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 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
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> |
5d1ea48b | 54 | #include <linux/swap_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
79bd9814 | 58 | #include <linux/file.h> |
08e552c6 | 59 | #include "internal.h" |
d1a4c0b3 | 60 | #include <net/sock.h> |
4bd2c1ee | 61 | #include <net/ip.h> |
d1a4c0b3 | 62 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 63 | #include "slab.h" |
8cdea7c0 | 64 | |
8697d331 BS |
65 | #include <asm/uaccess.h> |
66 | ||
cc8e970c KM |
67 | #include <trace/events/vmscan.h> |
68 | ||
073219e9 TH |
69 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
70 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 71 | |
a181b0e8 | 72 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 73 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 74 | |
c255a458 | 75 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 76 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 77 | int do_swap_account __read_mostly; |
a42c390c MH |
78 | |
79 | /* for remember boot option*/ | |
c255a458 | 80 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
81 | static int really_do_swap_account __initdata = 1; |
82 | #else | |
ada4ba59 | 83 | static int really_do_swap_account __initdata; |
a42c390c MH |
84 | #endif |
85 | ||
c077719b | 86 | #else |
a0db00fc | 87 | #define do_swap_account 0 |
c077719b KH |
88 | #endif |
89 | ||
90 | ||
af7c4b0e JW |
91 | static const char * const mem_cgroup_stat_names[] = { |
92 | "cache", | |
93 | "rss", | |
b070e65c | 94 | "rss_huge", |
af7c4b0e | 95 | "mapped_file", |
3ea67d06 | 96 | "writeback", |
af7c4b0e JW |
97 | "swap", |
98 | }; | |
99 | ||
af7c4b0e JW |
100 | static const char * const mem_cgroup_events_names[] = { |
101 | "pgpgin", | |
102 | "pgpgout", | |
103 | "pgfault", | |
104 | "pgmajfault", | |
105 | }; | |
106 | ||
58cf188e SZ |
107 | static const char * const mem_cgroup_lru_names[] = { |
108 | "inactive_anon", | |
109 | "active_anon", | |
110 | "inactive_file", | |
111 | "active_file", | |
112 | "unevictable", | |
113 | }; | |
114 | ||
7a159cc9 JW |
115 | /* |
116 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
117 | * it will be incremated by the number of pages. This counter is used for | |
118 | * for trigger some periodic events. This is straightforward and better | |
119 | * than using jiffies etc. to handle periodic memcg event. | |
120 | */ | |
121 | enum mem_cgroup_events_target { | |
122 | MEM_CGROUP_TARGET_THRESH, | |
bb4cc1a8 | 123 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 124 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
125 | MEM_CGROUP_NTARGETS, |
126 | }; | |
a0db00fc KS |
127 | #define THRESHOLDS_EVENTS_TARGET 128 |
128 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
129 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 130 | |
d52aa412 | 131 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 132 | long count[MEM_CGROUP_STAT_NSTATS]; |
241994ed | 133 | unsigned long events[MEMCG_NR_EVENTS]; |
13114716 | 134 | unsigned long nr_page_events; |
7a159cc9 | 135 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
136 | }; |
137 | ||
5ac8fb31 JW |
138 | struct reclaim_iter { |
139 | struct mem_cgroup *position; | |
527a5ec9 JW |
140 | /* scan generation, increased every round-trip */ |
141 | unsigned int generation; | |
142 | }; | |
143 | ||
6d12e2d8 KH |
144 | /* |
145 | * per-zone information in memory controller. | |
146 | */ | |
6d12e2d8 | 147 | struct mem_cgroup_per_zone { |
6290df54 | 148 | struct lruvec lruvec; |
1eb49272 | 149 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 150 | |
5ac8fb31 | 151 | struct reclaim_iter iter[DEF_PRIORITY + 1]; |
527a5ec9 | 152 | |
bb4cc1a8 | 153 | struct rb_node tree_node; /* RB tree node */ |
3e32cb2e | 154 | unsigned long usage_in_excess;/* Set to the value by which */ |
bb4cc1a8 AM |
155 | /* the soft limit is exceeded*/ |
156 | bool on_tree; | |
d79154bb | 157 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 158 | /* use container_of */ |
6d12e2d8 | 159 | }; |
6d12e2d8 KH |
160 | |
161 | struct mem_cgroup_per_node { | |
162 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
163 | }; | |
164 | ||
bb4cc1a8 AM |
165 | /* |
166 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
167 | * their hierarchy representation | |
168 | */ | |
169 | ||
170 | struct mem_cgroup_tree_per_zone { | |
171 | struct rb_root rb_root; | |
172 | spinlock_t lock; | |
173 | }; | |
174 | ||
175 | struct mem_cgroup_tree_per_node { | |
176 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
177 | }; | |
178 | ||
179 | struct mem_cgroup_tree { | |
180 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
181 | }; | |
182 | ||
183 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
184 | ||
2e72b634 KS |
185 | struct mem_cgroup_threshold { |
186 | struct eventfd_ctx *eventfd; | |
3e32cb2e | 187 | unsigned long threshold; |
2e72b634 KS |
188 | }; |
189 | ||
9490ff27 | 190 | /* For threshold */ |
2e72b634 | 191 | struct mem_cgroup_threshold_ary { |
748dad36 | 192 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 193 | int current_threshold; |
2e72b634 KS |
194 | /* Size of entries[] */ |
195 | unsigned int size; | |
196 | /* Array of thresholds */ | |
197 | struct mem_cgroup_threshold entries[0]; | |
198 | }; | |
2c488db2 KS |
199 | |
200 | struct mem_cgroup_thresholds { | |
201 | /* Primary thresholds array */ | |
202 | struct mem_cgroup_threshold_ary *primary; | |
203 | /* | |
204 | * Spare threshold array. | |
205 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
206 | * It must be able to store at least primary->size - 1 entries. | |
207 | */ | |
208 | struct mem_cgroup_threshold_ary *spare; | |
209 | }; | |
210 | ||
9490ff27 KH |
211 | /* for OOM */ |
212 | struct mem_cgroup_eventfd_list { | |
213 | struct list_head list; | |
214 | struct eventfd_ctx *eventfd; | |
215 | }; | |
2e72b634 | 216 | |
79bd9814 TH |
217 | /* |
218 | * cgroup_event represents events which userspace want to receive. | |
219 | */ | |
3bc942f3 | 220 | struct mem_cgroup_event { |
79bd9814 | 221 | /* |
59b6f873 | 222 | * memcg which the event belongs to. |
79bd9814 | 223 | */ |
59b6f873 | 224 | struct mem_cgroup *memcg; |
79bd9814 TH |
225 | /* |
226 | * eventfd to signal userspace about the event. | |
227 | */ | |
228 | struct eventfd_ctx *eventfd; | |
229 | /* | |
230 | * Each of these stored in a list by the cgroup. | |
231 | */ | |
232 | struct list_head list; | |
fba94807 TH |
233 | /* |
234 | * register_event() callback will be used to add new userspace | |
235 | * waiter for changes related to this event. Use eventfd_signal() | |
236 | * on eventfd to send notification to userspace. | |
237 | */ | |
59b6f873 | 238 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 239 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
240 | /* |
241 | * unregister_event() callback will be called when userspace closes | |
242 | * the eventfd or on cgroup removing. This callback must be set, | |
243 | * if you want provide notification functionality. | |
244 | */ | |
59b6f873 | 245 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 246 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
247 | /* |
248 | * All fields below needed to unregister event when | |
249 | * userspace closes eventfd. | |
250 | */ | |
251 | poll_table pt; | |
252 | wait_queue_head_t *wqh; | |
253 | wait_queue_t wait; | |
254 | struct work_struct remove; | |
255 | }; | |
256 | ||
c0ff4b85 R |
257 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
258 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 259 | |
8cdea7c0 BS |
260 | /* |
261 | * The memory controller data structure. The memory controller controls both | |
262 | * page cache and RSS per cgroup. We would eventually like to provide | |
263 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
264 | * to help the administrator determine what knobs to tune. | |
265 | * | |
266 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
267 | * we hit the water mark. May be even add a low water mark, such that |
268 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
269 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
270 | */ |
271 | struct mem_cgroup { | |
272 | struct cgroup_subsys_state css; | |
3e32cb2e JW |
273 | |
274 | /* Accounted resources */ | |
275 | struct page_counter memory; | |
276 | struct page_counter memsw; | |
277 | struct page_counter kmem; | |
278 | ||
241994ed JW |
279 | /* Normal memory consumption range */ |
280 | unsigned long low; | |
281 | unsigned long high; | |
282 | ||
3e32cb2e | 283 | unsigned long soft_limit; |
59927fb9 | 284 | |
70ddf637 AV |
285 | /* vmpressure notifications */ |
286 | struct vmpressure vmpressure; | |
287 | ||
2f7dd7a4 JW |
288 | /* css_online() has been completed */ |
289 | int initialized; | |
290 | ||
18f59ea7 BS |
291 | /* |
292 | * Should the accounting and control be hierarchical, per subtree? | |
293 | */ | |
294 | bool use_hierarchy; | |
79dfdacc MH |
295 | |
296 | bool oom_lock; | |
297 | atomic_t under_oom; | |
3812c8c8 | 298 | atomic_t oom_wakeups; |
79dfdacc | 299 | |
1f4c025b | 300 | int swappiness; |
3c11ecf4 KH |
301 | /* OOM-Killer disable */ |
302 | int oom_kill_disable; | |
a7885eb8 | 303 | |
2e72b634 KS |
304 | /* protect arrays of thresholds */ |
305 | struct mutex thresholds_lock; | |
306 | ||
307 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 308 | struct mem_cgroup_thresholds thresholds; |
907860ed | 309 | |
2e72b634 | 310 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 311 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 312 | |
9490ff27 KH |
313 | /* For oom notifier event fd */ |
314 | struct list_head oom_notify; | |
185efc0f | 315 | |
7dc74be0 DN |
316 | /* |
317 | * Should we move charges of a task when a task is moved into this | |
318 | * mem_cgroup ? And what type of charges should we move ? | |
319 | */ | |
f894ffa8 | 320 | unsigned long move_charge_at_immigrate; |
619d094b KH |
321 | /* |
322 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
323 | */ | |
6de22619 | 324 | atomic_t moving_account; |
312734c0 | 325 | /* taken only while moving_account > 0 */ |
6de22619 JW |
326 | spinlock_t move_lock; |
327 | struct task_struct *move_lock_task; | |
328 | unsigned long move_lock_flags; | |
d52aa412 | 329 | /* |
c62b1a3b | 330 | * percpu counter. |
d52aa412 | 331 | */ |
3a7951b4 | 332 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
333 | /* |
334 | * used when a cpu is offlined or other synchronizations | |
335 | * See mem_cgroup_read_stat(). | |
336 | */ | |
337 | struct mem_cgroup_stat_cpu nocpu_base; | |
338 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 339 | |
4bd2c1ee | 340 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
2e685cad | 341 | struct cg_proto tcp_mem; |
d1a4c0b3 | 342 | #endif |
2633d7a0 | 343 | #if defined(CONFIG_MEMCG_KMEM) |
2633d7a0 GC |
344 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ |
345 | int kmemcg_id; | |
346 | #endif | |
45cf7ebd GC |
347 | |
348 | int last_scanned_node; | |
349 | #if MAX_NUMNODES > 1 | |
350 | nodemask_t scan_nodes; | |
351 | atomic_t numainfo_events; | |
352 | atomic_t numainfo_updating; | |
353 | #endif | |
70ddf637 | 354 | |
fba94807 TH |
355 | /* List of events which userspace want to receive */ |
356 | struct list_head event_list; | |
357 | spinlock_t event_list_lock; | |
358 | ||
54f72fe0 JW |
359 | struct mem_cgroup_per_node *nodeinfo[0]; |
360 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
361 | }; |
362 | ||
510fc4e1 | 363 | #ifdef CONFIG_MEMCG_KMEM |
7de37682 GC |
364 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) |
365 | { | |
900a38f0 | 366 | return memcg->kmemcg_id >= 0; |
7de37682 | 367 | } |
510fc4e1 GC |
368 | #endif |
369 | ||
7dc74be0 DN |
370 | /* Stuffs for move charges at task migration. */ |
371 | /* | |
1dfab5ab | 372 | * Types of charges to be moved. |
7dc74be0 | 373 | */ |
1dfab5ab JW |
374 | #define MOVE_ANON 0x1U |
375 | #define MOVE_FILE 0x2U | |
376 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 377 | |
4ffef5fe DN |
378 | /* "mc" and its members are protected by cgroup_mutex */ |
379 | static struct move_charge_struct { | |
b1dd693e | 380 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
381 | struct mem_cgroup *from; |
382 | struct mem_cgroup *to; | |
1dfab5ab | 383 | unsigned long flags; |
4ffef5fe | 384 | unsigned long precharge; |
854ffa8d | 385 | unsigned long moved_charge; |
483c30b5 | 386 | unsigned long moved_swap; |
8033b97c DN |
387 | struct task_struct *moving_task; /* a task moving charges */ |
388 | wait_queue_head_t waitq; /* a waitq for other context */ | |
389 | } mc = { | |
2bd9bb20 | 390 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
391 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
392 | }; | |
4ffef5fe | 393 | |
4e416953 BS |
394 | /* |
395 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
396 | * limit reclaim to prevent infinite loops, if they ever occur. | |
397 | */ | |
a0db00fc | 398 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 399 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 400 | |
217bc319 KH |
401 | enum charge_type { |
402 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 403 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 404 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 405 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
406 | NR_CHARGE_TYPE, |
407 | }; | |
408 | ||
8c7c6e34 | 409 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
410 | enum res_type { |
411 | _MEM, | |
412 | _MEMSWAP, | |
413 | _OOM_TYPE, | |
510fc4e1 | 414 | _KMEM, |
86ae53e1 GC |
415 | }; |
416 | ||
a0db00fc KS |
417 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
418 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 419 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
420 | /* Used for OOM nofiier */ |
421 | #define OOM_CONTROL (0) | |
8c7c6e34 | 422 | |
0999821b GC |
423 | /* |
424 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
425 | * As a consequence, any change that needs to protect against new child cgroups | |
426 | * appearing has to hold it as well. | |
427 | */ | |
428 | static DEFINE_MUTEX(memcg_create_mutex); | |
429 | ||
b2145145 WL |
430 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
431 | { | |
a7c6d554 | 432 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
433 | } |
434 | ||
70ddf637 AV |
435 | /* Some nice accessors for the vmpressure. */ |
436 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
437 | { | |
438 | if (!memcg) | |
439 | memcg = root_mem_cgroup; | |
440 | return &memcg->vmpressure; | |
441 | } | |
442 | ||
443 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
444 | { | |
445 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
446 | } | |
447 | ||
7ffc0edc MH |
448 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
449 | { | |
450 | return (memcg == root_mem_cgroup); | |
451 | } | |
452 | ||
4219b2da LZ |
453 | /* |
454 | * We restrict the id in the range of [1, 65535], so it can fit into | |
455 | * an unsigned short. | |
456 | */ | |
457 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
458 | ||
34c00c31 LZ |
459 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
460 | { | |
15a4c835 | 461 | return memcg->css.id; |
34c00c31 LZ |
462 | } |
463 | ||
464 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
465 | { | |
466 | struct cgroup_subsys_state *css; | |
467 | ||
7d699ddb | 468 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
469 | return mem_cgroup_from_css(css); |
470 | } | |
471 | ||
e1aab161 | 472 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 473 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 474 | |
e1aab161 GC |
475 | void sock_update_memcg(struct sock *sk) |
476 | { | |
376be5ff | 477 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 478 | struct mem_cgroup *memcg; |
3f134619 | 479 | struct cg_proto *cg_proto; |
e1aab161 GC |
480 | |
481 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
482 | ||
f3f511e1 GC |
483 | /* Socket cloning can throw us here with sk_cgrp already |
484 | * filled. It won't however, necessarily happen from | |
485 | * process context. So the test for root memcg given | |
486 | * the current task's memcg won't help us in this case. | |
487 | * | |
488 | * Respecting the original socket's memcg is a better | |
489 | * decision in this case. | |
490 | */ | |
491 | if (sk->sk_cgrp) { | |
492 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 493 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
494 | return; |
495 | } | |
496 | ||
e1aab161 GC |
497 | rcu_read_lock(); |
498 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 499 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 500 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
501 | memcg_proto_active(cg_proto) && |
502 | css_tryget_online(&memcg->css)) { | |
3f134619 | 503 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
504 | } |
505 | rcu_read_unlock(); | |
506 | } | |
507 | } | |
508 | EXPORT_SYMBOL(sock_update_memcg); | |
509 | ||
510 | void sock_release_memcg(struct sock *sk) | |
511 | { | |
376be5ff | 512 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
513 | struct mem_cgroup *memcg; |
514 | WARN_ON(!sk->sk_cgrp->memcg); | |
515 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 516 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
517 | } |
518 | } | |
d1a4c0b3 GC |
519 | |
520 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
521 | { | |
522 | if (!memcg || mem_cgroup_is_root(memcg)) | |
523 | return NULL; | |
524 | ||
2e685cad | 525 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
526 | } |
527 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 528 | |
3f134619 GC |
529 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
530 | { | |
2e685cad | 531 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
532 | return; |
533 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
534 | } | |
535 | #else | |
536 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
537 | { | |
538 | } | |
539 | #endif | |
540 | ||
a8964b9b | 541 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
542 | /* |
543 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
b8627835 LZ |
544 | * The main reason for not using cgroup id for this: |
545 | * this works better in sparse environments, where we have a lot of memcgs, | |
546 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
547 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
548 | * 200 entry array for that. | |
55007d84 GC |
549 | * |
550 | * The current size of the caches array is stored in | |
551 | * memcg_limited_groups_array_size. It will double each time we have to | |
552 | * increase it. | |
553 | */ | |
554 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
555 | int memcg_limited_groups_array_size; |
556 | ||
55007d84 GC |
557 | /* |
558 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
559 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
560 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
561 | * tunable, but that is strictly not necessary. | |
562 | * | |
b8627835 | 563 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
564 | * this constant directly from cgroup, but it is understandable that this is |
565 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 566 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
567 | * increase ours as well if it increases. |
568 | */ | |
569 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 570 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 571 | |
d7f25f8a GC |
572 | /* |
573 | * A lot of the calls to the cache allocation functions are expected to be | |
574 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
575 | * conditional to this static branch, we'll have to allow modules that does | |
576 | * kmem_cache_alloc and the such to see this symbol as well | |
577 | */ | |
a8964b9b | 578 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 579 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 580 | |
f3bb3043 VD |
581 | static void memcg_free_cache_id(int id); |
582 | ||
a8964b9b GC |
583 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
584 | { | |
55007d84 | 585 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 586 | static_key_slow_dec(&memcg_kmem_enabled_key); |
f3bb3043 | 587 | memcg_free_cache_id(memcg->kmemcg_id); |
55007d84 | 588 | } |
bea207c8 GC |
589 | /* |
590 | * This check can't live in kmem destruction function, | |
591 | * since the charges will outlive the cgroup | |
592 | */ | |
3e32cb2e | 593 | WARN_ON(page_counter_read(&memcg->kmem)); |
a8964b9b GC |
594 | } |
595 | #else | |
596 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
597 | { | |
598 | } | |
599 | #endif /* CONFIG_MEMCG_KMEM */ | |
600 | ||
601 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
602 | { | |
603 | disarm_sock_keys(memcg); | |
604 | disarm_kmem_keys(memcg); | |
605 | } | |
606 | ||
f64c3f54 | 607 | static struct mem_cgroup_per_zone * |
e231875b | 608 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 609 | { |
e231875b JZ |
610 | int nid = zone_to_nid(zone); |
611 | int zid = zone_idx(zone); | |
612 | ||
54f72fe0 | 613 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
614 | } |
615 | ||
c0ff4b85 | 616 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 617 | { |
c0ff4b85 | 618 | return &memcg->css; |
d324236b WF |
619 | } |
620 | ||
f64c3f54 | 621 | static struct mem_cgroup_per_zone * |
e231875b | 622 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 623 | { |
97a6c37b JW |
624 | int nid = page_to_nid(page); |
625 | int zid = page_zonenum(page); | |
f64c3f54 | 626 | |
e231875b | 627 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
628 | } |
629 | ||
bb4cc1a8 AM |
630 | static struct mem_cgroup_tree_per_zone * |
631 | soft_limit_tree_node_zone(int nid, int zid) | |
632 | { | |
633 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
634 | } | |
635 | ||
636 | static struct mem_cgroup_tree_per_zone * | |
637 | soft_limit_tree_from_page(struct page *page) | |
638 | { | |
639 | int nid = page_to_nid(page); | |
640 | int zid = page_zonenum(page); | |
641 | ||
642 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
643 | } | |
644 | ||
cf2c8127 JW |
645 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
646 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 647 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
648 | { |
649 | struct rb_node **p = &mctz->rb_root.rb_node; | |
650 | struct rb_node *parent = NULL; | |
651 | struct mem_cgroup_per_zone *mz_node; | |
652 | ||
653 | if (mz->on_tree) | |
654 | return; | |
655 | ||
656 | mz->usage_in_excess = new_usage_in_excess; | |
657 | if (!mz->usage_in_excess) | |
658 | return; | |
659 | while (*p) { | |
660 | parent = *p; | |
661 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
662 | tree_node); | |
663 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
664 | p = &(*p)->rb_left; | |
665 | /* | |
666 | * We can't avoid mem cgroups that are over their soft | |
667 | * limit by the same amount | |
668 | */ | |
669 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
670 | p = &(*p)->rb_right; | |
671 | } | |
672 | rb_link_node(&mz->tree_node, parent, p); | |
673 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
674 | mz->on_tree = true; | |
675 | } | |
676 | ||
cf2c8127 JW |
677 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
678 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
679 | { |
680 | if (!mz->on_tree) | |
681 | return; | |
682 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
683 | mz->on_tree = false; | |
684 | } | |
685 | ||
cf2c8127 JW |
686 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
687 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 688 | { |
0a31bc97 JW |
689 | unsigned long flags; |
690 | ||
691 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 692 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 693 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
694 | } |
695 | ||
3e32cb2e JW |
696 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
697 | { | |
698 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
699 | unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit); | |
700 | unsigned long excess = 0; | |
701 | ||
702 | if (nr_pages > soft_limit) | |
703 | excess = nr_pages - soft_limit; | |
704 | ||
705 | return excess; | |
706 | } | |
bb4cc1a8 AM |
707 | |
708 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
709 | { | |
3e32cb2e | 710 | unsigned long excess; |
bb4cc1a8 AM |
711 | struct mem_cgroup_per_zone *mz; |
712 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 713 | |
e231875b | 714 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
715 | /* |
716 | * Necessary to update all ancestors when hierarchy is used. | |
717 | * because their event counter is not touched. | |
718 | */ | |
719 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 720 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 721 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
722 | /* |
723 | * We have to update the tree if mz is on RB-tree or | |
724 | * mem is over its softlimit. | |
725 | */ | |
726 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
727 | unsigned long flags; |
728 | ||
729 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
730 | /* if on-tree, remove it */ |
731 | if (mz->on_tree) | |
cf2c8127 | 732 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
733 | /* |
734 | * Insert again. mz->usage_in_excess will be updated. | |
735 | * If excess is 0, no tree ops. | |
736 | */ | |
cf2c8127 | 737 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 738 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
739 | } |
740 | } | |
741 | } | |
742 | ||
743 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
744 | { | |
bb4cc1a8 | 745 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
746 | struct mem_cgroup_per_zone *mz; |
747 | int nid, zid; | |
bb4cc1a8 | 748 | |
e231875b JZ |
749 | for_each_node(nid) { |
750 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
751 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
752 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 753 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
754 | } |
755 | } | |
756 | } | |
757 | ||
758 | static struct mem_cgroup_per_zone * | |
759 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
760 | { | |
761 | struct rb_node *rightmost = NULL; | |
762 | struct mem_cgroup_per_zone *mz; | |
763 | ||
764 | retry: | |
765 | mz = NULL; | |
766 | rightmost = rb_last(&mctz->rb_root); | |
767 | if (!rightmost) | |
768 | goto done; /* Nothing to reclaim from */ | |
769 | ||
770 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
771 | /* | |
772 | * Remove the node now but someone else can add it back, | |
773 | * we will to add it back at the end of reclaim to its correct | |
774 | * position in the tree. | |
775 | */ | |
cf2c8127 | 776 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 777 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 778 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
779 | goto retry; |
780 | done: | |
781 | return mz; | |
782 | } | |
783 | ||
784 | static struct mem_cgroup_per_zone * | |
785 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
786 | { | |
787 | struct mem_cgroup_per_zone *mz; | |
788 | ||
0a31bc97 | 789 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 790 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 791 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
792 | return mz; |
793 | } | |
794 | ||
711d3d2c KH |
795 | /* |
796 | * Implementation Note: reading percpu statistics for memcg. | |
797 | * | |
798 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
799 | * synchronization to implement "quick" read. There are trade-off between | |
800 | * reading cost and precision of value. Then, we may have a chance to implement | |
801 | * a periodic synchronizion of counter in memcg's counter. | |
802 | * | |
803 | * But this _read() function is used for user interface now. The user accounts | |
804 | * memory usage by memory cgroup and he _always_ requires exact value because | |
805 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
806 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
807 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
808 | * | |
809 | * If there are kernel internal actions which can make use of some not-exact | |
810 | * value, and reading all cpu value can be performance bottleneck in some | |
811 | * common workload, threashold and synchonization as vmstat[] should be | |
812 | * implemented. | |
813 | */ | |
c0ff4b85 | 814 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 815 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 816 | { |
7a159cc9 | 817 | long val = 0; |
c62b1a3b | 818 | int cpu; |
c62b1a3b | 819 | |
711d3d2c KH |
820 | get_online_cpus(); |
821 | for_each_online_cpu(cpu) | |
c0ff4b85 | 822 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 823 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
824 | spin_lock(&memcg->pcp_counter_lock); |
825 | val += memcg->nocpu_base.count[idx]; | |
826 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
827 | #endif |
828 | put_online_cpus(); | |
c62b1a3b KH |
829 | return val; |
830 | } | |
831 | ||
c0ff4b85 | 832 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
833 | enum mem_cgroup_events_index idx) |
834 | { | |
835 | unsigned long val = 0; | |
836 | int cpu; | |
837 | ||
9c567512 | 838 | get_online_cpus(); |
e9f8974f | 839 | for_each_online_cpu(cpu) |
c0ff4b85 | 840 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 841 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
842 | spin_lock(&memcg->pcp_counter_lock); |
843 | val += memcg->nocpu_base.events[idx]; | |
844 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 845 | #endif |
9c567512 | 846 | put_online_cpus(); |
e9f8974f JW |
847 | return val; |
848 | } | |
849 | ||
c0ff4b85 | 850 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 851 | struct page *page, |
0a31bc97 | 852 | int nr_pages) |
d52aa412 | 853 | { |
b2402857 KH |
854 | /* |
855 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
856 | * counted as CACHE even if it's on ANON LRU. | |
857 | */ | |
0a31bc97 | 858 | if (PageAnon(page)) |
b2402857 | 859 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 860 | nr_pages); |
d52aa412 | 861 | else |
b2402857 | 862 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 863 | nr_pages); |
55e462b0 | 864 | |
b070e65c DR |
865 | if (PageTransHuge(page)) |
866 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
867 | nr_pages); | |
868 | ||
e401f176 KH |
869 | /* pagein of a big page is an event. So, ignore page size */ |
870 | if (nr_pages > 0) | |
c0ff4b85 | 871 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 872 | else { |
c0ff4b85 | 873 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
874 | nr_pages = -nr_pages; /* for event */ |
875 | } | |
e401f176 | 876 | |
13114716 | 877 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
878 | } |
879 | ||
e231875b | 880 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
881 | { |
882 | struct mem_cgroup_per_zone *mz; | |
883 | ||
884 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
885 | return mz->lru_size[lru]; | |
886 | } | |
887 | ||
e231875b JZ |
888 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
889 | int nid, | |
890 | unsigned int lru_mask) | |
bb2a0de9 | 891 | { |
e231875b | 892 | unsigned long nr = 0; |
889976db YH |
893 | int zid; |
894 | ||
e231875b | 895 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 896 | |
e231875b JZ |
897 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
898 | struct mem_cgroup_per_zone *mz; | |
899 | enum lru_list lru; | |
900 | ||
901 | for_each_lru(lru) { | |
902 | if (!(BIT(lru) & lru_mask)) | |
903 | continue; | |
904 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
905 | nr += mz->lru_size[lru]; | |
906 | } | |
907 | } | |
908 | return nr; | |
889976db | 909 | } |
bb2a0de9 | 910 | |
c0ff4b85 | 911 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 912 | unsigned int lru_mask) |
6d12e2d8 | 913 | { |
e231875b | 914 | unsigned long nr = 0; |
889976db | 915 | int nid; |
6d12e2d8 | 916 | |
31aaea4a | 917 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
918 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
919 | return nr; | |
d52aa412 KH |
920 | } |
921 | ||
f53d7ce3 JW |
922 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
923 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
924 | { |
925 | unsigned long val, next; | |
926 | ||
13114716 | 927 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 928 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 929 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
930 | if ((long)next - (long)val < 0) { |
931 | switch (target) { | |
932 | case MEM_CGROUP_TARGET_THRESH: | |
933 | next = val + THRESHOLDS_EVENTS_TARGET; | |
934 | break; | |
bb4cc1a8 AM |
935 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
936 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
937 | break; | |
f53d7ce3 JW |
938 | case MEM_CGROUP_TARGET_NUMAINFO: |
939 | next = val + NUMAINFO_EVENTS_TARGET; | |
940 | break; | |
941 | default: | |
942 | break; | |
943 | } | |
944 | __this_cpu_write(memcg->stat->targets[target], next); | |
945 | return true; | |
7a159cc9 | 946 | } |
f53d7ce3 | 947 | return false; |
d2265e6f KH |
948 | } |
949 | ||
950 | /* | |
951 | * Check events in order. | |
952 | * | |
953 | */ | |
c0ff4b85 | 954 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
955 | { |
956 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
957 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
958 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 959 | bool do_softlimit; |
82b3f2a7 | 960 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 961 | |
bb4cc1a8 AM |
962 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
963 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
964 | #if MAX_NUMNODES > 1 |
965 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
966 | MEM_CGROUP_TARGET_NUMAINFO); | |
967 | #endif | |
c0ff4b85 | 968 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
969 | if (unlikely(do_softlimit)) |
970 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 971 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 972 | if (unlikely(do_numainfo)) |
c0ff4b85 | 973 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 974 | #endif |
0a31bc97 | 975 | } |
d2265e6f KH |
976 | } |
977 | ||
cf475ad2 | 978 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 979 | { |
31a78f23 BS |
980 | /* |
981 | * mm_update_next_owner() may clear mm->owner to NULL | |
982 | * if it races with swapoff, page migration, etc. | |
983 | * So this can be called with p == NULL. | |
984 | */ | |
985 | if (unlikely(!p)) | |
986 | return NULL; | |
987 | ||
073219e9 | 988 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
989 | } |
990 | ||
df381975 | 991 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 992 | { |
c0ff4b85 | 993 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 994 | |
54595fe2 KH |
995 | rcu_read_lock(); |
996 | do { | |
6f6acb00 MH |
997 | /* |
998 | * Page cache insertions can happen withou an | |
999 | * actual mm context, e.g. during disk probing | |
1000 | * on boot, loopback IO, acct() writes etc. | |
1001 | */ | |
1002 | if (unlikely(!mm)) | |
df381975 | 1003 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1004 | else { |
1005 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1006 | if (unlikely(!memcg)) | |
1007 | memcg = root_mem_cgroup; | |
1008 | } | |
ec903c0c | 1009 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1010 | rcu_read_unlock(); |
c0ff4b85 | 1011 | return memcg; |
54595fe2 KH |
1012 | } |
1013 | ||
5660048c JW |
1014 | /** |
1015 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1016 | * @root: hierarchy root | |
1017 | * @prev: previously returned memcg, NULL on first invocation | |
1018 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1019 | * | |
1020 | * Returns references to children of the hierarchy below @root, or | |
1021 | * @root itself, or %NULL after a full round-trip. | |
1022 | * | |
1023 | * Caller must pass the return value in @prev on subsequent | |
1024 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1025 | * to cancel a hierarchy walk before the round-trip is complete. | |
1026 | * | |
1027 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1028 | * divide up the memcgs in the hierarchy among all concurrent | |
1029 | * reclaimers operating on the same zone and priority. | |
1030 | */ | |
694fbc0f | 1031 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1032 | struct mem_cgroup *prev, |
694fbc0f | 1033 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1034 | { |
5ac8fb31 JW |
1035 | struct reclaim_iter *uninitialized_var(iter); |
1036 | struct cgroup_subsys_state *css = NULL; | |
9f3a0d09 | 1037 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1038 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1039 | |
694fbc0f AM |
1040 | if (mem_cgroup_disabled()) |
1041 | return NULL; | |
5660048c | 1042 | |
9f3a0d09 JW |
1043 | if (!root) |
1044 | root = root_mem_cgroup; | |
7d74b06f | 1045 | |
9f3a0d09 | 1046 | if (prev && !reclaim) |
5ac8fb31 | 1047 | pos = prev; |
14067bb3 | 1048 | |
9f3a0d09 JW |
1049 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1050 | if (prev) | |
5ac8fb31 | 1051 | goto out; |
694fbc0f | 1052 | return root; |
9f3a0d09 | 1053 | } |
14067bb3 | 1054 | |
542f85f9 | 1055 | rcu_read_lock(); |
5f578161 | 1056 | |
5ac8fb31 JW |
1057 | if (reclaim) { |
1058 | struct mem_cgroup_per_zone *mz; | |
1059 | ||
1060 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
1061 | iter = &mz->iter[reclaim->priority]; | |
1062 | ||
1063 | if (prev && reclaim->generation != iter->generation) | |
1064 | goto out_unlock; | |
1065 | ||
1066 | do { | |
1067 | pos = ACCESS_ONCE(iter->position); | |
1068 | /* | |
1069 | * A racing update may change the position and | |
1070 | * put the last reference, hence css_tryget(), | |
1071 | * or retry to see the updated position. | |
1072 | */ | |
1073 | } while (pos && !css_tryget(&pos->css)); | |
1074 | } | |
1075 | ||
1076 | if (pos) | |
1077 | css = &pos->css; | |
1078 | ||
1079 | for (;;) { | |
1080 | css = css_next_descendant_pre(css, &root->css); | |
1081 | if (!css) { | |
1082 | /* | |
1083 | * Reclaimers share the hierarchy walk, and a | |
1084 | * new one might jump in right at the end of | |
1085 | * the hierarchy - make sure they see at least | |
1086 | * one group and restart from the beginning. | |
1087 | */ | |
1088 | if (!prev) | |
1089 | continue; | |
1090 | break; | |
527a5ec9 | 1091 | } |
7d74b06f | 1092 | |
5ac8fb31 JW |
1093 | /* |
1094 | * Verify the css and acquire a reference. The root | |
1095 | * is provided by the caller, so we know it's alive | |
1096 | * and kicking, and don't take an extra reference. | |
1097 | */ | |
1098 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1099 | |
5ac8fb31 JW |
1100 | if (css == &root->css) |
1101 | break; | |
14067bb3 | 1102 | |
b2052564 | 1103 | if (css_tryget(css)) { |
5ac8fb31 JW |
1104 | /* |
1105 | * Make sure the memcg is initialized: | |
1106 | * mem_cgroup_css_online() orders the the | |
1107 | * initialization against setting the flag. | |
1108 | */ | |
1109 | if (smp_load_acquire(&memcg->initialized)) | |
1110 | break; | |
542f85f9 | 1111 | |
5ac8fb31 | 1112 | css_put(css); |
527a5ec9 | 1113 | } |
9f3a0d09 | 1114 | |
5ac8fb31 | 1115 | memcg = NULL; |
9f3a0d09 | 1116 | } |
5ac8fb31 JW |
1117 | |
1118 | if (reclaim) { | |
1119 | if (cmpxchg(&iter->position, pos, memcg) == pos) { | |
1120 | if (memcg) | |
1121 | css_get(&memcg->css); | |
1122 | if (pos) | |
1123 | css_put(&pos->css); | |
1124 | } | |
1125 | ||
1126 | /* | |
1127 | * pairs with css_tryget when dereferencing iter->position | |
1128 | * above. | |
1129 | */ | |
1130 | if (pos) | |
1131 | css_put(&pos->css); | |
1132 | ||
1133 | if (!memcg) | |
1134 | iter->generation++; | |
1135 | else if (!prev) | |
1136 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1137 | } |
5ac8fb31 | 1138 | |
542f85f9 MH |
1139 | out_unlock: |
1140 | rcu_read_unlock(); | |
5ac8fb31 | 1141 | out: |
c40046f3 MH |
1142 | if (prev && prev != root) |
1143 | css_put(&prev->css); | |
1144 | ||
9f3a0d09 | 1145 | return memcg; |
14067bb3 | 1146 | } |
7d74b06f | 1147 | |
5660048c JW |
1148 | /** |
1149 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1150 | * @root: hierarchy root | |
1151 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1152 | */ | |
1153 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1154 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1155 | { |
1156 | if (!root) | |
1157 | root = root_mem_cgroup; | |
1158 | if (prev && prev != root) | |
1159 | css_put(&prev->css); | |
1160 | } | |
7d74b06f | 1161 | |
9f3a0d09 JW |
1162 | /* |
1163 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1164 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1165 | * be used for reference counting. | |
1166 | */ | |
1167 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1168 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1169 | iter != NULL; \ |
527a5ec9 | 1170 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1171 | |
9f3a0d09 | 1172 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1173 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1174 | iter != NULL; \ |
527a5ec9 | 1175 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1176 | |
68ae564b | 1177 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1178 | { |
c0ff4b85 | 1179 | struct mem_cgroup *memcg; |
456f998e | 1180 | |
456f998e | 1181 | rcu_read_lock(); |
c0ff4b85 R |
1182 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1183 | if (unlikely(!memcg)) | |
456f998e YH |
1184 | goto out; |
1185 | ||
1186 | switch (idx) { | |
456f998e | 1187 | case PGFAULT: |
0e574a93 JW |
1188 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1189 | break; | |
1190 | case PGMAJFAULT: | |
1191 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1192 | break; |
1193 | default: | |
1194 | BUG(); | |
1195 | } | |
1196 | out: | |
1197 | rcu_read_unlock(); | |
1198 | } | |
68ae564b | 1199 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1200 | |
925b7673 JW |
1201 | /** |
1202 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1203 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1204 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1205 | * |
1206 | * Returns the lru list vector holding pages for the given @zone and | |
1207 | * @mem. This can be the global zone lruvec, if the memory controller | |
1208 | * is disabled. | |
1209 | */ | |
1210 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1211 | struct mem_cgroup *memcg) | |
1212 | { | |
1213 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1214 | struct lruvec *lruvec; |
925b7673 | 1215 | |
bea8c150 HD |
1216 | if (mem_cgroup_disabled()) { |
1217 | lruvec = &zone->lruvec; | |
1218 | goto out; | |
1219 | } | |
925b7673 | 1220 | |
e231875b | 1221 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1222 | lruvec = &mz->lruvec; |
1223 | out: | |
1224 | /* | |
1225 | * Since a node can be onlined after the mem_cgroup was created, | |
1226 | * we have to be prepared to initialize lruvec->zone here; | |
1227 | * and if offlined then reonlined, we need to reinitialize it. | |
1228 | */ | |
1229 | if (unlikely(lruvec->zone != zone)) | |
1230 | lruvec->zone = zone; | |
1231 | return lruvec; | |
925b7673 JW |
1232 | } |
1233 | ||
925b7673 | 1234 | /** |
dfe0e773 | 1235 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1236 | * @page: the page |
fa9add64 | 1237 | * @zone: zone of the page |
dfe0e773 JW |
1238 | * |
1239 | * This function is only safe when following the LRU page isolation | |
1240 | * and putback protocol: the LRU lock must be held, and the page must | |
1241 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1242 | */ |
fa9add64 | 1243 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1244 | { |
08e552c6 | 1245 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1246 | struct mem_cgroup *memcg; |
bea8c150 | 1247 | struct lruvec *lruvec; |
6d12e2d8 | 1248 | |
bea8c150 HD |
1249 | if (mem_cgroup_disabled()) { |
1250 | lruvec = &zone->lruvec; | |
1251 | goto out; | |
1252 | } | |
925b7673 | 1253 | |
1306a85a | 1254 | memcg = page->mem_cgroup; |
7512102c | 1255 | /* |
dfe0e773 | 1256 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1257 | * possibly migrated - before they are charged. |
7512102c | 1258 | */ |
29833315 JW |
1259 | if (!memcg) |
1260 | memcg = root_mem_cgroup; | |
7512102c | 1261 | |
e231875b | 1262 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1263 | lruvec = &mz->lruvec; |
1264 | out: | |
1265 | /* | |
1266 | * Since a node can be onlined after the mem_cgroup was created, | |
1267 | * we have to be prepared to initialize lruvec->zone here; | |
1268 | * and if offlined then reonlined, we need to reinitialize it. | |
1269 | */ | |
1270 | if (unlikely(lruvec->zone != zone)) | |
1271 | lruvec->zone = zone; | |
1272 | return lruvec; | |
08e552c6 | 1273 | } |
b69408e8 | 1274 | |
925b7673 | 1275 | /** |
fa9add64 HD |
1276 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1277 | * @lruvec: mem_cgroup per zone lru vector | |
1278 | * @lru: index of lru list the page is sitting on | |
1279 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1280 | * |
fa9add64 HD |
1281 | * This function must be called when a page is added to or removed from an |
1282 | * lru list. | |
3f58a829 | 1283 | */ |
fa9add64 HD |
1284 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1285 | int nr_pages) | |
3f58a829 MK |
1286 | { |
1287 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1288 | unsigned long *lru_size; |
3f58a829 MK |
1289 | |
1290 | if (mem_cgroup_disabled()) | |
1291 | return; | |
1292 | ||
fa9add64 HD |
1293 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1294 | lru_size = mz->lru_size + lru; | |
1295 | *lru_size += nr_pages; | |
1296 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1297 | } |
544122e5 | 1298 | |
2314b42d | 1299 | bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root) |
3e92041d | 1300 | { |
2314b42d | 1301 | if (root == memcg) |
91c63734 | 1302 | return true; |
2314b42d | 1303 | if (!root->use_hierarchy) |
91c63734 | 1304 | return false; |
2314b42d | 1305 | return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); |
c3ac9a8a JW |
1306 | } |
1307 | ||
2314b42d | 1308 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1309 | { |
2314b42d | 1310 | struct mem_cgroup *task_memcg; |
158e0a2d | 1311 | struct task_struct *p; |
ffbdccf5 | 1312 | bool ret; |
4c4a2214 | 1313 | |
158e0a2d | 1314 | p = find_lock_task_mm(task); |
de077d22 | 1315 | if (p) { |
2314b42d | 1316 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1317 | task_unlock(p); |
1318 | } else { | |
1319 | /* | |
1320 | * All threads may have already detached their mm's, but the oom | |
1321 | * killer still needs to detect if they have already been oom | |
1322 | * killed to prevent needlessly killing additional tasks. | |
1323 | */ | |
ffbdccf5 | 1324 | rcu_read_lock(); |
2314b42d JW |
1325 | task_memcg = mem_cgroup_from_task(task); |
1326 | css_get(&task_memcg->css); | |
ffbdccf5 | 1327 | rcu_read_unlock(); |
de077d22 | 1328 | } |
2314b42d JW |
1329 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1330 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1331 | return ret; |
1332 | } | |
1333 | ||
c56d5c7d | 1334 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1335 | { |
9b272977 | 1336 | unsigned long inactive_ratio; |
14797e23 | 1337 | unsigned long inactive; |
9b272977 | 1338 | unsigned long active; |
c772be93 | 1339 | unsigned long gb; |
14797e23 | 1340 | |
4d7dcca2 HD |
1341 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1342 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1343 | |
c772be93 KM |
1344 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1345 | if (gb) | |
1346 | inactive_ratio = int_sqrt(10 * gb); | |
1347 | else | |
1348 | inactive_ratio = 1; | |
1349 | ||
9b272977 | 1350 | return inactive * inactive_ratio < active; |
14797e23 KM |
1351 | } |
1352 | ||
90cbc250 VD |
1353 | bool mem_cgroup_lruvec_online(struct lruvec *lruvec) |
1354 | { | |
1355 | struct mem_cgroup_per_zone *mz; | |
1356 | struct mem_cgroup *memcg; | |
1357 | ||
1358 | if (mem_cgroup_disabled()) | |
1359 | return true; | |
1360 | ||
1361 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
1362 | memcg = mz->memcg; | |
1363 | ||
1364 | return !!(memcg->css.flags & CSS_ONLINE); | |
1365 | } | |
1366 | ||
3e32cb2e | 1367 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1368 | container_of(counter, struct mem_cgroup, member) |
1369 | ||
19942822 | 1370 | /** |
9d11ea9f | 1371 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1372 | * @memcg: the memory cgroup |
19942822 | 1373 | * |
9d11ea9f | 1374 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1375 | * pages. |
19942822 | 1376 | */ |
c0ff4b85 | 1377 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1378 | { |
3e32cb2e JW |
1379 | unsigned long margin = 0; |
1380 | unsigned long count; | |
1381 | unsigned long limit; | |
9d11ea9f | 1382 | |
3e32cb2e JW |
1383 | count = page_counter_read(&memcg->memory); |
1384 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1385 | if (count < limit) | |
1386 | margin = limit - count; | |
1387 | ||
1388 | if (do_swap_account) { | |
1389 | count = page_counter_read(&memcg->memsw); | |
1390 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1391 | if (count <= limit) | |
1392 | margin = min(margin, limit - count); | |
1393 | } | |
1394 | ||
1395 | return margin; | |
19942822 JW |
1396 | } |
1397 | ||
1f4c025b | 1398 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1399 | { |
a7885eb8 | 1400 | /* root ? */ |
14208b0e | 1401 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1402 | return vm_swappiness; |
1403 | ||
bf1ff263 | 1404 | return memcg->swappiness; |
a7885eb8 KM |
1405 | } |
1406 | ||
32047e2a | 1407 | /* |
bdcbb659 | 1408 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1409 | * |
bdcbb659 QH |
1410 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1411 | * moving cgroups. This is for waiting at high-memory pressure | |
1412 | * caused by "move". | |
32047e2a | 1413 | */ |
c0ff4b85 | 1414 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1415 | { |
2bd9bb20 KH |
1416 | struct mem_cgroup *from; |
1417 | struct mem_cgroup *to; | |
4b534334 | 1418 | bool ret = false; |
2bd9bb20 KH |
1419 | /* |
1420 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1421 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1422 | */ | |
1423 | spin_lock(&mc.lock); | |
1424 | from = mc.from; | |
1425 | to = mc.to; | |
1426 | if (!from) | |
1427 | goto unlock; | |
3e92041d | 1428 | |
2314b42d JW |
1429 | ret = mem_cgroup_is_descendant(from, memcg) || |
1430 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1431 | unlock: |
1432 | spin_unlock(&mc.lock); | |
4b534334 KH |
1433 | return ret; |
1434 | } | |
1435 | ||
c0ff4b85 | 1436 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1437 | { |
1438 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1439 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1440 | DEFINE_WAIT(wait); |
1441 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1442 | /* moving charge context might have finished. */ | |
1443 | if (mc.moving_task) | |
1444 | schedule(); | |
1445 | finish_wait(&mc.waitq, &wait); | |
1446 | return true; | |
1447 | } | |
1448 | } | |
1449 | return false; | |
1450 | } | |
1451 | ||
58cf188e | 1452 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1453 | /** |
58cf188e | 1454 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1455 | * @memcg: The memory cgroup that went over limit |
1456 | * @p: Task that is going to be killed | |
1457 | * | |
1458 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1459 | * enabled | |
1460 | */ | |
1461 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1462 | { | |
e61734c5 | 1463 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1464 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1465 | struct mem_cgroup *iter; |
1466 | unsigned int i; | |
e222432b | 1467 | |
58cf188e | 1468 | if (!p) |
e222432b BS |
1469 | return; |
1470 | ||
08088cb9 | 1471 | mutex_lock(&oom_info_lock); |
e222432b BS |
1472 | rcu_read_lock(); |
1473 | ||
e61734c5 TH |
1474 | pr_info("Task in "); |
1475 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
0346dadb | 1476 | pr_cont(" killed as a result of limit of "); |
e61734c5 | 1477 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1478 | pr_cont("\n"); |
e222432b | 1479 | |
e222432b BS |
1480 | rcu_read_unlock(); |
1481 | ||
3e32cb2e JW |
1482 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1483 | K((u64)page_counter_read(&memcg->memory)), | |
1484 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1485 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1486 | K((u64)page_counter_read(&memcg->memsw)), | |
1487 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1488 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1489 | K((u64)page_counter_read(&memcg->kmem)), | |
1490 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1491 | |
1492 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1493 | pr_info("Memory cgroup stats for "); |
1494 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1495 | pr_cont(":"); |
1496 | ||
1497 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1498 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1499 | continue; | |
1500 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1501 | K(mem_cgroup_read_stat(iter, i))); | |
1502 | } | |
1503 | ||
1504 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1505 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1506 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1507 | ||
1508 | pr_cont("\n"); | |
1509 | } | |
08088cb9 | 1510 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1511 | } |
1512 | ||
81d39c20 KH |
1513 | /* |
1514 | * This function returns the number of memcg under hierarchy tree. Returns | |
1515 | * 1(self count) if no children. | |
1516 | */ | |
c0ff4b85 | 1517 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1518 | { |
1519 | int num = 0; | |
7d74b06f KH |
1520 | struct mem_cgroup *iter; |
1521 | ||
c0ff4b85 | 1522 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1523 | num++; |
81d39c20 KH |
1524 | return num; |
1525 | } | |
1526 | ||
a63d83f4 DR |
1527 | /* |
1528 | * Return the memory (and swap, if configured) limit for a memcg. | |
1529 | */ | |
3e32cb2e | 1530 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1531 | { |
3e32cb2e | 1532 | unsigned long limit; |
f3e8eb70 | 1533 | |
3e32cb2e | 1534 | limit = memcg->memory.limit; |
9a5a8f19 | 1535 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1536 | unsigned long memsw_limit; |
9a5a8f19 | 1537 | |
3e32cb2e JW |
1538 | memsw_limit = memcg->memsw.limit; |
1539 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1540 | } |
9a5a8f19 | 1541 | return limit; |
a63d83f4 DR |
1542 | } |
1543 | ||
19965460 DR |
1544 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1545 | int order) | |
9cbb78bb DR |
1546 | { |
1547 | struct mem_cgroup *iter; | |
1548 | unsigned long chosen_points = 0; | |
1549 | unsigned long totalpages; | |
1550 | unsigned int points = 0; | |
1551 | struct task_struct *chosen = NULL; | |
1552 | ||
876aafbf | 1553 | /* |
465adcf1 DR |
1554 | * If current has a pending SIGKILL or is exiting, then automatically |
1555 | * select it. The goal is to allow it to allocate so that it may | |
1556 | * quickly exit and free its memory. | |
876aafbf | 1557 | */ |
d003f371 | 1558 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
49550b60 | 1559 | mark_tsk_oom_victim(current); |
876aafbf DR |
1560 | return; |
1561 | } | |
1562 | ||
1563 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1564 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1565 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1566 | struct css_task_iter it; |
9cbb78bb DR |
1567 | struct task_struct *task; |
1568 | ||
72ec7029 TH |
1569 | css_task_iter_start(&iter->css, &it); |
1570 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1571 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1572 | false)) { | |
1573 | case OOM_SCAN_SELECT: | |
1574 | if (chosen) | |
1575 | put_task_struct(chosen); | |
1576 | chosen = task; | |
1577 | chosen_points = ULONG_MAX; | |
1578 | get_task_struct(chosen); | |
1579 | /* fall through */ | |
1580 | case OOM_SCAN_CONTINUE: | |
1581 | continue; | |
1582 | case OOM_SCAN_ABORT: | |
72ec7029 | 1583 | css_task_iter_end(&it); |
9cbb78bb DR |
1584 | mem_cgroup_iter_break(memcg, iter); |
1585 | if (chosen) | |
1586 | put_task_struct(chosen); | |
1587 | return; | |
1588 | case OOM_SCAN_OK: | |
1589 | break; | |
1590 | }; | |
1591 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1592 | if (!points || points < chosen_points) |
1593 | continue; | |
1594 | /* Prefer thread group leaders for display purposes */ | |
1595 | if (points == chosen_points && | |
1596 | thread_group_leader(chosen)) | |
1597 | continue; | |
1598 | ||
1599 | if (chosen) | |
1600 | put_task_struct(chosen); | |
1601 | chosen = task; | |
1602 | chosen_points = points; | |
1603 | get_task_struct(chosen); | |
9cbb78bb | 1604 | } |
72ec7029 | 1605 | css_task_iter_end(&it); |
9cbb78bb DR |
1606 | } |
1607 | ||
1608 | if (!chosen) | |
1609 | return; | |
1610 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1611 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1612 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1613 | } |
1614 | ||
ae6e71d3 MC |
1615 | #if MAX_NUMNODES > 1 |
1616 | ||
4d0c066d KH |
1617 | /** |
1618 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1619 | * @memcg: the target memcg |
4d0c066d KH |
1620 | * @nid: the node ID to be checked. |
1621 | * @noswap : specify true here if the user wants flle only information. | |
1622 | * | |
1623 | * This function returns whether the specified memcg contains any | |
1624 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1625 | * pages in the node. | |
1626 | */ | |
c0ff4b85 | 1627 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1628 | int nid, bool noswap) |
1629 | { | |
c0ff4b85 | 1630 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1631 | return true; |
1632 | if (noswap || !total_swap_pages) | |
1633 | return false; | |
c0ff4b85 | 1634 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1635 | return true; |
1636 | return false; | |
1637 | ||
1638 | } | |
889976db YH |
1639 | |
1640 | /* | |
1641 | * Always updating the nodemask is not very good - even if we have an empty | |
1642 | * list or the wrong list here, we can start from some node and traverse all | |
1643 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1644 | * | |
1645 | */ | |
c0ff4b85 | 1646 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1647 | { |
1648 | int nid; | |
453a9bf3 KH |
1649 | /* |
1650 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1651 | * pagein/pageout changes since the last update. | |
1652 | */ | |
c0ff4b85 | 1653 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1654 | return; |
c0ff4b85 | 1655 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1656 | return; |
1657 | ||
889976db | 1658 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1659 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1660 | |
31aaea4a | 1661 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1662 | |
c0ff4b85 R |
1663 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1664 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1665 | } |
453a9bf3 | 1666 | |
c0ff4b85 R |
1667 | atomic_set(&memcg->numainfo_events, 0); |
1668 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1669 | } |
1670 | ||
1671 | /* | |
1672 | * Selecting a node where we start reclaim from. Because what we need is just | |
1673 | * reducing usage counter, start from anywhere is O,K. Considering | |
1674 | * memory reclaim from current node, there are pros. and cons. | |
1675 | * | |
1676 | * Freeing memory from current node means freeing memory from a node which | |
1677 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1678 | * hit limits, it will see a contention on a node. But freeing from remote | |
1679 | * node means more costs for memory reclaim because of memory latency. | |
1680 | * | |
1681 | * Now, we use round-robin. Better algorithm is welcomed. | |
1682 | */ | |
c0ff4b85 | 1683 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1684 | { |
1685 | int node; | |
1686 | ||
c0ff4b85 R |
1687 | mem_cgroup_may_update_nodemask(memcg); |
1688 | node = memcg->last_scanned_node; | |
889976db | 1689 | |
c0ff4b85 | 1690 | node = next_node(node, memcg->scan_nodes); |
889976db | 1691 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1692 | node = first_node(memcg->scan_nodes); |
889976db YH |
1693 | /* |
1694 | * We call this when we hit limit, not when pages are added to LRU. | |
1695 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1696 | * memcg is too small and all pages are not on LRU. In that case, | |
1697 | * we use curret node. | |
1698 | */ | |
1699 | if (unlikely(node == MAX_NUMNODES)) | |
1700 | node = numa_node_id(); | |
1701 | ||
c0ff4b85 | 1702 | memcg->last_scanned_node = node; |
889976db YH |
1703 | return node; |
1704 | } | |
889976db | 1705 | #else |
c0ff4b85 | 1706 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1707 | { |
1708 | return 0; | |
1709 | } | |
1710 | #endif | |
1711 | ||
0608f43d AM |
1712 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1713 | struct zone *zone, | |
1714 | gfp_t gfp_mask, | |
1715 | unsigned long *total_scanned) | |
1716 | { | |
1717 | struct mem_cgroup *victim = NULL; | |
1718 | int total = 0; | |
1719 | int loop = 0; | |
1720 | unsigned long excess; | |
1721 | unsigned long nr_scanned; | |
1722 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1723 | .zone = zone, | |
1724 | .priority = 0, | |
1725 | }; | |
1726 | ||
3e32cb2e | 1727 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1728 | |
1729 | while (1) { | |
1730 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1731 | if (!victim) { | |
1732 | loop++; | |
1733 | if (loop >= 2) { | |
1734 | /* | |
1735 | * If we have not been able to reclaim | |
1736 | * anything, it might because there are | |
1737 | * no reclaimable pages under this hierarchy | |
1738 | */ | |
1739 | if (!total) | |
1740 | break; | |
1741 | /* | |
1742 | * We want to do more targeted reclaim. | |
1743 | * excess >> 2 is not to excessive so as to | |
1744 | * reclaim too much, nor too less that we keep | |
1745 | * coming back to reclaim from this cgroup | |
1746 | */ | |
1747 | if (total >= (excess >> 2) || | |
1748 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1749 | break; | |
1750 | } | |
1751 | continue; | |
1752 | } | |
0608f43d AM |
1753 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1754 | zone, &nr_scanned); | |
1755 | *total_scanned += nr_scanned; | |
3e32cb2e | 1756 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1757 | break; |
6d61ef40 | 1758 | } |
0608f43d AM |
1759 | mem_cgroup_iter_break(root_memcg, victim); |
1760 | return total; | |
6d61ef40 BS |
1761 | } |
1762 | ||
0056f4e6 JW |
1763 | #ifdef CONFIG_LOCKDEP |
1764 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1765 | .name = "memcg_oom_lock", | |
1766 | }; | |
1767 | #endif | |
1768 | ||
fb2a6fc5 JW |
1769 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1770 | ||
867578cb KH |
1771 | /* |
1772 | * Check OOM-Killer is already running under our hierarchy. | |
1773 | * If someone is running, return false. | |
1774 | */ | |
fb2a6fc5 | 1775 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1776 | { |
79dfdacc | 1777 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1778 | |
fb2a6fc5 JW |
1779 | spin_lock(&memcg_oom_lock); |
1780 | ||
9f3a0d09 | 1781 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1782 | if (iter->oom_lock) { |
79dfdacc MH |
1783 | /* |
1784 | * this subtree of our hierarchy is already locked | |
1785 | * so we cannot give a lock. | |
1786 | */ | |
79dfdacc | 1787 | failed = iter; |
9f3a0d09 JW |
1788 | mem_cgroup_iter_break(memcg, iter); |
1789 | break; | |
23751be0 JW |
1790 | } else |
1791 | iter->oom_lock = true; | |
7d74b06f | 1792 | } |
867578cb | 1793 | |
fb2a6fc5 JW |
1794 | if (failed) { |
1795 | /* | |
1796 | * OK, we failed to lock the whole subtree so we have | |
1797 | * to clean up what we set up to the failing subtree | |
1798 | */ | |
1799 | for_each_mem_cgroup_tree(iter, memcg) { | |
1800 | if (iter == failed) { | |
1801 | mem_cgroup_iter_break(memcg, iter); | |
1802 | break; | |
1803 | } | |
1804 | iter->oom_lock = false; | |
79dfdacc | 1805 | } |
0056f4e6 JW |
1806 | } else |
1807 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1808 | |
1809 | spin_unlock(&memcg_oom_lock); | |
1810 | ||
1811 | return !failed; | |
a636b327 | 1812 | } |
0b7f569e | 1813 | |
fb2a6fc5 | 1814 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1815 | { |
7d74b06f KH |
1816 | struct mem_cgroup *iter; |
1817 | ||
fb2a6fc5 | 1818 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1819 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1820 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1821 | iter->oom_lock = false; |
fb2a6fc5 | 1822 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1823 | } |
1824 | ||
c0ff4b85 | 1825 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1826 | { |
1827 | struct mem_cgroup *iter; | |
1828 | ||
c0ff4b85 | 1829 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1830 | atomic_inc(&iter->under_oom); |
1831 | } | |
1832 | ||
c0ff4b85 | 1833 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1834 | { |
1835 | struct mem_cgroup *iter; | |
1836 | ||
867578cb KH |
1837 | /* |
1838 | * When a new child is created while the hierarchy is under oom, | |
1839 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1840 | * atomic_add_unless() here. | |
1841 | */ | |
c0ff4b85 | 1842 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1843 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1844 | } |
1845 | ||
867578cb KH |
1846 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1847 | ||
dc98df5a | 1848 | struct oom_wait_info { |
d79154bb | 1849 | struct mem_cgroup *memcg; |
dc98df5a KH |
1850 | wait_queue_t wait; |
1851 | }; | |
1852 | ||
1853 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1854 | unsigned mode, int sync, void *arg) | |
1855 | { | |
d79154bb HD |
1856 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1857 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1858 | struct oom_wait_info *oom_wait_info; |
1859 | ||
1860 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1861 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1862 | |
2314b42d JW |
1863 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1864 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1865 | return 0; |
dc98df5a KH |
1866 | return autoremove_wake_function(wait, mode, sync, arg); |
1867 | } | |
1868 | ||
c0ff4b85 | 1869 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1870 | { |
3812c8c8 | 1871 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
1872 | /* for filtering, pass "memcg" as argument. */ |
1873 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1874 | } |
1875 | ||
c0ff4b85 | 1876 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1877 | { |
c0ff4b85 R |
1878 | if (memcg && atomic_read(&memcg->under_oom)) |
1879 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1880 | } |
1881 | ||
3812c8c8 | 1882 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1883 | { |
3812c8c8 JW |
1884 | if (!current->memcg_oom.may_oom) |
1885 | return; | |
867578cb | 1886 | /* |
49426420 JW |
1887 | * We are in the middle of the charge context here, so we |
1888 | * don't want to block when potentially sitting on a callstack | |
1889 | * that holds all kinds of filesystem and mm locks. | |
1890 | * | |
1891 | * Also, the caller may handle a failed allocation gracefully | |
1892 | * (like optional page cache readahead) and so an OOM killer | |
1893 | * invocation might not even be necessary. | |
1894 | * | |
1895 | * That's why we don't do anything here except remember the | |
1896 | * OOM context and then deal with it at the end of the page | |
1897 | * fault when the stack is unwound, the locks are released, | |
1898 | * and when we know whether the fault was overall successful. | |
867578cb | 1899 | */ |
49426420 JW |
1900 | css_get(&memcg->css); |
1901 | current->memcg_oom.memcg = memcg; | |
1902 | current->memcg_oom.gfp_mask = mask; | |
1903 | current->memcg_oom.order = order; | |
3812c8c8 JW |
1904 | } |
1905 | ||
1906 | /** | |
1907 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1908 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1909 | * |
49426420 JW |
1910 | * This has to be called at the end of a page fault if the memcg OOM |
1911 | * handler was enabled. | |
3812c8c8 | 1912 | * |
49426420 | 1913 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1914 | * sleep on a waitqueue until the userspace task resolves the |
1915 | * situation. Sleeping directly in the charge context with all kinds | |
1916 | * of locks held is not a good idea, instead we remember an OOM state | |
1917 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1918 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1919 | * |
1920 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1921 | * completed, %false otherwise. |
3812c8c8 | 1922 | */ |
49426420 | 1923 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1924 | { |
49426420 | 1925 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 1926 | struct oom_wait_info owait; |
49426420 | 1927 | bool locked; |
3812c8c8 JW |
1928 | |
1929 | /* OOM is global, do not handle */ | |
3812c8c8 | 1930 | if (!memcg) |
49426420 | 1931 | return false; |
3812c8c8 | 1932 | |
c32b3cbe | 1933 | if (!handle || oom_killer_disabled) |
49426420 | 1934 | goto cleanup; |
3812c8c8 JW |
1935 | |
1936 | owait.memcg = memcg; | |
1937 | owait.wait.flags = 0; | |
1938 | owait.wait.func = memcg_oom_wake_function; | |
1939 | owait.wait.private = current; | |
1940 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1941 | |
3812c8c8 | 1942 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1943 | mem_cgroup_mark_under_oom(memcg); |
1944 | ||
1945 | locked = mem_cgroup_oom_trylock(memcg); | |
1946 | ||
1947 | if (locked) | |
1948 | mem_cgroup_oom_notify(memcg); | |
1949 | ||
1950 | if (locked && !memcg->oom_kill_disable) { | |
1951 | mem_cgroup_unmark_under_oom(memcg); | |
1952 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1953 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
1954 | current->memcg_oom.order); | |
1955 | } else { | |
3812c8c8 | 1956 | schedule(); |
49426420 JW |
1957 | mem_cgroup_unmark_under_oom(memcg); |
1958 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1959 | } | |
1960 | ||
1961 | if (locked) { | |
fb2a6fc5 JW |
1962 | mem_cgroup_oom_unlock(memcg); |
1963 | /* | |
1964 | * There is no guarantee that an OOM-lock contender | |
1965 | * sees the wakeups triggered by the OOM kill | |
1966 | * uncharges. Wake any sleepers explicitely. | |
1967 | */ | |
1968 | memcg_oom_recover(memcg); | |
1969 | } | |
49426420 JW |
1970 | cleanup: |
1971 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 1972 | css_put(&memcg->css); |
867578cb | 1973 | return true; |
0b7f569e KH |
1974 | } |
1975 | ||
d7365e78 JW |
1976 | /** |
1977 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1978 | * @page: page that is going to change accounted state | |
32047e2a | 1979 | * |
d7365e78 JW |
1980 | * This function must mark the beginning of an accounted page state |
1981 | * change to prevent double accounting when the page is concurrently | |
1982 | * being moved to another memcg: | |
32047e2a | 1983 | * |
6de22619 | 1984 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1985 | * if (TestClearPageState(page)) |
1986 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1987 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1988 | */ |
6de22619 | 1989 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1990 | { |
1991 | struct mem_cgroup *memcg; | |
6de22619 | 1992 | unsigned long flags; |
89c06bd5 | 1993 | |
6de22619 JW |
1994 | /* |
1995 | * The RCU lock is held throughout the transaction. The fast | |
1996 | * path can get away without acquiring the memcg->move_lock | |
1997 | * because page moving starts with an RCU grace period. | |
1998 | * | |
1999 | * The RCU lock also protects the memcg from being freed when | |
2000 | * the page state that is going to change is the only thing | |
2001 | * preventing the page from being uncharged. | |
2002 | * E.g. end-writeback clearing PageWriteback(), which allows | |
2003 | * migration to go ahead and uncharge the page before the | |
2004 | * account transaction might be complete. | |
2005 | */ | |
d7365e78 JW |
2006 | rcu_read_lock(); |
2007 | ||
2008 | if (mem_cgroup_disabled()) | |
2009 | return NULL; | |
89c06bd5 | 2010 | again: |
1306a85a | 2011 | memcg = page->mem_cgroup; |
29833315 | 2012 | if (unlikely(!memcg)) |
d7365e78 JW |
2013 | return NULL; |
2014 | ||
bdcbb659 | 2015 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2016 | return memcg; |
89c06bd5 | 2017 | |
6de22619 | 2018 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 2019 | if (memcg != page->mem_cgroup) { |
6de22619 | 2020 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2021 | goto again; |
2022 | } | |
6de22619 JW |
2023 | |
2024 | /* | |
2025 | * When charge migration first begins, we can have locked and | |
2026 | * unlocked page stat updates happening concurrently. Track | |
2027 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
2028 | */ | |
2029 | memcg->move_lock_task = current; | |
2030 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
2031 | |
2032 | return memcg; | |
89c06bd5 KH |
2033 | } |
2034 | ||
d7365e78 JW |
2035 | /** |
2036 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2037 | * @memcg: the memcg that was accounted against | |
d7365e78 | 2038 | */ |
6de22619 | 2039 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 2040 | { |
6de22619 JW |
2041 | if (memcg && memcg->move_lock_task == current) { |
2042 | unsigned long flags = memcg->move_lock_flags; | |
2043 | ||
2044 | memcg->move_lock_task = NULL; | |
2045 | memcg->move_lock_flags = 0; | |
2046 | ||
2047 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2048 | } | |
89c06bd5 | 2049 | |
d7365e78 | 2050 | rcu_read_unlock(); |
89c06bd5 KH |
2051 | } |
2052 | ||
d7365e78 JW |
2053 | /** |
2054 | * mem_cgroup_update_page_stat - update page state statistics | |
2055 | * @memcg: memcg to account against | |
2056 | * @idx: page state item to account | |
2057 | * @val: number of pages (positive or negative) | |
2058 | * | |
2059 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2060 | */ | |
2061 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2062 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2063 | { |
658b72c5 | 2064 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2065 | |
d7365e78 JW |
2066 | if (memcg) |
2067 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2068 | } |
26174efd | 2069 | |
cdec2e42 KH |
2070 | /* |
2071 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2072 | * TODO: maybe necessary to use big numbers in big irons. | |
2073 | */ | |
7ec99d62 | 2074 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2075 | struct memcg_stock_pcp { |
2076 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2077 | unsigned int nr_pages; |
cdec2e42 | 2078 | struct work_struct work; |
26fe6168 | 2079 | unsigned long flags; |
a0db00fc | 2080 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2081 | }; |
2082 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2083 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2084 | |
a0956d54 SS |
2085 | /** |
2086 | * consume_stock: Try to consume stocked charge on this cpu. | |
2087 | * @memcg: memcg to consume from. | |
2088 | * @nr_pages: how many pages to charge. | |
2089 | * | |
2090 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2091 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2092 | * service an allocation will refill the stock. | |
2093 | * | |
2094 | * returns true if successful, false otherwise. | |
cdec2e42 | 2095 | */ |
a0956d54 | 2096 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2097 | { |
2098 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2099 | bool ret = false; |
cdec2e42 | 2100 | |
a0956d54 | 2101 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2102 | return ret; |
a0956d54 | 2103 | |
cdec2e42 | 2104 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2105 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2106 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2107 | ret = true; |
2108 | } | |
cdec2e42 KH |
2109 | put_cpu_var(memcg_stock); |
2110 | return ret; | |
2111 | } | |
2112 | ||
2113 | /* | |
3e32cb2e | 2114 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2115 | */ |
2116 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2117 | { | |
2118 | struct mem_cgroup *old = stock->cached; | |
2119 | ||
11c9ea4e | 2120 | if (stock->nr_pages) { |
3e32cb2e | 2121 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2122 | if (do_swap_account) |
3e32cb2e | 2123 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2124 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2125 | stock->nr_pages = 0; |
cdec2e42 KH |
2126 | } |
2127 | stock->cached = NULL; | |
cdec2e42 KH |
2128 | } |
2129 | ||
2130 | /* | |
2131 | * This must be called under preempt disabled or must be called by | |
2132 | * a thread which is pinned to local cpu. | |
2133 | */ | |
2134 | static void drain_local_stock(struct work_struct *dummy) | |
2135 | { | |
7c8e0181 | 2136 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2137 | drain_stock(stock); |
26fe6168 | 2138 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2139 | } |
2140 | ||
e4777496 MH |
2141 | static void __init memcg_stock_init(void) |
2142 | { | |
2143 | int cpu; | |
2144 | ||
2145 | for_each_possible_cpu(cpu) { | |
2146 | struct memcg_stock_pcp *stock = | |
2147 | &per_cpu(memcg_stock, cpu); | |
2148 | INIT_WORK(&stock->work, drain_local_stock); | |
2149 | } | |
2150 | } | |
2151 | ||
cdec2e42 | 2152 | /* |
3e32cb2e | 2153 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2154 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2155 | */ |
c0ff4b85 | 2156 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2157 | { |
2158 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2159 | ||
c0ff4b85 | 2160 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2161 | drain_stock(stock); |
c0ff4b85 | 2162 | stock->cached = memcg; |
cdec2e42 | 2163 | } |
11c9ea4e | 2164 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2165 | put_cpu_var(memcg_stock); |
2166 | } | |
2167 | ||
2168 | /* | |
c0ff4b85 | 2169 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2170 | * of the hierarchy under it. |
cdec2e42 | 2171 | */ |
6d3d6aa2 | 2172 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2173 | { |
26fe6168 | 2174 | int cpu, curcpu; |
d38144b7 | 2175 | |
6d3d6aa2 JW |
2176 | /* If someone's already draining, avoid adding running more workers. */ |
2177 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2178 | return; | |
cdec2e42 | 2179 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2180 | get_online_cpus(); |
5af12d0e | 2181 | curcpu = get_cpu(); |
cdec2e42 KH |
2182 | for_each_online_cpu(cpu) { |
2183 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2184 | struct mem_cgroup *memcg; |
26fe6168 | 2185 | |
c0ff4b85 R |
2186 | memcg = stock->cached; |
2187 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2188 | continue; |
2314b42d | 2189 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 2190 | continue; |
d1a05b69 MH |
2191 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2192 | if (cpu == curcpu) | |
2193 | drain_local_stock(&stock->work); | |
2194 | else | |
2195 | schedule_work_on(cpu, &stock->work); | |
2196 | } | |
cdec2e42 | 2197 | } |
5af12d0e | 2198 | put_cpu(); |
f894ffa8 | 2199 | put_online_cpus(); |
9f50fad6 | 2200 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2201 | } |
2202 | ||
711d3d2c KH |
2203 | /* |
2204 | * This function drains percpu counter value from DEAD cpu and | |
2205 | * move it to local cpu. Note that this function can be preempted. | |
2206 | */ | |
c0ff4b85 | 2207 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2208 | { |
2209 | int i; | |
2210 | ||
c0ff4b85 | 2211 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2212 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2213 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2214 | |
c0ff4b85 R |
2215 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2216 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2217 | } |
e9f8974f | 2218 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2219 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2220 | |
c0ff4b85 R |
2221 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2222 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2223 | } |
c0ff4b85 | 2224 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2225 | } |
2226 | ||
0db0628d | 2227 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2228 | unsigned long action, |
2229 | void *hcpu) | |
2230 | { | |
2231 | int cpu = (unsigned long)hcpu; | |
2232 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2233 | struct mem_cgroup *iter; |
cdec2e42 | 2234 | |
619d094b | 2235 | if (action == CPU_ONLINE) |
1489ebad | 2236 | return NOTIFY_OK; |
1489ebad | 2237 | |
d833049b | 2238 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2239 | return NOTIFY_OK; |
711d3d2c | 2240 | |
9f3a0d09 | 2241 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2242 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2243 | ||
cdec2e42 KH |
2244 | stock = &per_cpu(memcg_stock, cpu); |
2245 | drain_stock(stock); | |
2246 | return NOTIFY_OK; | |
2247 | } | |
2248 | ||
00501b53 JW |
2249 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2250 | unsigned int nr_pages) | |
8a9f3ccd | 2251 | { |
7ec99d62 | 2252 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2253 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2254 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2255 | struct page_counter *counter; |
6539cc05 | 2256 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2257 | bool may_swap = true; |
2258 | bool drained = false; | |
05b84301 | 2259 | int ret = 0; |
a636b327 | 2260 | |
ce00a967 JW |
2261 | if (mem_cgroup_is_root(memcg)) |
2262 | goto done; | |
6539cc05 | 2263 | retry: |
b6b6cc72 MH |
2264 | if (consume_stock(memcg, nr_pages)) |
2265 | goto done; | |
8a9f3ccd | 2266 | |
3fbe7244 | 2267 | if (!do_swap_account || |
3e32cb2e JW |
2268 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2269 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2270 | goto done_restock; |
3fbe7244 | 2271 | if (do_swap_account) |
3e32cb2e JW |
2272 | page_counter_uncharge(&memcg->memsw, batch); |
2273 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2274 | } else { |
3e32cb2e | 2275 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2276 | may_swap = false; |
3fbe7244 | 2277 | } |
7a81b88c | 2278 | |
6539cc05 JW |
2279 | if (batch > nr_pages) { |
2280 | batch = nr_pages; | |
2281 | goto retry; | |
2282 | } | |
6d61ef40 | 2283 | |
06b078fc JW |
2284 | /* |
2285 | * Unlike in global OOM situations, memcg is not in a physical | |
2286 | * memory shortage. Allow dying and OOM-killed tasks to | |
2287 | * bypass the last charges so that they can exit quickly and | |
2288 | * free their memory. | |
2289 | */ | |
2290 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2291 | fatal_signal_pending(current) || | |
2292 | current->flags & PF_EXITING)) | |
2293 | goto bypass; | |
2294 | ||
2295 | if (unlikely(task_in_memcg_oom(current))) | |
2296 | goto nomem; | |
2297 | ||
6539cc05 JW |
2298 | if (!(gfp_mask & __GFP_WAIT)) |
2299 | goto nomem; | |
4b534334 | 2300 | |
241994ed JW |
2301 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2302 | ||
b70a2a21 JW |
2303 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2304 | gfp_mask, may_swap); | |
6539cc05 | 2305 | |
61e02c74 | 2306 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2307 | goto retry; |
28c34c29 | 2308 | |
b70a2a21 | 2309 | if (!drained) { |
6d3d6aa2 | 2310 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2311 | drained = true; |
2312 | goto retry; | |
2313 | } | |
2314 | ||
28c34c29 JW |
2315 | if (gfp_mask & __GFP_NORETRY) |
2316 | goto nomem; | |
6539cc05 JW |
2317 | /* |
2318 | * Even though the limit is exceeded at this point, reclaim | |
2319 | * may have been able to free some pages. Retry the charge | |
2320 | * before killing the task. | |
2321 | * | |
2322 | * Only for regular pages, though: huge pages are rather | |
2323 | * unlikely to succeed so close to the limit, and we fall back | |
2324 | * to regular pages anyway in case of failure. | |
2325 | */ | |
61e02c74 | 2326 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2327 | goto retry; |
2328 | /* | |
2329 | * At task move, charge accounts can be doubly counted. So, it's | |
2330 | * better to wait until the end of task_move if something is going on. | |
2331 | */ | |
2332 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2333 | goto retry; | |
2334 | ||
9b130619 JW |
2335 | if (nr_retries--) |
2336 | goto retry; | |
2337 | ||
06b078fc JW |
2338 | if (gfp_mask & __GFP_NOFAIL) |
2339 | goto bypass; | |
2340 | ||
6539cc05 JW |
2341 | if (fatal_signal_pending(current)) |
2342 | goto bypass; | |
2343 | ||
241994ed JW |
2344 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2345 | ||
61e02c74 | 2346 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2347 | nomem: |
6d1fdc48 | 2348 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2349 | return -ENOMEM; |
867578cb | 2350 | bypass: |
ce00a967 | 2351 | return -EINTR; |
6539cc05 JW |
2352 | |
2353 | done_restock: | |
e8ea14cc | 2354 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2355 | if (batch > nr_pages) |
2356 | refill_stock(memcg, batch - nr_pages); | |
241994ed JW |
2357 | /* |
2358 | * If the hierarchy is above the normal consumption range, | |
2359 | * make the charging task trim their excess contribution. | |
2360 | */ | |
2361 | do { | |
2362 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2363 | continue; | |
2364 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
2365 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
2366 | } while ((memcg = parent_mem_cgroup(memcg))); | |
6539cc05 | 2367 | done: |
05b84301 | 2368 | return ret; |
7a81b88c | 2369 | } |
8a9f3ccd | 2370 | |
00501b53 | 2371 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2372 | { |
ce00a967 JW |
2373 | if (mem_cgroup_is_root(memcg)) |
2374 | return; | |
2375 | ||
3e32cb2e | 2376 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2377 | if (do_swap_account) |
3e32cb2e | 2378 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2379 | |
e8ea14cc | 2380 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2381 | } |
2382 | ||
a3b2d692 KH |
2383 | /* |
2384 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2385 | * rcu_read_lock(). The caller is responsible for calling |
2386 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2387 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2388 | */ |
2389 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2390 | { | |
a3b2d692 KH |
2391 | /* ID 0 is unused ID */ |
2392 | if (!id) | |
2393 | return NULL; | |
34c00c31 | 2394 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2395 | } |
2396 | ||
0a31bc97 JW |
2397 | /* |
2398 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2399 | * @page: the page | |
2400 | * | |
2401 | * Look up, get a css reference, and return the memcg that owns @page. | |
2402 | * | |
2403 | * The page must be locked to prevent racing with swap-in and page | |
2404 | * cache charges. If coming from an unlocked page table, the caller | |
2405 | * must ensure the page is on the LRU or this can race with charging. | |
2406 | */ | |
e42d9d5d | 2407 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2408 | { |
29833315 | 2409 | struct mem_cgroup *memcg; |
a3b2d692 | 2410 | unsigned short id; |
b5a84319 KH |
2411 | swp_entry_t ent; |
2412 | ||
309381fe | 2413 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2414 | |
1306a85a | 2415 | memcg = page->mem_cgroup; |
29833315 JW |
2416 | if (memcg) { |
2417 | if (!css_tryget_online(&memcg->css)) | |
c0ff4b85 | 2418 | memcg = NULL; |
e42d9d5d | 2419 | } else if (PageSwapCache(page)) { |
3c776e64 | 2420 | ent.val = page_private(page); |
9fb4b7cc | 2421 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2422 | rcu_read_lock(); |
c0ff4b85 | 2423 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2424 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2425 | memcg = NULL; |
a3b2d692 | 2426 | rcu_read_unlock(); |
3c776e64 | 2427 | } |
c0ff4b85 | 2428 | return memcg; |
b5a84319 KH |
2429 | } |
2430 | ||
0a31bc97 JW |
2431 | static void lock_page_lru(struct page *page, int *isolated) |
2432 | { | |
2433 | struct zone *zone = page_zone(page); | |
2434 | ||
2435 | spin_lock_irq(&zone->lru_lock); | |
2436 | if (PageLRU(page)) { | |
2437 | struct lruvec *lruvec; | |
2438 | ||
2439 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2440 | ClearPageLRU(page); | |
2441 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2442 | *isolated = 1; | |
2443 | } else | |
2444 | *isolated = 0; | |
2445 | } | |
2446 | ||
2447 | static void unlock_page_lru(struct page *page, int isolated) | |
2448 | { | |
2449 | struct zone *zone = page_zone(page); | |
2450 | ||
2451 | if (isolated) { | |
2452 | struct lruvec *lruvec; | |
2453 | ||
2454 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2455 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2456 | SetPageLRU(page); | |
2457 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2458 | } | |
2459 | spin_unlock_irq(&zone->lru_lock); | |
2460 | } | |
2461 | ||
00501b53 | 2462 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2463 | bool lrucare) |
7a81b88c | 2464 | { |
0a31bc97 | 2465 | int isolated; |
9ce70c02 | 2466 | |
1306a85a | 2467 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2468 | |
2469 | /* | |
2470 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2471 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2472 | */ | |
0a31bc97 JW |
2473 | if (lrucare) |
2474 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2475 | |
0a31bc97 JW |
2476 | /* |
2477 | * Nobody should be changing or seriously looking at | |
1306a85a | 2478 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2479 | * |
2480 | * - the page is uncharged | |
2481 | * | |
2482 | * - the page is off-LRU | |
2483 | * | |
2484 | * - an anonymous fault has exclusive page access, except for | |
2485 | * a locked page table | |
2486 | * | |
2487 | * - a page cache insertion, a swapin fault, or a migration | |
2488 | * have the page locked | |
2489 | */ | |
1306a85a | 2490 | page->mem_cgroup = memcg; |
9ce70c02 | 2491 | |
0a31bc97 JW |
2492 | if (lrucare) |
2493 | unlock_page_lru(page, isolated); | |
7a81b88c | 2494 | } |
66e1707b | 2495 | |
7ae1e1d0 | 2496 | #ifdef CONFIG_MEMCG_KMEM |
dbf22eb6 VD |
2497 | int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2498 | unsigned long nr_pages) | |
7ae1e1d0 | 2499 | { |
3e32cb2e | 2500 | struct page_counter *counter; |
7ae1e1d0 | 2501 | int ret = 0; |
7ae1e1d0 | 2502 | |
3e32cb2e JW |
2503 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2504 | if (ret < 0) | |
7ae1e1d0 GC |
2505 | return ret; |
2506 | ||
3e32cb2e | 2507 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2508 | if (ret == -EINTR) { |
2509 | /* | |
00501b53 JW |
2510 | * try_charge() chose to bypass to root due to OOM kill or |
2511 | * fatal signal. Since our only options are to either fail | |
2512 | * the allocation or charge it to this cgroup, do it as a | |
2513 | * temporary condition. But we can't fail. From a kmem/slab | |
2514 | * perspective, the cache has already been selected, by | |
2515 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2516 | * our minds. |
2517 | * | |
2518 | * This condition will only trigger if the task entered | |
00501b53 JW |
2519 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2520 | * during try_charge() above. Tasks that were already dying | |
2521 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2522 | * directed to the root cgroup in memcontrol.h |
2523 | */ | |
3e32cb2e | 2524 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2525 | if (do_swap_account) |
3e32cb2e | 2526 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2527 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2528 | ret = 0; |
2529 | } else if (ret) | |
3e32cb2e | 2530 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2531 | |
2532 | return ret; | |
2533 | } | |
2534 | ||
dbf22eb6 | 2535 | void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages) |
7ae1e1d0 | 2536 | { |
3e32cb2e | 2537 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2538 | if (do_swap_account) |
3e32cb2e | 2539 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2540 | |
64f21993 | 2541 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7de37682 | 2542 | |
e8ea14cc | 2543 | css_put_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2544 | } |
2545 | ||
2633d7a0 GC |
2546 | /* |
2547 | * helper for acessing a memcg's index. It will be used as an index in the | |
2548 | * child cache array in kmem_cache, and also to derive its name. This function | |
2549 | * will return -1 when this is not a kmem-limited memcg. | |
2550 | */ | |
2551 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2552 | { | |
2553 | return memcg ? memcg->kmemcg_id : -1; | |
2554 | } | |
2555 | ||
f3bb3043 | 2556 | static int memcg_alloc_cache_id(void) |
55007d84 | 2557 | { |
f3bb3043 VD |
2558 | int id, size; |
2559 | int err; | |
2560 | ||
2561 | id = ida_simple_get(&kmem_limited_groups, | |
2562 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2563 | if (id < 0) | |
2564 | return id; | |
55007d84 | 2565 | |
f3bb3043 VD |
2566 | if (id < memcg_limited_groups_array_size) |
2567 | return id; | |
2568 | ||
2569 | /* | |
2570 | * There's no space for the new id in memcg_caches arrays, | |
2571 | * so we have to grow them. | |
2572 | */ | |
2573 | ||
2574 | size = 2 * (id + 1); | |
55007d84 GC |
2575 | if (size < MEMCG_CACHES_MIN_SIZE) |
2576 | size = MEMCG_CACHES_MIN_SIZE; | |
2577 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2578 | size = MEMCG_CACHES_MAX_SIZE; | |
2579 | ||
f3bb3043 | 2580 | err = memcg_update_all_caches(size); |
f3bb3043 VD |
2581 | if (err) { |
2582 | ida_simple_remove(&kmem_limited_groups, id); | |
2583 | return err; | |
2584 | } | |
2585 | return id; | |
2586 | } | |
2587 | ||
2588 | static void memcg_free_cache_id(int id) | |
2589 | { | |
2590 | ida_simple_remove(&kmem_limited_groups, id); | |
55007d84 GC |
2591 | } |
2592 | ||
2593 | /* | |
2594 | * We should update the current array size iff all caches updates succeed. This | |
2595 | * can only be done from the slab side. The slab mutex needs to be held when | |
2596 | * calling this. | |
2597 | */ | |
2598 | void memcg_update_array_size(int num) | |
2599 | { | |
f3bb3043 | 2600 | memcg_limited_groups_array_size = num; |
55007d84 GC |
2601 | } |
2602 | ||
d5b3cf71 | 2603 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2604 | struct mem_cgroup *memcg; |
2605 | struct kmem_cache *cachep; | |
2606 | struct work_struct work; | |
2607 | }; | |
2608 | ||
d5b3cf71 | 2609 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2610 | { |
d5b3cf71 VD |
2611 | struct memcg_kmem_cache_create_work *cw = |
2612 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2613 | struct mem_cgroup *memcg = cw->memcg; |
2614 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2615 | |
d5b3cf71 | 2616 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2617 | |
5722d094 | 2618 | css_put(&memcg->css); |
d7f25f8a GC |
2619 | kfree(cw); |
2620 | } | |
2621 | ||
2622 | /* | |
2623 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2624 | */ |
d5b3cf71 VD |
2625 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2626 | struct kmem_cache *cachep) | |
d7f25f8a | 2627 | { |
d5b3cf71 | 2628 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2629 | |
776ed0f0 | 2630 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2631 | if (!cw) |
d7f25f8a | 2632 | return; |
8135be5a VD |
2633 | |
2634 | css_get(&memcg->css); | |
d7f25f8a GC |
2635 | |
2636 | cw->memcg = memcg; | |
2637 | cw->cachep = cachep; | |
d5b3cf71 | 2638 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2639 | |
d7f25f8a GC |
2640 | schedule_work(&cw->work); |
2641 | } | |
2642 | ||
d5b3cf71 VD |
2643 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2644 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2645 | { |
2646 | /* | |
2647 | * We need to stop accounting when we kmalloc, because if the | |
2648 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2649 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2650 | * |
2651 | * However, it is better to enclose the whole function. Depending on | |
2652 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2653 | * trigger an allocation. This too, will make us recurse. Because at | |
2654 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2655 | * the safest choice is to do it like this, wrapping the whole function. | |
2656 | */ | |
6f185c29 | 2657 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2658 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2659 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2660 | } |
c67a8a68 | 2661 | |
d7f25f8a GC |
2662 | /* |
2663 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2664 | * We try to use the current memcg's version of the cache. | |
2665 | * | |
2666 | * If the cache does not exist yet, if we are the first user of it, | |
2667 | * we either create it immediately, if possible, or create it asynchronously | |
2668 | * in a workqueue. | |
2669 | * In the latter case, we will let the current allocation go through with | |
2670 | * the original cache. | |
2671 | * | |
2672 | * Can't be called in interrupt context or from kernel threads. | |
2673 | * This function needs to be called with rcu_read_lock() held. | |
2674 | */ | |
056b7cce | 2675 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2676 | { |
2677 | struct mem_cgroup *memcg; | |
959c8963 | 2678 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
2679 | |
2680 | VM_BUG_ON(!cachep->memcg_params); | |
2681 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
2682 | ||
9d100c5e | 2683 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2684 | return cachep; |
2685 | ||
8135be5a | 2686 | memcg = get_mem_cgroup_from_mm(current->mm); |
cf2b8fbf | 2687 | if (!memcg_kmem_is_active(memcg)) |
ca0dde97 | 2688 | goto out; |
d7f25f8a | 2689 | |
959c8963 | 2690 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
8135be5a VD |
2691 | if (likely(memcg_cachep)) |
2692 | return memcg_cachep; | |
ca0dde97 LZ |
2693 | |
2694 | /* | |
2695 | * If we are in a safe context (can wait, and not in interrupt | |
2696 | * context), we could be be predictable and return right away. | |
2697 | * This would guarantee that the allocation being performed | |
2698 | * already belongs in the new cache. | |
2699 | * | |
2700 | * However, there are some clashes that can arrive from locking. | |
2701 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2702 | * memcg_create_kmem_cache, this means no further allocation |
2703 | * could happen with the slab_mutex held. So it's better to | |
2704 | * defer everything. | |
ca0dde97 | 2705 | */ |
d5b3cf71 | 2706 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2707 | out: |
8135be5a | 2708 | css_put(&memcg->css); |
ca0dde97 | 2709 | return cachep; |
d7f25f8a | 2710 | } |
d7f25f8a | 2711 | |
8135be5a VD |
2712 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2713 | { | |
2714 | if (!is_root_cache(cachep)) | |
2715 | css_put(&cachep->memcg_params->memcg->css); | |
2716 | } | |
2717 | ||
7ae1e1d0 GC |
2718 | /* |
2719 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2720 | * possible for the given order. But the page is not allocated yet, so we'll | |
2721 | * need a further commit step to do the final arrangements. | |
2722 | * | |
2723 | * It is possible for the task to switch cgroups in this mean time, so at | |
2724 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2725 | * the handle argument to return to the caller which cgroup we should commit | |
2726 | * against. We could also return the memcg directly and avoid the pointer | |
2727 | * passing, but a boolean return value gives better semantics considering | |
2728 | * the compiled-out case as well. | |
2729 | * | |
2730 | * Returning true means the allocation is possible. | |
2731 | */ | |
2732 | bool | |
2733 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2734 | { | |
2735 | struct mem_cgroup *memcg; | |
2736 | int ret; | |
2737 | ||
2738 | *_memcg = NULL; | |
6d42c232 | 2739 | |
df381975 | 2740 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 2741 | |
cf2b8fbf | 2742 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
2743 | css_put(&memcg->css); |
2744 | return true; | |
2745 | } | |
2746 | ||
3e32cb2e | 2747 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
2748 | if (!ret) |
2749 | *_memcg = memcg; | |
7ae1e1d0 GC |
2750 | |
2751 | css_put(&memcg->css); | |
2752 | return (ret == 0); | |
2753 | } | |
2754 | ||
2755 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
2756 | int order) | |
2757 | { | |
7ae1e1d0 GC |
2758 | VM_BUG_ON(mem_cgroup_is_root(memcg)); |
2759 | ||
2760 | /* The page allocation failed. Revert */ | |
2761 | if (!page) { | |
3e32cb2e | 2762 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
2763 | return; |
2764 | } | |
1306a85a | 2765 | page->mem_cgroup = memcg; |
7ae1e1d0 GC |
2766 | } |
2767 | ||
2768 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
2769 | { | |
1306a85a | 2770 | struct mem_cgroup *memcg = page->mem_cgroup; |
7ae1e1d0 | 2771 | |
7ae1e1d0 GC |
2772 | if (!memcg) |
2773 | return; | |
2774 | ||
309381fe | 2775 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2776 | |
3e32cb2e | 2777 | memcg_uncharge_kmem(memcg, 1 << order); |
1306a85a | 2778 | page->mem_cgroup = NULL; |
7ae1e1d0 GC |
2779 | } |
2780 | #endif /* CONFIG_MEMCG_KMEM */ | |
2781 | ||
ca3e0214 KH |
2782 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2783 | ||
ca3e0214 KH |
2784 | /* |
2785 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2786 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2787 | * charge/uncharge will be never happen and move_account() is done under | |
2788 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2789 | */ |
e94c8a9c | 2790 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2791 | { |
e94c8a9c | 2792 | int i; |
ca3e0214 | 2793 | |
3d37c4a9 KH |
2794 | if (mem_cgroup_disabled()) |
2795 | return; | |
b070e65c | 2796 | |
29833315 | 2797 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2798 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2799 | |
1306a85a | 2800 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2801 | HPAGE_PMD_NR); |
ca3e0214 | 2802 | } |
12d27107 | 2803 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2804 | |
f817ed48 | 2805 | /** |
de3638d9 | 2806 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2807 | * @page: the page |
7ec99d62 | 2808 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2809 | * @from: mem_cgroup which the page is moved from. |
2810 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2811 | * | |
2812 | * The caller must confirm following. | |
08e552c6 | 2813 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2814 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2815 | * |
2f3479b1 KH |
2816 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2817 | * from old cgroup. | |
f817ed48 | 2818 | */ |
7ec99d62 JW |
2819 | static int mem_cgroup_move_account(struct page *page, |
2820 | unsigned int nr_pages, | |
7ec99d62 | 2821 | struct mem_cgroup *from, |
2f3479b1 | 2822 | struct mem_cgroup *to) |
f817ed48 | 2823 | { |
de3638d9 JW |
2824 | unsigned long flags; |
2825 | int ret; | |
987eba66 | 2826 | |
f817ed48 | 2827 | VM_BUG_ON(from == to); |
309381fe | 2828 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
2829 | /* |
2830 | * The page is isolated from LRU. So, collapse function | |
2831 | * will not handle this page. But page splitting can happen. | |
2832 | * Do this check under compound_page_lock(). The caller should | |
2833 | * hold it. | |
2834 | */ | |
2835 | ret = -EBUSY; | |
7ec99d62 | 2836 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2837 | goto out; |
2838 | ||
0a31bc97 | 2839 | /* |
1306a85a | 2840 | * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup |
0a31bc97 JW |
2841 | * of its source page while we change it: page migration takes |
2842 | * both pages off the LRU, but page cache replacement doesn't. | |
2843 | */ | |
2844 | if (!trylock_page(page)) | |
2845 | goto out; | |
de3638d9 JW |
2846 | |
2847 | ret = -EINVAL; | |
1306a85a | 2848 | if (page->mem_cgroup != from) |
0a31bc97 | 2849 | goto out_unlock; |
de3638d9 | 2850 | |
354a4783 | 2851 | spin_lock_irqsave(&from->move_lock, flags); |
f817ed48 | 2852 | |
0a31bc97 | 2853 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
2854 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
2855 | nr_pages); | |
2856 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
2857 | nr_pages); | |
2858 | } | |
3ea67d06 | 2859 | |
59d1d256 JW |
2860 | if (PageWriteback(page)) { |
2861 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
2862 | nr_pages); | |
2863 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
2864 | nr_pages); | |
2865 | } | |
3ea67d06 | 2866 | |
0a31bc97 | 2867 | /* |
1306a85a | 2868 | * It is safe to change page->mem_cgroup here because the page |
0a31bc97 JW |
2869 | * is referenced, charged, and isolated - we can't race with |
2870 | * uncharging, charging, migration, or LRU putback. | |
2871 | */ | |
d69b042f | 2872 | |
854ffa8d | 2873 | /* caller should have done css_get */ |
1306a85a | 2874 | page->mem_cgroup = to; |
354a4783 JW |
2875 | spin_unlock_irqrestore(&from->move_lock, flags); |
2876 | ||
de3638d9 | 2877 | ret = 0; |
0a31bc97 JW |
2878 | |
2879 | local_irq_disable(); | |
2880 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 2881 | memcg_check_events(to, page); |
0a31bc97 | 2882 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 2883 | memcg_check_events(from, page); |
0a31bc97 JW |
2884 | local_irq_enable(); |
2885 | out_unlock: | |
2886 | unlock_page(page); | |
de3638d9 | 2887 | out: |
f817ed48 KH |
2888 | return ret; |
2889 | } | |
2890 | ||
c255a458 | 2891 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2892 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2893 | bool charge) | |
d13d1443 | 2894 | { |
0a31bc97 JW |
2895 | int val = (charge) ? 1 : -1; |
2896 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2897 | } |
02491447 DN |
2898 | |
2899 | /** | |
2900 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2901 | * @entry: swap entry to be moved | |
2902 | * @from: mem_cgroup which the entry is moved from | |
2903 | * @to: mem_cgroup which the entry is moved to | |
2904 | * | |
2905 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2906 | * as the mem_cgroup's id of @from. | |
2907 | * | |
2908 | * Returns 0 on success, -EINVAL on failure. | |
2909 | * | |
3e32cb2e | 2910 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2911 | * both res and memsw, and called css_get(). |
2912 | */ | |
2913 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2914 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2915 | { |
2916 | unsigned short old_id, new_id; | |
2917 | ||
34c00c31 LZ |
2918 | old_id = mem_cgroup_id(from); |
2919 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2920 | |
2921 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2922 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2923 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2924 | return 0; |
2925 | } | |
2926 | return -EINVAL; | |
2927 | } | |
2928 | #else | |
2929 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2930 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2931 | { |
2932 | return -EINVAL; | |
2933 | } | |
8c7c6e34 | 2934 | #endif |
d13d1443 | 2935 | |
3e32cb2e | 2936 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2937 | |
d38d2a75 | 2938 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2939 | unsigned long limit) |
628f4235 | 2940 | { |
3e32cb2e JW |
2941 | unsigned long curusage; |
2942 | unsigned long oldusage; | |
2943 | bool enlarge = false; | |
81d39c20 | 2944 | int retry_count; |
3e32cb2e | 2945 | int ret; |
81d39c20 KH |
2946 | |
2947 | /* | |
2948 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2949 | * is depends on callers. We set our retry-count to be function | |
2950 | * of # of children which we should visit in this loop. | |
2951 | */ | |
3e32cb2e JW |
2952 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2953 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2954 | |
3e32cb2e | 2955 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2956 | |
3e32cb2e | 2957 | do { |
628f4235 KH |
2958 | if (signal_pending(current)) { |
2959 | ret = -EINTR; | |
2960 | break; | |
2961 | } | |
3e32cb2e JW |
2962 | |
2963 | mutex_lock(&memcg_limit_mutex); | |
2964 | if (limit > memcg->memsw.limit) { | |
2965 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2966 | ret = -EINVAL; |
628f4235 KH |
2967 | break; |
2968 | } | |
3e32cb2e JW |
2969 | if (limit > memcg->memory.limit) |
2970 | enlarge = true; | |
2971 | ret = page_counter_limit(&memcg->memory, limit); | |
2972 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2973 | |
2974 | if (!ret) | |
2975 | break; | |
2976 | ||
b70a2a21 JW |
2977 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2978 | ||
3e32cb2e | 2979 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2980 | /* Usage is reduced ? */ |
f894ffa8 | 2981 | if (curusage >= oldusage) |
81d39c20 KH |
2982 | retry_count--; |
2983 | else | |
2984 | oldusage = curusage; | |
3e32cb2e JW |
2985 | } while (retry_count); |
2986 | ||
3c11ecf4 KH |
2987 | if (!ret && enlarge) |
2988 | memcg_oom_recover(memcg); | |
14797e23 | 2989 | |
8c7c6e34 KH |
2990 | return ret; |
2991 | } | |
2992 | ||
338c8431 | 2993 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2994 | unsigned long limit) |
8c7c6e34 | 2995 | { |
3e32cb2e JW |
2996 | unsigned long curusage; |
2997 | unsigned long oldusage; | |
2998 | bool enlarge = false; | |
81d39c20 | 2999 | int retry_count; |
3e32cb2e | 3000 | int ret; |
8c7c6e34 | 3001 | |
81d39c20 | 3002 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
3003 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3004 | mem_cgroup_count_children(memcg); | |
3005 | ||
3006 | oldusage = page_counter_read(&memcg->memsw); | |
3007 | ||
3008 | do { | |
8c7c6e34 KH |
3009 | if (signal_pending(current)) { |
3010 | ret = -EINTR; | |
3011 | break; | |
3012 | } | |
3e32cb2e JW |
3013 | |
3014 | mutex_lock(&memcg_limit_mutex); | |
3015 | if (limit < memcg->memory.limit) { | |
3016 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3017 | ret = -EINVAL; |
8c7c6e34 KH |
3018 | break; |
3019 | } | |
3e32cb2e JW |
3020 | if (limit > memcg->memsw.limit) |
3021 | enlarge = true; | |
3022 | ret = page_counter_limit(&memcg->memsw, limit); | |
3023 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3024 | |
3025 | if (!ret) | |
3026 | break; | |
3027 | ||
b70a2a21 JW |
3028 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3029 | ||
3e32cb2e | 3030 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3031 | /* Usage is reduced ? */ |
8c7c6e34 | 3032 | if (curusage >= oldusage) |
628f4235 | 3033 | retry_count--; |
81d39c20 KH |
3034 | else |
3035 | oldusage = curusage; | |
3e32cb2e JW |
3036 | } while (retry_count); |
3037 | ||
3c11ecf4 KH |
3038 | if (!ret && enlarge) |
3039 | memcg_oom_recover(memcg); | |
3e32cb2e | 3040 | |
628f4235 KH |
3041 | return ret; |
3042 | } | |
3043 | ||
0608f43d AM |
3044 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3045 | gfp_t gfp_mask, | |
3046 | unsigned long *total_scanned) | |
3047 | { | |
3048 | unsigned long nr_reclaimed = 0; | |
3049 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3050 | unsigned long reclaimed; | |
3051 | int loop = 0; | |
3052 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3053 | unsigned long excess; |
0608f43d AM |
3054 | unsigned long nr_scanned; |
3055 | ||
3056 | if (order > 0) | |
3057 | return 0; | |
3058 | ||
3059 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3060 | /* | |
3061 | * This loop can run a while, specially if mem_cgroup's continuously | |
3062 | * keep exceeding their soft limit and putting the system under | |
3063 | * pressure | |
3064 | */ | |
3065 | do { | |
3066 | if (next_mz) | |
3067 | mz = next_mz; | |
3068 | else | |
3069 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3070 | if (!mz) | |
3071 | break; | |
3072 | ||
3073 | nr_scanned = 0; | |
3074 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3075 | gfp_mask, &nr_scanned); | |
3076 | nr_reclaimed += reclaimed; | |
3077 | *total_scanned += nr_scanned; | |
0a31bc97 | 3078 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3079 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3080 | |
3081 | /* | |
3082 | * If we failed to reclaim anything from this memory cgroup | |
3083 | * it is time to move on to the next cgroup | |
3084 | */ | |
3085 | next_mz = NULL; | |
bc2f2e7f VD |
3086 | if (!reclaimed) |
3087 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3088 | ||
3e32cb2e | 3089 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3090 | /* |
3091 | * One school of thought says that we should not add | |
3092 | * back the node to the tree if reclaim returns 0. | |
3093 | * But our reclaim could return 0, simply because due | |
3094 | * to priority we are exposing a smaller subset of | |
3095 | * memory to reclaim from. Consider this as a longer | |
3096 | * term TODO. | |
3097 | */ | |
3098 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3099 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3100 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3101 | css_put(&mz->memcg->css); |
3102 | loop++; | |
3103 | /* | |
3104 | * Could not reclaim anything and there are no more | |
3105 | * mem cgroups to try or we seem to be looping without | |
3106 | * reclaiming anything. | |
3107 | */ | |
3108 | if (!nr_reclaimed && | |
3109 | (next_mz == NULL || | |
3110 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3111 | break; | |
3112 | } while (!nr_reclaimed); | |
3113 | if (next_mz) | |
3114 | css_put(&next_mz->memcg->css); | |
3115 | return nr_reclaimed; | |
3116 | } | |
3117 | ||
ea280e7b TH |
3118 | /* |
3119 | * Test whether @memcg has children, dead or alive. Note that this | |
3120 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3121 | * returns %true if there are child csses according to the cgroup | |
3122 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3123 | */ | |
b5f99b53 GC |
3124 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3125 | { | |
ea280e7b TH |
3126 | bool ret; |
3127 | ||
696ac172 | 3128 | /* |
ea280e7b TH |
3129 | * The lock does not prevent addition or deletion of children, but |
3130 | * it prevents a new child from being initialized based on this | |
3131 | * parent in css_online(), so it's enough to decide whether | |
3132 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3133 | */ |
ea280e7b TH |
3134 | lockdep_assert_held(&memcg_create_mutex); |
3135 | ||
3136 | rcu_read_lock(); | |
3137 | ret = css_next_child(NULL, &memcg->css); | |
3138 | rcu_read_unlock(); | |
3139 | return ret; | |
b5f99b53 GC |
3140 | } |
3141 | ||
c26251f9 MH |
3142 | /* |
3143 | * Reclaims as many pages from the given memcg as possible and moves | |
3144 | * the rest to the parent. | |
3145 | * | |
3146 | * Caller is responsible for holding css reference for memcg. | |
3147 | */ | |
3148 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3149 | { | |
3150 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3151 | |
c1e862c1 KH |
3152 | /* we call try-to-free pages for make this cgroup empty */ |
3153 | lru_add_drain_all(); | |
f817ed48 | 3154 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3155 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3156 | int progress; |
c1e862c1 | 3157 | |
c26251f9 MH |
3158 | if (signal_pending(current)) |
3159 | return -EINTR; | |
3160 | ||
b70a2a21 JW |
3161 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3162 | GFP_KERNEL, true); | |
c1e862c1 | 3163 | if (!progress) { |
f817ed48 | 3164 | nr_retries--; |
c1e862c1 | 3165 | /* maybe some writeback is necessary */ |
8aa7e847 | 3166 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3167 | } |
f817ed48 KH |
3168 | |
3169 | } | |
ab5196c2 MH |
3170 | |
3171 | return 0; | |
cc847582 KH |
3172 | } |
3173 | ||
6770c64e TH |
3174 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3175 | char *buf, size_t nbytes, | |
3176 | loff_t off) | |
c1e862c1 | 3177 | { |
6770c64e | 3178 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3179 | |
d8423011 MH |
3180 | if (mem_cgroup_is_root(memcg)) |
3181 | return -EINVAL; | |
6770c64e | 3182 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3183 | } |
3184 | ||
182446d0 TH |
3185 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3186 | struct cftype *cft) | |
18f59ea7 | 3187 | { |
182446d0 | 3188 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3189 | } |
3190 | ||
182446d0 TH |
3191 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3192 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3193 | { |
3194 | int retval = 0; | |
182446d0 | 3195 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3196 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3197 | |
0999821b | 3198 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3199 | |
3200 | if (memcg->use_hierarchy == val) | |
3201 | goto out; | |
3202 | ||
18f59ea7 | 3203 | /* |
af901ca1 | 3204 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3205 | * in the child subtrees. If it is unset, then the change can |
3206 | * occur, provided the current cgroup has no children. | |
3207 | * | |
3208 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3209 | * set if there are no children. | |
3210 | */ | |
c0ff4b85 | 3211 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3212 | (val == 1 || val == 0)) { |
ea280e7b | 3213 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3214 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3215 | else |
3216 | retval = -EBUSY; | |
3217 | } else | |
3218 | retval = -EINVAL; | |
567fb435 GC |
3219 | |
3220 | out: | |
0999821b | 3221 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3222 | |
3223 | return retval; | |
3224 | } | |
3225 | ||
3e32cb2e JW |
3226 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3227 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3228 | { |
3229 | struct mem_cgroup *iter; | |
3230 | long val = 0; | |
3231 | ||
3232 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3233 | for_each_mem_cgroup_tree(iter, memcg) | |
3234 | val += mem_cgroup_read_stat(iter, idx); | |
3235 | ||
3236 | if (val < 0) /* race ? */ | |
3237 | val = 0; | |
3238 | return val; | |
3239 | } | |
3240 | ||
3241 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3242 | { | |
3243 | u64 val; | |
3244 | ||
3e32cb2e JW |
3245 | if (mem_cgroup_is_root(memcg)) { |
3246 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3247 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3248 | if (swap) | |
3249 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3250 | } else { | |
ce00a967 | 3251 | if (!swap) |
3e32cb2e | 3252 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3253 | else |
3e32cb2e | 3254 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3255 | } |
ce00a967 JW |
3256 | return val << PAGE_SHIFT; |
3257 | } | |
3258 | ||
3e32cb2e JW |
3259 | enum { |
3260 | RES_USAGE, | |
3261 | RES_LIMIT, | |
3262 | RES_MAX_USAGE, | |
3263 | RES_FAILCNT, | |
3264 | RES_SOFT_LIMIT, | |
3265 | }; | |
ce00a967 | 3266 | |
791badbd | 3267 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3268 | struct cftype *cft) |
8cdea7c0 | 3269 | { |
182446d0 | 3270 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3271 | struct page_counter *counter; |
af36f906 | 3272 | |
3e32cb2e | 3273 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3274 | case _MEM: |
3e32cb2e JW |
3275 | counter = &memcg->memory; |
3276 | break; | |
8c7c6e34 | 3277 | case _MEMSWAP: |
3e32cb2e JW |
3278 | counter = &memcg->memsw; |
3279 | break; | |
510fc4e1 | 3280 | case _KMEM: |
3e32cb2e | 3281 | counter = &memcg->kmem; |
510fc4e1 | 3282 | break; |
8c7c6e34 KH |
3283 | default: |
3284 | BUG(); | |
8c7c6e34 | 3285 | } |
3e32cb2e JW |
3286 | |
3287 | switch (MEMFILE_ATTR(cft->private)) { | |
3288 | case RES_USAGE: | |
3289 | if (counter == &memcg->memory) | |
3290 | return mem_cgroup_usage(memcg, false); | |
3291 | if (counter == &memcg->memsw) | |
3292 | return mem_cgroup_usage(memcg, true); | |
3293 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3294 | case RES_LIMIT: | |
3295 | return (u64)counter->limit * PAGE_SIZE; | |
3296 | case RES_MAX_USAGE: | |
3297 | return (u64)counter->watermark * PAGE_SIZE; | |
3298 | case RES_FAILCNT: | |
3299 | return counter->failcnt; | |
3300 | case RES_SOFT_LIMIT: | |
3301 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3302 | default: | |
3303 | BUG(); | |
3304 | } | |
8cdea7c0 | 3305 | } |
510fc4e1 | 3306 | |
510fc4e1 | 3307 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
3308 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
3309 | unsigned long nr_pages) | |
d6441637 VD |
3310 | { |
3311 | int err = 0; | |
3312 | int memcg_id; | |
3313 | ||
3314 | if (memcg_kmem_is_active(memcg)) | |
3315 | return 0; | |
3316 | ||
510fc4e1 GC |
3317 | /* |
3318 | * For simplicity, we won't allow this to be disabled. It also can't | |
3319 | * be changed if the cgroup has children already, or if tasks had | |
3320 | * already joined. | |
3321 | * | |
3322 | * If tasks join before we set the limit, a person looking at | |
3323 | * kmem.usage_in_bytes will have no way to determine when it took | |
3324 | * place, which makes the value quite meaningless. | |
3325 | * | |
3326 | * After it first became limited, changes in the value of the limit are | |
3327 | * of course permitted. | |
510fc4e1 | 3328 | */ |
0999821b | 3329 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
3330 | if (cgroup_has_tasks(memcg->css.cgroup) || |
3331 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
3332 | err = -EBUSY; |
3333 | mutex_unlock(&memcg_create_mutex); | |
3334 | if (err) | |
3335 | goto out; | |
510fc4e1 | 3336 | |
f3bb3043 | 3337 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
3338 | if (memcg_id < 0) { |
3339 | err = memcg_id; | |
3340 | goto out; | |
3341 | } | |
3342 | ||
d6441637 | 3343 | /* |
900a38f0 VD |
3344 | * We couldn't have accounted to this cgroup, because it hasn't got |
3345 | * activated yet, so this should succeed. | |
d6441637 | 3346 | */ |
3e32cb2e | 3347 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
3348 | VM_BUG_ON(err); |
3349 | ||
3350 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
3351 | /* | |
900a38f0 VD |
3352 | * A memory cgroup is considered kmem-active as soon as it gets |
3353 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
3354 | * guarantee no one starts accounting before all call sites are |
3355 | * patched. | |
3356 | */ | |
900a38f0 | 3357 | memcg->kmemcg_id = memcg_id; |
510fc4e1 | 3358 | out: |
d6441637 | 3359 | return err; |
d6441637 VD |
3360 | } |
3361 | ||
d6441637 | 3362 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3363 | unsigned long limit) |
d6441637 VD |
3364 | { |
3365 | int ret; | |
3366 | ||
3e32cb2e | 3367 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 3368 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 3369 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 3370 | else |
3e32cb2e JW |
3371 | ret = page_counter_limit(&memcg->kmem, limit); |
3372 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
3373 | return ret; |
3374 | } | |
3375 | ||
55007d84 | 3376 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 3377 | { |
55007d84 | 3378 | int ret = 0; |
510fc4e1 | 3379 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 3380 | |
d6441637 VD |
3381 | if (!parent) |
3382 | return 0; | |
55007d84 | 3383 | |
8c0145b6 | 3384 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 3385 | /* |
d6441637 VD |
3386 | * If the parent cgroup is not kmem-active now, it cannot be activated |
3387 | * after this point, because it has at least one child already. | |
55007d84 | 3388 | */ |
d6441637 | 3389 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3390 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3391 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3392 | return ret; |
510fc4e1 | 3393 | } |
d6441637 VD |
3394 | #else |
3395 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3396 | unsigned long limit) |
d6441637 VD |
3397 | { |
3398 | return -EINVAL; | |
3399 | } | |
6d043990 | 3400 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3401 | |
628f4235 KH |
3402 | /* |
3403 | * The user of this function is... | |
3404 | * RES_LIMIT. | |
3405 | */ | |
451af504 TH |
3406 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3407 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3408 | { |
451af504 | 3409 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3410 | unsigned long nr_pages; |
628f4235 KH |
3411 | int ret; |
3412 | ||
451af504 | 3413 | buf = strstrip(buf); |
650c5e56 | 3414 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3415 | if (ret) |
3416 | return ret; | |
af36f906 | 3417 | |
3e32cb2e | 3418 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3419 | case RES_LIMIT: |
4b3bde4c BS |
3420 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3421 | ret = -EINVAL; | |
3422 | break; | |
3423 | } | |
3e32cb2e JW |
3424 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3425 | case _MEM: | |
3426 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3427 | break; |
3e32cb2e JW |
3428 | case _MEMSWAP: |
3429 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3430 | break; |
3e32cb2e JW |
3431 | case _KMEM: |
3432 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3433 | break; | |
3434 | } | |
296c81d8 | 3435 | break; |
3e32cb2e JW |
3436 | case RES_SOFT_LIMIT: |
3437 | memcg->soft_limit = nr_pages; | |
3438 | ret = 0; | |
628f4235 KH |
3439 | break; |
3440 | } | |
451af504 | 3441 | return ret ?: nbytes; |
8cdea7c0 BS |
3442 | } |
3443 | ||
6770c64e TH |
3444 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3445 | size_t nbytes, loff_t off) | |
c84872e1 | 3446 | { |
6770c64e | 3447 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3448 | struct page_counter *counter; |
c84872e1 | 3449 | |
3e32cb2e JW |
3450 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3451 | case _MEM: | |
3452 | counter = &memcg->memory; | |
3453 | break; | |
3454 | case _MEMSWAP: | |
3455 | counter = &memcg->memsw; | |
3456 | break; | |
3457 | case _KMEM: | |
3458 | counter = &memcg->kmem; | |
3459 | break; | |
3460 | default: | |
3461 | BUG(); | |
3462 | } | |
af36f906 | 3463 | |
3e32cb2e | 3464 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3465 | case RES_MAX_USAGE: |
3e32cb2e | 3466 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3467 | break; |
3468 | case RES_FAILCNT: | |
3e32cb2e | 3469 | counter->failcnt = 0; |
29f2a4da | 3470 | break; |
3e32cb2e JW |
3471 | default: |
3472 | BUG(); | |
29f2a4da | 3473 | } |
f64c3f54 | 3474 | |
6770c64e | 3475 | return nbytes; |
c84872e1 PE |
3476 | } |
3477 | ||
182446d0 | 3478 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3479 | struct cftype *cft) |
3480 | { | |
182446d0 | 3481 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3482 | } |
3483 | ||
02491447 | 3484 | #ifdef CONFIG_MMU |
182446d0 | 3485 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3486 | struct cftype *cft, u64 val) |
3487 | { | |
182446d0 | 3488 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3489 | |
1dfab5ab | 3490 | if (val & ~MOVE_MASK) |
7dc74be0 | 3491 | return -EINVAL; |
ee5e8472 | 3492 | |
7dc74be0 | 3493 | /* |
ee5e8472 GC |
3494 | * No kind of locking is needed in here, because ->can_attach() will |
3495 | * check this value once in the beginning of the process, and then carry | |
3496 | * on with stale data. This means that changes to this value will only | |
3497 | * affect task migrations starting after the change. | |
7dc74be0 | 3498 | */ |
c0ff4b85 | 3499 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3500 | return 0; |
3501 | } | |
02491447 | 3502 | #else |
182446d0 | 3503 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3504 | struct cftype *cft, u64 val) |
3505 | { | |
3506 | return -ENOSYS; | |
3507 | } | |
3508 | #endif | |
7dc74be0 | 3509 | |
406eb0c9 | 3510 | #ifdef CONFIG_NUMA |
2da8ca82 | 3511 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3512 | { |
25485de6 GT |
3513 | struct numa_stat { |
3514 | const char *name; | |
3515 | unsigned int lru_mask; | |
3516 | }; | |
3517 | ||
3518 | static const struct numa_stat stats[] = { | |
3519 | { "total", LRU_ALL }, | |
3520 | { "file", LRU_ALL_FILE }, | |
3521 | { "anon", LRU_ALL_ANON }, | |
3522 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3523 | }; | |
3524 | const struct numa_stat *stat; | |
406eb0c9 | 3525 | int nid; |
25485de6 | 3526 | unsigned long nr; |
2da8ca82 | 3527 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3528 | |
25485de6 GT |
3529 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3530 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3531 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3532 | for_each_node_state(nid, N_MEMORY) { | |
3533 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3534 | stat->lru_mask); | |
3535 | seq_printf(m, " N%d=%lu", nid, nr); | |
3536 | } | |
3537 | seq_putc(m, '\n'); | |
406eb0c9 | 3538 | } |
406eb0c9 | 3539 | |
071aee13 YH |
3540 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3541 | struct mem_cgroup *iter; | |
3542 | ||
3543 | nr = 0; | |
3544 | for_each_mem_cgroup_tree(iter, memcg) | |
3545 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3546 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3547 | for_each_node_state(nid, N_MEMORY) { | |
3548 | nr = 0; | |
3549 | for_each_mem_cgroup_tree(iter, memcg) | |
3550 | nr += mem_cgroup_node_nr_lru_pages( | |
3551 | iter, nid, stat->lru_mask); | |
3552 | seq_printf(m, " N%d=%lu", nid, nr); | |
3553 | } | |
3554 | seq_putc(m, '\n'); | |
406eb0c9 | 3555 | } |
406eb0c9 | 3556 | |
406eb0c9 YH |
3557 | return 0; |
3558 | } | |
3559 | #endif /* CONFIG_NUMA */ | |
3560 | ||
2da8ca82 | 3561 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3562 | { |
2da8ca82 | 3563 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3564 | unsigned long memory, memsw; |
af7c4b0e JW |
3565 | struct mem_cgroup *mi; |
3566 | unsigned int i; | |
406eb0c9 | 3567 | |
0ca44b14 GT |
3568 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3569 | MEM_CGROUP_STAT_NSTATS); | |
3570 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3571 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3572 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3573 | ||
af7c4b0e | 3574 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3575 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3576 | continue; |
af7c4b0e JW |
3577 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
3578 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 3579 | } |
7b854121 | 3580 | |
af7c4b0e JW |
3581 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3582 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3583 | mem_cgroup_read_events(memcg, i)); | |
3584 | ||
3585 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3586 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3587 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3588 | ||
14067bb3 | 3589 | /* Hierarchical information */ |
3e32cb2e JW |
3590 | memory = memsw = PAGE_COUNTER_MAX; |
3591 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3592 | memory = min(memory, mi->memory.limit); | |
3593 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3594 | } |
3e32cb2e JW |
3595 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3596 | (u64)memory * PAGE_SIZE); | |
3597 | if (do_swap_account) | |
3598 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3599 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3600 | |
af7c4b0e JW |
3601 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
3602 | long long val = 0; | |
3603 | ||
bff6bb83 | 3604 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3605 | continue; |
af7c4b0e JW |
3606 | for_each_mem_cgroup_tree(mi, memcg) |
3607 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
3608 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
3609 | } | |
3610 | ||
3611 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3612 | unsigned long long val = 0; | |
3613 | ||
3614 | for_each_mem_cgroup_tree(mi, memcg) | |
3615 | val += mem_cgroup_read_events(mi, i); | |
3616 | seq_printf(m, "total_%s %llu\n", | |
3617 | mem_cgroup_events_names[i], val); | |
3618 | } | |
3619 | ||
3620 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3621 | unsigned long long val = 0; | |
3622 | ||
3623 | for_each_mem_cgroup_tree(mi, memcg) | |
3624 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3625 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3626 | } |
14067bb3 | 3627 | |
7f016ee8 | 3628 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3629 | { |
3630 | int nid, zid; | |
3631 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3632 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3633 | unsigned long recent_rotated[2] = {0, 0}; |
3634 | unsigned long recent_scanned[2] = {0, 0}; | |
3635 | ||
3636 | for_each_online_node(nid) | |
3637 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3638 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3639 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3640 | |
89abfab1 HD |
3641 | recent_rotated[0] += rstat->recent_rotated[0]; |
3642 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3643 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3644 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3645 | } |
78ccf5b5 JW |
3646 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3647 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3648 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3649 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3650 | } |
3651 | #endif | |
3652 | ||
d2ceb9b7 KH |
3653 | return 0; |
3654 | } | |
3655 | ||
182446d0 TH |
3656 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3657 | struct cftype *cft) | |
a7885eb8 | 3658 | { |
182446d0 | 3659 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3660 | |
1f4c025b | 3661 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3662 | } |
3663 | ||
182446d0 TH |
3664 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3665 | struct cftype *cft, u64 val) | |
a7885eb8 | 3666 | { |
182446d0 | 3667 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3668 | |
3dae7fec | 3669 | if (val > 100) |
a7885eb8 KM |
3670 | return -EINVAL; |
3671 | ||
14208b0e | 3672 | if (css->parent) |
3dae7fec JW |
3673 | memcg->swappiness = val; |
3674 | else | |
3675 | vm_swappiness = val; | |
068b38c1 | 3676 | |
a7885eb8 KM |
3677 | return 0; |
3678 | } | |
3679 | ||
2e72b634 KS |
3680 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3681 | { | |
3682 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3683 | unsigned long usage; |
2e72b634 KS |
3684 | int i; |
3685 | ||
3686 | rcu_read_lock(); | |
3687 | if (!swap) | |
2c488db2 | 3688 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3689 | else |
2c488db2 | 3690 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3691 | |
3692 | if (!t) | |
3693 | goto unlock; | |
3694 | ||
ce00a967 | 3695 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3696 | |
3697 | /* | |
748dad36 | 3698 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3699 | * If it's not true, a threshold was crossed after last |
3700 | * call of __mem_cgroup_threshold(). | |
3701 | */ | |
5407a562 | 3702 | i = t->current_threshold; |
2e72b634 KS |
3703 | |
3704 | /* | |
3705 | * Iterate backward over array of thresholds starting from | |
3706 | * current_threshold and check if a threshold is crossed. | |
3707 | * If none of thresholds below usage is crossed, we read | |
3708 | * only one element of the array here. | |
3709 | */ | |
3710 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3711 | eventfd_signal(t->entries[i].eventfd, 1); | |
3712 | ||
3713 | /* i = current_threshold + 1 */ | |
3714 | i++; | |
3715 | ||
3716 | /* | |
3717 | * Iterate forward over array of thresholds starting from | |
3718 | * current_threshold+1 and check if a threshold is crossed. | |
3719 | * If none of thresholds above usage is crossed, we read | |
3720 | * only one element of the array here. | |
3721 | */ | |
3722 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3723 | eventfd_signal(t->entries[i].eventfd, 1); | |
3724 | ||
3725 | /* Update current_threshold */ | |
5407a562 | 3726 | t->current_threshold = i - 1; |
2e72b634 KS |
3727 | unlock: |
3728 | rcu_read_unlock(); | |
3729 | } | |
3730 | ||
3731 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3732 | { | |
ad4ca5f4 KS |
3733 | while (memcg) { |
3734 | __mem_cgroup_threshold(memcg, false); | |
3735 | if (do_swap_account) | |
3736 | __mem_cgroup_threshold(memcg, true); | |
3737 | ||
3738 | memcg = parent_mem_cgroup(memcg); | |
3739 | } | |
2e72b634 KS |
3740 | } |
3741 | ||
3742 | static int compare_thresholds(const void *a, const void *b) | |
3743 | { | |
3744 | const struct mem_cgroup_threshold *_a = a; | |
3745 | const struct mem_cgroup_threshold *_b = b; | |
3746 | ||
2bff24a3 GT |
3747 | if (_a->threshold > _b->threshold) |
3748 | return 1; | |
3749 | ||
3750 | if (_a->threshold < _b->threshold) | |
3751 | return -1; | |
3752 | ||
3753 | return 0; | |
2e72b634 KS |
3754 | } |
3755 | ||
c0ff4b85 | 3756 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3757 | { |
3758 | struct mem_cgroup_eventfd_list *ev; | |
3759 | ||
2bcf2e92 MH |
3760 | spin_lock(&memcg_oom_lock); |
3761 | ||
c0ff4b85 | 3762 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3763 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3764 | |
3765 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3766 | return 0; |
3767 | } | |
3768 | ||
c0ff4b85 | 3769 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3770 | { |
7d74b06f KH |
3771 | struct mem_cgroup *iter; |
3772 | ||
c0ff4b85 | 3773 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3774 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3775 | } |
3776 | ||
59b6f873 | 3777 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3778 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3779 | { |
2c488db2 KS |
3780 | struct mem_cgroup_thresholds *thresholds; |
3781 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3782 | unsigned long threshold; |
3783 | unsigned long usage; | |
2c488db2 | 3784 | int i, size, ret; |
2e72b634 | 3785 | |
650c5e56 | 3786 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3787 | if (ret) |
3788 | return ret; | |
3789 | ||
3790 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3791 | |
05b84301 | 3792 | if (type == _MEM) { |
2c488db2 | 3793 | thresholds = &memcg->thresholds; |
ce00a967 | 3794 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3795 | } else if (type == _MEMSWAP) { |
2c488db2 | 3796 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3797 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3798 | } else |
2e72b634 KS |
3799 | BUG(); |
3800 | ||
2e72b634 | 3801 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3802 | if (thresholds->primary) |
2e72b634 KS |
3803 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3804 | ||
2c488db2 | 3805 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3806 | |
3807 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3808 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3809 | GFP_KERNEL); |
2c488db2 | 3810 | if (!new) { |
2e72b634 KS |
3811 | ret = -ENOMEM; |
3812 | goto unlock; | |
3813 | } | |
2c488db2 | 3814 | new->size = size; |
2e72b634 KS |
3815 | |
3816 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3817 | if (thresholds->primary) { |
3818 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3819 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3820 | } |
3821 | ||
2e72b634 | 3822 | /* Add new threshold */ |
2c488db2 KS |
3823 | new->entries[size - 1].eventfd = eventfd; |
3824 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3825 | |
3826 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3827 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3828 | compare_thresholds, NULL); |
3829 | ||
3830 | /* Find current threshold */ | |
2c488db2 | 3831 | new->current_threshold = -1; |
2e72b634 | 3832 | for (i = 0; i < size; i++) { |
748dad36 | 3833 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3834 | /* |
2c488db2 KS |
3835 | * new->current_threshold will not be used until |
3836 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3837 | * it here. |
3838 | */ | |
2c488db2 | 3839 | ++new->current_threshold; |
748dad36 SZ |
3840 | } else |
3841 | break; | |
2e72b634 KS |
3842 | } |
3843 | ||
2c488db2 KS |
3844 | /* Free old spare buffer and save old primary buffer as spare */ |
3845 | kfree(thresholds->spare); | |
3846 | thresholds->spare = thresholds->primary; | |
3847 | ||
3848 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3849 | |
907860ed | 3850 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3851 | synchronize_rcu(); |
3852 | ||
2e72b634 KS |
3853 | unlock: |
3854 | mutex_unlock(&memcg->thresholds_lock); | |
3855 | ||
3856 | return ret; | |
3857 | } | |
3858 | ||
59b6f873 | 3859 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3860 | struct eventfd_ctx *eventfd, const char *args) |
3861 | { | |
59b6f873 | 3862 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3863 | } |
3864 | ||
59b6f873 | 3865 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3866 | struct eventfd_ctx *eventfd, const char *args) |
3867 | { | |
59b6f873 | 3868 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3869 | } |
3870 | ||
59b6f873 | 3871 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3872 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3873 | { |
2c488db2 KS |
3874 | struct mem_cgroup_thresholds *thresholds; |
3875 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3876 | unsigned long usage; |
2c488db2 | 3877 | int i, j, size; |
2e72b634 KS |
3878 | |
3879 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3880 | |
3881 | if (type == _MEM) { | |
2c488db2 | 3882 | thresholds = &memcg->thresholds; |
ce00a967 | 3883 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3884 | } else if (type == _MEMSWAP) { |
2c488db2 | 3885 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3886 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3887 | } else |
2e72b634 KS |
3888 | BUG(); |
3889 | ||
371528ca AV |
3890 | if (!thresholds->primary) |
3891 | goto unlock; | |
3892 | ||
2e72b634 KS |
3893 | /* Check if a threshold crossed before removing */ |
3894 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3895 | ||
3896 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3897 | size = 0; |
3898 | for (i = 0; i < thresholds->primary->size; i++) { | |
3899 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3900 | size++; |
3901 | } | |
3902 | ||
2c488db2 | 3903 | new = thresholds->spare; |
907860ed | 3904 | |
2e72b634 KS |
3905 | /* Set thresholds array to NULL if we don't have thresholds */ |
3906 | if (!size) { | |
2c488db2 KS |
3907 | kfree(new); |
3908 | new = NULL; | |
907860ed | 3909 | goto swap_buffers; |
2e72b634 KS |
3910 | } |
3911 | ||
2c488db2 | 3912 | new->size = size; |
2e72b634 KS |
3913 | |
3914 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3915 | new->current_threshold = -1; |
3916 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3917 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3918 | continue; |
3919 | ||
2c488db2 | 3920 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3921 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3922 | /* |
2c488db2 | 3923 | * new->current_threshold will not be used |
2e72b634 KS |
3924 | * until rcu_assign_pointer(), so it's safe to increment |
3925 | * it here. | |
3926 | */ | |
2c488db2 | 3927 | ++new->current_threshold; |
2e72b634 KS |
3928 | } |
3929 | j++; | |
3930 | } | |
3931 | ||
907860ed | 3932 | swap_buffers: |
2c488db2 KS |
3933 | /* Swap primary and spare array */ |
3934 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
3935 | /* If all events are unregistered, free the spare array */ |
3936 | if (!new) { | |
3937 | kfree(thresholds->spare); | |
3938 | thresholds->spare = NULL; | |
3939 | } | |
3940 | ||
2c488db2 | 3941 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3942 | |
907860ed | 3943 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3944 | synchronize_rcu(); |
371528ca | 3945 | unlock: |
2e72b634 | 3946 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3947 | } |
c1e862c1 | 3948 | |
59b6f873 | 3949 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3950 | struct eventfd_ctx *eventfd) |
3951 | { | |
59b6f873 | 3952 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3953 | } |
3954 | ||
59b6f873 | 3955 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3956 | struct eventfd_ctx *eventfd) |
3957 | { | |
59b6f873 | 3958 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3959 | } |
3960 | ||
59b6f873 | 3961 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3962 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3963 | { |
9490ff27 | 3964 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3965 | |
9490ff27 KH |
3966 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3967 | if (!event) | |
3968 | return -ENOMEM; | |
3969 | ||
1af8efe9 | 3970 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3971 | |
3972 | event->eventfd = eventfd; | |
3973 | list_add(&event->list, &memcg->oom_notify); | |
3974 | ||
3975 | /* already in OOM ? */ | |
79dfdacc | 3976 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 3977 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3978 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3979 | |
3980 | return 0; | |
3981 | } | |
3982 | ||
59b6f873 | 3983 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3984 | struct eventfd_ctx *eventfd) |
9490ff27 | 3985 | { |
9490ff27 | 3986 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3987 | |
1af8efe9 | 3988 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3989 | |
c0ff4b85 | 3990 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3991 | if (ev->eventfd == eventfd) { |
3992 | list_del(&ev->list); | |
3993 | kfree(ev); | |
3994 | } | |
3995 | } | |
3996 | ||
1af8efe9 | 3997 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3998 | } |
3999 | ||
2da8ca82 | 4000 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4001 | { |
2da8ca82 | 4002 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4003 | |
791badbd TH |
4004 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4005 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4006 | return 0; |
4007 | } | |
4008 | ||
182446d0 | 4009 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4010 | struct cftype *cft, u64 val) |
4011 | { | |
182446d0 | 4012 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4013 | |
4014 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4015 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4016 | return -EINVAL; |
4017 | ||
c0ff4b85 | 4018 | memcg->oom_kill_disable = val; |
4d845ebf | 4019 | if (!val) |
c0ff4b85 | 4020 | memcg_oom_recover(memcg); |
3dae7fec | 4021 | |
3c11ecf4 KH |
4022 | return 0; |
4023 | } | |
4024 | ||
c255a458 | 4025 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4026 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4027 | { |
55007d84 GC |
4028 | int ret; |
4029 | ||
55007d84 GC |
4030 | ret = memcg_propagate_kmem(memcg); |
4031 | if (ret) | |
4032 | return ret; | |
2633d7a0 | 4033 | |
1d62e436 | 4034 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4035 | } |
e5671dfa | 4036 | |
10d5ebf4 | 4037 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4038 | { |
d5b3cf71 | 4039 | memcg_destroy_kmem_caches(memcg); |
1d62e436 | 4040 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 4041 | } |
e5671dfa | 4042 | #else |
cbe128e3 | 4043 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4044 | { |
4045 | return 0; | |
4046 | } | |
d1a4c0b3 | 4047 | |
10d5ebf4 LZ |
4048 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4049 | { | |
4050 | } | |
e5671dfa GC |
4051 | #endif |
4052 | ||
3bc942f3 TH |
4053 | /* |
4054 | * DO NOT USE IN NEW FILES. | |
4055 | * | |
4056 | * "cgroup.event_control" implementation. | |
4057 | * | |
4058 | * This is way over-engineered. It tries to support fully configurable | |
4059 | * events for each user. Such level of flexibility is completely | |
4060 | * unnecessary especially in the light of the planned unified hierarchy. | |
4061 | * | |
4062 | * Please deprecate this and replace with something simpler if at all | |
4063 | * possible. | |
4064 | */ | |
4065 | ||
79bd9814 TH |
4066 | /* |
4067 | * Unregister event and free resources. | |
4068 | * | |
4069 | * Gets called from workqueue. | |
4070 | */ | |
3bc942f3 | 4071 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4072 | { |
3bc942f3 TH |
4073 | struct mem_cgroup_event *event = |
4074 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4075 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4076 | |
4077 | remove_wait_queue(event->wqh, &event->wait); | |
4078 | ||
59b6f873 | 4079 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4080 | |
4081 | /* Notify userspace the event is going away. */ | |
4082 | eventfd_signal(event->eventfd, 1); | |
4083 | ||
4084 | eventfd_ctx_put(event->eventfd); | |
4085 | kfree(event); | |
59b6f873 | 4086 | css_put(&memcg->css); |
79bd9814 TH |
4087 | } |
4088 | ||
4089 | /* | |
4090 | * Gets called on POLLHUP on eventfd when user closes it. | |
4091 | * | |
4092 | * Called with wqh->lock held and interrupts disabled. | |
4093 | */ | |
3bc942f3 TH |
4094 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4095 | int sync, void *key) | |
79bd9814 | 4096 | { |
3bc942f3 TH |
4097 | struct mem_cgroup_event *event = |
4098 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4099 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4100 | unsigned long flags = (unsigned long)key; |
4101 | ||
4102 | if (flags & POLLHUP) { | |
4103 | /* | |
4104 | * If the event has been detached at cgroup removal, we | |
4105 | * can simply return knowing the other side will cleanup | |
4106 | * for us. | |
4107 | * | |
4108 | * We can't race against event freeing since the other | |
4109 | * side will require wqh->lock via remove_wait_queue(), | |
4110 | * which we hold. | |
4111 | */ | |
fba94807 | 4112 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4113 | if (!list_empty(&event->list)) { |
4114 | list_del_init(&event->list); | |
4115 | /* | |
4116 | * We are in atomic context, but cgroup_event_remove() | |
4117 | * may sleep, so we have to call it in workqueue. | |
4118 | */ | |
4119 | schedule_work(&event->remove); | |
4120 | } | |
fba94807 | 4121 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4122 | } |
4123 | ||
4124 | return 0; | |
4125 | } | |
4126 | ||
3bc942f3 | 4127 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4128 | wait_queue_head_t *wqh, poll_table *pt) |
4129 | { | |
3bc942f3 TH |
4130 | struct mem_cgroup_event *event = |
4131 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4132 | |
4133 | event->wqh = wqh; | |
4134 | add_wait_queue(wqh, &event->wait); | |
4135 | } | |
4136 | ||
4137 | /* | |
3bc942f3 TH |
4138 | * DO NOT USE IN NEW FILES. |
4139 | * | |
79bd9814 TH |
4140 | * Parse input and register new cgroup event handler. |
4141 | * | |
4142 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4143 | * Interpretation of args is defined by control file implementation. | |
4144 | */ | |
451af504 TH |
4145 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4146 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4147 | { |
451af504 | 4148 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4149 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4150 | struct mem_cgroup_event *event; |
79bd9814 TH |
4151 | struct cgroup_subsys_state *cfile_css; |
4152 | unsigned int efd, cfd; | |
4153 | struct fd efile; | |
4154 | struct fd cfile; | |
fba94807 | 4155 | const char *name; |
79bd9814 TH |
4156 | char *endp; |
4157 | int ret; | |
4158 | ||
451af504 TH |
4159 | buf = strstrip(buf); |
4160 | ||
4161 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4162 | if (*endp != ' ') |
4163 | return -EINVAL; | |
451af504 | 4164 | buf = endp + 1; |
79bd9814 | 4165 | |
451af504 | 4166 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4167 | if ((*endp != ' ') && (*endp != '\0')) |
4168 | return -EINVAL; | |
451af504 | 4169 | buf = endp + 1; |
79bd9814 TH |
4170 | |
4171 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4172 | if (!event) | |
4173 | return -ENOMEM; | |
4174 | ||
59b6f873 | 4175 | event->memcg = memcg; |
79bd9814 | 4176 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4177 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4178 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4179 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4180 | |
4181 | efile = fdget(efd); | |
4182 | if (!efile.file) { | |
4183 | ret = -EBADF; | |
4184 | goto out_kfree; | |
4185 | } | |
4186 | ||
4187 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4188 | if (IS_ERR(event->eventfd)) { | |
4189 | ret = PTR_ERR(event->eventfd); | |
4190 | goto out_put_efile; | |
4191 | } | |
4192 | ||
4193 | cfile = fdget(cfd); | |
4194 | if (!cfile.file) { | |
4195 | ret = -EBADF; | |
4196 | goto out_put_eventfd; | |
4197 | } | |
4198 | ||
4199 | /* the process need read permission on control file */ | |
4200 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4201 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4202 | if (ret < 0) | |
4203 | goto out_put_cfile; | |
4204 | ||
fba94807 TH |
4205 | /* |
4206 | * Determine the event callbacks and set them in @event. This used | |
4207 | * to be done via struct cftype but cgroup core no longer knows | |
4208 | * about these events. The following is crude but the whole thing | |
4209 | * is for compatibility anyway. | |
3bc942f3 TH |
4210 | * |
4211 | * DO NOT ADD NEW FILES. | |
fba94807 | 4212 | */ |
b583043e | 4213 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4214 | |
4215 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4216 | event->register_event = mem_cgroup_usage_register_event; | |
4217 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4218 | } else if (!strcmp(name, "memory.oom_control")) { | |
4219 | event->register_event = mem_cgroup_oom_register_event; | |
4220 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4221 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4222 | event->register_event = vmpressure_register_event; | |
4223 | event->unregister_event = vmpressure_unregister_event; | |
4224 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4225 | event->register_event = memsw_cgroup_usage_register_event; |
4226 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4227 | } else { |
4228 | ret = -EINVAL; | |
4229 | goto out_put_cfile; | |
4230 | } | |
4231 | ||
79bd9814 | 4232 | /* |
b5557c4c TH |
4233 | * Verify @cfile should belong to @css. Also, remaining events are |
4234 | * automatically removed on cgroup destruction but the removal is | |
4235 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4236 | */ |
b583043e | 4237 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4238 | &memory_cgrp_subsys); |
79bd9814 | 4239 | ret = -EINVAL; |
5a17f543 | 4240 | if (IS_ERR(cfile_css)) |
79bd9814 | 4241 | goto out_put_cfile; |
5a17f543 TH |
4242 | if (cfile_css != css) { |
4243 | css_put(cfile_css); | |
79bd9814 | 4244 | goto out_put_cfile; |
5a17f543 | 4245 | } |
79bd9814 | 4246 | |
451af504 | 4247 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4248 | if (ret) |
4249 | goto out_put_css; | |
4250 | ||
4251 | efile.file->f_op->poll(efile.file, &event->pt); | |
4252 | ||
fba94807 TH |
4253 | spin_lock(&memcg->event_list_lock); |
4254 | list_add(&event->list, &memcg->event_list); | |
4255 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4256 | |
4257 | fdput(cfile); | |
4258 | fdput(efile); | |
4259 | ||
451af504 | 4260 | return nbytes; |
79bd9814 TH |
4261 | |
4262 | out_put_css: | |
b5557c4c | 4263 | css_put(css); |
79bd9814 TH |
4264 | out_put_cfile: |
4265 | fdput(cfile); | |
4266 | out_put_eventfd: | |
4267 | eventfd_ctx_put(event->eventfd); | |
4268 | out_put_efile: | |
4269 | fdput(efile); | |
4270 | out_kfree: | |
4271 | kfree(event); | |
4272 | ||
4273 | return ret; | |
4274 | } | |
4275 | ||
241994ed | 4276 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4277 | { |
0eea1030 | 4278 | .name = "usage_in_bytes", |
8c7c6e34 | 4279 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4280 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4281 | }, |
c84872e1 PE |
4282 | { |
4283 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4284 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4285 | .write = mem_cgroup_reset, |
791badbd | 4286 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4287 | }, |
8cdea7c0 | 4288 | { |
0eea1030 | 4289 | .name = "limit_in_bytes", |
8c7c6e34 | 4290 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4291 | .write = mem_cgroup_write, |
791badbd | 4292 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4293 | }, |
296c81d8 BS |
4294 | { |
4295 | .name = "soft_limit_in_bytes", | |
4296 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4297 | .write = mem_cgroup_write, |
791badbd | 4298 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4299 | }, |
8cdea7c0 BS |
4300 | { |
4301 | .name = "failcnt", | |
8c7c6e34 | 4302 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4303 | .write = mem_cgroup_reset, |
791badbd | 4304 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4305 | }, |
d2ceb9b7 KH |
4306 | { |
4307 | .name = "stat", | |
2da8ca82 | 4308 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4309 | }, |
c1e862c1 KH |
4310 | { |
4311 | .name = "force_empty", | |
6770c64e | 4312 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4313 | }, |
18f59ea7 BS |
4314 | { |
4315 | .name = "use_hierarchy", | |
4316 | .write_u64 = mem_cgroup_hierarchy_write, | |
4317 | .read_u64 = mem_cgroup_hierarchy_read, | |
4318 | }, | |
79bd9814 | 4319 | { |
3bc942f3 | 4320 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4321 | .write = memcg_write_event_control, |
79bd9814 TH |
4322 | .flags = CFTYPE_NO_PREFIX, |
4323 | .mode = S_IWUGO, | |
4324 | }, | |
a7885eb8 KM |
4325 | { |
4326 | .name = "swappiness", | |
4327 | .read_u64 = mem_cgroup_swappiness_read, | |
4328 | .write_u64 = mem_cgroup_swappiness_write, | |
4329 | }, | |
7dc74be0 DN |
4330 | { |
4331 | .name = "move_charge_at_immigrate", | |
4332 | .read_u64 = mem_cgroup_move_charge_read, | |
4333 | .write_u64 = mem_cgroup_move_charge_write, | |
4334 | }, | |
9490ff27 KH |
4335 | { |
4336 | .name = "oom_control", | |
2da8ca82 | 4337 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4338 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4339 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4340 | }, | |
70ddf637 AV |
4341 | { |
4342 | .name = "pressure_level", | |
70ddf637 | 4343 | }, |
406eb0c9 YH |
4344 | #ifdef CONFIG_NUMA |
4345 | { | |
4346 | .name = "numa_stat", | |
2da8ca82 | 4347 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4348 | }, |
4349 | #endif | |
510fc4e1 GC |
4350 | #ifdef CONFIG_MEMCG_KMEM |
4351 | { | |
4352 | .name = "kmem.limit_in_bytes", | |
4353 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4354 | .write = mem_cgroup_write, |
791badbd | 4355 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4356 | }, |
4357 | { | |
4358 | .name = "kmem.usage_in_bytes", | |
4359 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4360 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4361 | }, |
4362 | { | |
4363 | .name = "kmem.failcnt", | |
4364 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4365 | .write = mem_cgroup_reset, |
791badbd | 4366 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4367 | }, |
4368 | { | |
4369 | .name = "kmem.max_usage_in_bytes", | |
4370 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4371 | .write = mem_cgroup_reset, |
791badbd | 4372 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4373 | }, |
749c5415 GC |
4374 | #ifdef CONFIG_SLABINFO |
4375 | { | |
4376 | .name = "kmem.slabinfo", | |
b047501c VD |
4377 | .seq_start = slab_start, |
4378 | .seq_next = slab_next, | |
4379 | .seq_stop = slab_stop, | |
4380 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4381 | }, |
4382 | #endif | |
8c7c6e34 | 4383 | #endif |
6bc10349 | 4384 | { }, /* terminate */ |
af36f906 | 4385 | }; |
8c7c6e34 | 4386 | |
2d11085e MH |
4387 | #ifdef CONFIG_MEMCG_SWAP |
4388 | static struct cftype memsw_cgroup_files[] = { | |
4389 | { | |
4390 | .name = "memsw.usage_in_bytes", | |
4391 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 4392 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4393 | }, |
4394 | { | |
4395 | .name = "memsw.max_usage_in_bytes", | |
4396 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 4397 | .write = mem_cgroup_reset, |
791badbd | 4398 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4399 | }, |
4400 | { | |
4401 | .name = "memsw.limit_in_bytes", | |
4402 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 4403 | .write = mem_cgroup_write, |
791badbd | 4404 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4405 | }, |
4406 | { | |
4407 | .name = "memsw.failcnt", | |
4408 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 4409 | .write = mem_cgroup_reset, |
791badbd | 4410 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4411 | }, |
4412 | { }, /* terminate */ | |
4413 | }; | |
4414 | #endif | |
c0ff4b85 | 4415 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4416 | { |
4417 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4418 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4419 | int zone, tmp = node; |
1ecaab2b KH |
4420 | /* |
4421 | * This routine is called against possible nodes. | |
4422 | * But it's BUG to call kmalloc() against offline node. | |
4423 | * | |
4424 | * TODO: this routine can waste much memory for nodes which will | |
4425 | * never be onlined. It's better to use memory hotplug callback | |
4426 | * function. | |
4427 | */ | |
41e3355d KH |
4428 | if (!node_state(node, N_NORMAL_MEMORY)) |
4429 | tmp = -1; | |
17295c88 | 4430 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4431 | if (!pn) |
4432 | return 1; | |
1ecaab2b | 4433 | |
1ecaab2b KH |
4434 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4435 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4436 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4437 | mz->usage_in_excess = 0; |
4438 | mz->on_tree = false; | |
d79154bb | 4439 | mz->memcg = memcg; |
1ecaab2b | 4440 | } |
54f72fe0 | 4441 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4442 | return 0; |
4443 | } | |
4444 | ||
c0ff4b85 | 4445 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4446 | { |
54f72fe0 | 4447 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4448 | } |
4449 | ||
33327948 KH |
4450 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4451 | { | |
d79154bb | 4452 | struct mem_cgroup *memcg; |
8ff69e2c | 4453 | size_t size; |
33327948 | 4454 | |
8ff69e2c VD |
4455 | size = sizeof(struct mem_cgroup); |
4456 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4457 | |
8ff69e2c | 4458 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4459 | if (!memcg) |
e7bbcdf3 DC |
4460 | return NULL; |
4461 | ||
d79154bb HD |
4462 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4463 | if (!memcg->stat) | |
d2e61b8d | 4464 | goto out_free; |
d79154bb HD |
4465 | spin_lock_init(&memcg->pcp_counter_lock); |
4466 | return memcg; | |
d2e61b8d DC |
4467 | |
4468 | out_free: | |
8ff69e2c | 4469 | kfree(memcg); |
d2e61b8d | 4470 | return NULL; |
33327948 KH |
4471 | } |
4472 | ||
59927fb9 | 4473 | /* |
c8b2a36f GC |
4474 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4475 | * (scanning all at force_empty is too costly...) | |
4476 | * | |
4477 | * Instead of clearing all references at force_empty, we remember | |
4478 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4479 | * it goes down to 0. | |
4480 | * | |
4481 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4482 | */ |
c8b2a36f GC |
4483 | |
4484 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4485 | { |
c8b2a36f | 4486 | int node; |
59927fb9 | 4487 | |
bb4cc1a8 | 4488 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4489 | |
4490 | for_each_node(node) | |
4491 | free_mem_cgroup_per_zone_info(memcg, node); | |
4492 | ||
4493 | free_percpu(memcg->stat); | |
4494 | ||
a8964b9b | 4495 | disarm_static_keys(memcg); |
8ff69e2c | 4496 | kfree(memcg); |
59927fb9 | 4497 | } |
3afe36b1 | 4498 | |
7bcc1bb1 DN |
4499 | /* |
4500 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4501 | */ | |
e1aab161 | 4502 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4503 | { |
3e32cb2e | 4504 | if (!memcg->memory.parent) |
7bcc1bb1 | 4505 | return NULL; |
3e32cb2e | 4506 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4507 | } |
e1aab161 | 4508 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4509 | |
bb4cc1a8 AM |
4510 | static void __init mem_cgroup_soft_limit_tree_init(void) |
4511 | { | |
9c608dbe | 4512 | int node; |
bb4cc1a8 AM |
4513 | |
4514 | for_each_node(node) { | |
9c608dbe JW |
4515 | struct mem_cgroup_tree_per_node *rtpn; |
4516 | int zone; | |
bb4cc1a8 | 4517 | |
9c608dbe JW |
4518 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, |
4519 | node_online(node) ? node : NUMA_NO_NODE); | |
bb4cc1a8 AM |
4520 | |
4521 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
9c608dbe JW |
4522 | struct mem_cgroup_tree_per_zone *rtpz; |
4523 | ||
bb4cc1a8 AM |
4524 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
4525 | rtpz->rb_root = RB_ROOT; | |
4526 | spin_lock_init(&rtpz->lock); | |
4527 | } | |
9c608dbe | 4528 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
bb4cc1a8 AM |
4529 | } |
4530 | } | |
4531 | ||
0eb253e2 | 4532 | static struct cgroup_subsys_state * __ref |
eb95419b | 4533 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4534 | { |
d142e3e6 | 4535 | struct mem_cgroup *memcg; |
04046e1a | 4536 | long error = -ENOMEM; |
6d12e2d8 | 4537 | int node; |
8cdea7c0 | 4538 | |
c0ff4b85 R |
4539 | memcg = mem_cgroup_alloc(); |
4540 | if (!memcg) | |
04046e1a | 4541 | return ERR_PTR(error); |
78fb7466 | 4542 | |
3ed28fa1 | 4543 | for_each_node(node) |
c0ff4b85 | 4544 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4545 | goto free_out; |
f64c3f54 | 4546 | |
c077719b | 4547 | /* root ? */ |
eb95419b | 4548 | if (parent_css == NULL) { |
a41c58a6 | 4549 | root_mem_cgroup = memcg; |
3e32cb2e | 4550 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4551 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4552 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4553 | page_counter_init(&memcg->memsw, NULL); |
4554 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4555 | } |
28dbc4b6 | 4556 | |
d142e3e6 GC |
4557 | memcg->last_scanned_node = MAX_NUMNODES; |
4558 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4559 | memcg->move_charge_at_immigrate = 0; |
4560 | mutex_init(&memcg->thresholds_lock); | |
4561 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4562 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4563 | INIT_LIST_HEAD(&memcg->event_list); |
4564 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4565 | #ifdef CONFIG_MEMCG_KMEM |
4566 | memcg->kmemcg_id = -1; | |
900a38f0 | 4567 | #endif |
d142e3e6 GC |
4568 | |
4569 | return &memcg->css; | |
4570 | ||
4571 | free_out: | |
4572 | __mem_cgroup_free(memcg); | |
4573 | return ERR_PTR(error); | |
4574 | } | |
4575 | ||
4576 | static int | |
eb95419b | 4577 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4578 | { |
eb95419b | 4579 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4580 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4581 | int ret; |
d142e3e6 | 4582 | |
15a4c835 | 4583 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4584 | return -ENOSPC; |
4585 | ||
63876986 | 4586 | if (!parent) |
d142e3e6 GC |
4587 | return 0; |
4588 | ||
0999821b | 4589 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4590 | |
4591 | memcg->use_hierarchy = parent->use_hierarchy; | |
4592 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4593 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4594 | ||
4595 | if (parent->use_hierarchy) { | |
3e32cb2e | 4596 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4597 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4598 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4599 | page_counter_init(&memcg->memsw, &parent->memsw); |
4600 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4601 | |
7bcc1bb1 | 4602 | /* |
8d76a979 LZ |
4603 | * No need to take a reference to the parent because cgroup |
4604 | * core guarantees its existence. | |
7bcc1bb1 | 4605 | */ |
18f59ea7 | 4606 | } else { |
3e32cb2e | 4607 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4608 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4609 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4610 | page_counter_init(&memcg->memsw, NULL); |
4611 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4612 | /* |
4613 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4614 | * much sense so let cgroup subsystem know about this | |
4615 | * unfortunate state in our controller. | |
4616 | */ | |
d142e3e6 | 4617 | if (parent != root_mem_cgroup) |
073219e9 | 4618 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4619 | } |
0999821b | 4620 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4621 | |
2f7dd7a4 JW |
4622 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4623 | if (ret) | |
4624 | return ret; | |
4625 | ||
4626 | /* | |
4627 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4628 | * orders reading memcg->initialized against its callers | |
4629 | * reading the memcg members. | |
4630 | */ | |
4631 | smp_store_release(&memcg->initialized, 1); | |
4632 | ||
4633 | return 0; | |
8cdea7c0 BS |
4634 | } |
4635 | ||
eb95419b | 4636 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4637 | { |
eb95419b | 4638 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4639 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4640 | |
4641 | /* | |
4642 | * Unregister events and notify userspace. | |
4643 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4644 | * directory to avoid race between userspace and kernelspace. | |
4645 | */ | |
fba94807 TH |
4646 | spin_lock(&memcg->event_list_lock); |
4647 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4648 | list_del_init(&event->list); |
4649 | schedule_work(&event->remove); | |
4650 | } | |
fba94807 | 4651 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4652 | |
33cb876e | 4653 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
4654 | } |
4655 | ||
eb95419b | 4656 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4657 | { |
eb95419b | 4658 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4659 | |
10d5ebf4 | 4660 | memcg_destroy_kmem(memcg); |
465939a1 | 4661 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4662 | } |
4663 | ||
1ced953b TH |
4664 | /** |
4665 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4666 | * @css: the target css | |
4667 | * | |
4668 | * Reset the states of the mem_cgroup associated with @css. This is | |
4669 | * invoked when the userland requests disabling on the default hierarchy | |
4670 | * but the memcg is pinned through dependency. The memcg should stop | |
4671 | * applying policies and should revert to the vanilla state as it may be | |
4672 | * made visible again. | |
4673 | * | |
4674 | * The current implementation only resets the essential configurations. | |
4675 | * This needs to be expanded to cover all the visible parts. | |
4676 | */ | |
4677 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4678 | { | |
4679 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4680 | ||
3e32cb2e JW |
4681 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4682 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4683 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4684 | memcg->low = 0; |
4685 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4686 | memcg->soft_limit = PAGE_COUNTER_MAX; |
1ced953b TH |
4687 | } |
4688 | ||
02491447 | 4689 | #ifdef CONFIG_MMU |
7dc74be0 | 4690 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4691 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4692 | { |
05b84301 | 4693 | int ret; |
9476db97 JW |
4694 | |
4695 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 4696 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 4697 | if (!ret) { |
854ffa8d | 4698 | mc.precharge += count; |
854ffa8d DN |
4699 | return ret; |
4700 | } | |
692e7c45 | 4701 | if (ret == -EINTR) { |
00501b53 | 4702 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
4703 | return ret; |
4704 | } | |
9476db97 JW |
4705 | |
4706 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4707 | while (count--) { |
00501b53 | 4708 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
4709 | /* |
4710 | * In case of failure, any residual charges against | |
4711 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
4712 | * later on. However, cancel any charges that are |
4713 | * bypassed to root right away or they'll be lost. | |
9476db97 | 4714 | */ |
692e7c45 | 4715 | if (ret == -EINTR) |
00501b53 | 4716 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 4717 | if (ret) |
38c5d72f | 4718 | return ret; |
854ffa8d | 4719 | mc.precharge++; |
9476db97 | 4720 | cond_resched(); |
854ffa8d | 4721 | } |
9476db97 | 4722 | return 0; |
4ffef5fe DN |
4723 | } |
4724 | ||
4725 | /** | |
8d32ff84 | 4726 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4727 | * @vma: the vma the pte to be checked belongs |
4728 | * @addr: the address corresponding to the pte to be checked | |
4729 | * @ptent: the pte to be checked | |
02491447 | 4730 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4731 | * |
4732 | * Returns | |
4733 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4734 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4735 | * move charge. if @target is not NULL, the page is stored in target->page | |
4736 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4737 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4738 | * target for charge migration. if @target is not NULL, the entry is stored | |
4739 | * in target->ent. | |
4ffef5fe DN |
4740 | * |
4741 | * Called with pte lock held. | |
4742 | */ | |
4ffef5fe DN |
4743 | union mc_target { |
4744 | struct page *page; | |
02491447 | 4745 | swp_entry_t ent; |
4ffef5fe DN |
4746 | }; |
4747 | ||
4ffef5fe | 4748 | enum mc_target_type { |
8d32ff84 | 4749 | MC_TARGET_NONE = 0, |
4ffef5fe | 4750 | MC_TARGET_PAGE, |
02491447 | 4751 | MC_TARGET_SWAP, |
4ffef5fe DN |
4752 | }; |
4753 | ||
90254a65 DN |
4754 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4755 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4756 | { |
90254a65 | 4757 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4758 | |
90254a65 DN |
4759 | if (!page || !page_mapped(page)) |
4760 | return NULL; | |
4761 | if (PageAnon(page)) { | |
1dfab5ab | 4762 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4763 | return NULL; |
1dfab5ab JW |
4764 | } else { |
4765 | if (!(mc.flags & MOVE_FILE)) | |
4766 | return NULL; | |
4767 | } | |
90254a65 DN |
4768 | if (!get_page_unless_zero(page)) |
4769 | return NULL; | |
4770 | ||
4771 | return page; | |
4772 | } | |
4773 | ||
4b91355e | 4774 | #ifdef CONFIG_SWAP |
90254a65 DN |
4775 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4776 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4777 | { | |
90254a65 DN |
4778 | struct page *page = NULL; |
4779 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4780 | ||
1dfab5ab | 4781 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4782 | return NULL; |
4b91355e KH |
4783 | /* |
4784 | * Because lookup_swap_cache() updates some statistics counter, | |
4785 | * we call find_get_page() with swapper_space directly. | |
4786 | */ | |
33806f06 | 4787 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
4788 | if (do_swap_account) |
4789 | entry->val = ent.val; | |
4790 | ||
4791 | return page; | |
4792 | } | |
4b91355e KH |
4793 | #else |
4794 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4795 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4796 | { | |
4797 | return NULL; | |
4798 | } | |
4799 | #endif | |
90254a65 | 4800 | |
87946a72 DN |
4801 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4802 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4803 | { | |
4804 | struct page *page = NULL; | |
87946a72 DN |
4805 | struct address_space *mapping; |
4806 | pgoff_t pgoff; | |
4807 | ||
4808 | if (!vma->vm_file) /* anonymous vma */ | |
4809 | return NULL; | |
1dfab5ab | 4810 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4811 | return NULL; |
4812 | ||
87946a72 | 4813 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4814 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4815 | |
4816 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4817 | #ifdef CONFIG_SWAP |
4818 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4819 | if (shmem_mapping(mapping)) { |
4820 | page = find_get_entry(mapping, pgoff); | |
4821 | if (radix_tree_exceptional_entry(page)) { | |
4822 | swp_entry_t swp = radix_to_swp_entry(page); | |
4823 | if (do_swap_account) | |
4824 | *entry = swp; | |
4825 | page = find_get_page(swap_address_space(swp), swp.val); | |
4826 | } | |
4827 | } else | |
4828 | page = find_get_page(mapping, pgoff); | |
4829 | #else | |
4830 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4831 | #endif |
87946a72 DN |
4832 | return page; |
4833 | } | |
4834 | ||
8d32ff84 | 4835 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4836 | unsigned long addr, pte_t ptent, union mc_target *target) |
4837 | { | |
4838 | struct page *page = NULL; | |
8d32ff84 | 4839 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4840 | swp_entry_t ent = { .val = 0 }; |
4841 | ||
4842 | if (pte_present(ptent)) | |
4843 | page = mc_handle_present_pte(vma, addr, ptent); | |
4844 | else if (is_swap_pte(ptent)) | |
4845 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4846 | else if (pte_none(ptent)) |
87946a72 | 4847 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4848 | |
4849 | if (!page && !ent.val) | |
8d32ff84 | 4850 | return ret; |
02491447 | 4851 | if (page) { |
02491447 | 4852 | /* |
0a31bc97 | 4853 | * Do only loose check w/o serialization. |
1306a85a | 4854 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4855 | * not under LRU exclusion. |
02491447 | 4856 | */ |
1306a85a | 4857 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4858 | ret = MC_TARGET_PAGE; |
4859 | if (target) | |
4860 | target->page = page; | |
4861 | } | |
4862 | if (!ret || !target) | |
4863 | put_page(page); | |
4864 | } | |
90254a65 DN |
4865 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4866 | if (ent.val && !ret && | |
34c00c31 | 4867 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4868 | ret = MC_TARGET_SWAP; |
4869 | if (target) | |
4870 | target->ent = ent; | |
4ffef5fe | 4871 | } |
4ffef5fe DN |
4872 | return ret; |
4873 | } | |
4874 | ||
12724850 NH |
4875 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4876 | /* | |
4877 | * We don't consider swapping or file mapped pages because THP does not | |
4878 | * support them for now. | |
4879 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4880 | */ | |
4881 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4882 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4883 | { | |
4884 | struct page *page = NULL; | |
12724850 NH |
4885 | enum mc_target_type ret = MC_TARGET_NONE; |
4886 | ||
4887 | page = pmd_page(pmd); | |
309381fe | 4888 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4889 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4890 | return ret; |
1306a85a | 4891 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4892 | ret = MC_TARGET_PAGE; |
4893 | if (target) { | |
4894 | get_page(page); | |
4895 | target->page = page; | |
4896 | } | |
4897 | } | |
4898 | return ret; | |
4899 | } | |
4900 | #else | |
4901 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4902 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4903 | { | |
4904 | return MC_TARGET_NONE; | |
4905 | } | |
4906 | #endif | |
4907 | ||
4ffef5fe DN |
4908 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4909 | unsigned long addr, unsigned long end, | |
4910 | struct mm_walk *walk) | |
4911 | { | |
4912 | struct vm_area_struct *vma = walk->private; | |
4913 | pte_t *pte; | |
4914 | spinlock_t *ptl; | |
4915 | ||
bf929152 | 4916 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
4917 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4918 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4919 | spin_unlock(ptl); |
1a5a9906 | 4920 | return 0; |
12724850 | 4921 | } |
03319327 | 4922 | |
45f83cef AA |
4923 | if (pmd_trans_unstable(pmd)) |
4924 | return 0; | |
4ffef5fe DN |
4925 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4926 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4927 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4928 | mc.precharge++; /* increment precharge temporarily */ |
4929 | pte_unmap_unlock(pte - 1, ptl); | |
4930 | cond_resched(); | |
4931 | ||
7dc74be0 DN |
4932 | return 0; |
4933 | } | |
4934 | ||
4ffef5fe DN |
4935 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4936 | { | |
4937 | unsigned long precharge; | |
4938 | struct vm_area_struct *vma; | |
4939 | ||
dfe076b0 | 4940 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
4941 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
4942 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
4943 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4944 | .mm = mm, | |
4945 | .private = vma, | |
4946 | }; | |
4947 | if (is_vm_hugetlb_page(vma)) | |
4948 | continue; | |
4ffef5fe DN |
4949 | walk_page_range(vma->vm_start, vma->vm_end, |
4950 | &mem_cgroup_count_precharge_walk); | |
4951 | } | |
dfe076b0 | 4952 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4953 | |
4954 | precharge = mc.precharge; | |
4955 | mc.precharge = 0; | |
4956 | ||
4957 | return precharge; | |
4958 | } | |
4959 | ||
4ffef5fe DN |
4960 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4961 | { | |
dfe076b0 DN |
4962 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4963 | ||
4964 | VM_BUG_ON(mc.moving_task); | |
4965 | mc.moving_task = current; | |
4966 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4967 | } |
4968 | ||
dfe076b0 DN |
4969 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4970 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4971 | { |
2bd9bb20 KH |
4972 | struct mem_cgroup *from = mc.from; |
4973 | struct mem_cgroup *to = mc.to; | |
4974 | ||
4ffef5fe | 4975 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4976 | if (mc.precharge) { |
00501b53 | 4977 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4978 | mc.precharge = 0; |
4979 | } | |
4980 | /* | |
4981 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4982 | * we must uncharge here. | |
4983 | */ | |
4984 | if (mc.moved_charge) { | |
00501b53 | 4985 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4986 | mc.moved_charge = 0; |
4ffef5fe | 4987 | } |
483c30b5 DN |
4988 | /* we must fixup refcnts and charges */ |
4989 | if (mc.moved_swap) { | |
483c30b5 | 4990 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4991 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4992 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4993 | |
05b84301 | 4994 | /* |
3e32cb2e JW |
4995 | * we charged both to->memory and to->memsw, so we |
4996 | * should uncharge to->memory. | |
05b84301 | 4997 | */ |
ce00a967 | 4998 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4999 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5000 | ||
e8ea14cc | 5001 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 5002 | |
4050377b | 5003 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
5004 | mc.moved_swap = 0; |
5005 | } | |
dfe076b0 DN |
5006 | memcg_oom_recover(from); |
5007 | memcg_oom_recover(to); | |
5008 | wake_up_all(&mc.waitq); | |
5009 | } | |
5010 | ||
5011 | static void mem_cgroup_clear_mc(void) | |
5012 | { | |
dfe076b0 DN |
5013 | /* |
5014 | * we must clear moving_task before waking up waiters at the end of | |
5015 | * task migration. | |
5016 | */ | |
5017 | mc.moving_task = NULL; | |
5018 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5019 | spin_lock(&mc.lock); |
4ffef5fe DN |
5020 | mc.from = NULL; |
5021 | mc.to = NULL; | |
2bd9bb20 | 5022 | spin_unlock(&mc.lock); |
4ffef5fe DN |
5023 | } |
5024 | ||
eb95419b | 5025 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5026 | struct cgroup_taskset *tset) |
7dc74be0 | 5027 | { |
2f7ee569 | 5028 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5029 | int ret = 0; |
eb95419b | 5030 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
1dfab5ab | 5031 | unsigned long move_flags; |
7dc74be0 | 5032 | |
ee5e8472 GC |
5033 | /* |
5034 | * We are now commited to this value whatever it is. Changes in this | |
5035 | * tunable will only affect upcoming migrations, not the current one. | |
5036 | * So we need to save it, and keep it going. | |
5037 | */ | |
1dfab5ab JW |
5038 | move_flags = ACCESS_ONCE(memcg->move_charge_at_immigrate); |
5039 | if (move_flags) { | |
7dc74be0 DN |
5040 | struct mm_struct *mm; |
5041 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5042 | ||
c0ff4b85 | 5043 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5044 | |
5045 | mm = get_task_mm(p); | |
5046 | if (!mm) | |
5047 | return 0; | |
7dc74be0 | 5048 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5049 | if (mm->owner == p) { |
5050 | VM_BUG_ON(mc.from); | |
5051 | VM_BUG_ON(mc.to); | |
5052 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5053 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5054 | VM_BUG_ON(mc.moved_swap); |
247b1447 | 5055 | |
2bd9bb20 | 5056 | spin_lock(&mc.lock); |
4ffef5fe | 5057 | mc.from = from; |
c0ff4b85 | 5058 | mc.to = memcg; |
1dfab5ab | 5059 | mc.flags = move_flags; |
2bd9bb20 | 5060 | spin_unlock(&mc.lock); |
dfe076b0 | 5061 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5062 | |
5063 | ret = mem_cgroup_precharge_mc(mm); | |
5064 | if (ret) | |
5065 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5066 | } |
5067 | mmput(mm); | |
7dc74be0 DN |
5068 | } |
5069 | return ret; | |
5070 | } | |
5071 | ||
eb95419b | 5072 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5073 | struct cgroup_taskset *tset) |
7dc74be0 | 5074 | { |
4e2f245d JW |
5075 | if (mc.to) |
5076 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5077 | } |
5078 | ||
4ffef5fe DN |
5079 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5080 | unsigned long addr, unsigned long end, | |
5081 | struct mm_walk *walk) | |
7dc74be0 | 5082 | { |
4ffef5fe DN |
5083 | int ret = 0; |
5084 | struct vm_area_struct *vma = walk->private; | |
5085 | pte_t *pte; | |
5086 | spinlock_t *ptl; | |
12724850 NH |
5087 | enum mc_target_type target_type; |
5088 | union mc_target target; | |
5089 | struct page *page; | |
4ffef5fe | 5090 | |
12724850 NH |
5091 | /* |
5092 | * We don't take compound_lock() here but no race with splitting thp | |
5093 | * happens because: | |
5094 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5095 | * under splitting, which means there's no concurrent thp split, | |
5096 | * - if another thread runs into split_huge_page() just after we | |
5097 | * entered this if-block, the thread must wait for page table lock | |
5098 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5099 | * part of thp split is not executed yet. | |
5100 | */ | |
bf929152 | 5101 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5102 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5103 | spin_unlock(ptl); |
12724850 NH |
5104 | return 0; |
5105 | } | |
5106 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5107 | if (target_type == MC_TARGET_PAGE) { | |
5108 | page = target.page; | |
5109 | if (!isolate_lru_page(page)) { | |
12724850 | 5110 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 5111 | mc.from, mc.to)) { |
12724850 NH |
5112 | mc.precharge -= HPAGE_PMD_NR; |
5113 | mc.moved_charge += HPAGE_PMD_NR; | |
5114 | } | |
5115 | putback_lru_page(page); | |
5116 | } | |
5117 | put_page(page); | |
5118 | } | |
bf929152 | 5119 | spin_unlock(ptl); |
1a5a9906 | 5120 | return 0; |
12724850 NH |
5121 | } |
5122 | ||
45f83cef AA |
5123 | if (pmd_trans_unstable(pmd)) |
5124 | return 0; | |
4ffef5fe DN |
5125 | retry: |
5126 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5127 | for (; addr != end; addr += PAGE_SIZE) { | |
5128 | pte_t ptent = *(pte++); | |
02491447 | 5129 | swp_entry_t ent; |
4ffef5fe DN |
5130 | |
5131 | if (!mc.precharge) | |
5132 | break; | |
5133 | ||
8d32ff84 | 5134 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5135 | case MC_TARGET_PAGE: |
5136 | page = target.page; | |
5137 | if (isolate_lru_page(page)) | |
5138 | goto put; | |
1306a85a | 5139 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 5140 | mc.precharge--; |
854ffa8d DN |
5141 | /* we uncharge from mc.from later. */ |
5142 | mc.moved_charge++; | |
4ffef5fe DN |
5143 | } |
5144 | putback_lru_page(page); | |
8d32ff84 | 5145 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5146 | put_page(page); |
5147 | break; | |
02491447 DN |
5148 | case MC_TARGET_SWAP: |
5149 | ent = target.ent; | |
e91cbb42 | 5150 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5151 | mc.precharge--; |
483c30b5 DN |
5152 | /* we fixup refcnts and charges later. */ |
5153 | mc.moved_swap++; | |
5154 | } | |
02491447 | 5155 | break; |
4ffef5fe DN |
5156 | default: |
5157 | break; | |
5158 | } | |
5159 | } | |
5160 | pte_unmap_unlock(pte - 1, ptl); | |
5161 | cond_resched(); | |
5162 | ||
5163 | if (addr != end) { | |
5164 | /* | |
5165 | * We have consumed all precharges we got in can_attach(). | |
5166 | * We try charge one by one, but don't do any additional | |
5167 | * charges to mc.to if we have failed in charge once in attach() | |
5168 | * phase. | |
5169 | */ | |
854ffa8d | 5170 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5171 | if (!ret) |
5172 | goto retry; | |
5173 | } | |
5174 | ||
5175 | return ret; | |
5176 | } | |
5177 | ||
5178 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5179 | { | |
5180 | struct vm_area_struct *vma; | |
5181 | ||
5182 | lru_add_drain_all(); | |
312722cb JW |
5183 | /* |
5184 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5185 | * move_lock while we're moving its pages to another memcg. | |
5186 | * Then wait for already started RCU-only updates to finish. | |
5187 | */ | |
5188 | atomic_inc(&mc.from->moving_account); | |
5189 | synchronize_rcu(); | |
dfe076b0 DN |
5190 | retry: |
5191 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5192 | /* | |
5193 | * Someone who are holding the mmap_sem might be waiting in | |
5194 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5195 | * and retry. Because we cancel precharges, we might not be able | |
5196 | * to move enough charges, but moving charge is a best-effort | |
5197 | * feature anyway, so it wouldn't be a big problem. | |
5198 | */ | |
5199 | __mem_cgroup_clear_mc(); | |
5200 | cond_resched(); | |
5201 | goto retry; | |
5202 | } | |
4ffef5fe DN |
5203 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5204 | int ret; | |
5205 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5206 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5207 | .mm = mm, | |
5208 | .private = vma, | |
5209 | }; | |
5210 | if (is_vm_hugetlb_page(vma)) | |
5211 | continue; | |
4ffef5fe DN |
5212 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5213 | &mem_cgroup_move_charge_walk); | |
5214 | if (ret) | |
5215 | /* | |
5216 | * means we have consumed all precharges and failed in | |
5217 | * doing additional charge. Just abandon here. | |
5218 | */ | |
5219 | break; | |
5220 | } | |
dfe076b0 | 5221 | up_read(&mm->mmap_sem); |
312722cb | 5222 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5223 | } |
5224 | ||
eb95419b | 5225 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5226 | struct cgroup_taskset *tset) |
67e465a7 | 5227 | { |
2f7ee569 | 5228 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5229 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5230 | |
dfe076b0 | 5231 | if (mm) { |
a433658c KM |
5232 | if (mc.to) |
5233 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5234 | mmput(mm); |
5235 | } | |
a433658c KM |
5236 | if (mc.to) |
5237 | mem_cgroup_clear_mc(); | |
67e465a7 | 5238 | } |
5cfb80a7 | 5239 | #else /* !CONFIG_MMU */ |
eb95419b | 5240 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5241 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5242 | { |
5243 | return 0; | |
5244 | } | |
eb95419b | 5245 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5246 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5247 | { |
5248 | } | |
eb95419b | 5249 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5250 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5251 | { |
5252 | } | |
5253 | #endif | |
67e465a7 | 5254 | |
f00baae7 TH |
5255 | /* |
5256 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5257 | * to verify whether we're attached to the default hierarchy on each mount |
5258 | * attempt. | |
f00baae7 | 5259 | */ |
eb95419b | 5260 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5261 | { |
5262 | /* | |
aa6ec29b | 5263 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5264 | * guarantees that @root doesn't have any children, so turning it |
5265 | * on for the root memcg is enough. | |
5266 | */ | |
aa6ec29b | 5267 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 5268 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
5269 | } |
5270 | ||
241994ed JW |
5271 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5272 | struct cftype *cft) | |
5273 | { | |
5274 | return mem_cgroup_usage(mem_cgroup_from_css(css), false); | |
5275 | } | |
5276 | ||
5277 | static int memory_low_show(struct seq_file *m, void *v) | |
5278 | { | |
5279 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5280 | unsigned long low = ACCESS_ONCE(memcg->low); | |
5281 | ||
5282 | if (low == PAGE_COUNTER_MAX) | |
5283 | seq_puts(m, "infinity\n"); | |
5284 | else | |
5285 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5286 | ||
5287 | return 0; | |
5288 | } | |
5289 | ||
5290 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5291 | char *buf, size_t nbytes, loff_t off) | |
5292 | { | |
5293 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5294 | unsigned long low; | |
5295 | int err; | |
5296 | ||
5297 | buf = strstrip(buf); | |
5298 | err = page_counter_memparse(buf, "infinity", &low); | |
5299 | if (err) | |
5300 | return err; | |
5301 | ||
5302 | memcg->low = low; | |
5303 | ||
5304 | return nbytes; | |
5305 | } | |
5306 | ||
5307 | static int memory_high_show(struct seq_file *m, void *v) | |
5308 | { | |
5309 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5310 | unsigned long high = ACCESS_ONCE(memcg->high); | |
5311 | ||
5312 | if (high == PAGE_COUNTER_MAX) | |
5313 | seq_puts(m, "infinity\n"); | |
5314 | else | |
5315 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5316 | ||
5317 | return 0; | |
5318 | } | |
5319 | ||
5320 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5321 | char *buf, size_t nbytes, loff_t off) | |
5322 | { | |
5323 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5324 | unsigned long high; | |
5325 | int err; | |
5326 | ||
5327 | buf = strstrip(buf); | |
5328 | err = page_counter_memparse(buf, "infinity", &high); | |
5329 | if (err) | |
5330 | return err; | |
5331 | ||
5332 | memcg->high = high; | |
5333 | ||
5334 | return nbytes; | |
5335 | } | |
5336 | ||
5337 | static int memory_max_show(struct seq_file *m, void *v) | |
5338 | { | |
5339 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5340 | unsigned long max = ACCESS_ONCE(memcg->memory.limit); | |
5341 | ||
5342 | if (max == PAGE_COUNTER_MAX) | |
5343 | seq_puts(m, "infinity\n"); | |
5344 | else | |
5345 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5346 | ||
5347 | return 0; | |
5348 | } | |
5349 | ||
5350 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5351 | char *buf, size_t nbytes, loff_t off) | |
5352 | { | |
5353 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5354 | unsigned long max; | |
5355 | int err; | |
5356 | ||
5357 | buf = strstrip(buf); | |
5358 | err = page_counter_memparse(buf, "infinity", &max); | |
5359 | if (err) | |
5360 | return err; | |
5361 | ||
5362 | err = mem_cgroup_resize_limit(memcg, max); | |
5363 | if (err) | |
5364 | return err; | |
5365 | ||
5366 | return nbytes; | |
5367 | } | |
5368 | ||
5369 | static int memory_events_show(struct seq_file *m, void *v) | |
5370 | { | |
5371 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5372 | ||
5373 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5374 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5375 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5376 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5377 | ||
5378 | return 0; | |
5379 | } | |
5380 | ||
5381 | static struct cftype memory_files[] = { | |
5382 | { | |
5383 | .name = "current", | |
5384 | .read_u64 = memory_current_read, | |
5385 | }, | |
5386 | { | |
5387 | .name = "low", | |
5388 | .flags = CFTYPE_NOT_ON_ROOT, | |
5389 | .seq_show = memory_low_show, | |
5390 | .write = memory_low_write, | |
5391 | }, | |
5392 | { | |
5393 | .name = "high", | |
5394 | .flags = CFTYPE_NOT_ON_ROOT, | |
5395 | .seq_show = memory_high_show, | |
5396 | .write = memory_high_write, | |
5397 | }, | |
5398 | { | |
5399 | .name = "max", | |
5400 | .flags = CFTYPE_NOT_ON_ROOT, | |
5401 | .seq_show = memory_max_show, | |
5402 | .write = memory_max_write, | |
5403 | }, | |
5404 | { | |
5405 | .name = "events", | |
5406 | .flags = CFTYPE_NOT_ON_ROOT, | |
5407 | .seq_show = memory_events_show, | |
5408 | }, | |
5409 | { } /* terminate */ | |
5410 | }; | |
5411 | ||
073219e9 | 5412 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5413 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5414 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5415 | .css_offline = mem_cgroup_css_offline, |
5416 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5417 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5418 | .can_attach = mem_cgroup_can_attach, |
5419 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5420 | .attach = mem_cgroup_move_task, |
f00baae7 | 5421 | .bind = mem_cgroup_bind, |
241994ed JW |
5422 | .dfl_cftypes = memory_files, |
5423 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5424 | .early_init = 0, |
8cdea7c0 | 5425 | }; |
c077719b | 5426 | |
c255a458 | 5427 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5428 | static int __init enable_swap_account(char *s) |
5429 | { | |
a2c8990a | 5430 | if (!strcmp(s, "1")) |
a42c390c | 5431 | really_do_swap_account = 1; |
a2c8990a | 5432 | else if (!strcmp(s, "0")) |
a42c390c MH |
5433 | really_do_swap_account = 0; |
5434 | return 1; | |
5435 | } | |
a2c8990a | 5436 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5437 | |
2d11085e MH |
5438 | static void __init memsw_file_init(void) |
5439 | { | |
2cf669a5 TH |
5440 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
5441 | memsw_cgroup_files)); | |
6acc8b02 MH |
5442 | } |
5443 | ||
5444 | static void __init enable_swap_cgroup(void) | |
5445 | { | |
5446 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5447 | do_swap_account = 1; | |
5448 | memsw_file_init(); | |
5449 | } | |
2d11085e | 5450 | } |
6acc8b02 | 5451 | |
2d11085e | 5452 | #else |
6acc8b02 | 5453 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
5454 | { |
5455 | } | |
c077719b | 5456 | #endif |
2d11085e | 5457 | |
241994ed JW |
5458 | /** |
5459 | * mem_cgroup_events - count memory events against a cgroup | |
5460 | * @memcg: the memory cgroup | |
5461 | * @idx: the event index | |
5462 | * @nr: the number of events to account for | |
5463 | */ | |
5464 | void mem_cgroup_events(struct mem_cgroup *memcg, | |
5465 | enum mem_cgroup_events_index idx, | |
5466 | unsigned int nr) | |
5467 | { | |
5468 | this_cpu_add(memcg->stat->events[idx], nr); | |
5469 | } | |
5470 | ||
5471 | /** | |
5472 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5473 | * @root: the highest ancestor to consider | |
5474 | * @memcg: the memory cgroup to check | |
5475 | * | |
5476 | * Returns %true if memory consumption of @memcg, and that of all | |
5477 | * configurable ancestors up to @root, is below the normal range. | |
5478 | */ | |
5479 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5480 | { | |
5481 | if (mem_cgroup_disabled()) | |
5482 | return false; | |
5483 | ||
5484 | /* | |
5485 | * The toplevel group doesn't have a configurable range, so | |
5486 | * it's never low when looked at directly, and it is not | |
5487 | * considered an ancestor when assessing the hierarchy. | |
5488 | */ | |
5489 | ||
5490 | if (memcg == root_mem_cgroup) | |
5491 | return false; | |
5492 | ||
5493 | if (page_counter_read(&memcg->memory) > memcg->low) | |
5494 | return false; | |
5495 | ||
5496 | while (memcg != root) { | |
5497 | memcg = parent_mem_cgroup(memcg); | |
5498 | ||
5499 | if (memcg == root_mem_cgroup) | |
5500 | break; | |
5501 | ||
5502 | if (page_counter_read(&memcg->memory) > memcg->low) | |
5503 | return false; | |
5504 | } | |
5505 | return true; | |
5506 | } | |
5507 | ||
0a31bc97 JW |
5508 | #ifdef CONFIG_MEMCG_SWAP |
5509 | /** | |
5510 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5511 | * @page: page whose memsw charge to transfer | |
5512 | * @entry: swap entry to move the charge to | |
5513 | * | |
5514 | * Transfer the memsw charge of @page to @entry. | |
5515 | */ | |
5516 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5517 | { | |
7bdd143c | 5518 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5519 | unsigned short oldid; |
5520 | ||
5521 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5522 | VM_BUG_ON_PAGE(page_count(page), page); | |
5523 | ||
5524 | if (!do_swap_account) | |
5525 | return; | |
5526 | ||
1306a85a | 5527 | memcg = page->mem_cgroup; |
0a31bc97 JW |
5528 | |
5529 | /* Readahead page, never charged */ | |
29833315 | 5530 | if (!memcg) |
0a31bc97 JW |
5531 | return; |
5532 | ||
7bdd143c | 5533 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
0a31bc97 | 5534 | VM_BUG_ON_PAGE(oldid, page); |
7bdd143c JW |
5535 | mem_cgroup_swap_statistics(memcg, true); |
5536 | ||
1306a85a | 5537 | page->mem_cgroup = NULL; |
0a31bc97 | 5538 | |
7bdd143c JW |
5539 | if (!mem_cgroup_is_root(memcg)) |
5540 | page_counter_uncharge(&memcg->memory, 1); | |
5541 | ||
5542 | /* XXX: caller holds IRQ-safe mapping->tree_lock */ | |
5543 | VM_BUG_ON(!irqs_disabled()); | |
0a31bc97 | 5544 | |
7bdd143c JW |
5545 | mem_cgroup_charge_statistics(memcg, page, -1); |
5546 | memcg_check_events(memcg, page); | |
0a31bc97 JW |
5547 | } |
5548 | ||
5549 | /** | |
5550 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5551 | * @entry: swap entry to uncharge | |
5552 | * | |
5553 | * Drop the memsw charge associated with @entry. | |
5554 | */ | |
5555 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5556 | { | |
5557 | struct mem_cgroup *memcg; | |
5558 | unsigned short id; | |
5559 | ||
5560 | if (!do_swap_account) | |
5561 | return; | |
5562 | ||
5563 | id = swap_cgroup_record(entry, 0); | |
5564 | rcu_read_lock(); | |
5565 | memcg = mem_cgroup_lookup(id); | |
5566 | if (memcg) { | |
ce00a967 | 5567 | if (!mem_cgroup_is_root(memcg)) |
3e32cb2e | 5568 | page_counter_uncharge(&memcg->memsw, 1); |
0a31bc97 JW |
5569 | mem_cgroup_swap_statistics(memcg, false); |
5570 | css_put(&memcg->css); | |
5571 | } | |
5572 | rcu_read_unlock(); | |
5573 | } | |
5574 | #endif | |
5575 | ||
00501b53 JW |
5576 | /** |
5577 | * mem_cgroup_try_charge - try charging a page | |
5578 | * @page: page to charge | |
5579 | * @mm: mm context of the victim | |
5580 | * @gfp_mask: reclaim mode | |
5581 | * @memcgp: charged memcg return | |
5582 | * | |
5583 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5584 | * pages according to @gfp_mask if necessary. | |
5585 | * | |
5586 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5587 | * Otherwise, an error code is returned. | |
5588 | * | |
5589 | * After page->mapping has been set up, the caller must finalize the | |
5590 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5591 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5592 | */ | |
5593 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5594 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5595 | { | |
5596 | struct mem_cgroup *memcg = NULL; | |
5597 | unsigned int nr_pages = 1; | |
5598 | int ret = 0; | |
5599 | ||
5600 | if (mem_cgroup_disabled()) | |
5601 | goto out; | |
5602 | ||
5603 | if (PageSwapCache(page)) { | |
00501b53 JW |
5604 | /* |
5605 | * Every swap fault against a single page tries to charge the | |
5606 | * page, bail as early as possible. shmem_unuse() encounters | |
5607 | * already charged pages, too. The USED bit is protected by | |
5608 | * the page lock, which serializes swap cache removal, which | |
5609 | * in turn serializes uncharging. | |
5610 | */ | |
1306a85a | 5611 | if (page->mem_cgroup) |
00501b53 JW |
5612 | goto out; |
5613 | } | |
5614 | ||
5615 | if (PageTransHuge(page)) { | |
5616 | nr_pages <<= compound_order(page); | |
5617 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5618 | } | |
5619 | ||
5620 | if (do_swap_account && PageSwapCache(page)) | |
5621 | memcg = try_get_mem_cgroup_from_page(page); | |
5622 | if (!memcg) | |
5623 | memcg = get_mem_cgroup_from_mm(mm); | |
5624 | ||
5625 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5626 | ||
5627 | css_put(&memcg->css); | |
5628 | ||
5629 | if (ret == -EINTR) { | |
5630 | memcg = root_mem_cgroup; | |
5631 | ret = 0; | |
5632 | } | |
5633 | out: | |
5634 | *memcgp = memcg; | |
5635 | return ret; | |
5636 | } | |
5637 | ||
5638 | /** | |
5639 | * mem_cgroup_commit_charge - commit a page charge | |
5640 | * @page: page to charge | |
5641 | * @memcg: memcg to charge the page to | |
5642 | * @lrucare: page might be on LRU already | |
5643 | * | |
5644 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5645 | * after page->mapping has been set up. This must happen atomically | |
5646 | * as part of the page instantiation, i.e. under the page table lock | |
5647 | * for anonymous pages, under the page lock for page and swap cache. | |
5648 | * | |
5649 | * In addition, the page must not be on the LRU during the commit, to | |
5650 | * prevent racing with task migration. If it might be, use @lrucare. | |
5651 | * | |
5652 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5653 | */ | |
5654 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5655 | bool lrucare) | |
5656 | { | |
5657 | unsigned int nr_pages = 1; | |
5658 | ||
5659 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5660 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5661 | ||
5662 | if (mem_cgroup_disabled()) | |
5663 | return; | |
5664 | /* | |
5665 | * Swap faults will attempt to charge the same page multiple | |
5666 | * times. But reuse_swap_page() might have removed the page | |
5667 | * from swapcache already, so we can't check PageSwapCache(). | |
5668 | */ | |
5669 | if (!memcg) | |
5670 | return; | |
5671 | ||
6abb5a86 JW |
5672 | commit_charge(page, memcg, lrucare); |
5673 | ||
00501b53 JW |
5674 | if (PageTransHuge(page)) { |
5675 | nr_pages <<= compound_order(page); | |
5676 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5677 | } | |
5678 | ||
6abb5a86 JW |
5679 | local_irq_disable(); |
5680 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5681 | memcg_check_events(memcg, page); | |
5682 | local_irq_enable(); | |
00501b53 JW |
5683 | |
5684 | if (do_swap_account && PageSwapCache(page)) { | |
5685 | swp_entry_t entry = { .val = page_private(page) }; | |
5686 | /* | |
5687 | * The swap entry might not get freed for a long time, | |
5688 | * let's not wait for it. The page already received a | |
5689 | * memory+swap charge, drop the swap entry duplicate. | |
5690 | */ | |
5691 | mem_cgroup_uncharge_swap(entry); | |
5692 | } | |
5693 | } | |
5694 | ||
5695 | /** | |
5696 | * mem_cgroup_cancel_charge - cancel a page charge | |
5697 | * @page: page to charge | |
5698 | * @memcg: memcg to charge the page to | |
5699 | * | |
5700 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5701 | */ | |
5702 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5703 | { | |
5704 | unsigned int nr_pages = 1; | |
5705 | ||
5706 | if (mem_cgroup_disabled()) | |
5707 | return; | |
5708 | /* | |
5709 | * Swap faults will attempt to charge the same page multiple | |
5710 | * times. But reuse_swap_page() might have removed the page | |
5711 | * from swapcache already, so we can't check PageSwapCache(). | |
5712 | */ | |
5713 | if (!memcg) | |
5714 | return; | |
5715 | ||
5716 | if (PageTransHuge(page)) { | |
5717 | nr_pages <<= compound_order(page); | |
5718 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5719 | } | |
5720 | ||
5721 | cancel_charge(memcg, nr_pages); | |
5722 | } | |
5723 | ||
747db954 | 5724 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5725 | unsigned long nr_anon, unsigned long nr_file, |
5726 | unsigned long nr_huge, struct page *dummy_page) | |
5727 | { | |
18eca2e6 | 5728 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5729 | unsigned long flags; |
5730 | ||
ce00a967 | 5731 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5732 | page_counter_uncharge(&memcg->memory, nr_pages); |
5733 | if (do_swap_account) | |
5734 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5735 | memcg_oom_recover(memcg); |
5736 | } | |
747db954 JW |
5737 | |
5738 | local_irq_save(flags); | |
5739 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5740 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5741 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5742 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5743 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5744 | memcg_check_events(memcg, dummy_page); |
5745 | local_irq_restore(flags); | |
e8ea14cc JW |
5746 | |
5747 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5748 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5749 | } |
5750 | ||
5751 | static void uncharge_list(struct list_head *page_list) | |
5752 | { | |
5753 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5754 | unsigned long nr_anon = 0; |
5755 | unsigned long nr_file = 0; | |
5756 | unsigned long nr_huge = 0; | |
5757 | unsigned long pgpgout = 0; | |
747db954 JW |
5758 | struct list_head *next; |
5759 | struct page *page; | |
5760 | ||
5761 | next = page_list->next; | |
5762 | do { | |
5763 | unsigned int nr_pages = 1; | |
747db954 JW |
5764 | |
5765 | page = list_entry(next, struct page, lru); | |
5766 | next = page->lru.next; | |
5767 | ||
5768 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5769 | VM_BUG_ON_PAGE(page_count(page), page); | |
5770 | ||
1306a85a | 5771 | if (!page->mem_cgroup) |
747db954 JW |
5772 | continue; |
5773 | ||
5774 | /* | |
5775 | * Nobody should be changing or seriously looking at | |
1306a85a | 5776 | * page->mem_cgroup at this point, we have fully |
29833315 | 5777 | * exclusive access to the page. |
747db954 JW |
5778 | */ |
5779 | ||
1306a85a | 5780 | if (memcg != page->mem_cgroup) { |
747db954 | 5781 | if (memcg) { |
18eca2e6 JW |
5782 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5783 | nr_huge, page); | |
5784 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5785 | } |
1306a85a | 5786 | memcg = page->mem_cgroup; |
747db954 JW |
5787 | } |
5788 | ||
5789 | if (PageTransHuge(page)) { | |
5790 | nr_pages <<= compound_order(page); | |
5791 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5792 | nr_huge += nr_pages; | |
5793 | } | |
5794 | ||
5795 | if (PageAnon(page)) | |
5796 | nr_anon += nr_pages; | |
5797 | else | |
5798 | nr_file += nr_pages; | |
5799 | ||
1306a85a | 5800 | page->mem_cgroup = NULL; |
747db954 JW |
5801 | |
5802 | pgpgout++; | |
5803 | } while (next != page_list); | |
5804 | ||
5805 | if (memcg) | |
18eca2e6 JW |
5806 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5807 | nr_huge, page); | |
747db954 JW |
5808 | } |
5809 | ||
0a31bc97 JW |
5810 | /** |
5811 | * mem_cgroup_uncharge - uncharge a page | |
5812 | * @page: page to uncharge | |
5813 | * | |
5814 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5815 | * mem_cgroup_commit_charge(). | |
5816 | */ | |
5817 | void mem_cgroup_uncharge(struct page *page) | |
5818 | { | |
0a31bc97 JW |
5819 | if (mem_cgroup_disabled()) |
5820 | return; | |
5821 | ||
747db954 | 5822 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5823 | if (!page->mem_cgroup) |
0a31bc97 JW |
5824 | return; |
5825 | ||
747db954 JW |
5826 | INIT_LIST_HEAD(&page->lru); |
5827 | uncharge_list(&page->lru); | |
5828 | } | |
0a31bc97 | 5829 | |
747db954 JW |
5830 | /** |
5831 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5832 | * @page_list: list of pages to uncharge | |
5833 | * | |
5834 | * Uncharge a list of pages previously charged with | |
5835 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5836 | */ | |
5837 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5838 | { | |
5839 | if (mem_cgroup_disabled()) | |
5840 | return; | |
0a31bc97 | 5841 | |
747db954 JW |
5842 | if (!list_empty(page_list)) |
5843 | uncharge_list(page_list); | |
0a31bc97 JW |
5844 | } |
5845 | ||
5846 | /** | |
5847 | * mem_cgroup_migrate - migrate a charge to another page | |
5848 | * @oldpage: currently charged page | |
5849 | * @newpage: page to transfer the charge to | |
f5e03a49 | 5850 | * @lrucare: either or both pages might be on the LRU already |
0a31bc97 JW |
5851 | * |
5852 | * Migrate the charge from @oldpage to @newpage. | |
5853 | * | |
5854 | * Both pages must be locked, @newpage->mapping must be set up. | |
5855 | */ | |
5856 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
5857 | bool lrucare) | |
5858 | { | |
29833315 | 5859 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5860 | int isolated; |
5861 | ||
5862 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5863 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
5864 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
5865 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
5866 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
5867 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5868 | newpage); | |
0a31bc97 JW |
5869 | |
5870 | if (mem_cgroup_disabled()) | |
5871 | return; | |
5872 | ||
5873 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5874 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5875 | return; |
5876 | ||
7d5e3245 JW |
5877 | /* |
5878 | * Swapcache readahead pages can get migrated before being | |
5879 | * charged, and migration from compaction can happen to an | |
5880 | * uncharged page when the PFN walker finds a page that | |
5881 | * reclaim just put back on the LRU but has not released yet. | |
5882 | */ | |
1306a85a | 5883 | memcg = oldpage->mem_cgroup; |
29833315 | 5884 | if (!memcg) |
0a31bc97 JW |
5885 | return; |
5886 | ||
0a31bc97 JW |
5887 | if (lrucare) |
5888 | lock_page_lru(oldpage, &isolated); | |
5889 | ||
1306a85a | 5890 | oldpage->mem_cgroup = NULL; |
0a31bc97 JW |
5891 | |
5892 | if (lrucare) | |
5893 | unlock_page_lru(oldpage, isolated); | |
5894 | ||
29833315 | 5895 | commit_charge(newpage, memcg, lrucare); |
0a31bc97 JW |
5896 | } |
5897 | ||
2d11085e | 5898 | /* |
1081312f MH |
5899 | * subsys_initcall() for memory controller. |
5900 | * | |
5901 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5902 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5903 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5904 | * be initialized from here. | |
2d11085e MH |
5905 | */ |
5906 | static int __init mem_cgroup_init(void) | |
5907 | { | |
5908 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 5909 | enable_swap_cgroup(); |
bb4cc1a8 | 5910 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 5911 | memcg_stock_init(); |
2d11085e MH |
5912 | return 0; |
5913 | } | |
5914 | subsys_initcall(mem_cgroup_init); |