882 | } | |
5883 | ||
182446d0 | 5884 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
5885 | struct cftype *cft, u64 val) |
5886 | { | |
182446d0 | 5887 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
63876986 | 5888 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(&memcg->css)); |
3c11ecf4 KH |
5889 | |
5890 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
63876986 | 5891 | if (!parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
5892 | return -EINVAL; |
5893 | ||
0999821b | 5894 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5895 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5896 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5897 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5898 | return -EINVAL; |
5899 | } | |
c0ff4b85 | 5900 | memcg->oom_kill_disable = val; |
4d845ebf | 5901 | if (!val) |
c0ff4b85 | 5902 | memcg_oom_recover(memcg); |
0999821b | 5903 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5904 | return 0; |
5905 | } | |
5906 | ||
c255a458 | 5907 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5908 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5909 | { |
55007d84 GC |
5910 | int ret; |
5911 | ||
2633d7a0 | 5912 | memcg->kmemcg_id = -1; |
55007d84 GC |
5913 | ret = memcg_propagate_kmem(memcg); |
5914 | if (ret) | |
5915 | return ret; | |
2633d7a0 | 5916 | |
1d62e436 | 5917 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5918 | } |
e5671dfa | 5919 | |
10d5ebf4 | 5920 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 5921 | { |
1d62e436 | 5922 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
5923 | } |
5924 | ||
5925 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
5926 | { | |
5927 | if (!memcg_kmem_is_active(memcg)) | |
5928 | return; | |
5929 | ||
5930 | /* | |
5931 | * kmem charges can outlive the cgroup. In the case of slab | |
5932 | * pages, for instance, a page contain objects from various | |
5933 | * processes. As we prevent from taking a reference for every | |
5934 | * such allocation we have to be careful when doing uncharge | |
5935 | * (see memcg_uncharge_kmem) and here during offlining. | |
5936 | * | |
5937 | * The idea is that that only the _last_ uncharge which sees | |
5938 | * the dead memcg will drop the last reference. An additional | |
5939 | * reference is taken here before the group is marked dead | |
5940 | * which is then paired with css_put during uncharge resp. here. | |
5941 | * | |
5942 | * Although this might sound strange as this path is called from | |
5943 | * css_offline() when the referencemight have dropped down to 0 | |
5944 | * and shouldn't be incremented anymore (css_tryget would fail) | |
5945 | * we do not have other options because of the kmem allocations | |
5946 | * lifetime. | |
5947 | */ | |
5948 | css_get(&memcg->css); | |
7de37682 GC |
5949 | |
5950 | memcg_kmem_mark_dead(memcg); | |
5951 | ||
5952 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5953 | return; | |
5954 | ||
7de37682 | 5955 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 5956 | css_put(&memcg->css); |
d1a4c0b3 | 5957 | } |
e5671dfa | 5958 | #else |
cbe128e3 | 5959 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5960 | { |
5961 | return 0; | |
5962 | } | |
d1a4c0b3 | 5963 | |
10d5ebf4 LZ |
5964 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
5965 | { | |
5966 | } | |
5967 | ||
5968 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
5969 | { |
5970 | } | |
e5671dfa GC |
5971 | #endif |
5972 | ||
3bc942f3 TH |
5973 | /* |
5974 | * DO NOT USE IN NEW FILES. | |
5975 | * | |
5976 | * "cgroup.event_control" implementation. | |
5977 | * | |
5978 | * This is way over-engineered. It tries to support fully configurable | |
5979 | * events for each user. Such level of flexibility is completely | |
5980 | * unnecessary especially in the light of the planned unified hierarchy. | |
5981 | * | |
5982 | * Please deprecate this and replace with something simpler if at all | |
5983 | * possible. | |
5984 | */ | |
5985 | ||
79bd9814 TH |
5986 | /* |
5987 | * Unregister event and free resources. | |
5988 | * | |
5989 | * Gets called from workqueue. | |
5990 | */ | |
3bc942f3 | 5991 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 5992 | { |
3bc942f3 TH |
5993 | struct mem_cgroup_event *event = |
5994 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 5995 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
5996 | |
5997 | remove_wait_queue(event->wqh, &event->wait); | |
5998 | ||
59b6f873 | 5999 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
6000 | |
6001 | /* Notify userspace the event is going away. */ | |
6002 | eventfd_signal(event->eventfd, 1); | |
6003 | ||
6004 | eventfd_ctx_put(event->eventfd); | |
6005 | kfree(event); | |
59b6f873 | 6006 | css_put(&memcg->css); |
79bd9814 TH |
6007 | } |
6008 | ||
6009 | /* | |
6010 | * Gets called on POLLHUP on eventfd when user closes it. | |
6011 | * | |
6012 | * Called with wqh->lock held and interrupts disabled. | |
6013 | */ | |
3bc942f3 TH |
6014 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
6015 | int sync, void *key) | |
79bd9814 | 6016 | { |
3bc942f3 TH |
6017 | struct mem_cgroup_event *event = |
6018 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 6019 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
6020 | unsigned long flags = (unsigned long)key; |
6021 | ||
6022 | if (flags & POLLHUP) { | |
6023 | /* | |
6024 | * If the event has been detached at cgroup removal, we | |
6025 | * can simply return knowing the other side will cleanup | |
6026 | * for us. | |
6027 | * | |
6028 | * We can't race against event freeing since the other | |
6029 | * side will require wqh->lock via remove_wait_queue(), | |
6030 | * which we hold. | |
6031 | */ | |
fba94807 | 6032 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
6033 | if (!list_empty(&event->list)) { |
6034 | list_del_init(&event->list); | |
6035 | /* | |
6036 | * We are in atomic context, but cgroup_event_remove() | |
6037 | * may sleep, so we have to call it in workqueue. | |
6038 | */ | |
6039 | schedule_work(&event->remove); | |
6040 | } | |
fba94807 | 6041 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
6042 | } |
6043 | ||
6044 | return 0; | |
6045 | } | |
6046 | ||
3bc942f3 | 6047 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
6048 | wait_queue_head_t *wqh, poll_table *pt) |
6049 | { | |
3bc942f3 TH |
6050 | struct mem_cgroup_event *event = |
6051 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
6052 | |
6053 | event->wqh = wqh; | |
6054 | add_wait_queue(wqh, &event->wait); | |
6055 | } | |
6056 | ||
6057 | /* | |
3bc942f3 TH |
6058 | * DO NOT USE IN NEW FILES. |
6059 | * | |
79bd9814 TH |
6060 | * Parse input and register new cgroup event handler. |
6061 | * | |
6062 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
6063 | * Interpretation of args is defined by control file implementation. | |
6064 | */ | |
3bc942f3 | 6065 | static int memcg_write_event_control(struct cgroup_subsys_state *css, |
4d3bb511 | 6066 | struct cftype *cft, char *buffer) |
79bd9814 | 6067 | { |
fba94807 | 6068 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 6069 | struct mem_cgroup_event *event; |
79bd9814 TH |
6070 | struct cgroup_subsys_state *cfile_css; |
6071 | unsigned int efd, cfd; | |
6072 | struct fd efile; | |
6073 | struct fd cfile; | |
fba94807 | 6074 | const char *name; |
79bd9814 TH |
6075 | char *endp; |
6076 | int ret; | |
6077 | ||
6078 | efd = simple_strtoul(buffer, &endp, 10); | |
6079 | if (*endp != ' ') | |
6080 | return -EINVAL; | |
6081 | buffer = endp + 1; | |
6082 | ||
6083 | cfd = simple_strtoul(buffer, &endp, 10); | |
6084 | if ((*endp != ' ') && (*endp != '\0')) | |
6085 | return -EINVAL; | |
6086 | buffer = endp + 1; | |
6087 | ||
6088 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
6089 | if (!event) | |
6090 | return -ENOMEM; | |
6091 | ||
59b6f873 | 6092 | event->memcg = memcg; |
79bd9814 | 6093 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
6094 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
6095 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
6096 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
6097 | |
6098 | efile = fdget(efd); | |
6099 | if (!efile.file) { | |
6100 | ret = -EBADF; | |
6101 | goto out_kfree; | |
6102 | } | |
6103 | ||
6104 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
6105 | if (IS_ERR(event->eventfd)) { | |
6106 | ret = PTR_ERR(event->eventfd); | |
6107 | goto out_put_efile; | |
6108 | } | |
6109 | ||
6110 | cfile = fdget(cfd); | |
6111 | if (!cfile.file) { | |
6112 | ret = -EBADF; | |
6113 | goto out_put_eventfd; | |
6114 | } | |
6115 | ||
6116 | /* the process need read permission on control file */ | |
6117 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
6118 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
6119 | if (ret < 0) | |
6120 | goto out_put_cfile; | |
6121 | ||
fba94807 TH |
6122 | /* |
6123 | * Determine the event callbacks and set them in @event. This used | |
6124 | * to be done via struct cftype but cgroup core no longer knows | |
6125 | * about these events. The following is crude but the whole thing | |
6126 | * is for compatibility anyway. | |
3bc942f3 TH |
6127 | * |
6128 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
6129 | */ |
6130 | name = cfile.file->f_dentry->d_name.name; | |
6131 | ||
6132 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
6133 | event->register_event = mem_cgroup_usage_register_event; | |
6134 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
6135 | } else if (!strcmp(name, "memory.oom_control")) { | |
6136 | event->register_event = mem_cgroup_oom_register_event; | |
6137 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
6138 | } else if (!strcmp(name, "memory.pressure_level")) { | |
6139 | event->register_event = vmpressure_register_event; | |
6140 | event->unregister_event = vmpressure_unregister_event; | |
6141 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
6142 | event->register_event = memsw_cgroup_usage_register_event; |
6143 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
6144 | } else { |
6145 | ret = -EINVAL; | |
6146 | goto out_put_cfile; | |
6147 | } | |
6148 | ||
79bd9814 | 6149 | /* |
b5557c4c TH |
6150 | * Verify @cfile should belong to @css. Also, remaining events are |
6151 | * automatically removed on cgroup destruction but the removal is | |
6152 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 6153 | */ |
5a17f543 TH |
6154 | cfile_css = css_tryget_from_dir(cfile.file->f_dentry->d_parent, |
6155 | &memory_cgrp_subsys); | |
79bd9814 | 6156 | ret = -EINVAL; |
5a17f543 | 6157 | if (IS_ERR(cfile_css)) |
79bd9814 | 6158 | goto out_put_cfile; |
5a17f543 TH |
6159 | if (cfile_css != css) { |
6160 | css_put(cfile_css); | |
79bd9814 | 6161 | goto out_put_cfile; |
5a17f543 | 6162 | } |
79bd9814 | 6163 | |
59b6f873 | 6164 | ret = event->register_event(memcg, event->eventfd, buffer); |
79bd9814 TH |
6165 | if (ret) |
6166 | goto out_put_css; | |
6167 | ||
6168 | efile.file->f_op->poll(efile.file, &event->pt); | |
6169 | ||
fba94807 TH |
6170 | spin_lock(&memcg->event_list_lock); |
6171 | list_add(&event->list, &memcg->event_list); | |
6172 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
6173 | |
6174 | fdput(cfile); | |
6175 | fdput(efile); | |
6176 | ||
6177 | return 0; | |
6178 | ||
6179 | out_put_css: | |
b5557c4c | 6180 | css_put(css); |
79bd9814 TH |
6181 | out_put_cfile: |
6182 | fdput(cfile); | |
6183 | out_put_eventfd: | |
6184 | eventfd_ctx_put(event->eventfd); | |
6185 | out_put_efile: | |
6186 | fdput(efile); | |
6187 | out_kfree: | |
6188 | kfree(event); | |
6189 | ||
6190 | return ret; | |
6191 | } | |
6192 | ||
8cdea7c0 BS |
6193 | static struct cftype mem_cgroup_files[] = { |
6194 | { | |
0eea1030 | 6195 | .name = "usage_in_bytes", |
8c7c6e34 | 6196 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 6197 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 6198 | }, |
c84872e1 PE |
6199 | { |
6200 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 6201 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 6202 | .trigger = mem_cgroup_reset, |
791badbd | 6203 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 6204 | }, |
8cdea7c0 | 6205 | { |
0eea1030 | 6206 | .name = "limit_in_bytes", |
8c7c6e34 | 6207 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 6208 | .write_string = mem_cgroup_write, |
791badbd | 6209 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 6210 | }, |
296c81d8 BS |
6211 | { |
6212 | .name = "soft_limit_in_bytes", | |
6213 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
6214 | .write_string = mem_cgroup_write, | |
791badbd | 6215 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 6216 | }, |
8cdea7c0 BS |
6217 | { |
6218 | .name = "failcnt", | |
8c7c6e34 | 6219 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 6220 | .trigger = mem_cgroup_reset, |
791badbd | 6221 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 6222 | }, |
d2ceb9b7 KH |
6223 | { |
6224 | .name = "stat", | |
2da8ca82 | 6225 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 6226 | }, |
c1e862c1 KH |
6227 | { |
6228 | .name = "force_empty", | |
6229 | .trigger = mem_cgroup_force_empty_write, | |
6230 | }, | |
18f59ea7 BS |
6231 | { |
6232 | .name = "use_hierarchy", | |
f00baae7 | 6233 | .flags = CFTYPE_INSANE, |
18f59ea7 BS |
6234 | .write_u64 = mem_cgroup_hierarchy_write, |
6235 | .read_u64 = mem_cgroup_hierarchy_read, | |
6236 | }, | |
79bd9814 | 6237 | { |
3bc942f3 TH |
6238 | .name = "cgroup.event_control", /* XXX: for compat */ |
6239 | .write_string = memcg_write_event_control, | |
79bd9814 TH |
6240 | .flags = CFTYPE_NO_PREFIX, |
6241 | .mode = S_IWUGO, | |
6242 | }, | |
a7885eb8 KM |
6243 | { |
6244 | .name = "swappiness", | |
6245 | .read_u64 = mem_cgroup_swappiness_read, | |
6246 | .write_u64 = mem_cgroup_swappiness_write, | |
6247 | }, | |
7dc74be0 DN |
6248 | { |
6249 | .name = "move_charge_at_immigrate", | |
6250 | .read_u64 = mem_cgroup_move_charge_read, | |
6251 | .write_u64 = mem_cgroup_move_charge_write, | |
6252 | }, | |
9490ff27 KH |
6253 | { |
6254 | .name = "oom_control", | |
2da8ca82 | 6255 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 6256 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
6257 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
6258 | }, | |
70ddf637 AV |
6259 | { |
6260 | .name = "pressure_level", | |
70ddf637 | 6261 | }, |
406eb0c9 YH |
6262 | #ifdef CONFIG_NUMA |
6263 | { | |
6264 | .name = "numa_stat", | |
2da8ca82 | 6265 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
6266 | }, |
6267 | #endif | |
510fc4e1 GC |
6268 | #ifdef CONFIG_MEMCG_KMEM |
6269 | { | |
6270 | .name = "kmem.limit_in_bytes", | |
6271 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
6272 | .write_string = mem_cgroup_write, | |
791badbd | 6273 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
6274 | }, |
6275 | { | |
6276 | .name = "kmem.usage_in_bytes", | |
6277 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 6278 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
6279 | }, |
6280 | { | |
6281 | .name = "kmem.failcnt", | |
6282 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6283 | .trigger = mem_cgroup_reset, | |
791badbd | 6284 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
6285 | }, |
6286 | { | |
6287 | .name = "kmem.max_usage_in_bytes", | |
6288 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6289 | .trigger = mem_cgroup_reset, | |
791badbd | 6290 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 6291 | }, |
749c5415 GC |
6292 | #ifdef CONFIG_SLABINFO |
6293 | { | |
6294 | .name = "kmem.slabinfo", | |
2da8ca82 | 6295 | .seq_show = mem_cgroup_slabinfo_read, |
749c5415 GC |
6296 | }, |
6297 | #endif | |
8c7c6e34 | 6298 | #endif |
6bc10349 | 6299 | { }, /* terminate */ |
af36f906 | 6300 | }; |
8c7c6e34 | 6301 | |
2d11085e MH |
6302 | #ifdef CONFIG_MEMCG_SWAP |
6303 | static struct cftype memsw_cgroup_files[] = { | |
6304 | { | |
6305 | .name = "memsw.usage_in_bytes", | |
6306 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 6307 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
6308 | }, |
6309 | { | |
6310 | .name = "memsw.max_usage_in_bytes", | |
6311 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6312 | .trigger = mem_cgroup_reset, | |
791badbd | 6313 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
6314 | }, |
6315 | { | |
6316 | .name = "memsw.limit_in_bytes", | |
6317 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6318 | .write_string = mem_cgroup_write, | |
791badbd | 6319 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
6320 | }, |
6321 | { | |
6322 | .name = "memsw.failcnt", | |
6323 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6324 | .trigger = mem_cgroup_reset, | |
791badbd | 6325 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
6326 | }, |
6327 | { }, /* terminate */ | |
6328 | }; | |
6329 | #endif | |
c0ff4b85 | 6330 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
6331 | { |
6332 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 6333 | struct mem_cgroup_per_zone *mz; |
41e3355d | 6334 | int zone, tmp = node; |
1ecaab2b KH |
6335 | /* |
6336 | * This routine is called against possible nodes. | |
6337 | * But it's BUG to call kmalloc() against offline node. | |
6338 | * | |
6339 | * TODO: this routine can waste much memory for nodes which will | |
6340 | * never be onlined. It's better to use memory hotplug callback | |
6341 | * function. | |
6342 | */ | |
41e3355d KH |
6343 | if (!node_state(node, N_NORMAL_MEMORY)) |
6344 | tmp = -1; | |
17295c88 | 6345 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
6346 | if (!pn) |
6347 | return 1; | |
1ecaab2b | 6348 | |
1ecaab2b KH |
6349 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
6350 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 6351 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
6352 | mz->usage_in_excess = 0; |
6353 | mz->on_tree = false; | |
d79154bb | 6354 | mz->memcg = memcg; |
1ecaab2b | 6355 | } |
54f72fe0 | 6356 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
6357 | return 0; |
6358 | } | |
6359 | ||
c0ff4b85 | 6360 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 6361 | { |
54f72fe0 | 6362 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
6363 | } |
6364 | ||
33327948 KH |
6365 | static struct mem_cgroup *mem_cgroup_alloc(void) |
6366 | { | |
d79154bb | 6367 | struct mem_cgroup *memcg; |
8ff69e2c | 6368 | size_t size; |
33327948 | 6369 | |
8ff69e2c VD |
6370 | size = sizeof(struct mem_cgroup); |
6371 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 6372 | |
8ff69e2c | 6373 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 6374 | if (!memcg) |
e7bbcdf3 DC |
6375 | return NULL; |
6376 | ||
d79154bb HD |
6377 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
6378 | if (!memcg->stat) | |
d2e61b8d | 6379 | goto out_free; |
d79154bb HD |
6380 | spin_lock_init(&memcg->pcp_counter_lock); |
6381 | return memcg; | |
d2e61b8d DC |
6382 | |
6383 | out_free: | |
8ff69e2c | 6384 | kfree(memcg); |
d2e61b8d | 6385 | return NULL; |
33327948 KH |
6386 | } |
6387 | ||
59927fb9 | 6388 | /* |
c8b2a36f GC |
6389 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
6390 | * (scanning all at force_empty is too costly...) | |
6391 | * | |
6392 | * Instead of clearing all references at force_empty, we remember | |
6393 | * the number of reference from swap_cgroup and free mem_cgroup when | |
6394 | * it goes down to 0. | |
6395 | * | |
6396 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 6397 | */ |
c8b2a36f GC |
6398 | |
6399 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 6400 | { |
c8b2a36f | 6401 | int node; |
59927fb9 | 6402 | |
bb4cc1a8 | 6403 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
6404 | |
6405 | for_each_node(node) | |
6406 | free_mem_cgroup_per_zone_info(memcg, node); | |
6407 | ||
6408 | free_percpu(memcg->stat); | |
6409 | ||
3f134619 GC |
6410 | /* |
6411 | * We need to make sure that (at least for now), the jump label | |
6412 | * destruction code runs outside of the cgroup lock. This is because | |
6413 | * get_online_cpus(), which is called from the static_branch update, | |
6414 | * can't be called inside the cgroup_lock. cpusets are the ones | |
6415 | * enforcing this dependency, so if they ever change, we might as well. | |
6416 | * | |
6417 | * schedule_work() will guarantee this happens. Be careful if you need | |
6418 | * to move this code around, and make sure it is outside | |
6419 | * the cgroup_lock. | |
6420 | */ | |
a8964b9b | 6421 | disarm_static_keys(memcg); |
8ff69e2c | 6422 | kfree(memcg); |
59927fb9 | 6423 | } |
3afe36b1 | 6424 | |
7bcc1bb1 DN |
6425 | /* |
6426 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6427 | */ | |
e1aab161 | 6428 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6429 | { |
c0ff4b85 | 6430 | if (!memcg->res.parent) |
7bcc1bb1 | 6431 | return NULL; |
c0ff4b85 | 6432 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6433 | } |
e1aab161 | 6434 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6435 | |
bb4cc1a8 AM |
6436 | static void __init mem_cgroup_soft_limit_tree_init(void) |
6437 | { | |
6438 | struct mem_cgroup_tree_per_node *rtpn; | |
6439 | struct mem_cgroup_tree_per_zone *rtpz; | |
6440 | int tmp, node, zone; | |
6441 | ||
6442 | for_each_node(node) { | |
6443 | tmp = node; | |
6444 | if (!node_state(node, N_NORMAL_MEMORY)) | |
6445 | tmp = -1; | |
6446 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
6447 | BUG_ON(!rtpn); | |
6448 | ||
6449 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
6450 | ||
6451 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
6452 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
6453 | rtpz->rb_root = RB_ROOT; | |
6454 | spin_lock_init(&rtpz->lock); | |
6455 | } | |
6456 | } | |
6457 | } | |
6458 | ||
0eb253e2 | 6459 | static struct cgroup_subsys_state * __ref |
eb95419b | 6460 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 6461 | { |
d142e3e6 | 6462 | struct mem_cgroup *memcg; |
04046e1a | 6463 | long error = -ENOMEM; |
6d12e2d8 | 6464 | int node; |
8cdea7c0 | 6465 | |
c0ff4b85 R |
6466 | memcg = mem_cgroup_alloc(); |
6467 | if (!memcg) | |
04046e1a | 6468 | return ERR_PTR(error); |
78fb7466 | 6469 | |
3ed28fa1 | 6470 | for_each_node(node) |
c0ff4b85 | 6471 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6472 | goto free_out; |
f64c3f54 | 6473 | |
c077719b | 6474 | /* root ? */ |
eb95419b | 6475 | if (parent_css == NULL) { |
a41c58a6 | 6476 | root_mem_cgroup = memcg; |
d142e3e6 GC |
6477 | res_counter_init(&memcg->res, NULL); |
6478 | res_counter_init(&memcg->memsw, NULL); | |
6479 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6480 | } |
28dbc4b6 | 6481 | |
d142e3e6 GC |
6482 | memcg->last_scanned_node = MAX_NUMNODES; |
6483 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
6484 | memcg->move_charge_at_immigrate = 0; |
6485 | mutex_init(&memcg->thresholds_lock); | |
6486 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 6487 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
6488 | INIT_LIST_HEAD(&memcg->event_list); |
6489 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
6490 | |
6491 | return &memcg->css; | |
6492 | ||
6493 | free_out: | |
6494 | __mem_cgroup_free(memcg); | |
6495 | return ERR_PTR(error); | |
6496 | } | |
6497 | ||
6498 | static int | |
eb95419b | 6499 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 6500 | { |
eb95419b TH |
6501 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6502 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css)); | |
d142e3e6 | 6503 | |
4219b2da LZ |
6504 | if (css->cgroup->id > MEM_CGROUP_ID_MAX) |
6505 | return -ENOSPC; | |
6506 | ||
63876986 | 6507 | if (!parent) |
d142e3e6 GC |
6508 | return 0; |
6509 | ||
0999821b | 6510 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6511 | |
6512 | memcg->use_hierarchy = parent->use_hierarchy; | |
6513 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6514 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6515 | ||
6516 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6517 | res_counter_init(&memcg->res, &parent->res); |
6518 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6519 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6520 | |
7bcc1bb1 | 6521 | /* |
8d76a979 LZ |
6522 | * No need to take a reference to the parent because cgroup |
6523 | * core guarantees its existence. | |
7bcc1bb1 | 6524 | */ |
18f59ea7 | 6525 | } else { |
c0ff4b85 R |
6526 | res_counter_init(&memcg->res, NULL); |
6527 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 6528 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
6529 | /* |
6530 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6531 | * much sense so let cgroup subsystem know about this | |
6532 | * unfortunate state in our controller. | |
6533 | */ | |
d142e3e6 | 6534 | if (parent != root_mem_cgroup) |
073219e9 | 6535 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 6536 | } |
0999821b | 6537 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 6538 | |
073219e9 | 6539 | return memcg_init_kmem(memcg, &memory_cgrp_subsys); |
8cdea7c0 BS |
6540 | } |
6541 | ||
5f578161 MH |
6542 | /* |
6543 | * Announce all parents that a group from their hierarchy is gone. | |
6544 | */ | |
6545 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6546 | { | |
6547 | struct mem_cgroup *parent = memcg; | |
6548 | ||
6549 | while ((parent = parent_mem_cgroup(parent))) | |
519ebea3 | 6550 | mem_cgroup_iter_invalidate(parent); |
5f578161 MH |
6551 | |
6552 | /* | |
6553 | * if the root memcg is not hierarchical we have to check it | |
6554 | * explicitely. | |
6555 | */ | |
6556 | if (!root_mem_cgroup->use_hierarchy) | |
519ebea3 | 6557 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
5f578161 MH |
6558 | } |
6559 | ||
eb95419b | 6560 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 6561 | { |
eb95419b | 6562 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 6563 | struct mem_cgroup_event *event, *tmp; |
4fb1a86f | 6564 | struct cgroup_subsys_state *iter; |
79bd9814 TH |
6565 | |
6566 | /* | |
6567 | * Unregister events and notify userspace. | |
6568 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
6569 | * directory to avoid race between userspace and kernelspace. | |
6570 | */ | |
fba94807 TH |
6571 | spin_lock(&memcg->event_list_lock); |
6572 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
6573 | list_del_init(&event->list); |
6574 | schedule_work(&event->remove); | |
6575 | } | |
fba94807 | 6576 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 6577 | |
10d5ebf4 LZ |
6578 | kmem_cgroup_css_offline(memcg); |
6579 | ||
5f578161 | 6580 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
4fb1a86f FB |
6581 | |
6582 | /* | |
6583 | * This requires that offlining is serialized. Right now that is | |
6584 | * guaranteed because css_killed_work_fn() holds the cgroup_mutex. | |
6585 | */ | |
6586 | css_for_each_descendant_post(iter, css) | |
6587 | mem_cgroup_reparent_charges(mem_cgroup_from_css(iter)); | |
6588 | ||
1f458cbf | 6589 | mem_cgroup_destroy_all_caches(memcg); |
33cb876e | 6590 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
6591 | } |
6592 | ||
eb95419b | 6593 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 6594 | { |
eb95419b | 6595 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
96f1c58d JW |
6596 | /* |
6597 | * XXX: css_offline() would be where we should reparent all | |
6598 | * memory to prepare the cgroup for destruction. However, | |
6599 | * memcg does not do css_tryget() and res_counter charging | |
6600 | * under the same RCU lock region, which means that charging | |
6601 | * could race with offlining. Offlining only happens to | |
6602 | * cgroups with no tasks in them but charges can show up | |
6603 | * without any tasks from the swapin path when the target | |
6604 | * memcg is looked up from the swapout record and not from the | |
6605 | * current task as it usually is. A race like this can leak | |
6606 | * charges and put pages with stale cgroup pointers into | |
6607 | * circulation: | |
6608 | * | |
6609 | * #0 #1 | |
6610 | * lookup_swap_cgroup_id() | |
6611 | * rcu_read_lock() | |
6612 | * mem_cgroup_lookup() | |
6613 | * css_tryget() | |
6614 | * rcu_read_unlock() | |
6615 | * disable css_tryget() | |
6616 | * call_rcu() | |
6617 | * offline_css() | |
6618 | * reparent_charges() | |
6619 | * res_counter_charge() | |
6620 | * css_put() | |
6621 | * css_free() | |
6622 | * pc->mem_cgroup = dead memcg | |
6623 | * add page to lru | |
6624 | * | |
6625 | * The bulk of the charges are still moved in offline_css() to | |
6626 | * avoid pinning a lot of pages in case a long-term reference | |
6627 | * like a swapout record is deferring the css_free() to long | |
6628 | * after offlining. But this makes sure we catch any charges | |
6629 | * made after offlining: | |
6630 | */ | |
6631 | mem_cgroup_reparent_charges(memcg); | |
c268e994 | 6632 | |
10d5ebf4 | 6633 | memcg_destroy_kmem(memcg); |
465939a1 | 6634 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
6635 | } |
6636 | ||
02491447 | 6637 | #ifdef CONFIG_MMU |
7dc74be0 | 6638 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6639 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6640 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6641 | { |
854ffa8d DN |
6642 | int ret = 0; |
6643 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6644 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6645 | |
c0ff4b85 | 6646 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6647 | mc.precharge += count; |
6648 | /* we don't need css_get for root */ | |
6649 | return ret; | |
6650 | } | |
6651 | /* try to charge at once */ | |
6652 | if (count > 1) { | |
6653 | struct res_counter *dummy; | |
6654 | /* | |
c0ff4b85 | 6655 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6656 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6657 | * are still under the same cgroup_mutex. So we can postpone | |
6658 | * css_get(). | |
6659 | */ | |
c0ff4b85 | 6660 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6661 | goto one_by_one; |
c0ff4b85 | 6662 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6663 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6664 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6665 | goto one_by_one; |
6666 | } | |
6667 | mc.precharge += count; | |
854ffa8d DN |
6668 | return ret; |
6669 | } | |
6670 | one_by_one: | |
6671 | /* fall back to one by one charge */ | |
6672 | while (count--) { | |
6673 | if (signal_pending(current)) { | |
6674 | ret = -EINTR; | |
6675 | break; | |
6676 | } | |
6677 | if (!batch_count--) { | |
6678 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6679 | cond_resched(); | |
6680 | } | |
c0ff4b85 R |
6681 | ret = __mem_cgroup_try_charge(NULL, |
6682 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6683 | if (ret) |
854ffa8d | 6684 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6685 | return ret; |
854ffa8d DN |
6686 | mc.precharge++; |
6687 | } | |
4ffef5fe DN |
6688 | return ret; |
6689 | } | |
6690 | ||
6691 | /** | |
8d32ff84 | 6692 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6693 | * @vma: the vma the pte to be checked belongs |
6694 | * @addr: the address corresponding to the pte to be checked | |
6695 | * @ptent: the pte to be checked | |
02491447 | 6696 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6697 | * |
6698 | * Returns | |
6699 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6700 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6701 | * move charge. if @target is not NULL, the page is stored in target->page | |
6702 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6703 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6704 | * target for charge migration. if @target is not NULL, the entry is stored | |
6705 | * in target->ent. | |
4ffef5fe DN |
6706 | * |
6707 | * Called with pte lock held. | |
6708 | */ | |
4ffef5fe DN |
6709 | union mc_target { |
6710 | struct page *page; | |
02491447 | 6711 | swp_entry_t ent; |
4ffef5fe DN |
6712 | }; |
6713 | ||
4ffef5fe | 6714 | enum mc_target_type { |
8d32ff84 | 6715 | MC_TARGET_NONE = 0, |
4ffef5fe | 6716 | MC_TARGET_PAGE, |
02491447 | 6717 | MC_TARGET_SWAP, |
4ffef5fe DN |
6718 | }; |
6719 | ||
90254a65 DN |
6720 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6721 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6722 | { |
90254a65 | 6723 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6724 | |
90254a65 DN |
6725 | if (!page || !page_mapped(page)) |
6726 | return NULL; | |
6727 | if (PageAnon(page)) { | |
6728 | /* we don't move shared anon */ | |
4b91355e | 6729 | if (!move_anon()) |
90254a65 | 6730 | return NULL; |
87946a72 DN |
6731 | } else if (!move_file()) |
6732 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6733 | return NULL; |
6734 | if (!get_page_unless_zero(page)) | |
6735 | return NULL; | |
6736 | ||
6737 | return page; | |
6738 | } | |
6739 | ||
4b91355e | 6740 | #ifdef CONFIG_SWAP |
90254a65 DN |
6741 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6742 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6743 | { | |
90254a65 DN |
6744 | struct page *page = NULL; |
6745 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6746 | ||
6747 | if (!move_anon() || non_swap_entry(ent)) | |
6748 | return NULL; | |
4b91355e KH |
6749 | /* |
6750 | * Because lookup_swap_cache() updates some statistics counter, | |
6751 | * we call find_get_page() with swapper_space directly. | |
6752 | */ | |
33806f06 | 6753 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6754 | if (do_swap_account) |
6755 | entry->val = ent.val; | |
6756 | ||
6757 | return page; | |
6758 | } | |
4b91355e KH |
6759 | #else |
6760 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6761 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6762 | { | |
6763 | return NULL; | |
6764 | } | |
6765 | #endif | |
90254a65 | 6766 | |
87946a72 DN |
6767 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6768 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6769 | { | |
6770 | struct page *page = NULL; | |
87946a72 DN |
6771 | struct address_space *mapping; |
6772 | pgoff_t pgoff; | |
6773 | ||
6774 | if (!vma->vm_file) /* anonymous vma */ | |
6775 | return NULL; | |
6776 | if (!move_file()) | |
6777 | return NULL; | |
6778 | ||
87946a72 DN |
6779 | mapping = vma->vm_file->f_mapping; |
6780 | if (pte_none(ptent)) | |
6781 | pgoff = linear_page_index(vma, addr); | |
6782 | else /* pte_file(ptent) is true */ | |
6783 | pgoff = pte_to_pgoff(ptent); | |
6784 | ||
6785 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6786 | page = find_get_page(mapping, pgoff); |
6787 | ||
6788 | #ifdef CONFIG_SWAP | |
6789 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6790 | if (radix_tree_exceptional_entry(page)) { | |
6791 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6792 | if (do_swap_account) |
aa3b1895 | 6793 | *entry = swap; |
33806f06 | 6794 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6795 | } |
aa3b1895 | 6796 | #endif |
87946a72 DN |
6797 | return page; |
6798 | } | |
6799 | ||
8d32ff84 | 6800 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6801 | unsigned long addr, pte_t ptent, union mc_target *target) |
6802 | { | |
6803 | struct page *page = NULL; | |
6804 | struct page_cgroup *pc; | |
8d32ff84 | 6805 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6806 | swp_entry_t ent = { .val = 0 }; |
6807 | ||
6808 | if (pte_present(ptent)) | |
6809 | page = mc_handle_present_pte(vma, addr, ptent); | |
6810 | else if (is_swap_pte(ptent)) | |
6811 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6812 | else if (pte_none(ptent) || pte_file(ptent)) |
6813 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6814 | |
6815 | if (!page && !ent.val) | |
8d32ff84 | 6816 | return ret; |
02491447 DN |
6817 | if (page) { |
6818 | pc = lookup_page_cgroup(page); | |
6819 | /* | |
6820 | * Do only loose check w/o page_cgroup lock. | |
6821 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6822 | * the lock. | |
6823 | */ | |
6824 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6825 | ret = MC_TARGET_PAGE; | |
6826 | if (target) | |
6827 | target->page = page; | |
6828 | } | |
6829 | if (!ret || !target) | |
6830 | put_page(page); | |
6831 | } | |
90254a65 DN |
6832 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6833 | if (ent.val && !ret && | |
34c00c31 | 6834 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6835 | ret = MC_TARGET_SWAP; |
6836 | if (target) | |
6837 | target->ent = ent; | |
4ffef5fe | 6838 | } |
4ffef5fe DN |
6839 | return ret; |
6840 | } | |
6841 | ||
12724850 NH |
6842 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6843 | /* | |
6844 | * We don't consider swapping or file mapped pages because THP does not | |
6845 | * support them for now. | |
6846 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6847 | */ | |
6848 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6849 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6850 | { | |
6851 | struct page *page = NULL; | |
6852 | struct page_cgroup *pc; | |
6853 | enum mc_target_type ret = MC_TARGET_NONE; | |
6854 | ||
6855 | page = pmd_page(pmd); | |
309381fe | 6856 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
6857 | if (!move_anon()) |
6858 | return ret; | |
6859 | pc = lookup_page_cgroup(page); | |
6860 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6861 | ret = MC_TARGET_PAGE; | |
6862 | if (target) { | |
6863 | get_page(page); | |
6864 | target->page = page; | |
6865 | } | |
6866 | } | |
6867 | return ret; | |
6868 | } | |
6869 | #else | |
6870 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6871 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6872 | { | |
6873 | return MC_TARGET_NONE; | |
6874 | } | |
6875 | #endif | |
6876 | ||
4ffef5fe DN |
6877 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6878 | unsigned long addr, unsigned long end, | |
6879 | struct mm_walk *walk) | |
6880 | { | |
6881 | struct vm_area_struct *vma = walk->private; | |
6882 | pte_t *pte; | |
6883 | spinlock_t *ptl; | |
6884 | ||
bf929152 | 6885 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
6886 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
6887 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 6888 | spin_unlock(ptl); |
1a5a9906 | 6889 | return 0; |
12724850 | 6890 | } |
03319327 | 6891 | |
45f83cef AA |
6892 | if (pmd_trans_unstable(pmd)) |
6893 | return 0; | |
4ffef5fe DN |
6894 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6895 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6896 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6897 | mc.precharge++; /* increment precharge temporarily */ |
6898 | pte_unmap_unlock(pte - 1, ptl); | |
6899 | cond_resched(); | |
6900 | ||
7dc74be0 DN |
6901 | return 0; |
6902 | } | |
6903 | ||
4ffef5fe DN |
6904 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6905 | { | |
6906 | unsigned long precharge; | |
6907 | struct vm_area_struct *vma; | |
6908 | ||
dfe076b0 | 6909 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6910 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6911 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6912 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6913 | .mm = mm, | |
6914 | .private = vma, | |
6915 | }; | |
6916 | if (is_vm_hugetlb_page(vma)) | |
6917 | continue; | |
4ffef5fe DN |
6918 | walk_page_range(vma->vm_start, vma->vm_end, |
6919 | &mem_cgroup_count_precharge_walk); | |
6920 | } | |
dfe076b0 | 6921 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6922 | |
6923 | precharge = mc.precharge; | |
6924 | mc.precharge = 0; | |
6925 | ||
6926 | return precharge; | |
6927 | } | |
6928 | ||
4ffef5fe DN |
6929 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6930 | { | |
dfe076b0 DN |
6931 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6932 | ||
6933 | VM_BUG_ON(mc.moving_task); | |
6934 | mc.moving_task = current; | |
6935 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6936 | } |
6937 | ||
dfe076b0 DN |
6938 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6939 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6940 | { |
2bd9bb20 KH |
6941 | struct mem_cgroup *from = mc.from; |
6942 | struct mem_cgroup *to = mc.to; | |
4050377b | 6943 | int i; |
2bd9bb20 | 6944 | |
4ffef5fe | 6945 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6946 | if (mc.precharge) { |
6947 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6948 | mc.precharge = 0; | |
6949 | } | |
6950 | /* | |
6951 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6952 | * we must uncharge here. | |
6953 | */ | |
6954 | if (mc.moved_charge) { | |
6955 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6956 | mc.moved_charge = 0; | |
4ffef5fe | 6957 | } |
483c30b5 DN |
6958 | /* we must fixup refcnts and charges */ |
6959 | if (mc.moved_swap) { | |
483c30b5 DN |
6960 | /* uncharge swap account from the old cgroup */ |
6961 | if (!mem_cgroup_is_root(mc.from)) | |
6962 | res_counter_uncharge(&mc.from->memsw, | |
6963 | PAGE_SIZE * mc.moved_swap); | |
4050377b LZ |
6964 | |
6965 | for (i = 0; i < mc.moved_swap; i++) | |
6966 | css_put(&mc.from->css); | |
483c30b5 DN |
6967 | |
6968 | if (!mem_cgroup_is_root(mc.to)) { | |
6969 | /* | |
6970 | * we charged both to->res and to->memsw, so we should | |
6971 | * uncharge to->res. | |
6972 | */ | |
6973 | res_counter_uncharge(&mc.to->res, | |
6974 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 | 6975 | } |
4050377b | 6976 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
6977 | mc.moved_swap = 0; |
6978 | } | |
dfe076b0 DN |
6979 | memcg_oom_recover(from); |
6980 | memcg_oom_recover(to); | |
6981 | wake_up_all(&mc.waitq); | |
6982 | } | |
6983 | ||
6984 | static void mem_cgroup_clear_mc(void) | |
6985 | { | |
6986 | struct mem_cgroup *from = mc.from; | |
6987 | ||
6988 | /* | |
6989 | * we must clear moving_task before waking up waiters at the end of | |
6990 | * task migration. | |
6991 | */ | |
6992 | mc.moving_task = NULL; | |
6993 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6994 | spin_lock(&mc.lock); |
4ffef5fe DN |
6995 | mc.from = NULL; |
6996 | mc.to = NULL; | |
2bd9bb20 | 6997 | spin_unlock(&mc.lock); |
32047e2a | 6998 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6999 | } |
7000 | ||
eb95419b | 7001 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7002 | struct cgroup_taskset *tset) |
7dc74be0 | 7003 | { |
2f7ee569 | 7004 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 7005 | int ret = 0; |
eb95419b | 7006 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 7007 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 7008 | |
ee5e8472 GC |
7009 | /* |
7010 | * We are now commited to this value whatever it is. Changes in this | |
7011 | * tunable will only affect upcoming migrations, not the current one. | |
7012 | * So we need to save it, and keep it going. | |
7013 | */ | |
7014 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
7015 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
7016 | struct mm_struct *mm; |
7017 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
7018 | ||
c0ff4b85 | 7019 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
7020 | |
7021 | mm = get_task_mm(p); | |
7022 | if (!mm) | |
7023 | return 0; | |
7dc74be0 | 7024 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
7025 | if (mm->owner == p) { |
7026 | VM_BUG_ON(mc.from); | |
7027 | VM_BUG_ON(mc.to); | |
7028 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 7029 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 7030 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 7031 | mem_cgroup_start_move(from); |
2bd9bb20 | 7032 | spin_lock(&mc.lock); |
4ffef5fe | 7033 | mc.from = from; |
c0ff4b85 | 7034 | mc.to = memcg; |
ee5e8472 | 7035 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 7036 | spin_unlock(&mc.lock); |
dfe076b0 | 7037 | /* We set mc.moving_task later */ |
4ffef5fe DN |
7038 | |
7039 | ret = mem_cgroup_precharge_mc(mm); | |
7040 | if (ret) | |
7041 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
7042 | } |
7043 | mmput(mm); | |
7dc74be0 DN |
7044 | } |
7045 | return ret; | |
7046 | } | |
7047 | ||
eb95419b | 7048 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7049 | struct cgroup_taskset *tset) |
7dc74be0 | 7050 | { |
4ffef5fe | 7051 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
7052 | } |
7053 | ||
4ffef5fe DN |
7054 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
7055 | unsigned long addr, unsigned long end, | |
7056 | struct mm_walk *walk) | |
7dc74be0 | 7057 | { |
4ffef5fe DN |
7058 | int ret = 0; |
7059 | struct vm_area_struct *vma = walk->private; | |
7060 | pte_t *pte; | |
7061 | spinlock_t *ptl; | |
12724850 NH |
7062 | enum mc_target_type target_type; |
7063 | union mc_target target; | |
7064 | struct page *page; | |
7065 | struct page_cgroup *pc; | |
4ffef5fe | 7066 | |
12724850 NH |
7067 | /* |
7068 | * We don't take compound_lock() here but no race with splitting thp | |
7069 | * happens because: | |
7070 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
7071 | * under splitting, which means there's no concurrent thp split, | |
7072 | * - if another thread runs into split_huge_page() just after we | |
7073 | * entered this if-block, the thread must wait for page table lock | |
7074 | * to be unlocked in __split_huge_page_splitting(), where the main | |
7075 | * part of thp split is not executed yet. | |
7076 | */ | |
bf929152 | 7077 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 7078 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 7079 | spin_unlock(ptl); |
12724850 NH |
7080 | return 0; |
7081 | } | |
7082 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
7083 | if (target_type == MC_TARGET_PAGE) { | |
7084 | page = target.page; | |
7085 | if (!isolate_lru_page(page)) { | |
7086 | pc = lookup_page_cgroup(page); | |
7087 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 7088 | pc, mc.from, mc.to)) { |
12724850 NH |
7089 | mc.precharge -= HPAGE_PMD_NR; |
7090 | mc.moved_charge += HPAGE_PMD_NR; | |
7091 | } | |
7092 | putback_lru_page(page); | |
7093 | } | |
7094 | put_page(page); | |
7095 | } | |
bf929152 | 7096 | spin_unlock(ptl); |
1a5a9906 | 7097 | return 0; |
12724850 NH |
7098 | } |
7099 | ||
45f83cef AA |
7100 | if (pmd_trans_unstable(pmd)) |
7101 | return 0; | |
4ffef5fe DN |
7102 | retry: |
7103 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
7104 | for (; addr != end; addr += PAGE_SIZE) { | |
7105 | pte_t ptent = *(pte++); | |
02491447 | 7106 | swp_entry_t ent; |
4ffef5fe DN |
7107 | |
7108 | if (!mc.precharge) | |
7109 | break; | |
7110 | ||
8d32ff84 | 7111 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
7112 | case MC_TARGET_PAGE: |
7113 | page = target.page; | |
7114 | if (isolate_lru_page(page)) | |
7115 | goto put; | |
7116 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 7117 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 7118 | mc.from, mc.to)) { |
4ffef5fe | 7119 | mc.precharge--; |
854ffa8d DN |
7120 | /* we uncharge from mc.from later. */ |
7121 | mc.moved_charge++; | |
4ffef5fe DN |
7122 | } |
7123 | putback_lru_page(page); | |
8d32ff84 | 7124 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
7125 | put_page(page); |
7126 | break; | |
02491447 DN |
7127 | case MC_TARGET_SWAP: |
7128 | ent = target.ent; | |
e91cbb42 | 7129 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 7130 | mc.precharge--; |
483c30b5 DN |
7131 | /* we fixup refcnts and charges later. */ |
7132 | mc.moved_swap++; | |
7133 | } | |
02491447 | 7134 | break; |
4ffef5fe DN |
7135 | default: |
7136 | break; | |
7137 | } | |
7138 | } | |
7139 | pte_unmap_unlock(pte - 1, ptl); | |
7140 | cond_resched(); | |
7141 | ||
7142 | if (addr != end) { | |
7143 | /* | |
7144 | * We have consumed all precharges we got in can_attach(). | |
7145 | * We try charge one by one, but don't do any additional | |
7146 | * charges to mc.to if we have failed in charge once in attach() | |
7147 | * phase. | |
7148 | */ | |
854ffa8d | 7149 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
7150 | if (!ret) |
7151 | goto retry; | |
7152 | } | |
7153 | ||
7154 | return ret; | |
7155 | } | |
7156 | ||
7157 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
7158 | { | |
7159 | struct vm_area_struct *vma; | |
7160 | ||
7161 | lru_add_drain_all(); | |
dfe076b0 DN |
7162 | retry: |
7163 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
7164 | /* | |
7165 | * Someone who are holding the mmap_sem might be waiting in | |
7166 | * waitq. So we cancel all extra charges, wake up all waiters, | |
7167 | * and retry. Because we cancel precharges, we might not be able | |
7168 | * to move enough charges, but moving charge is a best-effort | |
7169 | * feature anyway, so it wouldn't be a big problem. | |
7170 | */ | |
7171 | __mem_cgroup_clear_mc(); | |
7172 | cond_resched(); | |
7173 | goto retry; | |
7174 | } | |
4ffef5fe DN |
7175 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
7176 | int ret; | |
7177 | struct mm_walk mem_cgroup_move_charge_walk = { | |
7178 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
7179 | .mm = mm, | |
7180 | .private = vma, | |
7181 | }; | |
7182 | if (is_vm_hugetlb_page(vma)) | |
7183 | continue; | |
4ffef5fe DN |
7184 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
7185 | &mem_cgroup_move_charge_walk); | |
7186 | if (ret) | |
7187 | /* | |
7188 | * means we have consumed all precharges and failed in | |
7189 | * doing additional charge. Just abandon here. | |
7190 | */ | |
7191 | break; | |
7192 | } | |
dfe076b0 | 7193 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
7194 | } |
7195 | ||
eb95419b | 7196 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 7197 | struct cgroup_taskset *tset) |
67e465a7 | 7198 | { |
2f7ee569 | 7199 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 7200 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 7201 | |
dfe076b0 | 7202 | if (mm) { |
a433658c KM |
7203 | if (mc.to) |
7204 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
7205 | mmput(mm); |
7206 | } | |
a433658c KM |
7207 | if (mc.to) |
7208 | mem_cgroup_clear_mc(); | |
67e465a7 | 7209 | } |
5cfb80a7 | 7210 | #else /* !CONFIG_MMU */ |
eb95419b | 7211 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7212 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7213 | { |
7214 | return 0; | |
7215 | } | |
eb95419b | 7216 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 7217 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7218 | { |
7219 | } | |
eb95419b | 7220 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 7221 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
7222 | { |
7223 | } | |
7224 | #endif | |
67e465a7 | 7225 | |
f00baae7 TH |
7226 | /* |
7227 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
7228 | * to verify sane_behavior flag on each mount attempt. | |
7229 | */ | |
eb95419b | 7230 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
7231 | { |
7232 | /* | |
7233 | * use_hierarchy is forced with sane_behavior. cgroup core | |
7234 | * guarantees that @root doesn't have any children, so turning it | |
7235 | * on for the root memcg is enough. | |
7236 | */ | |
eb95419b TH |
7237 | if (cgroup_sane_behavior(root_css->cgroup)) |
7238 | mem_cgroup_from_css(root_css)->use_hierarchy = true; | |
f00baae7 TH |
7239 | } |
7240 | ||
073219e9 | 7241 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 7242 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 7243 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
7244 | .css_offline = mem_cgroup_css_offline, |
7245 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
7246 | .can_attach = mem_cgroup_can_attach, |
7247 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 7248 | .attach = mem_cgroup_move_task, |
f00baae7 | 7249 | .bind = mem_cgroup_bind, |
6bc10349 | 7250 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 7251 | .early_init = 0, |
8cdea7c0 | 7252 | }; |
c077719b | 7253 | |
c255a458 | 7254 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
7255 | static int __init enable_swap_account(char *s) |
7256 | { | |
a2c8990a | 7257 | if (!strcmp(s, "1")) |
a42c390c | 7258 | really_do_swap_account = 1; |
a2c8990a | 7259 | else if (!strcmp(s, "0")) |
a42c390c MH |
7260 | really_do_swap_account = 0; |
7261 | return 1; | |
7262 | } | |
a2c8990a | 7263 | __setup("swapaccount=", enable_swap_account); |
c077719b | 7264 | |
2d11085e MH |
7265 | static void __init memsw_file_init(void) |
7266 | { | |
073219e9 | 7267 | WARN_ON(cgroup_add_cftypes(&memory_cgrp_subsys, memsw_cgroup_files)); |
6acc8b02 MH |
7268 | } |
7269 | ||
7270 | static void __init enable_swap_cgroup(void) | |
7271 | { | |
7272 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
7273 | do_swap_account = 1; | |
7274 | memsw_file_init(); | |
7275 | } | |
2d11085e | 7276 | } |
6acc8b02 | 7277 | |
2d11085e | 7278 | #else |
6acc8b02 | 7279 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
7280 | { |
7281 | } | |
c077719b | 7282 | #endif |
2d11085e MH |
7283 | |
7284 | /* | |
1081312f MH |
7285 | * subsys_initcall() for memory controller. |
7286 | * | |
7287 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
7288 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
7289 | * everything that doesn't depend on a specific mem_cgroup structure should | |
7290 | * be initialized from here. | |
2d11085e MH |
7291 | */ |
7292 | static int __init mem_cgroup_init(void) | |
7293 | { | |
7294 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 7295 | enable_swap_cgroup(); |
bb4cc1a8 | 7296 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 7297 | memcg_stock_init(); |
2d11085e MH |
7298 | return 0; |
7299 | } | |
7300 | subsys_initcall(mem_cgroup_init); |