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 | |
21afa38e | 75 | /* Whether the swap controller is active */ |
c255a458 | 76 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 77 | int do_swap_account __read_mostly; |
c077719b | 78 | #else |
a0db00fc | 79 | #define do_swap_account 0 |
c077719b KH |
80 | #endif |
81 | ||
af7c4b0e JW |
82 | static const char * const mem_cgroup_stat_names[] = { |
83 | "cache", | |
84 | "rss", | |
b070e65c | 85 | "rss_huge", |
af7c4b0e | 86 | "mapped_file", |
3ea67d06 | 87 | "writeback", |
af7c4b0e JW |
88 | "swap", |
89 | }; | |
90 | ||
af7c4b0e JW |
91 | static const char * const mem_cgroup_events_names[] = { |
92 | "pgpgin", | |
93 | "pgpgout", | |
94 | "pgfault", | |
95 | "pgmajfault", | |
96 | }; | |
97 | ||
58cf188e SZ |
98 | static const char * const mem_cgroup_lru_names[] = { |
99 | "inactive_anon", | |
100 | "active_anon", | |
101 | "inactive_file", | |
102 | "active_file", | |
103 | "unevictable", | |
104 | }; | |
105 | ||
7a159cc9 JW |
106 | /* |
107 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
108 | * it will be incremated by the number of pages. This counter is used for | |
109 | * for trigger some periodic events. This is straightforward and better | |
110 | * than using jiffies etc. to handle periodic memcg event. | |
111 | */ | |
112 | enum mem_cgroup_events_target { | |
113 | MEM_CGROUP_TARGET_THRESH, | |
bb4cc1a8 | 114 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 115 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
116 | MEM_CGROUP_NTARGETS, |
117 | }; | |
a0db00fc KS |
118 | #define THRESHOLDS_EVENTS_TARGET 128 |
119 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
120 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 121 | |
d52aa412 | 122 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 123 | long count[MEM_CGROUP_STAT_NSTATS]; |
241994ed | 124 | unsigned long events[MEMCG_NR_EVENTS]; |
13114716 | 125 | unsigned long nr_page_events; |
7a159cc9 | 126 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
127 | }; |
128 | ||
5ac8fb31 JW |
129 | struct reclaim_iter { |
130 | struct mem_cgroup *position; | |
527a5ec9 JW |
131 | /* scan generation, increased every round-trip */ |
132 | unsigned int generation; | |
133 | }; | |
134 | ||
6d12e2d8 KH |
135 | /* |
136 | * per-zone information in memory controller. | |
137 | */ | |
6d12e2d8 | 138 | struct mem_cgroup_per_zone { |
6290df54 | 139 | struct lruvec lruvec; |
1eb49272 | 140 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 141 | |
5ac8fb31 | 142 | struct reclaim_iter iter[DEF_PRIORITY + 1]; |
527a5ec9 | 143 | |
bb4cc1a8 | 144 | struct rb_node tree_node; /* RB tree node */ |
3e32cb2e | 145 | unsigned long usage_in_excess;/* Set to the value by which */ |
bb4cc1a8 AM |
146 | /* the soft limit is exceeded*/ |
147 | bool on_tree; | |
d79154bb | 148 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 149 | /* use container_of */ |
6d12e2d8 | 150 | }; |
6d12e2d8 KH |
151 | |
152 | struct mem_cgroup_per_node { | |
153 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
154 | }; | |
155 | ||
bb4cc1a8 AM |
156 | /* |
157 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
158 | * their hierarchy representation | |
159 | */ | |
160 | ||
161 | struct mem_cgroup_tree_per_zone { | |
162 | struct rb_root rb_root; | |
163 | spinlock_t lock; | |
164 | }; | |
165 | ||
166 | struct mem_cgroup_tree_per_node { | |
167 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
168 | }; | |
169 | ||
170 | struct mem_cgroup_tree { | |
171 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
172 | }; | |
173 | ||
174 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
175 | ||
2e72b634 KS |
176 | struct mem_cgroup_threshold { |
177 | struct eventfd_ctx *eventfd; | |
3e32cb2e | 178 | unsigned long threshold; |
2e72b634 KS |
179 | }; |
180 | ||
9490ff27 | 181 | /* For threshold */ |
2e72b634 | 182 | struct mem_cgroup_threshold_ary { |
748dad36 | 183 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 184 | int current_threshold; |
2e72b634 KS |
185 | /* Size of entries[] */ |
186 | unsigned int size; | |
187 | /* Array of thresholds */ | |
188 | struct mem_cgroup_threshold entries[0]; | |
189 | }; | |
2c488db2 KS |
190 | |
191 | struct mem_cgroup_thresholds { | |
192 | /* Primary thresholds array */ | |
193 | struct mem_cgroup_threshold_ary *primary; | |
194 | /* | |
195 | * Spare threshold array. | |
196 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
197 | * It must be able to store at least primary->size - 1 entries. | |
198 | */ | |
199 | struct mem_cgroup_threshold_ary *spare; | |
200 | }; | |
201 | ||
9490ff27 KH |
202 | /* for OOM */ |
203 | struct mem_cgroup_eventfd_list { | |
204 | struct list_head list; | |
205 | struct eventfd_ctx *eventfd; | |
206 | }; | |
2e72b634 | 207 | |
79bd9814 TH |
208 | /* |
209 | * cgroup_event represents events which userspace want to receive. | |
210 | */ | |
3bc942f3 | 211 | struct mem_cgroup_event { |
79bd9814 | 212 | /* |
59b6f873 | 213 | * memcg which the event belongs to. |
79bd9814 | 214 | */ |
59b6f873 | 215 | struct mem_cgroup *memcg; |
79bd9814 TH |
216 | /* |
217 | * eventfd to signal userspace about the event. | |
218 | */ | |
219 | struct eventfd_ctx *eventfd; | |
220 | /* | |
221 | * Each of these stored in a list by the cgroup. | |
222 | */ | |
223 | struct list_head list; | |
fba94807 TH |
224 | /* |
225 | * register_event() callback will be used to add new userspace | |
226 | * waiter for changes related to this event. Use eventfd_signal() | |
227 | * on eventfd to send notification to userspace. | |
228 | */ | |
59b6f873 | 229 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 230 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
231 | /* |
232 | * unregister_event() callback will be called when userspace closes | |
233 | * the eventfd or on cgroup removing. This callback must be set, | |
234 | * if you want provide notification functionality. | |
235 | */ | |
59b6f873 | 236 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 237 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
238 | /* |
239 | * All fields below needed to unregister event when | |
240 | * userspace closes eventfd. | |
241 | */ | |
242 | poll_table pt; | |
243 | wait_queue_head_t *wqh; | |
244 | wait_queue_t wait; | |
245 | struct work_struct remove; | |
246 | }; | |
247 | ||
c0ff4b85 R |
248 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
249 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 250 | |
8cdea7c0 BS |
251 | /* |
252 | * The memory controller data structure. The memory controller controls both | |
253 | * page cache and RSS per cgroup. We would eventually like to provide | |
254 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
255 | * to help the administrator determine what knobs to tune. | |
256 | * | |
257 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
258 | * we hit the water mark. May be even add a low water mark, such that |
259 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
260 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
261 | */ |
262 | struct mem_cgroup { | |
263 | struct cgroup_subsys_state css; | |
3e32cb2e JW |
264 | |
265 | /* Accounted resources */ | |
266 | struct page_counter memory; | |
267 | struct page_counter memsw; | |
268 | struct page_counter kmem; | |
269 | ||
241994ed JW |
270 | /* Normal memory consumption range */ |
271 | unsigned long low; | |
272 | unsigned long high; | |
273 | ||
3e32cb2e | 274 | unsigned long soft_limit; |
59927fb9 | 275 | |
70ddf637 AV |
276 | /* vmpressure notifications */ |
277 | struct vmpressure vmpressure; | |
278 | ||
2f7dd7a4 JW |
279 | /* css_online() has been completed */ |
280 | int initialized; | |
281 | ||
18f59ea7 BS |
282 | /* |
283 | * Should the accounting and control be hierarchical, per subtree? | |
284 | */ | |
285 | bool use_hierarchy; | |
79dfdacc MH |
286 | |
287 | bool oom_lock; | |
288 | atomic_t under_oom; | |
3812c8c8 | 289 | atomic_t oom_wakeups; |
79dfdacc | 290 | |
1f4c025b | 291 | int swappiness; |
3c11ecf4 KH |
292 | /* OOM-Killer disable */ |
293 | int oom_kill_disable; | |
a7885eb8 | 294 | |
2e72b634 KS |
295 | /* protect arrays of thresholds */ |
296 | struct mutex thresholds_lock; | |
297 | ||
298 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 299 | struct mem_cgroup_thresholds thresholds; |
907860ed | 300 | |
2e72b634 | 301 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 302 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 303 | |
9490ff27 KH |
304 | /* For oom notifier event fd */ |
305 | struct list_head oom_notify; | |
185efc0f | 306 | |
7dc74be0 DN |
307 | /* |
308 | * Should we move charges of a task when a task is moved into this | |
309 | * mem_cgroup ? And what type of charges should we move ? | |
310 | */ | |
f894ffa8 | 311 | unsigned long move_charge_at_immigrate; |
619d094b KH |
312 | /* |
313 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
314 | */ | |
6de22619 | 315 | atomic_t moving_account; |
312734c0 | 316 | /* taken only while moving_account > 0 */ |
6de22619 JW |
317 | spinlock_t move_lock; |
318 | struct task_struct *move_lock_task; | |
319 | unsigned long move_lock_flags; | |
d52aa412 | 320 | /* |
c62b1a3b | 321 | * percpu counter. |
d52aa412 | 322 | */ |
3a7951b4 | 323 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
324 | /* |
325 | * used when a cpu is offlined or other synchronizations | |
326 | * See mem_cgroup_read_stat(). | |
327 | */ | |
328 | struct mem_cgroup_stat_cpu nocpu_base; | |
329 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 330 | |
4bd2c1ee | 331 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
2e685cad | 332 | struct cg_proto tcp_mem; |
d1a4c0b3 | 333 | #endif |
2633d7a0 | 334 | #if defined(CONFIG_MEMCG_KMEM) |
f7ce3190 | 335 | /* Index in the kmem_cache->memcg_params.memcg_caches array */ |
2633d7a0 | 336 | int kmemcg_id; |
2788cf0c | 337 | bool kmem_acct_activated; |
2a4db7eb | 338 | bool kmem_acct_active; |
2633d7a0 | 339 | #endif |
45cf7ebd GC |
340 | |
341 | int last_scanned_node; | |
342 | #if MAX_NUMNODES > 1 | |
343 | nodemask_t scan_nodes; | |
344 | atomic_t numainfo_events; | |
345 | atomic_t numainfo_updating; | |
346 | #endif | |
70ddf637 | 347 | |
fba94807 TH |
348 | /* List of events which userspace want to receive */ |
349 | struct list_head event_list; | |
350 | spinlock_t event_list_lock; | |
351 | ||
54f72fe0 JW |
352 | struct mem_cgroup_per_node *nodeinfo[0]; |
353 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
354 | }; |
355 | ||
510fc4e1 | 356 | #ifdef CONFIG_MEMCG_KMEM |
cb731d6c | 357 | bool memcg_kmem_is_active(struct mem_cgroup *memcg) |
7de37682 | 358 | { |
2a4db7eb | 359 | return memcg->kmem_acct_active; |
7de37682 | 360 | } |
510fc4e1 GC |
361 | #endif |
362 | ||
7dc74be0 DN |
363 | /* Stuffs for move charges at task migration. */ |
364 | /* | |
1dfab5ab | 365 | * Types of charges to be moved. |
7dc74be0 | 366 | */ |
1dfab5ab JW |
367 | #define MOVE_ANON 0x1U |
368 | #define MOVE_FILE 0x2U | |
369 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 370 | |
4ffef5fe DN |
371 | /* "mc" and its members are protected by cgroup_mutex */ |
372 | static struct move_charge_struct { | |
b1dd693e | 373 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
374 | struct mem_cgroup *from; |
375 | struct mem_cgroup *to; | |
1dfab5ab | 376 | unsigned long flags; |
4ffef5fe | 377 | unsigned long precharge; |
854ffa8d | 378 | unsigned long moved_charge; |
483c30b5 | 379 | unsigned long moved_swap; |
8033b97c DN |
380 | struct task_struct *moving_task; /* a task moving charges */ |
381 | wait_queue_head_t waitq; /* a waitq for other context */ | |
382 | } mc = { | |
2bd9bb20 | 383 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
384 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
385 | }; | |
4ffef5fe | 386 | |
4e416953 BS |
387 | /* |
388 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
389 | * limit reclaim to prevent infinite loops, if they ever occur. | |
390 | */ | |
a0db00fc | 391 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 392 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 393 | |
217bc319 KH |
394 | enum charge_type { |
395 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 396 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 397 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 398 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
399 | NR_CHARGE_TYPE, |
400 | }; | |
401 | ||
8c7c6e34 | 402 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
403 | enum res_type { |
404 | _MEM, | |
405 | _MEMSWAP, | |
406 | _OOM_TYPE, | |
510fc4e1 | 407 | _KMEM, |
86ae53e1 GC |
408 | }; |
409 | ||
a0db00fc KS |
410 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
411 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 412 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
413 | /* Used for OOM nofiier */ |
414 | #define OOM_CONTROL (0) | |
8c7c6e34 | 415 | |
0999821b GC |
416 | /* |
417 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
418 | * As a consequence, any change that needs to protect against new child cgroups | |
419 | * appearing has to hold it as well. | |
420 | */ | |
421 | static DEFINE_MUTEX(memcg_create_mutex); | |
422 | ||
b2145145 WL |
423 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
424 | { | |
a7c6d554 | 425 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
426 | } |
427 | ||
70ddf637 AV |
428 | /* Some nice accessors for the vmpressure. */ |
429 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
430 | { | |
431 | if (!memcg) | |
432 | memcg = root_mem_cgroup; | |
433 | return &memcg->vmpressure; | |
434 | } | |
435 | ||
436 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
437 | { | |
438 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
439 | } | |
440 | ||
7ffc0edc MH |
441 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
442 | { | |
443 | return (memcg == root_mem_cgroup); | |
444 | } | |
445 | ||
4219b2da LZ |
446 | /* |
447 | * We restrict the id in the range of [1, 65535], so it can fit into | |
448 | * an unsigned short. | |
449 | */ | |
450 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
451 | ||
34c00c31 LZ |
452 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
453 | { | |
15a4c835 | 454 | return memcg->css.id; |
34c00c31 LZ |
455 | } |
456 | ||
457 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
458 | { | |
459 | struct cgroup_subsys_state *css; | |
460 | ||
7d699ddb | 461 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
462 | return mem_cgroup_from_css(css); |
463 | } | |
464 | ||
e1aab161 | 465 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 466 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 467 | |
e1aab161 GC |
468 | void sock_update_memcg(struct sock *sk) |
469 | { | |
376be5ff | 470 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 471 | struct mem_cgroup *memcg; |
3f134619 | 472 | struct cg_proto *cg_proto; |
e1aab161 GC |
473 | |
474 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
475 | ||
f3f511e1 GC |
476 | /* Socket cloning can throw us here with sk_cgrp already |
477 | * filled. It won't however, necessarily happen from | |
478 | * process context. So the test for root memcg given | |
479 | * the current task's memcg won't help us in this case. | |
480 | * | |
481 | * Respecting the original socket's memcg is a better | |
482 | * decision in this case. | |
483 | */ | |
484 | if (sk->sk_cgrp) { | |
485 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 486 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
487 | return; |
488 | } | |
489 | ||
e1aab161 GC |
490 | rcu_read_lock(); |
491 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 492 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 493 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
494 | memcg_proto_active(cg_proto) && |
495 | css_tryget_online(&memcg->css)) { | |
3f134619 | 496 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
497 | } |
498 | rcu_read_unlock(); | |
499 | } | |
500 | } | |
501 | EXPORT_SYMBOL(sock_update_memcg); | |
502 | ||
503 | void sock_release_memcg(struct sock *sk) | |
504 | { | |
376be5ff | 505 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
506 | struct mem_cgroup *memcg; |
507 | WARN_ON(!sk->sk_cgrp->memcg); | |
508 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 509 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
510 | } |
511 | } | |
d1a4c0b3 GC |
512 | |
513 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
514 | { | |
515 | if (!memcg || mem_cgroup_is_root(memcg)) | |
516 | return NULL; | |
517 | ||
2e685cad | 518 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
519 | } |
520 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 521 | |
3f134619 GC |
522 | #endif |
523 | ||
a8964b9b | 524 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 525 | /* |
f7ce3190 | 526 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
527 | * The main reason for not using cgroup id for this: |
528 | * this works better in sparse environments, where we have a lot of memcgs, | |
529 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
530 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
531 | * 200 entry array for that. | |
55007d84 | 532 | * |
dbcf73e2 VD |
533 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
534 | * will double each time we have to increase it. | |
55007d84 | 535 | */ |
dbcf73e2 VD |
536 | static DEFINE_IDA(memcg_cache_ida); |
537 | int memcg_nr_cache_ids; | |
749c5415 | 538 | |
05257a1a VD |
539 | /* Protects memcg_nr_cache_ids */ |
540 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
541 | ||
542 | void memcg_get_cache_ids(void) | |
543 | { | |
544 | down_read(&memcg_cache_ids_sem); | |
545 | } | |
546 | ||
547 | void memcg_put_cache_ids(void) | |
548 | { | |
549 | up_read(&memcg_cache_ids_sem); | |
550 | } | |
551 | ||
55007d84 GC |
552 | /* |
553 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
554 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
555 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
556 | * tunable, but that is strictly not necessary. | |
557 | * | |
b8627835 | 558 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
559 | * this constant directly from cgroup, but it is understandable that this is |
560 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 561 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
562 | * increase ours as well if it increases. |
563 | */ | |
564 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 565 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 566 | |
d7f25f8a GC |
567 | /* |
568 | * A lot of the calls to the cache allocation functions are expected to be | |
569 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
570 | * conditional to this static branch, we'll have to allow modules that does | |
571 | * kmem_cache_alloc and the such to see this symbol as well | |
572 | */ | |
a8964b9b | 573 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 574 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 575 | |
a8964b9b GC |
576 | #endif /* CONFIG_MEMCG_KMEM */ |
577 | ||
f64c3f54 | 578 | static struct mem_cgroup_per_zone * |
e231875b | 579 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 580 | { |
e231875b JZ |
581 | int nid = zone_to_nid(zone); |
582 | int zid = zone_idx(zone); | |
583 | ||
54f72fe0 | 584 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
585 | } |
586 | ||
c0ff4b85 | 587 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 588 | { |
c0ff4b85 | 589 | return &memcg->css; |
d324236b WF |
590 | } |
591 | ||
f64c3f54 | 592 | static struct mem_cgroup_per_zone * |
e231875b | 593 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 594 | { |
97a6c37b JW |
595 | int nid = page_to_nid(page); |
596 | int zid = page_zonenum(page); | |
f64c3f54 | 597 | |
e231875b | 598 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
599 | } |
600 | ||
bb4cc1a8 AM |
601 | static struct mem_cgroup_tree_per_zone * |
602 | soft_limit_tree_node_zone(int nid, int zid) | |
603 | { | |
604 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
605 | } | |
606 | ||
607 | static struct mem_cgroup_tree_per_zone * | |
608 | soft_limit_tree_from_page(struct page *page) | |
609 | { | |
610 | int nid = page_to_nid(page); | |
611 | int zid = page_zonenum(page); | |
612 | ||
613 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
614 | } | |
615 | ||
cf2c8127 JW |
616 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
617 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 618 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
619 | { |
620 | struct rb_node **p = &mctz->rb_root.rb_node; | |
621 | struct rb_node *parent = NULL; | |
622 | struct mem_cgroup_per_zone *mz_node; | |
623 | ||
624 | if (mz->on_tree) | |
625 | return; | |
626 | ||
627 | mz->usage_in_excess = new_usage_in_excess; | |
628 | if (!mz->usage_in_excess) | |
629 | return; | |
630 | while (*p) { | |
631 | parent = *p; | |
632 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
633 | tree_node); | |
634 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
635 | p = &(*p)->rb_left; | |
636 | /* | |
637 | * We can't avoid mem cgroups that are over their soft | |
638 | * limit by the same amount | |
639 | */ | |
640 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
641 | p = &(*p)->rb_right; | |
642 | } | |
643 | rb_link_node(&mz->tree_node, parent, p); | |
644 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
645 | mz->on_tree = true; | |
646 | } | |
647 | ||
cf2c8127 JW |
648 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
649 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
650 | { |
651 | if (!mz->on_tree) | |
652 | return; | |
653 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
654 | mz->on_tree = false; | |
655 | } | |
656 | ||
cf2c8127 JW |
657 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
658 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 659 | { |
0a31bc97 JW |
660 | unsigned long flags; |
661 | ||
662 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 663 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 664 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
665 | } |
666 | ||
3e32cb2e JW |
667 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
668 | { | |
669 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
670 | unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit); | |
671 | unsigned long excess = 0; | |
672 | ||
673 | if (nr_pages > soft_limit) | |
674 | excess = nr_pages - soft_limit; | |
675 | ||
676 | return excess; | |
677 | } | |
bb4cc1a8 AM |
678 | |
679 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
680 | { | |
3e32cb2e | 681 | unsigned long excess; |
bb4cc1a8 AM |
682 | struct mem_cgroup_per_zone *mz; |
683 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 684 | |
e231875b | 685 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
686 | /* |
687 | * Necessary to update all ancestors when hierarchy is used. | |
688 | * because their event counter is not touched. | |
689 | */ | |
690 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 691 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 692 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
693 | /* |
694 | * We have to update the tree if mz is on RB-tree or | |
695 | * mem is over its softlimit. | |
696 | */ | |
697 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
698 | unsigned long flags; |
699 | ||
700 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
701 | /* if on-tree, remove it */ |
702 | if (mz->on_tree) | |
cf2c8127 | 703 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
704 | /* |
705 | * Insert again. mz->usage_in_excess will be updated. | |
706 | * If excess is 0, no tree ops. | |
707 | */ | |
cf2c8127 | 708 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 709 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
710 | } |
711 | } | |
712 | } | |
713 | ||
714 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
715 | { | |
bb4cc1a8 | 716 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
717 | struct mem_cgroup_per_zone *mz; |
718 | int nid, zid; | |
bb4cc1a8 | 719 | |
e231875b JZ |
720 | for_each_node(nid) { |
721 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
722 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
723 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 724 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
725 | } |
726 | } | |
727 | } | |
728 | ||
729 | static struct mem_cgroup_per_zone * | |
730 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
731 | { | |
732 | struct rb_node *rightmost = NULL; | |
733 | struct mem_cgroup_per_zone *mz; | |
734 | ||
735 | retry: | |
736 | mz = NULL; | |
737 | rightmost = rb_last(&mctz->rb_root); | |
738 | if (!rightmost) | |
739 | goto done; /* Nothing to reclaim from */ | |
740 | ||
741 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
742 | /* | |
743 | * Remove the node now but someone else can add it back, | |
744 | * we will to add it back at the end of reclaim to its correct | |
745 | * position in the tree. | |
746 | */ | |
cf2c8127 | 747 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 748 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 749 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
750 | goto retry; |
751 | done: | |
752 | return mz; | |
753 | } | |
754 | ||
755 | static struct mem_cgroup_per_zone * | |
756 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
757 | { | |
758 | struct mem_cgroup_per_zone *mz; | |
759 | ||
0a31bc97 | 760 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 761 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 762 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
763 | return mz; |
764 | } | |
765 | ||
711d3d2c KH |
766 | /* |
767 | * Implementation Note: reading percpu statistics for memcg. | |
768 | * | |
769 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
770 | * synchronization to implement "quick" read. There are trade-off between | |
771 | * reading cost and precision of value. Then, we may have a chance to implement | |
772 | * a periodic synchronizion of counter in memcg's counter. | |
773 | * | |
774 | * But this _read() function is used for user interface now. The user accounts | |
775 | * memory usage by memory cgroup and he _always_ requires exact value because | |
776 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
777 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
778 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
779 | * | |
780 | * If there are kernel internal actions which can make use of some not-exact | |
781 | * value, and reading all cpu value can be performance bottleneck in some | |
782 | * common workload, threashold and synchonization as vmstat[] should be | |
783 | * implemented. | |
784 | */ | |
c0ff4b85 | 785 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 786 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 787 | { |
7a159cc9 | 788 | long val = 0; |
c62b1a3b | 789 | int cpu; |
c62b1a3b | 790 | |
711d3d2c KH |
791 | get_online_cpus(); |
792 | for_each_online_cpu(cpu) | |
c0ff4b85 | 793 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 794 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
795 | spin_lock(&memcg->pcp_counter_lock); |
796 | val += memcg->nocpu_base.count[idx]; | |
797 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
798 | #endif |
799 | put_online_cpus(); | |
c62b1a3b KH |
800 | return val; |
801 | } | |
802 | ||
c0ff4b85 | 803 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
804 | enum mem_cgroup_events_index idx) |
805 | { | |
806 | unsigned long val = 0; | |
807 | int cpu; | |
808 | ||
9c567512 | 809 | get_online_cpus(); |
e9f8974f | 810 | for_each_online_cpu(cpu) |
c0ff4b85 | 811 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 812 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
813 | spin_lock(&memcg->pcp_counter_lock); |
814 | val += memcg->nocpu_base.events[idx]; | |
815 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 816 | #endif |
9c567512 | 817 | put_online_cpus(); |
e9f8974f JW |
818 | return val; |
819 | } | |
820 | ||
c0ff4b85 | 821 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 822 | struct page *page, |
0a31bc97 | 823 | int nr_pages) |
d52aa412 | 824 | { |
b2402857 KH |
825 | /* |
826 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
827 | * counted as CACHE even if it's on ANON LRU. | |
828 | */ | |
0a31bc97 | 829 | if (PageAnon(page)) |
b2402857 | 830 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 831 | nr_pages); |
d52aa412 | 832 | else |
b2402857 | 833 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 834 | nr_pages); |
55e462b0 | 835 | |
b070e65c DR |
836 | if (PageTransHuge(page)) |
837 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
838 | nr_pages); | |
839 | ||
e401f176 KH |
840 | /* pagein of a big page is an event. So, ignore page size */ |
841 | if (nr_pages > 0) | |
c0ff4b85 | 842 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 843 | else { |
c0ff4b85 | 844 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
845 | nr_pages = -nr_pages; /* for event */ |
846 | } | |
e401f176 | 847 | |
13114716 | 848 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
849 | } |
850 | ||
e231875b | 851 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
852 | { |
853 | struct mem_cgroup_per_zone *mz; | |
854 | ||
855 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
856 | return mz->lru_size[lru]; | |
857 | } | |
858 | ||
e231875b JZ |
859 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
860 | int nid, | |
861 | unsigned int lru_mask) | |
bb2a0de9 | 862 | { |
e231875b | 863 | unsigned long nr = 0; |
889976db YH |
864 | int zid; |
865 | ||
e231875b | 866 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 867 | |
e231875b JZ |
868 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
869 | struct mem_cgroup_per_zone *mz; | |
870 | enum lru_list lru; | |
871 | ||
872 | for_each_lru(lru) { | |
873 | if (!(BIT(lru) & lru_mask)) | |
874 | continue; | |
875 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
876 | nr += mz->lru_size[lru]; | |
877 | } | |
878 | } | |
879 | return nr; | |
889976db | 880 | } |
bb2a0de9 | 881 | |
c0ff4b85 | 882 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 883 | unsigned int lru_mask) |
6d12e2d8 | 884 | { |
e231875b | 885 | unsigned long nr = 0; |
889976db | 886 | int nid; |
6d12e2d8 | 887 | |
31aaea4a | 888 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
889 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
890 | return nr; | |
d52aa412 KH |
891 | } |
892 | ||
f53d7ce3 JW |
893 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
894 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
895 | { |
896 | unsigned long val, next; | |
897 | ||
13114716 | 898 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 899 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 900 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
901 | if ((long)next - (long)val < 0) { |
902 | switch (target) { | |
903 | case MEM_CGROUP_TARGET_THRESH: | |
904 | next = val + THRESHOLDS_EVENTS_TARGET; | |
905 | break; | |
bb4cc1a8 AM |
906 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
907 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
908 | break; | |
f53d7ce3 JW |
909 | case MEM_CGROUP_TARGET_NUMAINFO: |
910 | next = val + NUMAINFO_EVENTS_TARGET; | |
911 | break; | |
912 | default: | |
913 | break; | |
914 | } | |
915 | __this_cpu_write(memcg->stat->targets[target], next); | |
916 | return true; | |
7a159cc9 | 917 | } |
f53d7ce3 | 918 | return false; |
d2265e6f KH |
919 | } |
920 | ||
921 | /* | |
922 | * Check events in order. | |
923 | * | |
924 | */ | |
c0ff4b85 | 925 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
926 | { |
927 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
928 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
929 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 930 | bool do_softlimit; |
82b3f2a7 | 931 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 932 | |
bb4cc1a8 AM |
933 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
934 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
935 | #if MAX_NUMNODES > 1 |
936 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
937 | MEM_CGROUP_TARGET_NUMAINFO); | |
938 | #endif | |
c0ff4b85 | 939 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
940 | if (unlikely(do_softlimit)) |
941 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 942 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 943 | if (unlikely(do_numainfo)) |
c0ff4b85 | 944 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 945 | #endif |
0a31bc97 | 946 | } |
d2265e6f KH |
947 | } |
948 | ||
cf475ad2 | 949 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 950 | { |
31a78f23 BS |
951 | /* |
952 | * mm_update_next_owner() may clear mm->owner to NULL | |
953 | * if it races with swapoff, page migration, etc. | |
954 | * So this can be called with p == NULL. | |
955 | */ | |
956 | if (unlikely(!p)) | |
957 | return NULL; | |
958 | ||
073219e9 | 959 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
960 | } |
961 | ||
df381975 | 962 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 963 | { |
c0ff4b85 | 964 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 965 | |
54595fe2 KH |
966 | rcu_read_lock(); |
967 | do { | |
6f6acb00 MH |
968 | /* |
969 | * Page cache insertions can happen withou an | |
970 | * actual mm context, e.g. during disk probing | |
971 | * on boot, loopback IO, acct() writes etc. | |
972 | */ | |
973 | if (unlikely(!mm)) | |
df381975 | 974 | memcg = root_mem_cgroup; |
6f6acb00 MH |
975 | else { |
976 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
977 | if (unlikely(!memcg)) | |
978 | memcg = root_mem_cgroup; | |
979 | } | |
ec903c0c | 980 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 981 | rcu_read_unlock(); |
c0ff4b85 | 982 | return memcg; |
54595fe2 KH |
983 | } |
984 | ||
5660048c JW |
985 | /** |
986 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
987 | * @root: hierarchy root | |
988 | * @prev: previously returned memcg, NULL on first invocation | |
989 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
990 | * | |
991 | * Returns references to children of the hierarchy below @root, or | |
992 | * @root itself, or %NULL after a full round-trip. | |
993 | * | |
994 | * Caller must pass the return value in @prev on subsequent | |
995 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
996 | * to cancel a hierarchy walk before the round-trip is complete. | |
997 | * | |
998 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
999 | * divide up the memcgs in the hierarchy among all concurrent | |
1000 | * reclaimers operating on the same zone and priority. | |
1001 | */ | |
694fbc0f | 1002 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1003 | struct mem_cgroup *prev, |
694fbc0f | 1004 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1005 | { |
5ac8fb31 JW |
1006 | struct reclaim_iter *uninitialized_var(iter); |
1007 | struct cgroup_subsys_state *css = NULL; | |
9f3a0d09 | 1008 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1009 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1010 | |
694fbc0f AM |
1011 | if (mem_cgroup_disabled()) |
1012 | return NULL; | |
5660048c | 1013 | |
9f3a0d09 JW |
1014 | if (!root) |
1015 | root = root_mem_cgroup; | |
7d74b06f | 1016 | |
9f3a0d09 | 1017 | if (prev && !reclaim) |
5ac8fb31 | 1018 | pos = prev; |
14067bb3 | 1019 | |
9f3a0d09 JW |
1020 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1021 | if (prev) | |
5ac8fb31 | 1022 | goto out; |
694fbc0f | 1023 | return root; |
9f3a0d09 | 1024 | } |
14067bb3 | 1025 | |
542f85f9 | 1026 | rcu_read_lock(); |
5f578161 | 1027 | |
5ac8fb31 JW |
1028 | if (reclaim) { |
1029 | struct mem_cgroup_per_zone *mz; | |
1030 | ||
1031 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
1032 | iter = &mz->iter[reclaim->priority]; | |
1033 | ||
1034 | if (prev && reclaim->generation != iter->generation) | |
1035 | goto out_unlock; | |
1036 | ||
1037 | do { | |
1038 | pos = ACCESS_ONCE(iter->position); | |
1039 | /* | |
1040 | * A racing update may change the position and | |
1041 | * put the last reference, hence css_tryget(), | |
1042 | * or retry to see the updated position. | |
1043 | */ | |
1044 | } while (pos && !css_tryget(&pos->css)); | |
1045 | } | |
1046 | ||
1047 | if (pos) | |
1048 | css = &pos->css; | |
1049 | ||
1050 | for (;;) { | |
1051 | css = css_next_descendant_pre(css, &root->css); | |
1052 | if (!css) { | |
1053 | /* | |
1054 | * Reclaimers share the hierarchy walk, and a | |
1055 | * new one might jump in right at the end of | |
1056 | * the hierarchy - make sure they see at least | |
1057 | * one group and restart from the beginning. | |
1058 | */ | |
1059 | if (!prev) | |
1060 | continue; | |
1061 | break; | |
527a5ec9 | 1062 | } |
7d74b06f | 1063 | |
5ac8fb31 JW |
1064 | /* |
1065 | * Verify the css and acquire a reference. The root | |
1066 | * is provided by the caller, so we know it's alive | |
1067 | * and kicking, and don't take an extra reference. | |
1068 | */ | |
1069 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1070 | |
5ac8fb31 JW |
1071 | if (css == &root->css) |
1072 | break; | |
14067bb3 | 1073 | |
b2052564 | 1074 | if (css_tryget(css)) { |
5ac8fb31 JW |
1075 | /* |
1076 | * Make sure the memcg is initialized: | |
1077 | * mem_cgroup_css_online() orders the the | |
1078 | * initialization against setting the flag. | |
1079 | */ | |
1080 | if (smp_load_acquire(&memcg->initialized)) | |
1081 | break; | |
542f85f9 | 1082 | |
5ac8fb31 | 1083 | css_put(css); |
527a5ec9 | 1084 | } |
9f3a0d09 | 1085 | |
5ac8fb31 | 1086 | memcg = NULL; |
9f3a0d09 | 1087 | } |
5ac8fb31 JW |
1088 | |
1089 | if (reclaim) { | |
1090 | if (cmpxchg(&iter->position, pos, memcg) == pos) { | |
1091 | if (memcg) | |
1092 | css_get(&memcg->css); | |
1093 | if (pos) | |
1094 | css_put(&pos->css); | |
1095 | } | |
1096 | ||
1097 | /* | |
1098 | * pairs with css_tryget when dereferencing iter->position | |
1099 | * above. | |
1100 | */ | |
1101 | if (pos) | |
1102 | css_put(&pos->css); | |
1103 | ||
1104 | if (!memcg) | |
1105 | iter->generation++; | |
1106 | else if (!prev) | |
1107 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1108 | } |
5ac8fb31 | 1109 | |
542f85f9 MH |
1110 | out_unlock: |
1111 | rcu_read_unlock(); | |
5ac8fb31 | 1112 | out: |
c40046f3 MH |
1113 | if (prev && prev != root) |
1114 | css_put(&prev->css); | |
1115 | ||
9f3a0d09 | 1116 | return memcg; |
14067bb3 | 1117 | } |
7d74b06f | 1118 | |
5660048c JW |
1119 | /** |
1120 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1121 | * @root: hierarchy root | |
1122 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1123 | */ | |
1124 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1125 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1126 | { |
1127 | if (!root) | |
1128 | root = root_mem_cgroup; | |
1129 | if (prev && prev != root) | |
1130 | css_put(&prev->css); | |
1131 | } | |
7d74b06f | 1132 | |
9f3a0d09 JW |
1133 | /* |
1134 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1135 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1136 | * be used for reference counting. | |
1137 | */ | |
1138 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1139 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1140 | iter != NULL; \ |
527a5ec9 | 1141 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1142 | |
9f3a0d09 | 1143 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1144 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1145 | iter != NULL; \ |
527a5ec9 | 1146 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1147 | |
68ae564b | 1148 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1149 | { |
c0ff4b85 | 1150 | struct mem_cgroup *memcg; |
456f998e | 1151 | |
456f998e | 1152 | rcu_read_lock(); |
c0ff4b85 R |
1153 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1154 | if (unlikely(!memcg)) | |
456f998e YH |
1155 | goto out; |
1156 | ||
1157 | switch (idx) { | |
456f998e | 1158 | case PGFAULT: |
0e574a93 JW |
1159 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1160 | break; | |
1161 | case PGMAJFAULT: | |
1162 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1163 | break; |
1164 | default: | |
1165 | BUG(); | |
1166 | } | |
1167 | out: | |
1168 | rcu_read_unlock(); | |
1169 | } | |
68ae564b | 1170 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1171 | |
925b7673 JW |
1172 | /** |
1173 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1174 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1175 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1176 | * |
1177 | * Returns the lru list vector holding pages for the given @zone and | |
1178 | * @mem. This can be the global zone lruvec, if the memory controller | |
1179 | * is disabled. | |
1180 | */ | |
1181 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1182 | struct mem_cgroup *memcg) | |
1183 | { | |
1184 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1185 | struct lruvec *lruvec; |
925b7673 | 1186 | |
bea8c150 HD |
1187 | if (mem_cgroup_disabled()) { |
1188 | lruvec = &zone->lruvec; | |
1189 | goto out; | |
1190 | } | |
925b7673 | 1191 | |
e231875b | 1192 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1193 | lruvec = &mz->lruvec; |
1194 | out: | |
1195 | /* | |
1196 | * Since a node can be onlined after the mem_cgroup was created, | |
1197 | * we have to be prepared to initialize lruvec->zone here; | |
1198 | * and if offlined then reonlined, we need to reinitialize it. | |
1199 | */ | |
1200 | if (unlikely(lruvec->zone != zone)) | |
1201 | lruvec->zone = zone; | |
1202 | return lruvec; | |
925b7673 JW |
1203 | } |
1204 | ||
925b7673 | 1205 | /** |
dfe0e773 | 1206 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1207 | * @page: the page |
fa9add64 | 1208 | * @zone: zone of the page |
dfe0e773 JW |
1209 | * |
1210 | * This function is only safe when following the LRU page isolation | |
1211 | * and putback protocol: the LRU lock must be held, and the page must | |
1212 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1213 | */ |
fa9add64 | 1214 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1215 | { |
08e552c6 | 1216 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1217 | struct mem_cgroup *memcg; |
bea8c150 | 1218 | struct lruvec *lruvec; |
6d12e2d8 | 1219 | |
bea8c150 HD |
1220 | if (mem_cgroup_disabled()) { |
1221 | lruvec = &zone->lruvec; | |
1222 | goto out; | |
1223 | } | |
925b7673 | 1224 | |
1306a85a | 1225 | memcg = page->mem_cgroup; |
7512102c | 1226 | /* |
dfe0e773 | 1227 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1228 | * possibly migrated - before they are charged. |
7512102c | 1229 | */ |
29833315 JW |
1230 | if (!memcg) |
1231 | memcg = root_mem_cgroup; | |
7512102c | 1232 | |
e231875b | 1233 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1234 | lruvec = &mz->lruvec; |
1235 | out: | |
1236 | /* | |
1237 | * Since a node can be onlined after the mem_cgroup was created, | |
1238 | * we have to be prepared to initialize lruvec->zone here; | |
1239 | * and if offlined then reonlined, we need to reinitialize it. | |
1240 | */ | |
1241 | if (unlikely(lruvec->zone != zone)) | |
1242 | lruvec->zone = zone; | |
1243 | return lruvec; | |
08e552c6 | 1244 | } |
b69408e8 | 1245 | |
925b7673 | 1246 | /** |
fa9add64 HD |
1247 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1248 | * @lruvec: mem_cgroup per zone lru vector | |
1249 | * @lru: index of lru list the page is sitting on | |
1250 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1251 | * |
fa9add64 HD |
1252 | * This function must be called when a page is added to or removed from an |
1253 | * lru list. | |
3f58a829 | 1254 | */ |
fa9add64 HD |
1255 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1256 | int nr_pages) | |
3f58a829 MK |
1257 | { |
1258 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1259 | unsigned long *lru_size; |
3f58a829 MK |
1260 | |
1261 | if (mem_cgroup_disabled()) | |
1262 | return; | |
1263 | ||
fa9add64 HD |
1264 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1265 | lru_size = mz->lru_size + lru; | |
1266 | *lru_size += nr_pages; | |
1267 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1268 | } |
544122e5 | 1269 | |
2314b42d | 1270 | bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root) |
3e92041d | 1271 | { |
2314b42d | 1272 | if (root == memcg) |
91c63734 | 1273 | return true; |
2314b42d | 1274 | if (!root->use_hierarchy) |
91c63734 | 1275 | return false; |
2314b42d | 1276 | return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); |
c3ac9a8a JW |
1277 | } |
1278 | ||
2314b42d | 1279 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1280 | { |
2314b42d | 1281 | struct mem_cgroup *task_memcg; |
158e0a2d | 1282 | struct task_struct *p; |
ffbdccf5 | 1283 | bool ret; |
4c4a2214 | 1284 | |
158e0a2d | 1285 | p = find_lock_task_mm(task); |
de077d22 | 1286 | if (p) { |
2314b42d | 1287 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1288 | task_unlock(p); |
1289 | } else { | |
1290 | /* | |
1291 | * All threads may have already detached their mm's, but the oom | |
1292 | * killer still needs to detect if they have already been oom | |
1293 | * killed to prevent needlessly killing additional tasks. | |
1294 | */ | |
ffbdccf5 | 1295 | rcu_read_lock(); |
2314b42d JW |
1296 | task_memcg = mem_cgroup_from_task(task); |
1297 | css_get(&task_memcg->css); | |
ffbdccf5 | 1298 | rcu_read_unlock(); |
de077d22 | 1299 | } |
2314b42d JW |
1300 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1301 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1302 | return ret; |
1303 | } | |
1304 | ||
c56d5c7d | 1305 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1306 | { |
9b272977 | 1307 | unsigned long inactive_ratio; |
14797e23 | 1308 | unsigned long inactive; |
9b272977 | 1309 | unsigned long active; |
c772be93 | 1310 | unsigned long gb; |
14797e23 | 1311 | |
4d7dcca2 HD |
1312 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1313 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1314 | |
c772be93 KM |
1315 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1316 | if (gb) | |
1317 | inactive_ratio = int_sqrt(10 * gb); | |
1318 | else | |
1319 | inactive_ratio = 1; | |
1320 | ||
9b272977 | 1321 | return inactive * inactive_ratio < active; |
14797e23 KM |
1322 | } |
1323 | ||
90cbc250 VD |
1324 | bool mem_cgroup_lruvec_online(struct lruvec *lruvec) |
1325 | { | |
1326 | struct mem_cgroup_per_zone *mz; | |
1327 | struct mem_cgroup *memcg; | |
1328 | ||
1329 | if (mem_cgroup_disabled()) | |
1330 | return true; | |
1331 | ||
1332 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
1333 | memcg = mz->memcg; | |
1334 | ||
1335 | return !!(memcg->css.flags & CSS_ONLINE); | |
1336 | } | |
1337 | ||
3e32cb2e | 1338 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1339 | container_of(counter, struct mem_cgroup, member) |
1340 | ||
19942822 | 1341 | /** |
9d11ea9f | 1342 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1343 | * @memcg: the memory cgroup |
19942822 | 1344 | * |
9d11ea9f | 1345 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1346 | * pages. |
19942822 | 1347 | */ |
c0ff4b85 | 1348 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1349 | { |
3e32cb2e JW |
1350 | unsigned long margin = 0; |
1351 | unsigned long count; | |
1352 | unsigned long limit; | |
9d11ea9f | 1353 | |
3e32cb2e JW |
1354 | count = page_counter_read(&memcg->memory); |
1355 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1356 | if (count < limit) | |
1357 | margin = limit - count; | |
1358 | ||
1359 | if (do_swap_account) { | |
1360 | count = page_counter_read(&memcg->memsw); | |
1361 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1362 | if (count <= limit) | |
1363 | margin = min(margin, limit - count); | |
1364 | } | |
1365 | ||
1366 | return margin; | |
19942822 JW |
1367 | } |
1368 | ||
1f4c025b | 1369 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1370 | { |
a7885eb8 | 1371 | /* root ? */ |
14208b0e | 1372 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1373 | return vm_swappiness; |
1374 | ||
bf1ff263 | 1375 | return memcg->swappiness; |
a7885eb8 KM |
1376 | } |
1377 | ||
32047e2a | 1378 | /* |
bdcbb659 | 1379 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1380 | * |
bdcbb659 QH |
1381 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1382 | * moving cgroups. This is for waiting at high-memory pressure | |
1383 | * caused by "move". | |
32047e2a | 1384 | */ |
c0ff4b85 | 1385 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1386 | { |
2bd9bb20 KH |
1387 | struct mem_cgroup *from; |
1388 | struct mem_cgroup *to; | |
4b534334 | 1389 | bool ret = false; |
2bd9bb20 KH |
1390 | /* |
1391 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1392 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1393 | */ | |
1394 | spin_lock(&mc.lock); | |
1395 | from = mc.from; | |
1396 | to = mc.to; | |
1397 | if (!from) | |
1398 | goto unlock; | |
3e92041d | 1399 | |
2314b42d JW |
1400 | ret = mem_cgroup_is_descendant(from, memcg) || |
1401 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1402 | unlock: |
1403 | spin_unlock(&mc.lock); | |
4b534334 KH |
1404 | return ret; |
1405 | } | |
1406 | ||
c0ff4b85 | 1407 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1408 | { |
1409 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1410 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1411 | DEFINE_WAIT(wait); |
1412 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1413 | /* moving charge context might have finished. */ | |
1414 | if (mc.moving_task) | |
1415 | schedule(); | |
1416 | finish_wait(&mc.waitq, &wait); | |
1417 | return true; | |
1418 | } | |
1419 | } | |
1420 | return false; | |
1421 | } | |
1422 | ||
58cf188e | 1423 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1424 | /** |
58cf188e | 1425 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1426 | * @memcg: The memory cgroup that went over limit |
1427 | * @p: Task that is going to be killed | |
1428 | * | |
1429 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1430 | * enabled | |
1431 | */ | |
1432 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1433 | { | |
e61734c5 | 1434 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1435 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1436 | struct mem_cgroup *iter; |
1437 | unsigned int i; | |
e222432b | 1438 | |
58cf188e | 1439 | if (!p) |
e222432b BS |
1440 | return; |
1441 | ||
08088cb9 | 1442 | mutex_lock(&oom_info_lock); |
e222432b BS |
1443 | rcu_read_lock(); |
1444 | ||
e61734c5 TH |
1445 | pr_info("Task in "); |
1446 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
0346dadb | 1447 | pr_cont(" killed as a result of limit of "); |
e61734c5 | 1448 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1449 | pr_cont("\n"); |
e222432b | 1450 | |
e222432b BS |
1451 | rcu_read_unlock(); |
1452 | ||
3e32cb2e JW |
1453 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1454 | K((u64)page_counter_read(&memcg->memory)), | |
1455 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1456 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1457 | K((u64)page_counter_read(&memcg->memsw)), | |
1458 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1459 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1460 | K((u64)page_counter_read(&memcg->kmem)), | |
1461 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1462 | |
1463 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1464 | pr_info("Memory cgroup stats for "); |
1465 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1466 | pr_cont(":"); |
1467 | ||
1468 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1469 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1470 | continue; | |
1471 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1472 | K(mem_cgroup_read_stat(iter, i))); | |
1473 | } | |
1474 | ||
1475 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1476 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1477 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1478 | ||
1479 | pr_cont("\n"); | |
1480 | } | |
08088cb9 | 1481 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1482 | } |
1483 | ||
81d39c20 KH |
1484 | /* |
1485 | * This function returns the number of memcg under hierarchy tree. Returns | |
1486 | * 1(self count) if no children. | |
1487 | */ | |
c0ff4b85 | 1488 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1489 | { |
1490 | int num = 0; | |
7d74b06f KH |
1491 | struct mem_cgroup *iter; |
1492 | ||
c0ff4b85 | 1493 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1494 | num++; |
81d39c20 KH |
1495 | return num; |
1496 | } | |
1497 | ||
a63d83f4 DR |
1498 | /* |
1499 | * Return the memory (and swap, if configured) limit for a memcg. | |
1500 | */ | |
3e32cb2e | 1501 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1502 | { |
3e32cb2e | 1503 | unsigned long limit; |
f3e8eb70 | 1504 | |
3e32cb2e | 1505 | limit = memcg->memory.limit; |
9a5a8f19 | 1506 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1507 | unsigned long memsw_limit; |
9a5a8f19 | 1508 | |
3e32cb2e JW |
1509 | memsw_limit = memcg->memsw.limit; |
1510 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1511 | } |
9a5a8f19 | 1512 | return limit; |
a63d83f4 DR |
1513 | } |
1514 | ||
19965460 DR |
1515 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1516 | int order) | |
9cbb78bb DR |
1517 | { |
1518 | struct mem_cgroup *iter; | |
1519 | unsigned long chosen_points = 0; | |
1520 | unsigned long totalpages; | |
1521 | unsigned int points = 0; | |
1522 | struct task_struct *chosen = NULL; | |
1523 | ||
876aafbf | 1524 | /* |
465adcf1 DR |
1525 | * If current has a pending SIGKILL or is exiting, then automatically |
1526 | * select it. The goal is to allow it to allocate so that it may | |
1527 | * quickly exit and free its memory. | |
876aafbf | 1528 | */ |
d003f371 | 1529 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
49550b60 | 1530 | mark_tsk_oom_victim(current); |
876aafbf DR |
1531 | return; |
1532 | } | |
1533 | ||
1534 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1535 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1536 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1537 | struct css_task_iter it; |
9cbb78bb DR |
1538 | struct task_struct *task; |
1539 | ||
72ec7029 TH |
1540 | css_task_iter_start(&iter->css, &it); |
1541 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1542 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1543 | false)) { | |
1544 | case OOM_SCAN_SELECT: | |
1545 | if (chosen) | |
1546 | put_task_struct(chosen); | |
1547 | chosen = task; | |
1548 | chosen_points = ULONG_MAX; | |
1549 | get_task_struct(chosen); | |
1550 | /* fall through */ | |
1551 | case OOM_SCAN_CONTINUE: | |
1552 | continue; | |
1553 | case OOM_SCAN_ABORT: | |
72ec7029 | 1554 | css_task_iter_end(&it); |
9cbb78bb DR |
1555 | mem_cgroup_iter_break(memcg, iter); |
1556 | if (chosen) | |
1557 | put_task_struct(chosen); | |
1558 | return; | |
1559 | case OOM_SCAN_OK: | |
1560 | break; | |
1561 | }; | |
1562 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1563 | if (!points || points < chosen_points) |
1564 | continue; | |
1565 | /* Prefer thread group leaders for display purposes */ | |
1566 | if (points == chosen_points && | |
1567 | thread_group_leader(chosen)) | |
1568 | continue; | |
1569 | ||
1570 | if (chosen) | |
1571 | put_task_struct(chosen); | |
1572 | chosen = task; | |
1573 | chosen_points = points; | |
1574 | get_task_struct(chosen); | |
9cbb78bb | 1575 | } |
72ec7029 | 1576 | css_task_iter_end(&it); |
9cbb78bb DR |
1577 | } |
1578 | ||
1579 | if (!chosen) | |
1580 | return; | |
1581 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1582 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1583 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1584 | } |
1585 | ||
ae6e71d3 MC |
1586 | #if MAX_NUMNODES > 1 |
1587 | ||
4d0c066d KH |
1588 | /** |
1589 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1590 | * @memcg: the target memcg |
4d0c066d KH |
1591 | * @nid: the node ID to be checked. |
1592 | * @noswap : specify true here if the user wants flle only information. | |
1593 | * | |
1594 | * This function returns whether the specified memcg contains any | |
1595 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1596 | * pages in the node. | |
1597 | */ | |
c0ff4b85 | 1598 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1599 | int nid, bool noswap) |
1600 | { | |
c0ff4b85 | 1601 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1602 | return true; |
1603 | if (noswap || !total_swap_pages) | |
1604 | return false; | |
c0ff4b85 | 1605 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1606 | return true; |
1607 | return false; | |
1608 | ||
1609 | } | |
889976db YH |
1610 | |
1611 | /* | |
1612 | * Always updating the nodemask is not very good - even if we have an empty | |
1613 | * list or the wrong list here, we can start from some node and traverse all | |
1614 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1615 | * | |
1616 | */ | |
c0ff4b85 | 1617 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1618 | { |
1619 | int nid; | |
453a9bf3 KH |
1620 | /* |
1621 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1622 | * pagein/pageout changes since the last update. | |
1623 | */ | |
c0ff4b85 | 1624 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1625 | return; |
c0ff4b85 | 1626 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1627 | return; |
1628 | ||
889976db | 1629 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1630 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1631 | |
31aaea4a | 1632 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1633 | |
c0ff4b85 R |
1634 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1635 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1636 | } |
453a9bf3 | 1637 | |
c0ff4b85 R |
1638 | atomic_set(&memcg->numainfo_events, 0); |
1639 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1640 | } |
1641 | ||
1642 | /* | |
1643 | * Selecting a node where we start reclaim from. Because what we need is just | |
1644 | * reducing usage counter, start from anywhere is O,K. Considering | |
1645 | * memory reclaim from current node, there are pros. and cons. | |
1646 | * | |
1647 | * Freeing memory from current node means freeing memory from a node which | |
1648 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1649 | * hit limits, it will see a contention on a node. But freeing from remote | |
1650 | * node means more costs for memory reclaim because of memory latency. | |
1651 | * | |
1652 | * Now, we use round-robin. Better algorithm is welcomed. | |
1653 | */ | |
c0ff4b85 | 1654 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1655 | { |
1656 | int node; | |
1657 | ||
c0ff4b85 R |
1658 | mem_cgroup_may_update_nodemask(memcg); |
1659 | node = memcg->last_scanned_node; | |
889976db | 1660 | |
c0ff4b85 | 1661 | node = next_node(node, memcg->scan_nodes); |
889976db | 1662 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1663 | node = first_node(memcg->scan_nodes); |
889976db YH |
1664 | /* |
1665 | * We call this when we hit limit, not when pages are added to LRU. | |
1666 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1667 | * memcg is too small and all pages are not on LRU. In that case, | |
1668 | * we use curret node. | |
1669 | */ | |
1670 | if (unlikely(node == MAX_NUMNODES)) | |
1671 | node = numa_node_id(); | |
1672 | ||
c0ff4b85 | 1673 | memcg->last_scanned_node = node; |
889976db YH |
1674 | return node; |
1675 | } | |
889976db | 1676 | #else |
c0ff4b85 | 1677 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1678 | { |
1679 | return 0; | |
1680 | } | |
1681 | #endif | |
1682 | ||
0608f43d AM |
1683 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1684 | struct zone *zone, | |
1685 | gfp_t gfp_mask, | |
1686 | unsigned long *total_scanned) | |
1687 | { | |
1688 | struct mem_cgroup *victim = NULL; | |
1689 | int total = 0; | |
1690 | int loop = 0; | |
1691 | unsigned long excess; | |
1692 | unsigned long nr_scanned; | |
1693 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1694 | .zone = zone, | |
1695 | .priority = 0, | |
1696 | }; | |
1697 | ||
3e32cb2e | 1698 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1699 | |
1700 | while (1) { | |
1701 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1702 | if (!victim) { | |
1703 | loop++; | |
1704 | if (loop >= 2) { | |
1705 | /* | |
1706 | * If we have not been able to reclaim | |
1707 | * anything, it might because there are | |
1708 | * no reclaimable pages under this hierarchy | |
1709 | */ | |
1710 | if (!total) | |
1711 | break; | |
1712 | /* | |
1713 | * We want to do more targeted reclaim. | |
1714 | * excess >> 2 is not to excessive so as to | |
1715 | * reclaim too much, nor too less that we keep | |
1716 | * coming back to reclaim from this cgroup | |
1717 | */ | |
1718 | if (total >= (excess >> 2) || | |
1719 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1720 | break; | |
1721 | } | |
1722 | continue; | |
1723 | } | |
0608f43d AM |
1724 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1725 | zone, &nr_scanned); | |
1726 | *total_scanned += nr_scanned; | |
3e32cb2e | 1727 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1728 | break; |
6d61ef40 | 1729 | } |
0608f43d AM |
1730 | mem_cgroup_iter_break(root_memcg, victim); |
1731 | return total; | |
6d61ef40 BS |
1732 | } |
1733 | ||
0056f4e6 JW |
1734 | #ifdef CONFIG_LOCKDEP |
1735 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1736 | .name = "memcg_oom_lock", | |
1737 | }; | |
1738 | #endif | |
1739 | ||
fb2a6fc5 JW |
1740 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1741 | ||
867578cb KH |
1742 | /* |
1743 | * Check OOM-Killer is already running under our hierarchy. | |
1744 | * If someone is running, return false. | |
1745 | */ | |
fb2a6fc5 | 1746 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1747 | { |
79dfdacc | 1748 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1749 | |
fb2a6fc5 JW |
1750 | spin_lock(&memcg_oom_lock); |
1751 | ||
9f3a0d09 | 1752 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1753 | if (iter->oom_lock) { |
79dfdacc MH |
1754 | /* |
1755 | * this subtree of our hierarchy is already locked | |
1756 | * so we cannot give a lock. | |
1757 | */ | |
79dfdacc | 1758 | failed = iter; |
9f3a0d09 JW |
1759 | mem_cgroup_iter_break(memcg, iter); |
1760 | break; | |
23751be0 JW |
1761 | } else |
1762 | iter->oom_lock = true; | |
7d74b06f | 1763 | } |
867578cb | 1764 | |
fb2a6fc5 JW |
1765 | if (failed) { |
1766 | /* | |
1767 | * OK, we failed to lock the whole subtree so we have | |
1768 | * to clean up what we set up to the failing subtree | |
1769 | */ | |
1770 | for_each_mem_cgroup_tree(iter, memcg) { | |
1771 | if (iter == failed) { | |
1772 | mem_cgroup_iter_break(memcg, iter); | |
1773 | break; | |
1774 | } | |
1775 | iter->oom_lock = false; | |
79dfdacc | 1776 | } |
0056f4e6 JW |
1777 | } else |
1778 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1779 | |
1780 | spin_unlock(&memcg_oom_lock); | |
1781 | ||
1782 | return !failed; | |
a636b327 | 1783 | } |
0b7f569e | 1784 | |
fb2a6fc5 | 1785 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1786 | { |
7d74b06f KH |
1787 | struct mem_cgroup *iter; |
1788 | ||
fb2a6fc5 | 1789 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1790 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1791 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1792 | iter->oom_lock = false; |
fb2a6fc5 | 1793 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1794 | } |
1795 | ||
c0ff4b85 | 1796 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1797 | { |
1798 | struct mem_cgroup *iter; | |
1799 | ||
c0ff4b85 | 1800 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1801 | atomic_inc(&iter->under_oom); |
1802 | } | |
1803 | ||
c0ff4b85 | 1804 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1805 | { |
1806 | struct mem_cgroup *iter; | |
1807 | ||
867578cb KH |
1808 | /* |
1809 | * When a new child is created while the hierarchy is under oom, | |
1810 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1811 | * atomic_add_unless() here. | |
1812 | */ | |
c0ff4b85 | 1813 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1814 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1815 | } |
1816 | ||
867578cb KH |
1817 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1818 | ||
dc98df5a | 1819 | struct oom_wait_info { |
d79154bb | 1820 | struct mem_cgroup *memcg; |
dc98df5a KH |
1821 | wait_queue_t wait; |
1822 | }; | |
1823 | ||
1824 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1825 | unsigned mode, int sync, void *arg) | |
1826 | { | |
d79154bb HD |
1827 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1828 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1829 | struct oom_wait_info *oom_wait_info; |
1830 | ||
1831 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1832 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1833 | |
2314b42d JW |
1834 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1835 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1836 | return 0; |
dc98df5a KH |
1837 | return autoremove_wake_function(wait, mode, sync, arg); |
1838 | } | |
1839 | ||
c0ff4b85 | 1840 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1841 | { |
3812c8c8 | 1842 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
1843 | /* for filtering, pass "memcg" as argument. */ |
1844 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1845 | } |
1846 | ||
c0ff4b85 | 1847 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1848 | { |
c0ff4b85 R |
1849 | if (memcg && atomic_read(&memcg->under_oom)) |
1850 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1851 | } |
1852 | ||
3812c8c8 | 1853 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1854 | { |
3812c8c8 JW |
1855 | if (!current->memcg_oom.may_oom) |
1856 | return; | |
867578cb | 1857 | /* |
49426420 JW |
1858 | * We are in the middle of the charge context here, so we |
1859 | * don't want to block when potentially sitting on a callstack | |
1860 | * that holds all kinds of filesystem and mm locks. | |
1861 | * | |
1862 | * Also, the caller may handle a failed allocation gracefully | |
1863 | * (like optional page cache readahead) and so an OOM killer | |
1864 | * invocation might not even be necessary. | |
1865 | * | |
1866 | * That's why we don't do anything here except remember the | |
1867 | * OOM context and then deal with it at the end of the page | |
1868 | * fault when the stack is unwound, the locks are released, | |
1869 | * and when we know whether the fault was overall successful. | |
867578cb | 1870 | */ |
49426420 JW |
1871 | css_get(&memcg->css); |
1872 | current->memcg_oom.memcg = memcg; | |
1873 | current->memcg_oom.gfp_mask = mask; | |
1874 | current->memcg_oom.order = order; | |
3812c8c8 JW |
1875 | } |
1876 | ||
1877 | /** | |
1878 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1879 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1880 | * |
49426420 JW |
1881 | * This has to be called at the end of a page fault if the memcg OOM |
1882 | * handler was enabled. | |
3812c8c8 | 1883 | * |
49426420 | 1884 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1885 | * sleep on a waitqueue until the userspace task resolves the |
1886 | * situation. Sleeping directly in the charge context with all kinds | |
1887 | * of locks held is not a good idea, instead we remember an OOM state | |
1888 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1889 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1890 | * |
1891 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1892 | * completed, %false otherwise. |
3812c8c8 | 1893 | */ |
49426420 | 1894 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1895 | { |
49426420 | 1896 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 1897 | struct oom_wait_info owait; |
49426420 | 1898 | bool locked; |
3812c8c8 JW |
1899 | |
1900 | /* OOM is global, do not handle */ | |
3812c8c8 | 1901 | if (!memcg) |
49426420 | 1902 | return false; |
3812c8c8 | 1903 | |
c32b3cbe | 1904 | if (!handle || oom_killer_disabled) |
49426420 | 1905 | goto cleanup; |
3812c8c8 JW |
1906 | |
1907 | owait.memcg = memcg; | |
1908 | owait.wait.flags = 0; | |
1909 | owait.wait.func = memcg_oom_wake_function; | |
1910 | owait.wait.private = current; | |
1911 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1912 | |
3812c8c8 | 1913 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1914 | mem_cgroup_mark_under_oom(memcg); |
1915 | ||
1916 | locked = mem_cgroup_oom_trylock(memcg); | |
1917 | ||
1918 | if (locked) | |
1919 | mem_cgroup_oom_notify(memcg); | |
1920 | ||
1921 | if (locked && !memcg->oom_kill_disable) { | |
1922 | mem_cgroup_unmark_under_oom(memcg); | |
1923 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1924 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
1925 | current->memcg_oom.order); | |
1926 | } else { | |
3812c8c8 | 1927 | schedule(); |
49426420 JW |
1928 | mem_cgroup_unmark_under_oom(memcg); |
1929 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1930 | } | |
1931 | ||
1932 | if (locked) { | |
fb2a6fc5 JW |
1933 | mem_cgroup_oom_unlock(memcg); |
1934 | /* | |
1935 | * There is no guarantee that an OOM-lock contender | |
1936 | * sees the wakeups triggered by the OOM kill | |
1937 | * uncharges. Wake any sleepers explicitely. | |
1938 | */ | |
1939 | memcg_oom_recover(memcg); | |
1940 | } | |
49426420 JW |
1941 | cleanup: |
1942 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 1943 | css_put(&memcg->css); |
867578cb | 1944 | return true; |
0b7f569e KH |
1945 | } |
1946 | ||
d7365e78 JW |
1947 | /** |
1948 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1949 | * @page: page that is going to change accounted state | |
32047e2a | 1950 | * |
d7365e78 JW |
1951 | * This function must mark the beginning of an accounted page state |
1952 | * change to prevent double accounting when the page is concurrently | |
1953 | * being moved to another memcg: | |
32047e2a | 1954 | * |
6de22619 | 1955 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1956 | * if (TestClearPageState(page)) |
1957 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1958 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1959 | */ |
6de22619 | 1960 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1961 | { |
1962 | struct mem_cgroup *memcg; | |
6de22619 | 1963 | unsigned long flags; |
89c06bd5 | 1964 | |
6de22619 JW |
1965 | /* |
1966 | * The RCU lock is held throughout the transaction. The fast | |
1967 | * path can get away without acquiring the memcg->move_lock | |
1968 | * because page moving starts with an RCU grace period. | |
1969 | * | |
1970 | * The RCU lock also protects the memcg from being freed when | |
1971 | * the page state that is going to change is the only thing | |
1972 | * preventing the page from being uncharged. | |
1973 | * E.g. end-writeback clearing PageWriteback(), which allows | |
1974 | * migration to go ahead and uncharge the page before the | |
1975 | * account transaction might be complete. | |
1976 | */ | |
d7365e78 JW |
1977 | rcu_read_lock(); |
1978 | ||
1979 | if (mem_cgroup_disabled()) | |
1980 | return NULL; | |
89c06bd5 | 1981 | again: |
1306a85a | 1982 | memcg = page->mem_cgroup; |
29833315 | 1983 | if (unlikely(!memcg)) |
d7365e78 JW |
1984 | return NULL; |
1985 | ||
bdcbb659 | 1986 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 1987 | return memcg; |
89c06bd5 | 1988 | |
6de22619 | 1989 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1990 | if (memcg != page->mem_cgroup) { |
6de22619 | 1991 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1992 | goto again; |
1993 | } | |
6de22619 JW |
1994 | |
1995 | /* | |
1996 | * When charge migration first begins, we can have locked and | |
1997 | * unlocked page stat updates happening concurrently. Track | |
1998 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
1999 | */ | |
2000 | memcg->move_lock_task = current; | |
2001 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
2002 | |
2003 | return memcg; | |
89c06bd5 KH |
2004 | } |
2005 | ||
d7365e78 JW |
2006 | /** |
2007 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2008 | * @memcg: the memcg that was accounted against | |
d7365e78 | 2009 | */ |
6de22619 | 2010 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 2011 | { |
6de22619 JW |
2012 | if (memcg && memcg->move_lock_task == current) { |
2013 | unsigned long flags = memcg->move_lock_flags; | |
2014 | ||
2015 | memcg->move_lock_task = NULL; | |
2016 | memcg->move_lock_flags = 0; | |
2017 | ||
2018 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2019 | } | |
89c06bd5 | 2020 | |
d7365e78 | 2021 | rcu_read_unlock(); |
89c06bd5 KH |
2022 | } |
2023 | ||
d7365e78 JW |
2024 | /** |
2025 | * mem_cgroup_update_page_stat - update page state statistics | |
2026 | * @memcg: memcg to account against | |
2027 | * @idx: page state item to account | |
2028 | * @val: number of pages (positive or negative) | |
2029 | * | |
2030 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2031 | */ | |
2032 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2033 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2034 | { |
658b72c5 | 2035 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2036 | |
d7365e78 JW |
2037 | if (memcg) |
2038 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2039 | } |
26174efd | 2040 | |
cdec2e42 KH |
2041 | /* |
2042 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2043 | * TODO: maybe necessary to use big numbers in big irons. | |
2044 | */ | |
7ec99d62 | 2045 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2046 | struct memcg_stock_pcp { |
2047 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2048 | unsigned int nr_pages; |
cdec2e42 | 2049 | struct work_struct work; |
26fe6168 | 2050 | unsigned long flags; |
a0db00fc | 2051 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2052 | }; |
2053 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2054 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2055 | |
a0956d54 SS |
2056 | /** |
2057 | * consume_stock: Try to consume stocked charge on this cpu. | |
2058 | * @memcg: memcg to consume from. | |
2059 | * @nr_pages: how many pages to charge. | |
2060 | * | |
2061 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2062 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2063 | * service an allocation will refill the stock. | |
2064 | * | |
2065 | * returns true if successful, false otherwise. | |
cdec2e42 | 2066 | */ |
a0956d54 | 2067 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2068 | { |
2069 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2070 | bool ret = false; |
cdec2e42 | 2071 | |
a0956d54 | 2072 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2073 | return ret; |
a0956d54 | 2074 | |
cdec2e42 | 2075 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2076 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2077 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2078 | ret = true; |
2079 | } | |
cdec2e42 KH |
2080 | put_cpu_var(memcg_stock); |
2081 | return ret; | |
2082 | } | |
2083 | ||
2084 | /* | |
3e32cb2e | 2085 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2086 | */ |
2087 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2088 | { | |
2089 | struct mem_cgroup *old = stock->cached; | |
2090 | ||
11c9ea4e | 2091 | if (stock->nr_pages) { |
3e32cb2e | 2092 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2093 | if (do_swap_account) |
3e32cb2e | 2094 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2095 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2096 | stock->nr_pages = 0; |
cdec2e42 KH |
2097 | } |
2098 | stock->cached = NULL; | |
cdec2e42 KH |
2099 | } |
2100 | ||
2101 | /* | |
2102 | * This must be called under preempt disabled or must be called by | |
2103 | * a thread which is pinned to local cpu. | |
2104 | */ | |
2105 | static void drain_local_stock(struct work_struct *dummy) | |
2106 | { | |
7c8e0181 | 2107 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2108 | drain_stock(stock); |
26fe6168 | 2109 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2110 | } |
2111 | ||
2112 | /* | |
3e32cb2e | 2113 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2114 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2115 | */ |
c0ff4b85 | 2116 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2117 | { |
2118 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2119 | ||
c0ff4b85 | 2120 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2121 | drain_stock(stock); |
c0ff4b85 | 2122 | stock->cached = memcg; |
cdec2e42 | 2123 | } |
11c9ea4e | 2124 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2125 | put_cpu_var(memcg_stock); |
2126 | } | |
2127 | ||
2128 | /* | |
c0ff4b85 | 2129 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2130 | * of the hierarchy under it. |
cdec2e42 | 2131 | */ |
6d3d6aa2 | 2132 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2133 | { |
26fe6168 | 2134 | int cpu, curcpu; |
d38144b7 | 2135 | |
6d3d6aa2 JW |
2136 | /* If someone's already draining, avoid adding running more workers. */ |
2137 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2138 | return; | |
cdec2e42 | 2139 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2140 | get_online_cpus(); |
5af12d0e | 2141 | curcpu = get_cpu(); |
cdec2e42 KH |
2142 | for_each_online_cpu(cpu) { |
2143 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2144 | struct mem_cgroup *memcg; |
26fe6168 | 2145 | |
c0ff4b85 R |
2146 | memcg = stock->cached; |
2147 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2148 | continue; |
2314b42d | 2149 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 2150 | continue; |
d1a05b69 MH |
2151 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2152 | if (cpu == curcpu) | |
2153 | drain_local_stock(&stock->work); | |
2154 | else | |
2155 | schedule_work_on(cpu, &stock->work); | |
2156 | } | |
cdec2e42 | 2157 | } |
5af12d0e | 2158 | put_cpu(); |
f894ffa8 | 2159 | put_online_cpus(); |
9f50fad6 | 2160 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2161 | } |
2162 | ||
711d3d2c KH |
2163 | /* |
2164 | * This function drains percpu counter value from DEAD cpu and | |
2165 | * move it to local cpu. Note that this function can be preempted. | |
2166 | */ | |
c0ff4b85 | 2167 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2168 | { |
2169 | int i; | |
2170 | ||
c0ff4b85 | 2171 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2172 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2173 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2174 | |
c0ff4b85 R |
2175 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2176 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2177 | } |
e9f8974f | 2178 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2179 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2180 | |
c0ff4b85 R |
2181 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2182 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2183 | } |
c0ff4b85 | 2184 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2185 | } |
2186 | ||
0db0628d | 2187 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2188 | unsigned long action, |
2189 | void *hcpu) | |
2190 | { | |
2191 | int cpu = (unsigned long)hcpu; | |
2192 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2193 | struct mem_cgroup *iter; |
cdec2e42 | 2194 | |
619d094b | 2195 | if (action == CPU_ONLINE) |
1489ebad | 2196 | return NOTIFY_OK; |
1489ebad | 2197 | |
d833049b | 2198 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2199 | return NOTIFY_OK; |
711d3d2c | 2200 | |
9f3a0d09 | 2201 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2202 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2203 | ||
cdec2e42 KH |
2204 | stock = &per_cpu(memcg_stock, cpu); |
2205 | drain_stock(stock); | |
2206 | return NOTIFY_OK; | |
2207 | } | |
2208 | ||
00501b53 JW |
2209 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2210 | unsigned int nr_pages) | |
8a9f3ccd | 2211 | { |
7ec99d62 | 2212 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2213 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2214 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2215 | struct page_counter *counter; |
6539cc05 | 2216 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2217 | bool may_swap = true; |
2218 | bool drained = false; | |
05b84301 | 2219 | int ret = 0; |
a636b327 | 2220 | |
ce00a967 JW |
2221 | if (mem_cgroup_is_root(memcg)) |
2222 | goto done; | |
6539cc05 | 2223 | retry: |
b6b6cc72 MH |
2224 | if (consume_stock(memcg, nr_pages)) |
2225 | goto done; | |
8a9f3ccd | 2226 | |
3fbe7244 | 2227 | if (!do_swap_account || |
3e32cb2e JW |
2228 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2229 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2230 | goto done_restock; |
3fbe7244 | 2231 | if (do_swap_account) |
3e32cb2e JW |
2232 | page_counter_uncharge(&memcg->memsw, batch); |
2233 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2234 | } else { |
3e32cb2e | 2235 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2236 | may_swap = false; |
3fbe7244 | 2237 | } |
7a81b88c | 2238 | |
6539cc05 JW |
2239 | if (batch > nr_pages) { |
2240 | batch = nr_pages; | |
2241 | goto retry; | |
2242 | } | |
6d61ef40 | 2243 | |
06b078fc JW |
2244 | /* |
2245 | * Unlike in global OOM situations, memcg is not in a physical | |
2246 | * memory shortage. Allow dying and OOM-killed tasks to | |
2247 | * bypass the last charges so that they can exit quickly and | |
2248 | * free their memory. | |
2249 | */ | |
2250 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2251 | fatal_signal_pending(current) || | |
2252 | current->flags & PF_EXITING)) | |
2253 | goto bypass; | |
2254 | ||
2255 | if (unlikely(task_in_memcg_oom(current))) | |
2256 | goto nomem; | |
2257 | ||
6539cc05 JW |
2258 | if (!(gfp_mask & __GFP_WAIT)) |
2259 | goto nomem; | |
4b534334 | 2260 | |
241994ed JW |
2261 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2262 | ||
b70a2a21 JW |
2263 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2264 | gfp_mask, may_swap); | |
6539cc05 | 2265 | |
61e02c74 | 2266 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2267 | goto retry; |
28c34c29 | 2268 | |
b70a2a21 | 2269 | if (!drained) { |
6d3d6aa2 | 2270 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2271 | drained = true; |
2272 | goto retry; | |
2273 | } | |
2274 | ||
28c34c29 JW |
2275 | if (gfp_mask & __GFP_NORETRY) |
2276 | goto nomem; | |
6539cc05 JW |
2277 | /* |
2278 | * Even though the limit is exceeded at this point, reclaim | |
2279 | * may have been able to free some pages. Retry the charge | |
2280 | * before killing the task. | |
2281 | * | |
2282 | * Only for regular pages, though: huge pages are rather | |
2283 | * unlikely to succeed so close to the limit, and we fall back | |
2284 | * to regular pages anyway in case of failure. | |
2285 | */ | |
61e02c74 | 2286 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2287 | goto retry; |
2288 | /* | |
2289 | * At task move, charge accounts can be doubly counted. So, it's | |
2290 | * better to wait until the end of task_move if something is going on. | |
2291 | */ | |
2292 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2293 | goto retry; | |
2294 | ||
9b130619 JW |
2295 | if (nr_retries--) |
2296 | goto retry; | |
2297 | ||
06b078fc JW |
2298 | if (gfp_mask & __GFP_NOFAIL) |
2299 | goto bypass; | |
2300 | ||
6539cc05 JW |
2301 | if (fatal_signal_pending(current)) |
2302 | goto bypass; | |
2303 | ||
241994ed JW |
2304 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2305 | ||
61e02c74 | 2306 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2307 | nomem: |
6d1fdc48 | 2308 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2309 | return -ENOMEM; |
867578cb | 2310 | bypass: |
ce00a967 | 2311 | return -EINTR; |
6539cc05 JW |
2312 | |
2313 | done_restock: | |
e8ea14cc | 2314 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2315 | if (batch > nr_pages) |
2316 | refill_stock(memcg, batch - nr_pages); | |
241994ed JW |
2317 | /* |
2318 | * If the hierarchy is above the normal consumption range, | |
2319 | * make the charging task trim their excess contribution. | |
2320 | */ | |
2321 | do { | |
2322 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2323 | continue; | |
2324 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
2325 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
2326 | } while ((memcg = parent_mem_cgroup(memcg))); | |
6539cc05 | 2327 | done: |
05b84301 | 2328 | return ret; |
7a81b88c | 2329 | } |
8a9f3ccd | 2330 | |
00501b53 | 2331 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2332 | { |
ce00a967 JW |
2333 | if (mem_cgroup_is_root(memcg)) |
2334 | return; | |
2335 | ||
3e32cb2e | 2336 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2337 | if (do_swap_account) |
3e32cb2e | 2338 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2339 | |
e8ea14cc | 2340 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2341 | } |
2342 | ||
a3b2d692 KH |
2343 | /* |
2344 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2345 | * rcu_read_lock(). The caller is responsible for calling |
2346 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2347 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2348 | */ |
2349 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2350 | { | |
a3b2d692 KH |
2351 | /* ID 0 is unused ID */ |
2352 | if (!id) | |
2353 | return NULL; | |
34c00c31 | 2354 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2355 | } |
2356 | ||
0a31bc97 JW |
2357 | /* |
2358 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2359 | * @page: the page | |
2360 | * | |
2361 | * Look up, get a css reference, and return the memcg that owns @page. | |
2362 | * | |
2363 | * The page must be locked to prevent racing with swap-in and page | |
2364 | * cache charges. If coming from an unlocked page table, the caller | |
2365 | * must ensure the page is on the LRU or this can race with charging. | |
2366 | */ | |
e42d9d5d | 2367 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2368 | { |
29833315 | 2369 | struct mem_cgroup *memcg; |
a3b2d692 | 2370 | unsigned short id; |
b5a84319 KH |
2371 | swp_entry_t ent; |
2372 | ||
309381fe | 2373 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2374 | |
1306a85a | 2375 | memcg = page->mem_cgroup; |
29833315 JW |
2376 | if (memcg) { |
2377 | if (!css_tryget_online(&memcg->css)) | |
c0ff4b85 | 2378 | memcg = NULL; |
e42d9d5d | 2379 | } else if (PageSwapCache(page)) { |
3c776e64 | 2380 | ent.val = page_private(page); |
9fb4b7cc | 2381 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2382 | rcu_read_lock(); |
c0ff4b85 | 2383 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2384 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2385 | memcg = NULL; |
a3b2d692 | 2386 | rcu_read_unlock(); |
3c776e64 | 2387 | } |
c0ff4b85 | 2388 | return memcg; |
b5a84319 KH |
2389 | } |
2390 | ||
0a31bc97 JW |
2391 | static void lock_page_lru(struct page *page, int *isolated) |
2392 | { | |
2393 | struct zone *zone = page_zone(page); | |
2394 | ||
2395 | spin_lock_irq(&zone->lru_lock); | |
2396 | if (PageLRU(page)) { | |
2397 | struct lruvec *lruvec; | |
2398 | ||
2399 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2400 | ClearPageLRU(page); | |
2401 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2402 | *isolated = 1; | |
2403 | } else | |
2404 | *isolated = 0; | |
2405 | } | |
2406 | ||
2407 | static void unlock_page_lru(struct page *page, int isolated) | |
2408 | { | |
2409 | struct zone *zone = page_zone(page); | |
2410 | ||
2411 | if (isolated) { | |
2412 | struct lruvec *lruvec; | |
2413 | ||
2414 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2415 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2416 | SetPageLRU(page); | |
2417 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2418 | } | |
2419 | spin_unlock_irq(&zone->lru_lock); | |
2420 | } | |
2421 | ||
00501b53 | 2422 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2423 | bool lrucare) |
7a81b88c | 2424 | { |
0a31bc97 | 2425 | int isolated; |
9ce70c02 | 2426 | |
1306a85a | 2427 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2428 | |
2429 | /* | |
2430 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2431 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2432 | */ | |
0a31bc97 JW |
2433 | if (lrucare) |
2434 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2435 | |
0a31bc97 JW |
2436 | /* |
2437 | * Nobody should be changing or seriously looking at | |
1306a85a | 2438 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2439 | * |
2440 | * - the page is uncharged | |
2441 | * | |
2442 | * - the page is off-LRU | |
2443 | * | |
2444 | * - an anonymous fault has exclusive page access, except for | |
2445 | * a locked page table | |
2446 | * | |
2447 | * - a page cache insertion, a swapin fault, or a migration | |
2448 | * have the page locked | |
2449 | */ | |
1306a85a | 2450 | page->mem_cgroup = memcg; |
9ce70c02 | 2451 | |
0a31bc97 JW |
2452 | if (lrucare) |
2453 | unlock_page_lru(page, isolated); | |
7a81b88c | 2454 | } |
66e1707b | 2455 | |
7ae1e1d0 | 2456 | #ifdef CONFIG_MEMCG_KMEM |
dbf22eb6 VD |
2457 | int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2458 | unsigned long nr_pages) | |
7ae1e1d0 | 2459 | { |
3e32cb2e | 2460 | struct page_counter *counter; |
7ae1e1d0 | 2461 | int ret = 0; |
7ae1e1d0 | 2462 | |
3e32cb2e JW |
2463 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2464 | if (ret < 0) | |
7ae1e1d0 GC |
2465 | return ret; |
2466 | ||
3e32cb2e | 2467 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2468 | if (ret == -EINTR) { |
2469 | /* | |
00501b53 JW |
2470 | * try_charge() chose to bypass to root due to OOM kill or |
2471 | * fatal signal. Since our only options are to either fail | |
2472 | * the allocation or charge it to this cgroup, do it as a | |
2473 | * temporary condition. But we can't fail. From a kmem/slab | |
2474 | * perspective, the cache has already been selected, by | |
2475 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2476 | * our minds. |
2477 | * | |
2478 | * This condition will only trigger if the task entered | |
00501b53 JW |
2479 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2480 | * during try_charge() above. Tasks that were already dying | |
2481 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2482 | * directed to the root cgroup in memcontrol.h |
2483 | */ | |
3e32cb2e | 2484 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2485 | if (do_swap_account) |
3e32cb2e | 2486 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2487 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2488 | ret = 0; |
2489 | } else if (ret) | |
3e32cb2e | 2490 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2491 | |
2492 | return ret; | |
2493 | } | |
2494 | ||
dbf22eb6 | 2495 | void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages) |
7ae1e1d0 | 2496 | { |
3e32cb2e | 2497 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2498 | if (do_swap_account) |
3e32cb2e | 2499 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2500 | |
64f21993 | 2501 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7de37682 | 2502 | |
e8ea14cc | 2503 | css_put_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2504 | } |
2505 | ||
2633d7a0 GC |
2506 | /* |
2507 | * helper for acessing a memcg's index. It will be used as an index in the | |
2508 | * child cache array in kmem_cache, and also to derive its name. This function | |
2509 | * will return -1 when this is not a kmem-limited memcg. | |
2510 | */ | |
2511 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2512 | { | |
2513 | return memcg ? memcg->kmemcg_id : -1; | |
2514 | } | |
2515 | ||
f3bb3043 | 2516 | static int memcg_alloc_cache_id(void) |
55007d84 | 2517 | { |
f3bb3043 VD |
2518 | int id, size; |
2519 | int err; | |
2520 | ||
dbcf73e2 | 2521 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2522 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2523 | if (id < 0) | |
2524 | return id; | |
55007d84 | 2525 | |
dbcf73e2 | 2526 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2527 | return id; |
2528 | ||
2529 | /* | |
2530 | * There's no space for the new id in memcg_caches arrays, | |
2531 | * so we have to grow them. | |
2532 | */ | |
05257a1a | 2533 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2534 | |
2535 | size = 2 * (id + 1); | |
55007d84 GC |
2536 | if (size < MEMCG_CACHES_MIN_SIZE) |
2537 | size = MEMCG_CACHES_MIN_SIZE; | |
2538 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2539 | size = MEMCG_CACHES_MAX_SIZE; | |
2540 | ||
f3bb3043 | 2541 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2542 | if (!err) |
2543 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2544 | if (!err) |
2545 | memcg_nr_cache_ids = size; | |
2546 | ||
2547 | up_write(&memcg_cache_ids_sem); | |
2548 | ||
f3bb3043 | 2549 | if (err) { |
dbcf73e2 | 2550 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2551 | return err; |
2552 | } | |
2553 | return id; | |
2554 | } | |
2555 | ||
2556 | static void memcg_free_cache_id(int id) | |
2557 | { | |
dbcf73e2 | 2558 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2559 | } |
2560 | ||
d5b3cf71 | 2561 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2562 | struct mem_cgroup *memcg; |
2563 | struct kmem_cache *cachep; | |
2564 | struct work_struct work; | |
2565 | }; | |
2566 | ||
d5b3cf71 | 2567 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2568 | { |
d5b3cf71 VD |
2569 | struct memcg_kmem_cache_create_work *cw = |
2570 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2571 | struct mem_cgroup *memcg = cw->memcg; |
2572 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2573 | |
d5b3cf71 | 2574 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2575 | |
5722d094 | 2576 | css_put(&memcg->css); |
d7f25f8a GC |
2577 | kfree(cw); |
2578 | } | |
2579 | ||
2580 | /* | |
2581 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2582 | */ |
d5b3cf71 VD |
2583 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2584 | struct kmem_cache *cachep) | |
d7f25f8a | 2585 | { |
d5b3cf71 | 2586 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2587 | |
776ed0f0 | 2588 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2589 | if (!cw) |
d7f25f8a | 2590 | return; |
8135be5a VD |
2591 | |
2592 | css_get(&memcg->css); | |
d7f25f8a GC |
2593 | |
2594 | cw->memcg = memcg; | |
2595 | cw->cachep = cachep; | |
d5b3cf71 | 2596 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2597 | |
d7f25f8a GC |
2598 | schedule_work(&cw->work); |
2599 | } | |
2600 | ||
d5b3cf71 VD |
2601 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2602 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2603 | { |
2604 | /* | |
2605 | * We need to stop accounting when we kmalloc, because if the | |
2606 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2607 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2608 | * |
2609 | * However, it is better to enclose the whole function. Depending on | |
2610 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2611 | * trigger an allocation. This too, will make us recurse. Because at | |
2612 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2613 | * the safest choice is to do it like this, wrapping the whole function. | |
2614 | */ | |
6f185c29 | 2615 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2616 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2617 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2618 | } |
c67a8a68 | 2619 | |
d7f25f8a GC |
2620 | /* |
2621 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2622 | * We try to use the current memcg's version of the cache. | |
2623 | * | |
2624 | * If the cache does not exist yet, if we are the first user of it, | |
2625 | * we either create it immediately, if possible, or create it asynchronously | |
2626 | * in a workqueue. | |
2627 | * In the latter case, we will let the current allocation go through with | |
2628 | * the original cache. | |
2629 | * | |
2630 | * Can't be called in interrupt context or from kernel threads. | |
2631 | * This function needs to be called with rcu_read_lock() held. | |
2632 | */ | |
056b7cce | 2633 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2634 | { |
2635 | struct mem_cgroup *memcg; | |
959c8963 | 2636 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2637 | int kmemcg_id; |
d7f25f8a | 2638 | |
f7ce3190 | 2639 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2640 | |
9d100c5e | 2641 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2642 | return cachep; |
2643 | ||
8135be5a | 2644 | memcg = get_mem_cgroup_from_mm(current->mm); |
2a4db7eb VD |
2645 | kmemcg_id = ACCESS_ONCE(memcg->kmemcg_id); |
2646 | if (kmemcg_id < 0) | |
ca0dde97 | 2647 | goto out; |
d7f25f8a | 2648 | |
2a4db7eb | 2649 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2650 | if (likely(memcg_cachep)) |
2651 | return memcg_cachep; | |
ca0dde97 LZ |
2652 | |
2653 | /* | |
2654 | * If we are in a safe context (can wait, and not in interrupt | |
2655 | * context), we could be be predictable and return right away. | |
2656 | * This would guarantee that the allocation being performed | |
2657 | * already belongs in the new cache. | |
2658 | * | |
2659 | * However, there are some clashes that can arrive from locking. | |
2660 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2661 | * memcg_create_kmem_cache, this means no further allocation |
2662 | * could happen with the slab_mutex held. So it's better to | |
2663 | * defer everything. | |
ca0dde97 | 2664 | */ |
d5b3cf71 | 2665 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2666 | out: |
8135be5a | 2667 | css_put(&memcg->css); |
ca0dde97 | 2668 | return cachep; |
d7f25f8a | 2669 | } |
d7f25f8a | 2670 | |
8135be5a VD |
2671 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2672 | { | |
2673 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2674 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2675 | } |
2676 | ||
7ae1e1d0 GC |
2677 | /* |
2678 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2679 | * possible for the given order. But the page is not allocated yet, so we'll | |
2680 | * need a further commit step to do the final arrangements. | |
2681 | * | |
2682 | * It is possible for the task to switch cgroups in this mean time, so at | |
2683 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2684 | * the handle argument to return to the caller which cgroup we should commit | |
2685 | * against. We could also return the memcg directly and avoid the pointer | |
2686 | * passing, but a boolean return value gives better semantics considering | |
2687 | * the compiled-out case as well. | |
2688 | * | |
2689 | * Returning true means the allocation is possible. | |
2690 | */ | |
2691 | bool | |
2692 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2693 | { | |
2694 | struct mem_cgroup *memcg; | |
2695 | int ret; | |
2696 | ||
2697 | *_memcg = NULL; | |
6d42c232 | 2698 | |
df381975 | 2699 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 2700 | |
cf2b8fbf | 2701 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
2702 | css_put(&memcg->css); |
2703 | return true; | |
2704 | } | |
2705 | ||
3e32cb2e | 2706 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
2707 | if (!ret) |
2708 | *_memcg = memcg; | |
7ae1e1d0 GC |
2709 | |
2710 | css_put(&memcg->css); | |
2711 | return (ret == 0); | |
2712 | } | |
2713 | ||
2714 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
2715 | int order) | |
2716 | { | |
7ae1e1d0 GC |
2717 | VM_BUG_ON(mem_cgroup_is_root(memcg)); |
2718 | ||
2719 | /* The page allocation failed. Revert */ | |
2720 | if (!page) { | |
3e32cb2e | 2721 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
2722 | return; |
2723 | } | |
1306a85a | 2724 | page->mem_cgroup = memcg; |
7ae1e1d0 GC |
2725 | } |
2726 | ||
2727 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
2728 | { | |
1306a85a | 2729 | struct mem_cgroup *memcg = page->mem_cgroup; |
7ae1e1d0 | 2730 | |
7ae1e1d0 GC |
2731 | if (!memcg) |
2732 | return; | |
2733 | ||
309381fe | 2734 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2735 | |
3e32cb2e | 2736 | memcg_uncharge_kmem(memcg, 1 << order); |
1306a85a | 2737 | page->mem_cgroup = NULL; |
7ae1e1d0 | 2738 | } |
60d3fd32 VD |
2739 | |
2740 | struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr) | |
2741 | { | |
2742 | struct mem_cgroup *memcg = NULL; | |
2743 | struct kmem_cache *cachep; | |
2744 | struct page *page; | |
2745 | ||
2746 | page = virt_to_head_page(ptr); | |
2747 | if (PageSlab(page)) { | |
2748 | cachep = page->slab_cache; | |
2749 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2750 | memcg = cachep->memcg_params.memcg; |
60d3fd32 VD |
2751 | } else |
2752 | /* page allocated by alloc_kmem_pages */ | |
2753 | memcg = page->mem_cgroup; | |
2754 | ||
2755 | return memcg; | |
2756 | } | |
7ae1e1d0 GC |
2757 | #endif /* CONFIG_MEMCG_KMEM */ |
2758 | ||
ca3e0214 KH |
2759 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2760 | ||
ca3e0214 KH |
2761 | /* |
2762 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2763 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2764 | * charge/uncharge will be never happen and move_account() is done under | |
2765 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2766 | */ |
e94c8a9c | 2767 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2768 | { |
e94c8a9c | 2769 | int i; |
ca3e0214 | 2770 | |
3d37c4a9 KH |
2771 | if (mem_cgroup_disabled()) |
2772 | return; | |
b070e65c | 2773 | |
29833315 | 2774 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2775 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2776 | |
1306a85a | 2777 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2778 | HPAGE_PMD_NR); |
ca3e0214 | 2779 | } |
12d27107 | 2780 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2781 | |
f817ed48 | 2782 | /** |
de3638d9 | 2783 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2784 | * @page: the page |
7ec99d62 | 2785 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2786 | * @from: mem_cgroup which the page is moved from. |
2787 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2788 | * | |
2789 | * The caller must confirm following. | |
08e552c6 | 2790 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2791 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2792 | * |
2f3479b1 KH |
2793 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2794 | * from old cgroup. | |
f817ed48 | 2795 | */ |
7ec99d62 JW |
2796 | static int mem_cgroup_move_account(struct page *page, |
2797 | unsigned int nr_pages, | |
7ec99d62 | 2798 | struct mem_cgroup *from, |
2f3479b1 | 2799 | struct mem_cgroup *to) |
f817ed48 | 2800 | { |
de3638d9 JW |
2801 | unsigned long flags; |
2802 | int ret; | |
987eba66 | 2803 | |
f817ed48 | 2804 | VM_BUG_ON(from == to); |
309381fe | 2805 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
2806 | /* |
2807 | * The page is isolated from LRU. So, collapse function | |
2808 | * will not handle this page. But page splitting can happen. | |
2809 | * Do this check under compound_page_lock(). The caller should | |
2810 | * hold it. | |
2811 | */ | |
2812 | ret = -EBUSY; | |
7ec99d62 | 2813 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2814 | goto out; |
2815 | ||
0a31bc97 | 2816 | /* |
1306a85a | 2817 | * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup |
0a31bc97 JW |
2818 | * of its source page while we change it: page migration takes |
2819 | * both pages off the LRU, but page cache replacement doesn't. | |
2820 | */ | |
2821 | if (!trylock_page(page)) | |
2822 | goto out; | |
de3638d9 JW |
2823 | |
2824 | ret = -EINVAL; | |
1306a85a | 2825 | if (page->mem_cgroup != from) |
0a31bc97 | 2826 | goto out_unlock; |
de3638d9 | 2827 | |
354a4783 | 2828 | spin_lock_irqsave(&from->move_lock, flags); |
f817ed48 | 2829 | |
0a31bc97 | 2830 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
2831 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
2832 | nr_pages); | |
2833 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
2834 | nr_pages); | |
2835 | } | |
3ea67d06 | 2836 | |
59d1d256 JW |
2837 | if (PageWriteback(page)) { |
2838 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
2839 | nr_pages); | |
2840 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
2841 | nr_pages); | |
2842 | } | |
3ea67d06 | 2843 | |
0a31bc97 | 2844 | /* |
1306a85a | 2845 | * It is safe to change page->mem_cgroup here because the page |
0a31bc97 JW |
2846 | * is referenced, charged, and isolated - we can't race with |
2847 | * uncharging, charging, migration, or LRU putback. | |
2848 | */ | |
d69b042f | 2849 | |
854ffa8d | 2850 | /* caller should have done css_get */ |
1306a85a | 2851 | page->mem_cgroup = to; |
354a4783 JW |
2852 | spin_unlock_irqrestore(&from->move_lock, flags); |
2853 | ||
de3638d9 | 2854 | ret = 0; |
0a31bc97 JW |
2855 | |
2856 | local_irq_disable(); | |
2857 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 2858 | memcg_check_events(to, page); |
0a31bc97 | 2859 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 2860 | memcg_check_events(from, page); |
0a31bc97 JW |
2861 | local_irq_enable(); |
2862 | out_unlock: | |
2863 | unlock_page(page); | |
de3638d9 | 2864 | out: |
f817ed48 KH |
2865 | return ret; |
2866 | } | |
2867 | ||
c255a458 | 2868 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2869 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2870 | bool charge) | |
d13d1443 | 2871 | { |
0a31bc97 JW |
2872 | int val = (charge) ? 1 : -1; |
2873 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2874 | } |
02491447 DN |
2875 | |
2876 | /** | |
2877 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2878 | * @entry: swap entry to be moved | |
2879 | * @from: mem_cgroup which the entry is moved from | |
2880 | * @to: mem_cgroup which the entry is moved to | |
2881 | * | |
2882 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2883 | * as the mem_cgroup's id of @from. | |
2884 | * | |
2885 | * Returns 0 on success, -EINVAL on failure. | |
2886 | * | |
3e32cb2e | 2887 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2888 | * both res and memsw, and called css_get(). |
2889 | */ | |
2890 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2891 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2892 | { |
2893 | unsigned short old_id, new_id; | |
2894 | ||
34c00c31 LZ |
2895 | old_id = mem_cgroup_id(from); |
2896 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2897 | |
2898 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2899 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2900 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2901 | return 0; |
2902 | } | |
2903 | return -EINVAL; | |
2904 | } | |
2905 | #else | |
2906 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2907 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2908 | { |
2909 | return -EINVAL; | |
2910 | } | |
8c7c6e34 | 2911 | #endif |
d13d1443 | 2912 | |
3e32cb2e | 2913 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2914 | |
d38d2a75 | 2915 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2916 | unsigned long limit) |
628f4235 | 2917 | { |
3e32cb2e JW |
2918 | unsigned long curusage; |
2919 | unsigned long oldusage; | |
2920 | bool enlarge = false; | |
81d39c20 | 2921 | int retry_count; |
3e32cb2e | 2922 | int ret; |
81d39c20 KH |
2923 | |
2924 | /* | |
2925 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2926 | * is depends on callers. We set our retry-count to be function | |
2927 | * of # of children which we should visit in this loop. | |
2928 | */ | |
3e32cb2e JW |
2929 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2930 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2931 | |
3e32cb2e | 2932 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2933 | |
3e32cb2e | 2934 | do { |
628f4235 KH |
2935 | if (signal_pending(current)) { |
2936 | ret = -EINTR; | |
2937 | break; | |
2938 | } | |
3e32cb2e JW |
2939 | |
2940 | mutex_lock(&memcg_limit_mutex); | |
2941 | if (limit > memcg->memsw.limit) { | |
2942 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2943 | ret = -EINVAL; |
628f4235 KH |
2944 | break; |
2945 | } | |
3e32cb2e JW |
2946 | if (limit > memcg->memory.limit) |
2947 | enlarge = true; | |
2948 | ret = page_counter_limit(&memcg->memory, limit); | |
2949 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2950 | |
2951 | if (!ret) | |
2952 | break; | |
2953 | ||
b70a2a21 JW |
2954 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2955 | ||
3e32cb2e | 2956 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2957 | /* Usage is reduced ? */ |
f894ffa8 | 2958 | if (curusage >= oldusage) |
81d39c20 KH |
2959 | retry_count--; |
2960 | else | |
2961 | oldusage = curusage; | |
3e32cb2e JW |
2962 | } while (retry_count); |
2963 | ||
3c11ecf4 KH |
2964 | if (!ret && enlarge) |
2965 | memcg_oom_recover(memcg); | |
14797e23 | 2966 | |
8c7c6e34 KH |
2967 | return ret; |
2968 | } | |
2969 | ||
338c8431 | 2970 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2971 | unsigned long limit) |
8c7c6e34 | 2972 | { |
3e32cb2e JW |
2973 | unsigned long curusage; |
2974 | unsigned long oldusage; | |
2975 | bool enlarge = false; | |
81d39c20 | 2976 | int retry_count; |
3e32cb2e | 2977 | int ret; |
8c7c6e34 | 2978 | |
81d39c20 | 2979 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2980 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2981 | mem_cgroup_count_children(memcg); | |
2982 | ||
2983 | oldusage = page_counter_read(&memcg->memsw); | |
2984 | ||
2985 | do { | |
8c7c6e34 KH |
2986 | if (signal_pending(current)) { |
2987 | ret = -EINTR; | |
2988 | break; | |
2989 | } | |
3e32cb2e JW |
2990 | |
2991 | mutex_lock(&memcg_limit_mutex); | |
2992 | if (limit < memcg->memory.limit) { | |
2993 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2994 | ret = -EINVAL; |
8c7c6e34 KH |
2995 | break; |
2996 | } | |
3e32cb2e JW |
2997 | if (limit > memcg->memsw.limit) |
2998 | enlarge = true; | |
2999 | ret = page_counter_limit(&memcg->memsw, limit); | |
3000 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3001 | |
3002 | if (!ret) | |
3003 | break; | |
3004 | ||
b70a2a21 JW |
3005 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3006 | ||
3e32cb2e | 3007 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3008 | /* Usage is reduced ? */ |
8c7c6e34 | 3009 | if (curusage >= oldusage) |
628f4235 | 3010 | retry_count--; |
81d39c20 KH |
3011 | else |
3012 | oldusage = curusage; | |
3e32cb2e JW |
3013 | } while (retry_count); |
3014 | ||
3c11ecf4 KH |
3015 | if (!ret && enlarge) |
3016 | memcg_oom_recover(memcg); | |
3e32cb2e | 3017 | |
628f4235 KH |
3018 | return ret; |
3019 | } | |
3020 | ||
0608f43d AM |
3021 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3022 | gfp_t gfp_mask, | |
3023 | unsigned long *total_scanned) | |
3024 | { | |
3025 | unsigned long nr_reclaimed = 0; | |
3026 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3027 | unsigned long reclaimed; | |
3028 | int loop = 0; | |
3029 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3030 | unsigned long excess; |
0608f43d AM |
3031 | unsigned long nr_scanned; |
3032 | ||
3033 | if (order > 0) | |
3034 | return 0; | |
3035 | ||
3036 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3037 | /* | |
3038 | * This loop can run a while, specially if mem_cgroup's continuously | |
3039 | * keep exceeding their soft limit and putting the system under | |
3040 | * pressure | |
3041 | */ | |
3042 | do { | |
3043 | if (next_mz) | |
3044 | mz = next_mz; | |
3045 | else | |
3046 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3047 | if (!mz) | |
3048 | break; | |
3049 | ||
3050 | nr_scanned = 0; | |
3051 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3052 | gfp_mask, &nr_scanned); | |
3053 | nr_reclaimed += reclaimed; | |
3054 | *total_scanned += nr_scanned; | |
0a31bc97 | 3055 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3056 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3057 | |
3058 | /* | |
3059 | * If we failed to reclaim anything from this memory cgroup | |
3060 | * it is time to move on to the next cgroup | |
3061 | */ | |
3062 | next_mz = NULL; | |
bc2f2e7f VD |
3063 | if (!reclaimed) |
3064 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3065 | ||
3e32cb2e | 3066 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3067 | /* |
3068 | * One school of thought says that we should not add | |
3069 | * back the node to the tree if reclaim returns 0. | |
3070 | * But our reclaim could return 0, simply because due | |
3071 | * to priority we are exposing a smaller subset of | |
3072 | * memory to reclaim from. Consider this as a longer | |
3073 | * term TODO. | |
3074 | */ | |
3075 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3076 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3077 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3078 | css_put(&mz->memcg->css); |
3079 | loop++; | |
3080 | /* | |
3081 | * Could not reclaim anything and there are no more | |
3082 | * mem cgroups to try or we seem to be looping without | |
3083 | * reclaiming anything. | |
3084 | */ | |
3085 | if (!nr_reclaimed && | |
3086 | (next_mz == NULL || | |
3087 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3088 | break; | |
3089 | } while (!nr_reclaimed); | |
3090 | if (next_mz) | |
3091 | css_put(&next_mz->memcg->css); | |
3092 | return nr_reclaimed; | |
3093 | } | |
3094 | ||
ea280e7b TH |
3095 | /* |
3096 | * Test whether @memcg has children, dead or alive. Note that this | |
3097 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3098 | * returns %true if there are child csses according to the cgroup | |
3099 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3100 | */ | |
b5f99b53 GC |
3101 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3102 | { | |
ea280e7b TH |
3103 | bool ret; |
3104 | ||
696ac172 | 3105 | /* |
ea280e7b TH |
3106 | * The lock does not prevent addition or deletion of children, but |
3107 | * it prevents a new child from being initialized based on this | |
3108 | * parent in css_online(), so it's enough to decide whether | |
3109 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3110 | */ |
ea280e7b TH |
3111 | lockdep_assert_held(&memcg_create_mutex); |
3112 | ||
3113 | rcu_read_lock(); | |
3114 | ret = css_next_child(NULL, &memcg->css); | |
3115 | rcu_read_unlock(); | |
3116 | return ret; | |
b5f99b53 GC |
3117 | } |
3118 | ||
c26251f9 MH |
3119 | /* |
3120 | * Reclaims as many pages from the given memcg as possible and moves | |
3121 | * the rest to the parent. | |
3122 | * | |
3123 | * Caller is responsible for holding css reference for memcg. | |
3124 | */ | |
3125 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3126 | { | |
3127 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3128 | |
c1e862c1 KH |
3129 | /* we call try-to-free pages for make this cgroup empty */ |
3130 | lru_add_drain_all(); | |
f817ed48 | 3131 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3132 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3133 | int progress; |
c1e862c1 | 3134 | |
c26251f9 MH |
3135 | if (signal_pending(current)) |
3136 | return -EINTR; | |
3137 | ||
b70a2a21 JW |
3138 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3139 | GFP_KERNEL, true); | |
c1e862c1 | 3140 | if (!progress) { |
f817ed48 | 3141 | nr_retries--; |
c1e862c1 | 3142 | /* maybe some writeback is necessary */ |
8aa7e847 | 3143 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3144 | } |
f817ed48 KH |
3145 | |
3146 | } | |
ab5196c2 MH |
3147 | |
3148 | return 0; | |
cc847582 KH |
3149 | } |
3150 | ||
6770c64e TH |
3151 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3152 | char *buf, size_t nbytes, | |
3153 | loff_t off) | |
c1e862c1 | 3154 | { |
6770c64e | 3155 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3156 | |
d8423011 MH |
3157 | if (mem_cgroup_is_root(memcg)) |
3158 | return -EINVAL; | |
6770c64e | 3159 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3160 | } |
3161 | ||
182446d0 TH |
3162 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3163 | struct cftype *cft) | |
18f59ea7 | 3164 | { |
182446d0 | 3165 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3166 | } |
3167 | ||
182446d0 TH |
3168 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3169 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3170 | { |
3171 | int retval = 0; | |
182446d0 | 3172 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3173 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3174 | |
0999821b | 3175 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3176 | |
3177 | if (memcg->use_hierarchy == val) | |
3178 | goto out; | |
3179 | ||
18f59ea7 | 3180 | /* |
af901ca1 | 3181 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3182 | * in the child subtrees. If it is unset, then the change can |
3183 | * occur, provided the current cgroup has no children. | |
3184 | * | |
3185 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3186 | * set if there are no children. | |
3187 | */ | |
c0ff4b85 | 3188 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3189 | (val == 1 || val == 0)) { |
ea280e7b | 3190 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3191 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3192 | else |
3193 | retval = -EBUSY; | |
3194 | } else | |
3195 | retval = -EINVAL; | |
567fb435 GC |
3196 | |
3197 | out: | |
0999821b | 3198 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3199 | |
3200 | return retval; | |
3201 | } | |
3202 | ||
3e32cb2e JW |
3203 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3204 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3205 | { |
3206 | struct mem_cgroup *iter; | |
3207 | long val = 0; | |
3208 | ||
3209 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3210 | for_each_mem_cgroup_tree(iter, memcg) | |
3211 | val += mem_cgroup_read_stat(iter, idx); | |
3212 | ||
3213 | if (val < 0) /* race ? */ | |
3214 | val = 0; | |
3215 | return val; | |
3216 | } | |
3217 | ||
3218 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3219 | { | |
3220 | u64 val; | |
3221 | ||
3e32cb2e JW |
3222 | if (mem_cgroup_is_root(memcg)) { |
3223 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3224 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3225 | if (swap) | |
3226 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3227 | } else { | |
ce00a967 | 3228 | if (!swap) |
3e32cb2e | 3229 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3230 | else |
3e32cb2e | 3231 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3232 | } |
ce00a967 JW |
3233 | return val << PAGE_SHIFT; |
3234 | } | |
3235 | ||
3e32cb2e JW |
3236 | enum { |
3237 | RES_USAGE, | |
3238 | RES_LIMIT, | |
3239 | RES_MAX_USAGE, | |
3240 | RES_FAILCNT, | |
3241 | RES_SOFT_LIMIT, | |
3242 | }; | |
ce00a967 | 3243 | |
791badbd | 3244 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3245 | struct cftype *cft) |
8cdea7c0 | 3246 | { |
182446d0 | 3247 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3248 | struct page_counter *counter; |
af36f906 | 3249 | |
3e32cb2e | 3250 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3251 | case _MEM: |
3e32cb2e JW |
3252 | counter = &memcg->memory; |
3253 | break; | |
8c7c6e34 | 3254 | case _MEMSWAP: |
3e32cb2e JW |
3255 | counter = &memcg->memsw; |
3256 | break; | |
510fc4e1 | 3257 | case _KMEM: |
3e32cb2e | 3258 | counter = &memcg->kmem; |
510fc4e1 | 3259 | break; |
8c7c6e34 KH |
3260 | default: |
3261 | BUG(); | |
8c7c6e34 | 3262 | } |
3e32cb2e JW |
3263 | |
3264 | switch (MEMFILE_ATTR(cft->private)) { | |
3265 | case RES_USAGE: | |
3266 | if (counter == &memcg->memory) | |
3267 | return mem_cgroup_usage(memcg, false); | |
3268 | if (counter == &memcg->memsw) | |
3269 | return mem_cgroup_usage(memcg, true); | |
3270 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3271 | case RES_LIMIT: | |
3272 | return (u64)counter->limit * PAGE_SIZE; | |
3273 | case RES_MAX_USAGE: | |
3274 | return (u64)counter->watermark * PAGE_SIZE; | |
3275 | case RES_FAILCNT: | |
3276 | return counter->failcnt; | |
3277 | case RES_SOFT_LIMIT: | |
3278 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3279 | default: | |
3280 | BUG(); | |
3281 | } | |
8cdea7c0 | 3282 | } |
510fc4e1 | 3283 | |
510fc4e1 | 3284 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
3285 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
3286 | unsigned long nr_pages) | |
d6441637 VD |
3287 | { |
3288 | int err = 0; | |
3289 | int memcg_id; | |
3290 | ||
2a4db7eb | 3291 | BUG_ON(memcg->kmemcg_id >= 0); |
2788cf0c | 3292 | BUG_ON(memcg->kmem_acct_activated); |
2a4db7eb | 3293 | BUG_ON(memcg->kmem_acct_active); |
d6441637 | 3294 | |
510fc4e1 GC |
3295 | /* |
3296 | * For simplicity, we won't allow this to be disabled. It also can't | |
3297 | * be changed if the cgroup has children already, or if tasks had | |
3298 | * already joined. | |
3299 | * | |
3300 | * If tasks join before we set the limit, a person looking at | |
3301 | * kmem.usage_in_bytes will have no way to determine when it took | |
3302 | * place, which makes the value quite meaningless. | |
3303 | * | |
3304 | * After it first became limited, changes in the value of the limit are | |
3305 | * of course permitted. | |
510fc4e1 | 3306 | */ |
0999821b | 3307 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
3308 | if (cgroup_has_tasks(memcg->css.cgroup) || |
3309 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
3310 | err = -EBUSY; |
3311 | mutex_unlock(&memcg_create_mutex); | |
3312 | if (err) | |
3313 | goto out; | |
510fc4e1 | 3314 | |
f3bb3043 | 3315 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
3316 | if (memcg_id < 0) { |
3317 | err = memcg_id; | |
3318 | goto out; | |
3319 | } | |
3320 | ||
d6441637 | 3321 | /* |
900a38f0 VD |
3322 | * We couldn't have accounted to this cgroup, because it hasn't got |
3323 | * activated yet, so this should succeed. | |
d6441637 | 3324 | */ |
3e32cb2e | 3325 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
3326 | VM_BUG_ON(err); |
3327 | ||
3328 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
3329 | /* | |
900a38f0 VD |
3330 | * A memory cgroup is considered kmem-active as soon as it gets |
3331 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
3332 | * guarantee no one starts accounting before all call sites are |
3333 | * patched. | |
3334 | */ | |
900a38f0 | 3335 | memcg->kmemcg_id = memcg_id; |
2788cf0c | 3336 | memcg->kmem_acct_activated = true; |
2a4db7eb | 3337 | memcg->kmem_acct_active = true; |
510fc4e1 | 3338 | out: |
d6441637 | 3339 | return err; |
d6441637 VD |
3340 | } |
3341 | ||
d6441637 | 3342 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3343 | unsigned long limit) |
d6441637 VD |
3344 | { |
3345 | int ret; | |
3346 | ||
3e32cb2e | 3347 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 3348 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 3349 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 3350 | else |
3e32cb2e JW |
3351 | ret = page_counter_limit(&memcg->kmem, limit); |
3352 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
3353 | return ret; |
3354 | } | |
3355 | ||
55007d84 | 3356 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 3357 | { |
55007d84 | 3358 | int ret = 0; |
510fc4e1 | 3359 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 3360 | |
d6441637 VD |
3361 | if (!parent) |
3362 | return 0; | |
55007d84 | 3363 | |
8c0145b6 | 3364 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 3365 | /* |
d6441637 VD |
3366 | * If the parent cgroup is not kmem-active now, it cannot be activated |
3367 | * after this point, because it has at least one child already. | |
55007d84 | 3368 | */ |
d6441637 | 3369 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3370 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3371 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3372 | return ret; |
510fc4e1 | 3373 | } |
d6441637 VD |
3374 | #else |
3375 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3376 | unsigned long limit) |
d6441637 VD |
3377 | { |
3378 | return -EINVAL; | |
3379 | } | |
6d043990 | 3380 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3381 | |
628f4235 KH |
3382 | /* |
3383 | * The user of this function is... | |
3384 | * RES_LIMIT. | |
3385 | */ | |
451af504 TH |
3386 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3387 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3388 | { |
451af504 | 3389 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3390 | unsigned long nr_pages; |
628f4235 KH |
3391 | int ret; |
3392 | ||
451af504 | 3393 | buf = strstrip(buf); |
650c5e56 | 3394 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3395 | if (ret) |
3396 | return ret; | |
af36f906 | 3397 | |
3e32cb2e | 3398 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3399 | case RES_LIMIT: |
4b3bde4c BS |
3400 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3401 | ret = -EINVAL; | |
3402 | break; | |
3403 | } | |
3e32cb2e JW |
3404 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3405 | case _MEM: | |
3406 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3407 | break; |
3e32cb2e JW |
3408 | case _MEMSWAP: |
3409 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3410 | break; |
3e32cb2e JW |
3411 | case _KMEM: |
3412 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3413 | break; | |
3414 | } | |
296c81d8 | 3415 | break; |
3e32cb2e JW |
3416 | case RES_SOFT_LIMIT: |
3417 | memcg->soft_limit = nr_pages; | |
3418 | ret = 0; | |
628f4235 KH |
3419 | break; |
3420 | } | |
451af504 | 3421 | return ret ?: nbytes; |
8cdea7c0 BS |
3422 | } |
3423 | ||
6770c64e TH |
3424 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3425 | size_t nbytes, loff_t off) | |
c84872e1 | 3426 | { |
6770c64e | 3427 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3428 | struct page_counter *counter; |
c84872e1 | 3429 | |
3e32cb2e JW |
3430 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3431 | case _MEM: | |
3432 | counter = &memcg->memory; | |
3433 | break; | |
3434 | case _MEMSWAP: | |
3435 | counter = &memcg->memsw; | |
3436 | break; | |
3437 | case _KMEM: | |
3438 | counter = &memcg->kmem; | |
3439 | break; | |
3440 | default: | |
3441 | BUG(); | |
3442 | } | |
af36f906 | 3443 | |
3e32cb2e | 3444 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3445 | case RES_MAX_USAGE: |
3e32cb2e | 3446 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3447 | break; |
3448 | case RES_FAILCNT: | |
3e32cb2e | 3449 | counter->failcnt = 0; |
29f2a4da | 3450 | break; |
3e32cb2e JW |
3451 | default: |
3452 | BUG(); | |
29f2a4da | 3453 | } |
f64c3f54 | 3454 | |
6770c64e | 3455 | return nbytes; |
c84872e1 PE |
3456 | } |
3457 | ||
182446d0 | 3458 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3459 | struct cftype *cft) |
3460 | { | |
182446d0 | 3461 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3462 | } |
3463 | ||
02491447 | 3464 | #ifdef CONFIG_MMU |
182446d0 | 3465 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3466 | struct cftype *cft, u64 val) |
3467 | { | |
182446d0 | 3468 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3469 | |
1dfab5ab | 3470 | if (val & ~MOVE_MASK) |
7dc74be0 | 3471 | return -EINVAL; |
ee5e8472 | 3472 | |
7dc74be0 | 3473 | /* |
ee5e8472 GC |
3474 | * No kind of locking is needed in here, because ->can_attach() will |
3475 | * check this value once in the beginning of the process, and then carry | |
3476 | * on with stale data. This means that changes to this value will only | |
3477 | * affect task migrations starting after the change. | |
7dc74be0 | 3478 | */ |
c0ff4b85 | 3479 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3480 | return 0; |
3481 | } | |
02491447 | 3482 | #else |
182446d0 | 3483 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3484 | struct cftype *cft, u64 val) |
3485 | { | |
3486 | return -ENOSYS; | |
3487 | } | |
3488 | #endif | |
7dc74be0 | 3489 | |
406eb0c9 | 3490 | #ifdef CONFIG_NUMA |
2da8ca82 | 3491 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3492 | { |
25485de6 GT |
3493 | struct numa_stat { |
3494 | const char *name; | |
3495 | unsigned int lru_mask; | |
3496 | }; | |
3497 | ||
3498 | static const struct numa_stat stats[] = { | |
3499 | { "total", LRU_ALL }, | |
3500 | { "file", LRU_ALL_FILE }, | |
3501 | { "anon", LRU_ALL_ANON }, | |
3502 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3503 | }; | |
3504 | const struct numa_stat *stat; | |
406eb0c9 | 3505 | int nid; |
25485de6 | 3506 | unsigned long nr; |
2da8ca82 | 3507 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3508 | |
25485de6 GT |
3509 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3510 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3511 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3512 | for_each_node_state(nid, N_MEMORY) { | |
3513 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3514 | stat->lru_mask); | |
3515 | seq_printf(m, " N%d=%lu", nid, nr); | |
3516 | } | |
3517 | seq_putc(m, '\n'); | |
406eb0c9 | 3518 | } |
406eb0c9 | 3519 | |
071aee13 YH |
3520 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3521 | struct mem_cgroup *iter; | |
3522 | ||
3523 | nr = 0; | |
3524 | for_each_mem_cgroup_tree(iter, memcg) | |
3525 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3526 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3527 | for_each_node_state(nid, N_MEMORY) { | |
3528 | nr = 0; | |
3529 | for_each_mem_cgroup_tree(iter, memcg) | |
3530 | nr += mem_cgroup_node_nr_lru_pages( | |
3531 | iter, nid, stat->lru_mask); | |
3532 | seq_printf(m, " N%d=%lu", nid, nr); | |
3533 | } | |
3534 | seq_putc(m, '\n'); | |
406eb0c9 | 3535 | } |
406eb0c9 | 3536 | |
406eb0c9 YH |
3537 | return 0; |
3538 | } | |
3539 | #endif /* CONFIG_NUMA */ | |
3540 | ||
2da8ca82 | 3541 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3542 | { |
2da8ca82 | 3543 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3544 | unsigned long memory, memsw; |
af7c4b0e JW |
3545 | struct mem_cgroup *mi; |
3546 | unsigned int i; | |
406eb0c9 | 3547 | |
0ca44b14 GT |
3548 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3549 | MEM_CGROUP_STAT_NSTATS); | |
3550 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3551 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3552 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3553 | ||
af7c4b0e | 3554 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3555 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3556 | continue; |
af7c4b0e JW |
3557 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
3558 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 3559 | } |
7b854121 | 3560 | |
af7c4b0e JW |
3561 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3562 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3563 | mem_cgroup_read_events(memcg, i)); | |
3564 | ||
3565 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3566 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3567 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3568 | ||
14067bb3 | 3569 | /* Hierarchical information */ |
3e32cb2e JW |
3570 | memory = memsw = PAGE_COUNTER_MAX; |
3571 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3572 | memory = min(memory, mi->memory.limit); | |
3573 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3574 | } |
3e32cb2e JW |
3575 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3576 | (u64)memory * PAGE_SIZE); | |
3577 | if (do_swap_account) | |
3578 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3579 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3580 | |
af7c4b0e JW |
3581 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
3582 | long long val = 0; | |
3583 | ||
bff6bb83 | 3584 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3585 | continue; |
af7c4b0e JW |
3586 | for_each_mem_cgroup_tree(mi, memcg) |
3587 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
3588 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
3589 | } | |
3590 | ||
3591 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3592 | unsigned long long val = 0; | |
3593 | ||
3594 | for_each_mem_cgroup_tree(mi, memcg) | |
3595 | val += mem_cgroup_read_events(mi, i); | |
3596 | seq_printf(m, "total_%s %llu\n", | |
3597 | mem_cgroup_events_names[i], val); | |
3598 | } | |
3599 | ||
3600 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3601 | unsigned long long val = 0; | |
3602 | ||
3603 | for_each_mem_cgroup_tree(mi, memcg) | |
3604 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3605 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3606 | } |
14067bb3 | 3607 | |
7f016ee8 | 3608 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3609 | { |
3610 | int nid, zid; | |
3611 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3612 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3613 | unsigned long recent_rotated[2] = {0, 0}; |
3614 | unsigned long recent_scanned[2] = {0, 0}; | |
3615 | ||
3616 | for_each_online_node(nid) | |
3617 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3618 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3619 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3620 | |
89abfab1 HD |
3621 | recent_rotated[0] += rstat->recent_rotated[0]; |
3622 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3623 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3624 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3625 | } |
78ccf5b5 JW |
3626 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3627 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3628 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3629 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3630 | } |
3631 | #endif | |
3632 | ||
d2ceb9b7 KH |
3633 | return 0; |
3634 | } | |
3635 | ||
182446d0 TH |
3636 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3637 | struct cftype *cft) | |
a7885eb8 | 3638 | { |
182446d0 | 3639 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3640 | |
1f4c025b | 3641 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3642 | } |
3643 | ||
182446d0 TH |
3644 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3645 | struct cftype *cft, u64 val) | |
a7885eb8 | 3646 | { |
182446d0 | 3647 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3648 | |
3dae7fec | 3649 | if (val > 100) |
a7885eb8 KM |
3650 | return -EINVAL; |
3651 | ||
14208b0e | 3652 | if (css->parent) |
3dae7fec JW |
3653 | memcg->swappiness = val; |
3654 | else | |
3655 | vm_swappiness = val; | |
068b38c1 | 3656 | |
a7885eb8 KM |
3657 | return 0; |
3658 | } | |
3659 | ||
2e72b634 KS |
3660 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3661 | { | |
3662 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3663 | unsigned long usage; |
2e72b634 KS |
3664 | int i; |
3665 | ||
3666 | rcu_read_lock(); | |
3667 | if (!swap) | |
2c488db2 | 3668 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3669 | else |
2c488db2 | 3670 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3671 | |
3672 | if (!t) | |
3673 | goto unlock; | |
3674 | ||
ce00a967 | 3675 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3676 | |
3677 | /* | |
748dad36 | 3678 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3679 | * If it's not true, a threshold was crossed after last |
3680 | * call of __mem_cgroup_threshold(). | |
3681 | */ | |
5407a562 | 3682 | i = t->current_threshold; |
2e72b634 KS |
3683 | |
3684 | /* | |
3685 | * Iterate backward over array of thresholds starting from | |
3686 | * current_threshold and check if a threshold is crossed. | |
3687 | * If none of thresholds below usage is crossed, we read | |
3688 | * only one element of the array here. | |
3689 | */ | |
3690 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3691 | eventfd_signal(t->entries[i].eventfd, 1); | |
3692 | ||
3693 | /* i = current_threshold + 1 */ | |
3694 | i++; | |
3695 | ||
3696 | /* | |
3697 | * Iterate forward over array of thresholds starting from | |
3698 | * current_threshold+1 and check if a threshold is crossed. | |
3699 | * If none of thresholds above usage is crossed, we read | |
3700 | * only one element of the array here. | |
3701 | */ | |
3702 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3703 | eventfd_signal(t->entries[i].eventfd, 1); | |
3704 | ||
3705 | /* Update current_threshold */ | |
5407a562 | 3706 | t->current_threshold = i - 1; |
2e72b634 KS |
3707 | unlock: |
3708 | rcu_read_unlock(); | |
3709 | } | |
3710 | ||
3711 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3712 | { | |
ad4ca5f4 KS |
3713 | while (memcg) { |
3714 | __mem_cgroup_threshold(memcg, false); | |
3715 | if (do_swap_account) | |
3716 | __mem_cgroup_threshold(memcg, true); | |
3717 | ||
3718 | memcg = parent_mem_cgroup(memcg); | |
3719 | } | |
2e72b634 KS |
3720 | } |
3721 | ||
3722 | static int compare_thresholds(const void *a, const void *b) | |
3723 | { | |
3724 | const struct mem_cgroup_threshold *_a = a; | |
3725 | const struct mem_cgroup_threshold *_b = b; | |
3726 | ||
2bff24a3 GT |
3727 | if (_a->threshold > _b->threshold) |
3728 | return 1; | |
3729 | ||
3730 | if (_a->threshold < _b->threshold) | |
3731 | return -1; | |
3732 | ||
3733 | return 0; | |
2e72b634 KS |
3734 | } |
3735 | ||
c0ff4b85 | 3736 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3737 | { |
3738 | struct mem_cgroup_eventfd_list *ev; | |
3739 | ||
2bcf2e92 MH |
3740 | spin_lock(&memcg_oom_lock); |
3741 | ||
c0ff4b85 | 3742 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3743 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3744 | |
3745 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3746 | return 0; |
3747 | } | |
3748 | ||
c0ff4b85 | 3749 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3750 | { |
7d74b06f KH |
3751 | struct mem_cgroup *iter; |
3752 | ||
c0ff4b85 | 3753 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3754 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3755 | } |
3756 | ||
59b6f873 | 3757 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3758 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3759 | { |
2c488db2 KS |
3760 | struct mem_cgroup_thresholds *thresholds; |
3761 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3762 | unsigned long threshold; |
3763 | unsigned long usage; | |
2c488db2 | 3764 | int i, size, ret; |
2e72b634 | 3765 | |
650c5e56 | 3766 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3767 | if (ret) |
3768 | return ret; | |
3769 | ||
3770 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3771 | |
05b84301 | 3772 | if (type == _MEM) { |
2c488db2 | 3773 | thresholds = &memcg->thresholds; |
ce00a967 | 3774 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3775 | } else if (type == _MEMSWAP) { |
2c488db2 | 3776 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3777 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3778 | } else |
2e72b634 KS |
3779 | BUG(); |
3780 | ||
2e72b634 | 3781 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3782 | if (thresholds->primary) |
2e72b634 KS |
3783 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3784 | ||
2c488db2 | 3785 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3786 | |
3787 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3788 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3789 | GFP_KERNEL); |
2c488db2 | 3790 | if (!new) { |
2e72b634 KS |
3791 | ret = -ENOMEM; |
3792 | goto unlock; | |
3793 | } | |
2c488db2 | 3794 | new->size = size; |
2e72b634 KS |
3795 | |
3796 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3797 | if (thresholds->primary) { |
3798 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3799 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3800 | } |
3801 | ||
2e72b634 | 3802 | /* Add new threshold */ |
2c488db2 KS |
3803 | new->entries[size - 1].eventfd = eventfd; |
3804 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3805 | |
3806 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3807 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3808 | compare_thresholds, NULL); |
3809 | ||
3810 | /* Find current threshold */ | |
2c488db2 | 3811 | new->current_threshold = -1; |
2e72b634 | 3812 | for (i = 0; i < size; i++) { |
748dad36 | 3813 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3814 | /* |
2c488db2 KS |
3815 | * new->current_threshold will not be used until |
3816 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3817 | * it here. |
3818 | */ | |
2c488db2 | 3819 | ++new->current_threshold; |
748dad36 SZ |
3820 | } else |
3821 | break; | |
2e72b634 KS |
3822 | } |
3823 | ||
2c488db2 KS |
3824 | /* Free old spare buffer and save old primary buffer as spare */ |
3825 | kfree(thresholds->spare); | |
3826 | thresholds->spare = thresholds->primary; | |
3827 | ||
3828 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3829 | |
907860ed | 3830 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3831 | synchronize_rcu(); |
3832 | ||
2e72b634 KS |
3833 | unlock: |
3834 | mutex_unlock(&memcg->thresholds_lock); | |
3835 | ||
3836 | return ret; | |
3837 | } | |
3838 | ||
59b6f873 | 3839 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3840 | struct eventfd_ctx *eventfd, const char *args) |
3841 | { | |
59b6f873 | 3842 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3843 | } |
3844 | ||
59b6f873 | 3845 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3846 | struct eventfd_ctx *eventfd, const char *args) |
3847 | { | |
59b6f873 | 3848 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3849 | } |
3850 | ||
59b6f873 | 3851 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3852 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3853 | { |
2c488db2 KS |
3854 | struct mem_cgroup_thresholds *thresholds; |
3855 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3856 | unsigned long usage; |
2c488db2 | 3857 | int i, j, size; |
2e72b634 KS |
3858 | |
3859 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3860 | |
3861 | if (type == _MEM) { | |
2c488db2 | 3862 | thresholds = &memcg->thresholds; |
ce00a967 | 3863 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3864 | } else if (type == _MEMSWAP) { |
2c488db2 | 3865 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3866 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3867 | } else |
2e72b634 KS |
3868 | BUG(); |
3869 | ||
371528ca AV |
3870 | if (!thresholds->primary) |
3871 | goto unlock; | |
3872 | ||
2e72b634 KS |
3873 | /* Check if a threshold crossed before removing */ |
3874 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3875 | ||
3876 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3877 | size = 0; |
3878 | for (i = 0; i < thresholds->primary->size; i++) { | |
3879 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3880 | size++; |
3881 | } | |
3882 | ||
2c488db2 | 3883 | new = thresholds->spare; |
907860ed | 3884 | |
2e72b634 KS |
3885 | /* Set thresholds array to NULL if we don't have thresholds */ |
3886 | if (!size) { | |
2c488db2 KS |
3887 | kfree(new); |
3888 | new = NULL; | |
907860ed | 3889 | goto swap_buffers; |
2e72b634 KS |
3890 | } |
3891 | ||
2c488db2 | 3892 | new->size = size; |
2e72b634 KS |
3893 | |
3894 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3895 | new->current_threshold = -1; |
3896 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3897 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3898 | continue; |
3899 | ||
2c488db2 | 3900 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3901 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3902 | /* |
2c488db2 | 3903 | * new->current_threshold will not be used |
2e72b634 KS |
3904 | * until rcu_assign_pointer(), so it's safe to increment |
3905 | * it here. | |
3906 | */ | |
2c488db2 | 3907 | ++new->current_threshold; |
2e72b634 KS |
3908 | } |
3909 | j++; | |
3910 | } | |
3911 | ||
907860ed | 3912 | swap_buffers: |
2c488db2 KS |
3913 | /* Swap primary and spare array */ |
3914 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
3915 | /* If all events are unregistered, free the spare array */ |
3916 | if (!new) { | |
3917 | kfree(thresholds->spare); | |
3918 | thresholds->spare = NULL; | |
3919 | } | |
3920 | ||
2c488db2 | 3921 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3922 | |
907860ed | 3923 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3924 | synchronize_rcu(); |
371528ca | 3925 | unlock: |
2e72b634 | 3926 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3927 | } |
c1e862c1 | 3928 | |
59b6f873 | 3929 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3930 | struct eventfd_ctx *eventfd) |
3931 | { | |
59b6f873 | 3932 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3933 | } |
3934 | ||
59b6f873 | 3935 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3936 | struct eventfd_ctx *eventfd) |
3937 | { | |
59b6f873 | 3938 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3939 | } |
3940 | ||
59b6f873 | 3941 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3942 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3943 | { |
9490ff27 | 3944 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3945 | |
9490ff27 KH |
3946 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3947 | if (!event) | |
3948 | return -ENOMEM; | |
3949 | ||
1af8efe9 | 3950 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3951 | |
3952 | event->eventfd = eventfd; | |
3953 | list_add(&event->list, &memcg->oom_notify); | |
3954 | ||
3955 | /* already in OOM ? */ | |
79dfdacc | 3956 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 3957 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3958 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3959 | |
3960 | return 0; | |
3961 | } | |
3962 | ||
59b6f873 | 3963 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3964 | struct eventfd_ctx *eventfd) |
9490ff27 | 3965 | { |
9490ff27 | 3966 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3967 | |
1af8efe9 | 3968 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3969 | |
c0ff4b85 | 3970 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3971 | if (ev->eventfd == eventfd) { |
3972 | list_del(&ev->list); | |
3973 | kfree(ev); | |
3974 | } | |
3975 | } | |
3976 | ||
1af8efe9 | 3977 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3978 | } |
3979 | ||
2da8ca82 | 3980 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3981 | { |
2da8ca82 | 3982 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3983 | |
791badbd TH |
3984 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
3985 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
3986 | return 0; |
3987 | } | |
3988 | ||
182446d0 | 3989 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3990 | struct cftype *cft, u64 val) |
3991 | { | |
182446d0 | 3992 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3993 | |
3994 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3995 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3996 | return -EINVAL; |
3997 | ||
c0ff4b85 | 3998 | memcg->oom_kill_disable = val; |
4d845ebf | 3999 | if (!val) |
c0ff4b85 | 4000 | memcg_oom_recover(memcg); |
3dae7fec | 4001 | |
3c11ecf4 KH |
4002 | return 0; |
4003 | } | |
4004 | ||
c255a458 | 4005 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4006 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4007 | { |
55007d84 GC |
4008 | int ret; |
4009 | ||
55007d84 GC |
4010 | ret = memcg_propagate_kmem(memcg); |
4011 | if (ret) | |
4012 | return ret; | |
2633d7a0 | 4013 | |
1d62e436 | 4014 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4015 | } |
e5671dfa | 4016 | |
2a4db7eb VD |
4017 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
4018 | { | |
2788cf0c VD |
4019 | struct cgroup_subsys_state *css; |
4020 | struct mem_cgroup *parent, *child; | |
4021 | int kmemcg_id; | |
4022 | ||
2a4db7eb VD |
4023 | if (!memcg->kmem_acct_active) |
4024 | return; | |
4025 | ||
4026 | /* | |
4027 | * Clear the 'active' flag before clearing memcg_caches arrays entries. | |
4028 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it | |
4029 | * guarantees no cache will be created for this cgroup after we are | |
4030 | * done (see memcg_create_kmem_cache()). | |
4031 | */ | |
4032 | memcg->kmem_acct_active = false; | |
4033 | ||
4034 | memcg_deactivate_kmem_caches(memcg); | |
2788cf0c VD |
4035 | |
4036 | kmemcg_id = memcg->kmemcg_id; | |
4037 | BUG_ON(kmemcg_id < 0); | |
4038 | ||
4039 | parent = parent_mem_cgroup(memcg); | |
4040 | if (!parent) | |
4041 | parent = root_mem_cgroup; | |
4042 | ||
4043 | /* | |
4044 | * Change kmemcg_id of this cgroup and all its descendants to the | |
4045 | * parent's id, and then move all entries from this cgroup's list_lrus | |
4046 | * to ones of the parent. After we have finished, all list_lrus | |
4047 | * corresponding to this cgroup are guaranteed to remain empty. The | |
4048 | * ordering is imposed by list_lru_node->lock taken by | |
4049 | * memcg_drain_all_list_lrus(). | |
4050 | */ | |
4051 | css_for_each_descendant_pre(css, &memcg->css) { | |
4052 | child = mem_cgroup_from_css(css); | |
4053 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
4054 | child->kmemcg_id = parent->kmemcg_id; | |
4055 | if (!memcg->use_hierarchy) | |
4056 | break; | |
4057 | } | |
4058 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
4059 | ||
4060 | memcg_free_cache_id(kmemcg_id); | |
2a4db7eb VD |
4061 | } |
4062 | ||
10d5ebf4 | 4063 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4064 | { |
f48b80a5 VD |
4065 | if (memcg->kmem_acct_activated) { |
4066 | memcg_destroy_kmem_caches(memcg); | |
4067 | static_key_slow_dec(&memcg_kmem_enabled_key); | |
4068 | WARN_ON(page_counter_read(&memcg->kmem)); | |
4069 | } | |
1d62e436 | 4070 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 4071 | } |
e5671dfa | 4072 | #else |
cbe128e3 | 4073 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4074 | { |
4075 | return 0; | |
4076 | } | |
d1a4c0b3 | 4077 | |
2a4db7eb VD |
4078 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
4079 | { | |
4080 | } | |
4081 | ||
10d5ebf4 LZ |
4082 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4083 | { | |
4084 | } | |
e5671dfa GC |
4085 | #endif |
4086 | ||
3bc942f3 TH |
4087 | /* |
4088 | * DO NOT USE IN NEW FILES. | |
4089 | * | |
4090 | * "cgroup.event_control" implementation. | |
4091 | * | |
4092 | * This is way over-engineered. It tries to support fully configurable | |
4093 | * events for each user. Such level of flexibility is completely | |
4094 | * unnecessary especially in the light of the planned unified hierarchy. | |
4095 | * | |
4096 | * Please deprecate this and replace with something simpler if at all | |
4097 | * possible. | |
4098 | */ | |
4099 | ||
79bd9814 TH |
4100 | /* |
4101 | * Unregister event and free resources. | |
4102 | * | |
4103 | * Gets called from workqueue. | |
4104 | */ | |
3bc942f3 | 4105 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4106 | { |
3bc942f3 TH |
4107 | struct mem_cgroup_event *event = |
4108 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4109 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4110 | |
4111 | remove_wait_queue(event->wqh, &event->wait); | |
4112 | ||
59b6f873 | 4113 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4114 | |
4115 | /* Notify userspace the event is going away. */ | |
4116 | eventfd_signal(event->eventfd, 1); | |
4117 | ||
4118 | eventfd_ctx_put(event->eventfd); | |
4119 | kfree(event); | |
59b6f873 | 4120 | css_put(&memcg->css); |
79bd9814 TH |
4121 | } |
4122 | ||
4123 | /* | |
4124 | * Gets called on POLLHUP on eventfd when user closes it. | |
4125 | * | |
4126 | * Called with wqh->lock held and interrupts disabled. | |
4127 | */ | |
3bc942f3 TH |
4128 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4129 | int sync, void *key) | |
79bd9814 | 4130 | { |
3bc942f3 TH |
4131 | struct mem_cgroup_event *event = |
4132 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4133 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4134 | unsigned long flags = (unsigned long)key; |
4135 | ||
4136 | if (flags & POLLHUP) { | |
4137 | /* | |
4138 | * If the event has been detached at cgroup removal, we | |
4139 | * can simply return knowing the other side will cleanup | |
4140 | * for us. | |
4141 | * | |
4142 | * We can't race against event freeing since the other | |
4143 | * side will require wqh->lock via remove_wait_queue(), | |
4144 | * which we hold. | |
4145 | */ | |
fba94807 | 4146 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4147 | if (!list_empty(&event->list)) { |
4148 | list_del_init(&event->list); | |
4149 | /* | |
4150 | * We are in atomic context, but cgroup_event_remove() | |
4151 | * may sleep, so we have to call it in workqueue. | |
4152 | */ | |
4153 | schedule_work(&event->remove); | |
4154 | } | |
fba94807 | 4155 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4156 | } |
4157 | ||
4158 | return 0; | |
4159 | } | |
4160 | ||
3bc942f3 | 4161 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4162 | wait_queue_head_t *wqh, poll_table *pt) |
4163 | { | |
3bc942f3 TH |
4164 | struct mem_cgroup_event *event = |
4165 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4166 | |
4167 | event->wqh = wqh; | |
4168 | add_wait_queue(wqh, &event->wait); | |
4169 | } | |
4170 | ||
4171 | /* | |
3bc942f3 TH |
4172 | * DO NOT USE IN NEW FILES. |
4173 | * | |
79bd9814 TH |
4174 | * Parse input and register new cgroup event handler. |
4175 | * | |
4176 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4177 | * Interpretation of args is defined by control file implementation. | |
4178 | */ | |
451af504 TH |
4179 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4180 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4181 | { |
451af504 | 4182 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4183 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4184 | struct mem_cgroup_event *event; |
79bd9814 TH |
4185 | struct cgroup_subsys_state *cfile_css; |
4186 | unsigned int efd, cfd; | |
4187 | struct fd efile; | |
4188 | struct fd cfile; | |
fba94807 | 4189 | const char *name; |
79bd9814 TH |
4190 | char *endp; |
4191 | int ret; | |
4192 | ||
451af504 TH |
4193 | buf = strstrip(buf); |
4194 | ||
4195 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4196 | if (*endp != ' ') |
4197 | return -EINVAL; | |
451af504 | 4198 | buf = endp + 1; |
79bd9814 | 4199 | |
451af504 | 4200 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4201 | if ((*endp != ' ') && (*endp != '\0')) |
4202 | return -EINVAL; | |
451af504 | 4203 | buf = endp + 1; |
79bd9814 TH |
4204 | |
4205 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4206 | if (!event) | |
4207 | return -ENOMEM; | |
4208 | ||
59b6f873 | 4209 | event->memcg = memcg; |
79bd9814 | 4210 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4211 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4212 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4213 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4214 | |
4215 | efile = fdget(efd); | |
4216 | if (!efile.file) { | |
4217 | ret = -EBADF; | |
4218 | goto out_kfree; | |
4219 | } | |
4220 | ||
4221 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4222 | if (IS_ERR(event->eventfd)) { | |
4223 | ret = PTR_ERR(event->eventfd); | |
4224 | goto out_put_efile; | |
4225 | } | |
4226 | ||
4227 | cfile = fdget(cfd); | |
4228 | if (!cfile.file) { | |
4229 | ret = -EBADF; | |
4230 | goto out_put_eventfd; | |
4231 | } | |
4232 | ||
4233 | /* the process need read permission on control file */ | |
4234 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4235 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4236 | if (ret < 0) | |
4237 | goto out_put_cfile; | |
4238 | ||
fba94807 TH |
4239 | /* |
4240 | * Determine the event callbacks and set them in @event. This used | |
4241 | * to be done via struct cftype but cgroup core no longer knows | |
4242 | * about these events. The following is crude but the whole thing | |
4243 | * is for compatibility anyway. | |
3bc942f3 TH |
4244 | * |
4245 | * DO NOT ADD NEW FILES. | |
fba94807 | 4246 | */ |
b583043e | 4247 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4248 | |
4249 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4250 | event->register_event = mem_cgroup_usage_register_event; | |
4251 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4252 | } else if (!strcmp(name, "memory.oom_control")) { | |
4253 | event->register_event = mem_cgroup_oom_register_event; | |
4254 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4255 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4256 | event->register_event = vmpressure_register_event; | |
4257 | event->unregister_event = vmpressure_unregister_event; | |
4258 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4259 | event->register_event = memsw_cgroup_usage_register_event; |
4260 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4261 | } else { |
4262 | ret = -EINVAL; | |
4263 | goto out_put_cfile; | |
4264 | } | |
4265 | ||
79bd9814 | 4266 | /* |
b5557c4c TH |
4267 | * Verify @cfile should belong to @css. Also, remaining events are |
4268 | * automatically removed on cgroup destruction but the removal is | |
4269 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4270 | */ |
b583043e | 4271 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4272 | &memory_cgrp_subsys); |
79bd9814 | 4273 | ret = -EINVAL; |
5a17f543 | 4274 | if (IS_ERR(cfile_css)) |
79bd9814 | 4275 | goto out_put_cfile; |
5a17f543 TH |
4276 | if (cfile_css != css) { |
4277 | css_put(cfile_css); | |
79bd9814 | 4278 | goto out_put_cfile; |
5a17f543 | 4279 | } |
79bd9814 | 4280 | |
451af504 | 4281 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4282 | if (ret) |
4283 | goto out_put_css; | |
4284 | ||
4285 | efile.file->f_op->poll(efile.file, &event->pt); | |
4286 | ||
fba94807 TH |
4287 | spin_lock(&memcg->event_list_lock); |
4288 | list_add(&event->list, &memcg->event_list); | |
4289 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4290 | |
4291 | fdput(cfile); | |
4292 | fdput(efile); | |
4293 | ||
451af504 | 4294 | return nbytes; |
79bd9814 TH |
4295 | |
4296 | out_put_css: | |
b5557c4c | 4297 | css_put(css); |
79bd9814 TH |
4298 | out_put_cfile: |
4299 | fdput(cfile); | |
4300 | out_put_eventfd: | |
4301 | eventfd_ctx_put(event->eventfd); | |
4302 | out_put_efile: | |
4303 | fdput(efile); | |
4304 | out_kfree: | |
4305 | kfree(event); | |
4306 | ||
4307 | return ret; | |
4308 | } | |
4309 | ||
241994ed | 4310 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4311 | { |
0eea1030 | 4312 | .name = "usage_in_bytes", |
8c7c6e34 | 4313 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4314 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4315 | }, |
c84872e1 PE |
4316 | { |
4317 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4318 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4319 | .write = mem_cgroup_reset, |
791badbd | 4320 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4321 | }, |
8cdea7c0 | 4322 | { |
0eea1030 | 4323 | .name = "limit_in_bytes", |
8c7c6e34 | 4324 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4325 | .write = mem_cgroup_write, |
791badbd | 4326 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4327 | }, |
296c81d8 BS |
4328 | { |
4329 | .name = "soft_limit_in_bytes", | |
4330 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4331 | .write = mem_cgroup_write, |
791badbd | 4332 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4333 | }, |
8cdea7c0 BS |
4334 | { |
4335 | .name = "failcnt", | |
8c7c6e34 | 4336 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4337 | .write = mem_cgroup_reset, |
791badbd | 4338 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4339 | }, |
d2ceb9b7 KH |
4340 | { |
4341 | .name = "stat", | |
2da8ca82 | 4342 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4343 | }, |
c1e862c1 KH |
4344 | { |
4345 | .name = "force_empty", | |
6770c64e | 4346 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4347 | }, |
18f59ea7 BS |
4348 | { |
4349 | .name = "use_hierarchy", | |
4350 | .write_u64 = mem_cgroup_hierarchy_write, | |
4351 | .read_u64 = mem_cgroup_hierarchy_read, | |
4352 | }, | |
79bd9814 | 4353 | { |
3bc942f3 | 4354 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4355 | .write = memcg_write_event_control, |
79bd9814 TH |
4356 | .flags = CFTYPE_NO_PREFIX, |
4357 | .mode = S_IWUGO, | |
4358 | }, | |
a7885eb8 KM |
4359 | { |
4360 | .name = "swappiness", | |
4361 | .read_u64 = mem_cgroup_swappiness_read, | |
4362 | .write_u64 = mem_cgroup_swappiness_write, | |
4363 | }, | |
7dc74be0 DN |
4364 | { |
4365 | .name = "move_charge_at_immigrate", | |
4366 | .read_u64 = mem_cgroup_move_charge_read, | |
4367 | .write_u64 = mem_cgroup_move_charge_write, | |
4368 | }, | |
9490ff27 KH |
4369 | { |
4370 | .name = "oom_control", | |
2da8ca82 | 4371 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4372 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4373 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4374 | }, | |
70ddf637 AV |
4375 | { |
4376 | .name = "pressure_level", | |
70ddf637 | 4377 | }, |
406eb0c9 YH |
4378 | #ifdef CONFIG_NUMA |
4379 | { | |
4380 | .name = "numa_stat", | |
2da8ca82 | 4381 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4382 | }, |
4383 | #endif | |
510fc4e1 GC |
4384 | #ifdef CONFIG_MEMCG_KMEM |
4385 | { | |
4386 | .name = "kmem.limit_in_bytes", | |
4387 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4388 | .write = mem_cgroup_write, |
791badbd | 4389 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4390 | }, |
4391 | { | |
4392 | .name = "kmem.usage_in_bytes", | |
4393 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4394 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4395 | }, |
4396 | { | |
4397 | .name = "kmem.failcnt", | |
4398 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4399 | .write = mem_cgroup_reset, |
791badbd | 4400 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4401 | }, |
4402 | { | |
4403 | .name = "kmem.max_usage_in_bytes", | |
4404 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4405 | .write = mem_cgroup_reset, |
791badbd | 4406 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4407 | }, |
749c5415 GC |
4408 | #ifdef CONFIG_SLABINFO |
4409 | { | |
4410 | .name = "kmem.slabinfo", | |
b047501c VD |
4411 | .seq_start = slab_start, |
4412 | .seq_next = slab_next, | |
4413 | .seq_stop = slab_stop, | |
4414 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4415 | }, |
4416 | #endif | |
8c7c6e34 | 4417 | #endif |
6bc10349 | 4418 | { }, /* terminate */ |
af36f906 | 4419 | }; |
8c7c6e34 | 4420 | |
c0ff4b85 | 4421 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4422 | { |
4423 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4424 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4425 | int zone, tmp = node; |
1ecaab2b KH |
4426 | /* |
4427 | * This routine is called against possible nodes. | |
4428 | * But it's BUG to call kmalloc() against offline node. | |
4429 | * | |
4430 | * TODO: this routine can waste much memory for nodes which will | |
4431 | * never be onlined. It's better to use memory hotplug callback | |
4432 | * function. | |
4433 | */ | |
41e3355d KH |
4434 | if (!node_state(node, N_NORMAL_MEMORY)) |
4435 | tmp = -1; | |
17295c88 | 4436 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4437 | if (!pn) |
4438 | return 1; | |
1ecaab2b | 4439 | |
1ecaab2b KH |
4440 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4441 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4442 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4443 | mz->usage_in_excess = 0; |
4444 | mz->on_tree = false; | |
d79154bb | 4445 | mz->memcg = memcg; |
1ecaab2b | 4446 | } |
54f72fe0 | 4447 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4448 | return 0; |
4449 | } | |
4450 | ||
c0ff4b85 | 4451 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4452 | { |
54f72fe0 | 4453 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4454 | } |
4455 | ||
33327948 KH |
4456 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4457 | { | |
d79154bb | 4458 | struct mem_cgroup *memcg; |
8ff69e2c | 4459 | size_t size; |
33327948 | 4460 | |
8ff69e2c VD |
4461 | size = sizeof(struct mem_cgroup); |
4462 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4463 | |
8ff69e2c | 4464 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4465 | if (!memcg) |
e7bbcdf3 DC |
4466 | return NULL; |
4467 | ||
d79154bb HD |
4468 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4469 | if (!memcg->stat) | |
d2e61b8d | 4470 | goto out_free; |
d79154bb HD |
4471 | spin_lock_init(&memcg->pcp_counter_lock); |
4472 | return memcg; | |
d2e61b8d DC |
4473 | |
4474 | out_free: | |
8ff69e2c | 4475 | kfree(memcg); |
d2e61b8d | 4476 | return NULL; |
33327948 KH |
4477 | } |
4478 | ||
59927fb9 | 4479 | /* |
c8b2a36f GC |
4480 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4481 | * (scanning all at force_empty is too costly...) | |
4482 | * | |
4483 | * Instead of clearing all references at force_empty, we remember | |
4484 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4485 | * it goes down to 0. | |
4486 | * | |
4487 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4488 | */ |
c8b2a36f GC |
4489 | |
4490 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4491 | { |
c8b2a36f | 4492 | int node; |
59927fb9 | 4493 | |
bb4cc1a8 | 4494 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4495 | |
4496 | for_each_node(node) | |
4497 | free_mem_cgroup_per_zone_info(memcg, node); | |
4498 | ||
4499 | free_percpu(memcg->stat); | |
8ff69e2c | 4500 | kfree(memcg); |
59927fb9 | 4501 | } |
3afe36b1 | 4502 | |
7bcc1bb1 DN |
4503 | /* |
4504 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4505 | */ | |
e1aab161 | 4506 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4507 | { |
3e32cb2e | 4508 | if (!memcg->memory.parent) |
7bcc1bb1 | 4509 | return NULL; |
3e32cb2e | 4510 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4511 | } |
e1aab161 | 4512 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4513 | |
0eb253e2 | 4514 | static struct cgroup_subsys_state * __ref |
eb95419b | 4515 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4516 | { |
d142e3e6 | 4517 | struct mem_cgroup *memcg; |
04046e1a | 4518 | long error = -ENOMEM; |
6d12e2d8 | 4519 | int node; |
8cdea7c0 | 4520 | |
c0ff4b85 R |
4521 | memcg = mem_cgroup_alloc(); |
4522 | if (!memcg) | |
04046e1a | 4523 | return ERR_PTR(error); |
78fb7466 | 4524 | |
3ed28fa1 | 4525 | for_each_node(node) |
c0ff4b85 | 4526 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4527 | goto free_out; |
f64c3f54 | 4528 | |
c077719b | 4529 | /* root ? */ |
eb95419b | 4530 | if (parent_css == NULL) { |
a41c58a6 | 4531 | root_mem_cgroup = memcg; |
3e32cb2e | 4532 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4533 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4534 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4535 | page_counter_init(&memcg->memsw, NULL); |
4536 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4537 | } |
28dbc4b6 | 4538 | |
d142e3e6 GC |
4539 | memcg->last_scanned_node = MAX_NUMNODES; |
4540 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4541 | memcg->move_charge_at_immigrate = 0; |
4542 | mutex_init(&memcg->thresholds_lock); | |
4543 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4544 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4545 | INIT_LIST_HEAD(&memcg->event_list); |
4546 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4547 | #ifdef CONFIG_MEMCG_KMEM |
4548 | memcg->kmemcg_id = -1; | |
900a38f0 | 4549 | #endif |
d142e3e6 GC |
4550 | |
4551 | return &memcg->css; | |
4552 | ||
4553 | free_out: | |
4554 | __mem_cgroup_free(memcg); | |
4555 | return ERR_PTR(error); | |
4556 | } | |
4557 | ||
4558 | static int | |
eb95419b | 4559 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4560 | { |
eb95419b | 4561 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4562 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4563 | int ret; |
d142e3e6 | 4564 | |
15a4c835 | 4565 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4566 | return -ENOSPC; |
4567 | ||
63876986 | 4568 | if (!parent) |
d142e3e6 GC |
4569 | return 0; |
4570 | ||
0999821b | 4571 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4572 | |
4573 | memcg->use_hierarchy = parent->use_hierarchy; | |
4574 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4575 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4576 | ||
4577 | if (parent->use_hierarchy) { | |
3e32cb2e | 4578 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4579 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4580 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4581 | page_counter_init(&memcg->memsw, &parent->memsw); |
4582 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4583 | |
7bcc1bb1 | 4584 | /* |
8d76a979 LZ |
4585 | * No need to take a reference to the parent because cgroup |
4586 | * core guarantees its existence. | |
7bcc1bb1 | 4587 | */ |
18f59ea7 | 4588 | } else { |
3e32cb2e | 4589 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4590 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4591 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4592 | page_counter_init(&memcg->memsw, NULL); |
4593 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4594 | /* |
4595 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4596 | * much sense so let cgroup subsystem know about this | |
4597 | * unfortunate state in our controller. | |
4598 | */ | |
d142e3e6 | 4599 | if (parent != root_mem_cgroup) |
073219e9 | 4600 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4601 | } |
0999821b | 4602 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4603 | |
2f7dd7a4 JW |
4604 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4605 | if (ret) | |
4606 | return ret; | |
4607 | ||
4608 | /* | |
4609 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4610 | * orders reading memcg->initialized against its callers | |
4611 | * reading the memcg members. | |
4612 | */ | |
4613 | smp_store_release(&memcg->initialized, 1); | |
4614 | ||
4615 | return 0; | |
8cdea7c0 BS |
4616 | } |
4617 | ||
eb95419b | 4618 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4619 | { |
eb95419b | 4620 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4621 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4622 | |
4623 | /* | |
4624 | * Unregister events and notify userspace. | |
4625 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4626 | * directory to avoid race between userspace and kernelspace. | |
4627 | */ | |
fba94807 TH |
4628 | spin_lock(&memcg->event_list_lock); |
4629 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4630 | list_del_init(&event->list); |
4631 | schedule_work(&event->remove); | |
4632 | } | |
fba94807 | 4633 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4634 | |
33cb876e | 4635 | vmpressure_cleanup(&memcg->vmpressure); |
2a4db7eb VD |
4636 | |
4637 | memcg_deactivate_kmem(memcg); | |
df878fb0 KH |
4638 | } |
4639 | ||
eb95419b | 4640 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4641 | { |
eb95419b | 4642 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4643 | |
10d5ebf4 | 4644 | memcg_destroy_kmem(memcg); |
465939a1 | 4645 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4646 | } |
4647 | ||
1ced953b TH |
4648 | /** |
4649 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4650 | * @css: the target css | |
4651 | * | |
4652 | * Reset the states of the mem_cgroup associated with @css. This is | |
4653 | * invoked when the userland requests disabling on the default hierarchy | |
4654 | * but the memcg is pinned through dependency. The memcg should stop | |
4655 | * applying policies and should revert to the vanilla state as it may be | |
4656 | * made visible again. | |
4657 | * | |
4658 | * The current implementation only resets the essential configurations. | |
4659 | * This needs to be expanded to cover all the visible parts. | |
4660 | */ | |
4661 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4662 | { | |
4663 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4664 | ||
3e32cb2e JW |
4665 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4666 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4667 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4668 | memcg->low = 0; |
4669 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4670 | memcg->soft_limit = PAGE_COUNTER_MAX; |
1ced953b TH |
4671 | } |
4672 | ||
02491447 | 4673 | #ifdef CONFIG_MMU |
7dc74be0 | 4674 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4675 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4676 | { |
05b84301 | 4677 | int ret; |
9476db97 JW |
4678 | |
4679 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 4680 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 4681 | if (!ret) { |
854ffa8d | 4682 | mc.precharge += count; |
854ffa8d DN |
4683 | return ret; |
4684 | } | |
692e7c45 | 4685 | if (ret == -EINTR) { |
00501b53 | 4686 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
4687 | return ret; |
4688 | } | |
9476db97 JW |
4689 | |
4690 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4691 | while (count--) { |
00501b53 | 4692 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
4693 | /* |
4694 | * In case of failure, any residual charges against | |
4695 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
4696 | * later on. However, cancel any charges that are |
4697 | * bypassed to root right away or they'll be lost. | |
9476db97 | 4698 | */ |
692e7c45 | 4699 | if (ret == -EINTR) |
00501b53 | 4700 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 4701 | if (ret) |
38c5d72f | 4702 | return ret; |
854ffa8d | 4703 | mc.precharge++; |
9476db97 | 4704 | cond_resched(); |
854ffa8d | 4705 | } |
9476db97 | 4706 | return 0; |
4ffef5fe DN |
4707 | } |
4708 | ||
4709 | /** | |
8d32ff84 | 4710 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4711 | * @vma: the vma the pte to be checked belongs |
4712 | * @addr: the address corresponding to the pte to be checked | |
4713 | * @ptent: the pte to be checked | |
02491447 | 4714 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4715 | * |
4716 | * Returns | |
4717 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4718 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4719 | * move charge. if @target is not NULL, the page is stored in target->page | |
4720 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4721 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4722 | * target for charge migration. if @target is not NULL, the entry is stored | |
4723 | * in target->ent. | |
4ffef5fe DN |
4724 | * |
4725 | * Called with pte lock held. | |
4726 | */ | |
4ffef5fe DN |
4727 | union mc_target { |
4728 | struct page *page; | |
02491447 | 4729 | swp_entry_t ent; |
4ffef5fe DN |
4730 | }; |
4731 | ||
4ffef5fe | 4732 | enum mc_target_type { |
8d32ff84 | 4733 | MC_TARGET_NONE = 0, |
4ffef5fe | 4734 | MC_TARGET_PAGE, |
02491447 | 4735 | MC_TARGET_SWAP, |
4ffef5fe DN |
4736 | }; |
4737 | ||
90254a65 DN |
4738 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4739 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4740 | { |
90254a65 | 4741 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4742 | |
90254a65 DN |
4743 | if (!page || !page_mapped(page)) |
4744 | return NULL; | |
4745 | if (PageAnon(page)) { | |
1dfab5ab | 4746 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4747 | return NULL; |
1dfab5ab JW |
4748 | } else { |
4749 | if (!(mc.flags & MOVE_FILE)) | |
4750 | return NULL; | |
4751 | } | |
90254a65 DN |
4752 | if (!get_page_unless_zero(page)) |
4753 | return NULL; | |
4754 | ||
4755 | return page; | |
4756 | } | |
4757 | ||
4b91355e | 4758 | #ifdef CONFIG_SWAP |
90254a65 DN |
4759 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4760 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4761 | { | |
90254a65 DN |
4762 | struct page *page = NULL; |
4763 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4764 | ||
1dfab5ab | 4765 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4766 | return NULL; |
4b91355e KH |
4767 | /* |
4768 | * Because lookup_swap_cache() updates some statistics counter, | |
4769 | * we call find_get_page() with swapper_space directly. | |
4770 | */ | |
33806f06 | 4771 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
4772 | if (do_swap_account) |
4773 | entry->val = ent.val; | |
4774 | ||
4775 | return page; | |
4776 | } | |
4b91355e KH |
4777 | #else |
4778 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4779 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4780 | { | |
4781 | return NULL; | |
4782 | } | |
4783 | #endif | |
90254a65 | 4784 | |
87946a72 DN |
4785 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4786 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4787 | { | |
4788 | struct page *page = NULL; | |
87946a72 DN |
4789 | struct address_space *mapping; |
4790 | pgoff_t pgoff; | |
4791 | ||
4792 | if (!vma->vm_file) /* anonymous vma */ | |
4793 | return NULL; | |
1dfab5ab | 4794 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4795 | return NULL; |
4796 | ||
87946a72 | 4797 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4798 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4799 | |
4800 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4801 | #ifdef CONFIG_SWAP |
4802 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4803 | if (shmem_mapping(mapping)) { |
4804 | page = find_get_entry(mapping, pgoff); | |
4805 | if (radix_tree_exceptional_entry(page)) { | |
4806 | swp_entry_t swp = radix_to_swp_entry(page); | |
4807 | if (do_swap_account) | |
4808 | *entry = swp; | |
4809 | page = find_get_page(swap_address_space(swp), swp.val); | |
4810 | } | |
4811 | } else | |
4812 | page = find_get_page(mapping, pgoff); | |
4813 | #else | |
4814 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4815 | #endif |
87946a72 DN |
4816 | return page; |
4817 | } | |
4818 | ||
8d32ff84 | 4819 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4820 | unsigned long addr, pte_t ptent, union mc_target *target) |
4821 | { | |
4822 | struct page *page = NULL; | |
8d32ff84 | 4823 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4824 | swp_entry_t ent = { .val = 0 }; |
4825 | ||
4826 | if (pte_present(ptent)) | |
4827 | page = mc_handle_present_pte(vma, addr, ptent); | |
4828 | else if (is_swap_pte(ptent)) | |
4829 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4830 | else if (pte_none(ptent)) |
87946a72 | 4831 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4832 | |
4833 | if (!page && !ent.val) | |
8d32ff84 | 4834 | return ret; |
02491447 | 4835 | if (page) { |
02491447 | 4836 | /* |
0a31bc97 | 4837 | * Do only loose check w/o serialization. |
1306a85a | 4838 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4839 | * not under LRU exclusion. |
02491447 | 4840 | */ |
1306a85a | 4841 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4842 | ret = MC_TARGET_PAGE; |
4843 | if (target) | |
4844 | target->page = page; | |
4845 | } | |
4846 | if (!ret || !target) | |
4847 | put_page(page); | |
4848 | } | |
90254a65 DN |
4849 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4850 | if (ent.val && !ret && | |
34c00c31 | 4851 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4852 | ret = MC_TARGET_SWAP; |
4853 | if (target) | |
4854 | target->ent = ent; | |
4ffef5fe | 4855 | } |
4ffef5fe DN |
4856 | return ret; |
4857 | } | |
4858 | ||
12724850 NH |
4859 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4860 | /* | |
4861 | * We don't consider swapping or file mapped pages because THP does not | |
4862 | * support them for now. | |
4863 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4864 | */ | |
4865 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4866 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4867 | { | |
4868 | struct page *page = NULL; | |
12724850 NH |
4869 | enum mc_target_type ret = MC_TARGET_NONE; |
4870 | ||
4871 | page = pmd_page(pmd); | |
309381fe | 4872 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4873 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4874 | return ret; |
1306a85a | 4875 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4876 | ret = MC_TARGET_PAGE; |
4877 | if (target) { | |
4878 | get_page(page); | |
4879 | target->page = page; | |
4880 | } | |
4881 | } | |
4882 | return ret; | |
4883 | } | |
4884 | #else | |
4885 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4886 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4887 | { | |
4888 | return MC_TARGET_NONE; | |
4889 | } | |
4890 | #endif | |
4891 | ||
4ffef5fe DN |
4892 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4893 | unsigned long addr, unsigned long end, | |
4894 | struct mm_walk *walk) | |
4895 | { | |
26bcd64a | 4896 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4897 | pte_t *pte; |
4898 | spinlock_t *ptl; | |
4899 | ||
bf929152 | 4900 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
4901 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4902 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4903 | spin_unlock(ptl); |
1a5a9906 | 4904 | return 0; |
12724850 | 4905 | } |
03319327 | 4906 | |
45f83cef AA |
4907 | if (pmd_trans_unstable(pmd)) |
4908 | return 0; | |
4ffef5fe DN |
4909 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4910 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4911 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4912 | mc.precharge++; /* increment precharge temporarily */ |
4913 | pte_unmap_unlock(pte - 1, ptl); | |
4914 | cond_resched(); | |
4915 | ||
7dc74be0 DN |
4916 | return 0; |
4917 | } | |
4918 | ||
4ffef5fe DN |
4919 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4920 | { | |
4921 | unsigned long precharge; | |
4ffef5fe | 4922 | |
26bcd64a NH |
4923 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4924 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4925 | .mm = mm, | |
4926 | }; | |
dfe076b0 | 4927 | down_read(&mm->mmap_sem); |
26bcd64a | 4928 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4929 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4930 | |
4931 | precharge = mc.precharge; | |
4932 | mc.precharge = 0; | |
4933 | ||
4934 | return precharge; | |
4935 | } | |
4936 | ||
4ffef5fe DN |
4937 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4938 | { | |
dfe076b0 DN |
4939 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4940 | ||
4941 | VM_BUG_ON(mc.moving_task); | |
4942 | mc.moving_task = current; | |
4943 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4944 | } |
4945 | ||
dfe076b0 DN |
4946 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4947 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4948 | { |
2bd9bb20 KH |
4949 | struct mem_cgroup *from = mc.from; |
4950 | struct mem_cgroup *to = mc.to; | |
4951 | ||
4ffef5fe | 4952 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4953 | if (mc.precharge) { |
00501b53 | 4954 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4955 | mc.precharge = 0; |
4956 | } | |
4957 | /* | |
4958 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4959 | * we must uncharge here. | |
4960 | */ | |
4961 | if (mc.moved_charge) { | |
00501b53 | 4962 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4963 | mc.moved_charge = 0; |
4ffef5fe | 4964 | } |
483c30b5 DN |
4965 | /* we must fixup refcnts and charges */ |
4966 | if (mc.moved_swap) { | |
483c30b5 | 4967 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4968 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4969 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4970 | |
05b84301 | 4971 | /* |
3e32cb2e JW |
4972 | * we charged both to->memory and to->memsw, so we |
4973 | * should uncharge to->memory. | |
05b84301 | 4974 | */ |
ce00a967 | 4975 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4976 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4977 | ||
e8ea14cc | 4978 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4979 | |
4050377b | 4980 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4981 | mc.moved_swap = 0; |
4982 | } | |
dfe076b0 DN |
4983 | memcg_oom_recover(from); |
4984 | memcg_oom_recover(to); | |
4985 | wake_up_all(&mc.waitq); | |
4986 | } | |
4987 | ||
4988 | static void mem_cgroup_clear_mc(void) | |
4989 | { | |
dfe076b0 DN |
4990 | /* |
4991 | * we must clear moving_task before waking up waiters at the end of | |
4992 | * task migration. | |
4993 | */ | |
4994 | mc.moving_task = NULL; | |
4995 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4996 | spin_lock(&mc.lock); |
4ffef5fe DN |
4997 | mc.from = NULL; |
4998 | mc.to = NULL; | |
2bd9bb20 | 4999 | spin_unlock(&mc.lock); |
4ffef5fe DN |
5000 | } |
5001 | ||
eb95419b | 5002 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5003 | struct cgroup_taskset *tset) |
7dc74be0 | 5004 | { |
2f7ee569 | 5005 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5006 | int ret = 0; |
eb95419b | 5007 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
1dfab5ab | 5008 | unsigned long move_flags; |
7dc74be0 | 5009 | |
ee5e8472 GC |
5010 | /* |
5011 | * We are now commited to this value whatever it is. Changes in this | |
5012 | * tunable will only affect upcoming migrations, not the current one. | |
5013 | * So we need to save it, and keep it going. | |
5014 | */ | |
1dfab5ab JW |
5015 | move_flags = ACCESS_ONCE(memcg->move_charge_at_immigrate); |
5016 | if (move_flags) { | |
7dc74be0 DN |
5017 | struct mm_struct *mm; |
5018 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5019 | ||
c0ff4b85 | 5020 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5021 | |
5022 | mm = get_task_mm(p); | |
5023 | if (!mm) | |
5024 | return 0; | |
7dc74be0 | 5025 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5026 | if (mm->owner == p) { |
5027 | VM_BUG_ON(mc.from); | |
5028 | VM_BUG_ON(mc.to); | |
5029 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5030 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5031 | VM_BUG_ON(mc.moved_swap); |
247b1447 | 5032 | |
2bd9bb20 | 5033 | spin_lock(&mc.lock); |
4ffef5fe | 5034 | mc.from = from; |
c0ff4b85 | 5035 | mc.to = memcg; |
1dfab5ab | 5036 | mc.flags = move_flags; |
2bd9bb20 | 5037 | spin_unlock(&mc.lock); |
dfe076b0 | 5038 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5039 | |
5040 | ret = mem_cgroup_precharge_mc(mm); | |
5041 | if (ret) | |
5042 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5043 | } |
5044 | mmput(mm); | |
7dc74be0 DN |
5045 | } |
5046 | return ret; | |
5047 | } | |
5048 | ||
eb95419b | 5049 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5050 | struct cgroup_taskset *tset) |
7dc74be0 | 5051 | { |
4e2f245d JW |
5052 | if (mc.to) |
5053 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5054 | } |
5055 | ||
4ffef5fe DN |
5056 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5057 | unsigned long addr, unsigned long end, | |
5058 | struct mm_walk *walk) | |
7dc74be0 | 5059 | { |
4ffef5fe | 5060 | int ret = 0; |
26bcd64a | 5061 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5062 | pte_t *pte; |
5063 | spinlock_t *ptl; | |
12724850 NH |
5064 | enum mc_target_type target_type; |
5065 | union mc_target target; | |
5066 | struct page *page; | |
4ffef5fe | 5067 | |
12724850 NH |
5068 | /* |
5069 | * We don't take compound_lock() here but no race with splitting thp | |
5070 | * happens because: | |
5071 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5072 | * under splitting, which means there's no concurrent thp split, | |
5073 | * - if another thread runs into split_huge_page() just after we | |
5074 | * entered this if-block, the thread must wait for page table lock | |
5075 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5076 | * part of thp split is not executed yet. | |
5077 | */ | |
bf929152 | 5078 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5079 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5080 | spin_unlock(ptl); |
12724850 NH |
5081 | return 0; |
5082 | } | |
5083 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5084 | if (target_type == MC_TARGET_PAGE) { | |
5085 | page = target.page; | |
5086 | if (!isolate_lru_page(page)) { | |
12724850 | 5087 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 5088 | mc.from, mc.to)) { |
12724850 NH |
5089 | mc.precharge -= HPAGE_PMD_NR; |
5090 | mc.moved_charge += HPAGE_PMD_NR; | |
5091 | } | |
5092 | putback_lru_page(page); | |
5093 | } | |
5094 | put_page(page); | |
5095 | } | |
bf929152 | 5096 | spin_unlock(ptl); |
1a5a9906 | 5097 | return 0; |
12724850 NH |
5098 | } |
5099 | ||
45f83cef AA |
5100 | if (pmd_trans_unstable(pmd)) |
5101 | return 0; | |
4ffef5fe DN |
5102 | retry: |
5103 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5104 | for (; addr != end; addr += PAGE_SIZE) { | |
5105 | pte_t ptent = *(pte++); | |
02491447 | 5106 | swp_entry_t ent; |
4ffef5fe DN |
5107 | |
5108 | if (!mc.precharge) | |
5109 | break; | |
5110 | ||
8d32ff84 | 5111 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5112 | case MC_TARGET_PAGE: |
5113 | page = target.page; | |
5114 | if (isolate_lru_page(page)) | |
5115 | goto put; | |
1306a85a | 5116 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 5117 | mc.precharge--; |
854ffa8d DN |
5118 | /* we uncharge from mc.from later. */ |
5119 | mc.moved_charge++; | |
4ffef5fe DN |
5120 | } |
5121 | putback_lru_page(page); | |
8d32ff84 | 5122 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5123 | put_page(page); |
5124 | break; | |
02491447 DN |
5125 | case MC_TARGET_SWAP: |
5126 | ent = target.ent; | |
e91cbb42 | 5127 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5128 | mc.precharge--; |
483c30b5 DN |
5129 | /* we fixup refcnts and charges later. */ |
5130 | mc.moved_swap++; | |
5131 | } | |
02491447 | 5132 | break; |
4ffef5fe DN |
5133 | default: |
5134 | break; | |
5135 | } | |
5136 | } | |
5137 | pte_unmap_unlock(pte - 1, ptl); | |
5138 | cond_resched(); | |
5139 | ||
5140 | if (addr != end) { | |
5141 | /* | |
5142 | * We have consumed all precharges we got in can_attach(). | |
5143 | * We try charge one by one, but don't do any additional | |
5144 | * charges to mc.to if we have failed in charge once in attach() | |
5145 | * phase. | |
5146 | */ | |
854ffa8d | 5147 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5148 | if (!ret) |
5149 | goto retry; | |
5150 | } | |
5151 | ||
5152 | return ret; | |
5153 | } | |
5154 | ||
5155 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5156 | { | |
26bcd64a NH |
5157 | struct mm_walk mem_cgroup_move_charge_walk = { |
5158 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5159 | .mm = mm, | |
5160 | }; | |
4ffef5fe DN |
5161 | |
5162 | lru_add_drain_all(); | |
312722cb JW |
5163 | /* |
5164 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5165 | * move_lock while we're moving its pages to another memcg. | |
5166 | * Then wait for already started RCU-only updates to finish. | |
5167 | */ | |
5168 | atomic_inc(&mc.from->moving_account); | |
5169 | synchronize_rcu(); | |
dfe076b0 DN |
5170 | retry: |
5171 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5172 | /* | |
5173 | * Someone who are holding the mmap_sem might be waiting in | |
5174 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5175 | * and retry. Because we cancel precharges, we might not be able | |
5176 | * to move enough charges, but moving charge is a best-effort | |
5177 | * feature anyway, so it wouldn't be a big problem. | |
5178 | */ | |
5179 | __mem_cgroup_clear_mc(); | |
5180 | cond_resched(); | |
5181 | goto retry; | |
5182 | } | |
26bcd64a NH |
5183 | /* |
5184 | * When we have consumed all precharges and failed in doing | |
5185 | * additional charge, the page walk just aborts. | |
5186 | */ | |
5187 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
dfe076b0 | 5188 | up_read(&mm->mmap_sem); |
312722cb | 5189 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5190 | } |
5191 | ||
eb95419b | 5192 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5193 | struct cgroup_taskset *tset) |
67e465a7 | 5194 | { |
2f7ee569 | 5195 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5196 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5197 | |
dfe076b0 | 5198 | if (mm) { |
a433658c KM |
5199 | if (mc.to) |
5200 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5201 | mmput(mm); |
5202 | } | |
a433658c KM |
5203 | if (mc.to) |
5204 | mem_cgroup_clear_mc(); | |
67e465a7 | 5205 | } |
5cfb80a7 | 5206 | #else /* !CONFIG_MMU */ |
eb95419b | 5207 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5208 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5209 | { |
5210 | return 0; | |
5211 | } | |
eb95419b | 5212 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5213 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5214 | { |
5215 | } | |
eb95419b | 5216 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5217 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5218 | { |
5219 | } | |
5220 | #endif | |
67e465a7 | 5221 | |
f00baae7 TH |
5222 | /* |
5223 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5224 | * to verify whether we're attached to the default hierarchy on each mount |
5225 | * attempt. | |
f00baae7 | 5226 | */ |
eb95419b | 5227 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5228 | { |
5229 | /* | |
aa6ec29b | 5230 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5231 | * guarantees that @root doesn't have any children, so turning it |
5232 | * on for the root memcg is enough. | |
5233 | */ | |
aa6ec29b | 5234 | if (cgroup_on_dfl(root_css->cgroup)) |
7feee590 VD |
5235 | root_mem_cgroup->use_hierarchy = true; |
5236 | else | |
5237 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5238 | } |
5239 | ||
241994ed JW |
5240 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5241 | struct cftype *cft) | |
5242 | { | |
5243 | return mem_cgroup_usage(mem_cgroup_from_css(css), false); | |
5244 | } | |
5245 | ||
5246 | static int memory_low_show(struct seq_file *m, void *v) | |
5247 | { | |
5248 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5249 | unsigned long low = ACCESS_ONCE(memcg->low); | |
5250 | ||
5251 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5252 | seq_puts(m, "max\n"); |
241994ed JW |
5253 | else |
5254 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5255 | ||
5256 | return 0; | |
5257 | } | |
5258 | ||
5259 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5260 | char *buf, size_t nbytes, loff_t off) | |
5261 | { | |
5262 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5263 | unsigned long low; | |
5264 | int err; | |
5265 | ||
5266 | buf = strstrip(buf); | |
d2973697 | 5267 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5268 | if (err) |
5269 | return err; | |
5270 | ||
5271 | memcg->low = low; | |
5272 | ||
5273 | return nbytes; | |
5274 | } | |
5275 | ||
5276 | static int memory_high_show(struct seq_file *m, void *v) | |
5277 | { | |
5278 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5279 | unsigned long high = ACCESS_ONCE(memcg->high); | |
5280 | ||
5281 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5282 | seq_puts(m, "max\n"); |
241994ed JW |
5283 | else |
5284 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5285 | ||
5286 | return 0; | |
5287 | } | |
5288 | ||
5289 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5290 | char *buf, size_t nbytes, loff_t off) | |
5291 | { | |
5292 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5293 | unsigned long high; | |
5294 | int err; | |
5295 | ||
5296 | buf = strstrip(buf); | |
d2973697 | 5297 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5298 | if (err) |
5299 | return err; | |
5300 | ||
5301 | memcg->high = high; | |
5302 | ||
5303 | return nbytes; | |
5304 | } | |
5305 | ||
5306 | static int memory_max_show(struct seq_file *m, void *v) | |
5307 | { | |
5308 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5309 | unsigned long max = ACCESS_ONCE(memcg->memory.limit); | |
5310 | ||
5311 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5312 | seq_puts(m, "max\n"); |
241994ed JW |
5313 | else |
5314 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5315 | ||
5316 | return 0; | |
5317 | } | |
5318 | ||
5319 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5320 | char *buf, size_t nbytes, loff_t off) | |
5321 | { | |
5322 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5323 | unsigned long max; | |
5324 | int err; | |
5325 | ||
5326 | buf = strstrip(buf); | |
d2973697 | 5327 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5328 | if (err) |
5329 | return err; | |
5330 | ||
5331 | err = mem_cgroup_resize_limit(memcg, max); | |
5332 | if (err) | |
5333 | return err; | |
5334 | ||
5335 | return nbytes; | |
5336 | } | |
5337 | ||
5338 | static int memory_events_show(struct seq_file *m, void *v) | |
5339 | { | |
5340 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5341 | ||
5342 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5343 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5344 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5345 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5346 | ||
5347 | return 0; | |
5348 | } | |
5349 | ||
5350 | static struct cftype memory_files[] = { | |
5351 | { | |
5352 | .name = "current", | |
5353 | .read_u64 = memory_current_read, | |
5354 | }, | |
5355 | { | |
5356 | .name = "low", | |
5357 | .flags = CFTYPE_NOT_ON_ROOT, | |
5358 | .seq_show = memory_low_show, | |
5359 | .write = memory_low_write, | |
5360 | }, | |
5361 | { | |
5362 | .name = "high", | |
5363 | .flags = CFTYPE_NOT_ON_ROOT, | |
5364 | .seq_show = memory_high_show, | |
5365 | .write = memory_high_write, | |
5366 | }, | |
5367 | { | |
5368 | .name = "max", | |
5369 | .flags = CFTYPE_NOT_ON_ROOT, | |
5370 | .seq_show = memory_max_show, | |
5371 | .write = memory_max_write, | |
5372 | }, | |
5373 | { | |
5374 | .name = "events", | |
5375 | .flags = CFTYPE_NOT_ON_ROOT, | |
5376 | .seq_show = memory_events_show, | |
5377 | }, | |
5378 | { } /* terminate */ | |
5379 | }; | |
5380 | ||
073219e9 | 5381 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5382 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5383 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5384 | .css_offline = mem_cgroup_css_offline, |
5385 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5386 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5387 | .can_attach = mem_cgroup_can_attach, |
5388 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5389 | .attach = mem_cgroup_move_task, |
f00baae7 | 5390 | .bind = mem_cgroup_bind, |
241994ed JW |
5391 | .dfl_cftypes = memory_files, |
5392 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5393 | .early_init = 0, |
8cdea7c0 | 5394 | }; |
c077719b | 5395 | |
241994ed JW |
5396 | /** |
5397 | * mem_cgroup_events - count memory events against a cgroup | |
5398 | * @memcg: the memory cgroup | |
5399 | * @idx: the event index | |
5400 | * @nr: the number of events to account for | |
5401 | */ | |
5402 | void mem_cgroup_events(struct mem_cgroup *memcg, | |
5403 | enum mem_cgroup_events_index idx, | |
5404 | unsigned int nr) | |
5405 | { | |
5406 | this_cpu_add(memcg->stat->events[idx], nr); | |
5407 | } | |
5408 | ||
5409 | /** | |
5410 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5411 | * @root: the highest ancestor to consider | |
5412 | * @memcg: the memory cgroup to check | |
5413 | * | |
5414 | * Returns %true if memory consumption of @memcg, and that of all | |
5415 | * configurable ancestors up to @root, is below the normal range. | |
5416 | */ | |
5417 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5418 | { | |
5419 | if (mem_cgroup_disabled()) | |
5420 | return false; | |
5421 | ||
5422 | /* | |
5423 | * The toplevel group doesn't have a configurable range, so | |
5424 | * it's never low when looked at directly, and it is not | |
5425 | * considered an ancestor when assessing the hierarchy. | |
5426 | */ | |
5427 | ||
5428 | if (memcg == root_mem_cgroup) | |
5429 | return false; | |
5430 | ||
4e54dede | 5431 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5432 | return false; |
5433 | ||
5434 | while (memcg != root) { | |
5435 | memcg = parent_mem_cgroup(memcg); | |
5436 | ||
5437 | if (memcg == root_mem_cgroup) | |
5438 | break; | |
5439 | ||
4e54dede | 5440 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5441 | return false; |
5442 | } | |
5443 | return true; | |
5444 | } | |
5445 | ||
00501b53 JW |
5446 | /** |
5447 | * mem_cgroup_try_charge - try charging a page | |
5448 | * @page: page to charge | |
5449 | * @mm: mm context of the victim | |
5450 | * @gfp_mask: reclaim mode | |
5451 | * @memcgp: charged memcg return | |
5452 | * | |
5453 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5454 | * pages according to @gfp_mask if necessary. | |
5455 | * | |
5456 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5457 | * Otherwise, an error code is returned. | |
5458 | * | |
5459 | * After page->mapping has been set up, the caller must finalize the | |
5460 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5461 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5462 | */ | |
5463 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5464 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5465 | { | |
5466 | struct mem_cgroup *memcg = NULL; | |
5467 | unsigned int nr_pages = 1; | |
5468 | int ret = 0; | |
5469 | ||
5470 | if (mem_cgroup_disabled()) | |
5471 | goto out; | |
5472 | ||
5473 | if (PageSwapCache(page)) { | |
00501b53 JW |
5474 | /* |
5475 | * Every swap fault against a single page tries to charge the | |
5476 | * page, bail as early as possible. shmem_unuse() encounters | |
5477 | * already charged pages, too. The USED bit is protected by | |
5478 | * the page lock, which serializes swap cache removal, which | |
5479 | * in turn serializes uncharging. | |
5480 | */ | |
1306a85a | 5481 | if (page->mem_cgroup) |
00501b53 JW |
5482 | goto out; |
5483 | } | |
5484 | ||
5485 | if (PageTransHuge(page)) { | |
5486 | nr_pages <<= compound_order(page); | |
5487 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5488 | } | |
5489 | ||
5490 | if (do_swap_account && PageSwapCache(page)) | |
5491 | memcg = try_get_mem_cgroup_from_page(page); | |
5492 | if (!memcg) | |
5493 | memcg = get_mem_cgroup_from_mm(mm); | |
5494 | ||
5495 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5496 | ||
5497 | css_put(&memcg->css); | |
5498 | ||
5499 | if (ret == -EINTR) { | |
5500 | memcg = root_mem_cgroup; | |
5501 | ret = 0; | |
5502 | } | |
5503 | out: | |
5504 | *memcgp = memcg; | |
5505 | return ret; | |
5506 | } | |
5507 | ||
5508 | /** | |
5509 | * mem_cgroup_commit_charge - commit a page charge | |
5510 | * @page: page to charge | |
5511 | * @memcg: memcg to charge the page to | |
5512 | * @lrucare: page might be on LRU already | |
5513 | * | |
5514 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5515 | * after page->mapping has been set up. This must happen atomically | |
5516 | * as part of the page instantiation, i.e. under the page table lock | |
5517 | * for anonymous pages, under the page lock for page and swap cache. | |
5518 | * | |
5519 | * In addition, the page must not be on the LRU during the commit, to | |
5520 | * prevent racing with task migration. If it might be, use @lrucare. | |
5521 | * | |
5522 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5523 | */ | |
5524 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5525 | bool lrucare) | |
5526 | { | |
5527 | unsigned int nr_pages = 1; | |
5528 | ||
5529 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5530 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5531 | ||
5532 | if (mem_cgroup_disabled()) | |
5533 | return; | |
5534 | /* | |
5535 | * Swap faults will attempt to charge the same page multiple | |
5536 | * times. But reuse_swap_page() might have removed the page | |
5537 | * from swapcache already, so we can't check PageSwapCache(). | |
5538 | */ | |
5539 | if (!memcg) | |
5540 | return; | |
5541 | ||
6abb5a86 JW |
5542 | commit_charge(page, memcg, lrucare); |
5543 | ||
00501b53 JW |
5544 | if (PageTransHuge(page)) { |
5545 | nr_pages <<= compound_order(page); | |
5546 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5547 | } | |
5548 | ||
6abb5a86 JW |
5549 | local_irq_disable(); |
5550 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5551 | memcg_check_events(memcg, page); | |
5552 | local_irq_enable(); | |
00501b53 JW |
5553 | |
5554 | if (do_swap_account && PageSwapCache(page)) { | |
5555 | swp_entry_t entry = { .val = page_private(page) }; | |
5556 | /* | |
5557 | * The swap entry might not get freed for a long time, | |
5558 | * let's not wait for it. The page already received a | |
5559 | * memory+swap charge, drop the swap entry duplicate. | |
5560 | */ | |
5561 | mem_cgroup_uncharge_swap(entry); | |
5562 | } | |
5563 | } | |
5564 | ||
5565 | /** | |
5566 | * mem_cgroup_cancel_charge - cancel a page charge | |
5567 | * @page: page to charge | |
5568 | * @memcg: memcg to charge the page to | |
5569 | * | |
5570 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5571 | */ | |
5572 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5573 | { | |
5574 | unsigned int nr_pages = 1; | |
5575 | ||
5576 | if (mem_cgroup_disabled()) | |
5577 | return; | |
5578 | /* | |
5579 | * Swap faults will attempt to charge the same page multiple | |
5580 | * times. But reuse_swap_page() might have removed the page | |
5581 | * from swapcache already, so we can't check PageSwapCache(). | |
5582 | */ | |
5583 | if (!memcg) | |
5584 | return; | |
5585 | ||
5586 | if (PageTransHuge(page)) { | |
5587 | nr_pages <<= compound_order(page); | |
5588 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5589 | } | |
5590 | ||
5591 | cancel_charge(memcg, nr_pages); | |
5592 | } | |
5593 | ||
747db954 | 5594 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5595 | unsigned long nr_anon, unsigned long nr_file, |
5596 | unsigned long nr_huge, struct page *dummy_page) | |
5597 | { | |
18eca2e6 | 5598 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5599 | unsigned long flags; |
5600 | ||
ce00a967 | 5601 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5602 | page_counter_uncharge(&memcg->memory, nr_pages); |
5603 | if (do_swap_account) | |
5604 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5605 | memcg_oom_recover(memcg); |
5606 | } | |
747db954 JW |
5607 | |
5608 | local_irq_save(flags); | |
5609 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5610 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5611 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5612 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5613 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5614 | memcg_check_events(memcg, dummy_page); |
5615 | local_irq_restore(flags); | |
e8ea14cc JW |
5616 | |
5617 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5618 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5619 | } |
5620 | ||
5621 | static void uncharge_list(struct list_head *page_list) | |
5622 | { | |
5623 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5624 | unsigned long nr_anon = 0; |
5625 | unsigned long nr_file = 0; | |
5626 | unsigned long nr_huge = 0; | |
5627 | unsigned long pgpgout = 0; | |
747db954 JW |
5628 | struct list_head *next; |
5629 | struct page *page; | |
5630 | ||
5631 | next = page_list->next; | |
5632 | do { | |
5633 | unsigned int nr_pages = 1; | |
747db954 JW |
5634 | |
5635 | page = list_entry(next, struct page, lru); | |
5636 | next = page->lru.next; | |
5637 | ||
5638 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5639 | VM_BUG_ON_PAGE(page_count(page), page); | |
5640 | ||
1306a85a | 5641 | if (!page->mem_cgroup) |
747db954 JW |
5642 | continue; |
5643 | ||
5644 | /* | |
5645 | * Nobody should be changing or seriously looking at | |
1306a85a | 5646 | * page->mem_cgroup at this point, we have fully |
29833315 | 5647 | * exclusive access to the page. |
747db954 JW |
5648 | */ |
5649 | ||
1306a85a | 5650 | if (memcg != page->mem_cgroup) { |
747db954 | 5651 | if (memcg) { |
18eca2e6 JW |
5652 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5653 | nr_huge, page); | |
5654 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5655 | } |
1306a85a | 5656 | memcg = page->mem_cgroup; |
747db954 JW |
5657 | } |
5658 | ||
5659 | if (PageTransHuge(page)) { | |
5660 | nr_pages <<= compound_order(page); | |
5661 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5662 | nr_huge += nr_pages; | |
5663 | } | |
5664 | ||
5665 | if (PageAnon(page)) | |
5666 | nr_anon += nr_pages; | |
5667 | else | |
5668 | nr_file += nr_pages; | |
5669 | ||
1306a85a | 5670 | page->mem_cgroup = NULL; |
747db954 JW |
5671 | |
5672 | pgpgout++; | |
5673 | } while (next != page_list); | |
5674 | ||
5675 | if (memcg) | |
18eca2e6 JW |
5676 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5677 | nr_huge, page); | |
747db954 JW |
5678 | } |
5679 | ||
0a31bc97 JW |
5680 | /** |
5681 | * mem_cgroup_uncharge - uncharge a page | |
5682 | * @page: page to uncharge | |
5683 | * | |
5684 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5685 | * mem_cgroup_commit_charge(). | |
5686 | */ | |
5687 | void mem_cgroup_uncharge(struct page *page) | |
5688 | { | |
0a31bc97 JW |
5689 | if (mem_cgroup_disabled()) |
5690 | return; | |
5691 | ||
747db954 | 5692 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5693 | if (!page->mem_cgroup) |
0a31bc97 JW |
5694 | return; |
5695 | ||
747db954 JW |
5696 | INIT_LIST_HEAD(&page->lru); |
5697 | uncharge_list(&page->lru); | |
5698 | } | |
0a31bc97 | 5699 | |
747db954 JW |
5700 | /** |
5701 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5702 | * @page_list: list of pages to uncharge | |
5703 | * | |
5704 | * Uncharge a list of pages previously charged with | |
5705 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5706 | */ | |
5707 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5708 | { | |
5709 | if (mem_cgroup_disabled()) | |
5710 | return; | |
0a31bc97 | 5711 | |
747db954 JW |
5712 | if (!list_empty(page_list)) |
5713 | uncharge_list(page_list); | |
0a31bc97 JW |
5714 | } |
5715 | ||
5716 | /** | |
5717 | * mem_cgroup_migrate - migrate a charge to another page | |
5718 | * @oldpage: currently charged page | |
5719 | * @newpage: page to transfer the charge to | |
f5e03a49 | 5720 | * @lrucare: either or both pages might be on the LRU already |
0a31bc97 JW |
5721 | * |
5722 | * Migrate the charge from @oldpage to @newpage. | |
5723 | * | |
5724 | * Both pages must be locked, @newpage->mapping must be set up. | |
5725 | */ | |
5726 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
5727 | bool lrucare) | |
5728 | { | |
29833315 | 5729 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5730 | int isolated; |
5731 | ||
5732 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5733 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
5734 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
5735 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
5736 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
5737 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5738 | newpage); | |
0a31bc97 JW |
5739 | |
5740 | if (mem_cgroup_disabled()) | |
5741 | return; | |
5742 | ||
5743 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5744 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5745 | return; |
5746 | ||
7d5e3245 JW |
5747 | /* |
5748 | * Swapcache readahead pages can get migrated before being | |
5749 | * charged, and migration from compaction can happen to an | |
5750 | * uncharged page when the PFN walker finds a page that | |
5751 | * reclaim just put back on the LRU but has not released yet. | |
5752 | */ | |
1306a85a | 5753 | memcg = oldpage->mem_cgroup; |
29833315 | 5754 | if (!memcg) |
0a31bc97 JW |
5755 | return; |
5756 | ||
0a31bc97 JW |
5757 | if (lrucare) |
5758 | lock_page_lru(oldpage, &isolated); | |
5759 | ||
1306a85a | 5760 | oldpage->mem_cgroup = NULL; |
0a31bc97 JW |
5761 | |
5762 | if (lrucare) | |
5763 | unlock_page_lru(oldpage, isolated); | |
5764 | ||
29833315 | 5765 | commit_charge(newpage, memcg, lrucare); |
0a31bc97 JW |
5766 | } |
5767 | ||
2d11085e | 5768 | /* |
1081312f MH |
5769 | * subsys_initcall() for memory controller. |
5770 | * | |
5771 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5772 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5773 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5774 | * be initialized from here. | |
2d11085e MH |
5775 | */ |
5776 | static int __init mem_cgroup_init(void) | |
5777 | { | |
95a045f6 JW |
5778 | int cpu, node; |
5779 | ||
2d11085e | 5780 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5781 | |
5782 | for_each_possible_cpu(cpu) | |
5783 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5784 | drain_local_stock); | |
5785 | ||
5786 | for_each_node(node) { | |
5787 | struct mem_cgroup_tree_per_node *rtpn; | |
5788 | int zone; | |
5789 | ||
5790 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5791 | node_online(node) ? node : NUMA_NO_NODE); | |
5792 | ||
5793 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5794 | struct mem_cgroup_tree_per_zone *rtpz; | |
5795 | ||
5796 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5797 | rtpz->rb_root = RB_ROOT; | |
5798 | spin_lock_init(&rtpz->lock); | |
5799 | } | |
5800 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5801 | } | |
5802 | ||
2d11085e MH |
5803 | return 0; |
5804 | } | |
5805 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5806 | |
5807 | #ifdef CONFIG_MEMCG_SWAP | |
5808 | /** | |
5809 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5810 | * @page: page whose memsw charge to transfer | |
5811 | * @entry: swap entry to move the charge to | |
5812 | * | |
5813 | * Transfer the memsw charge of @page to @entry. | |
5814 | */ | |
5815 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5816 | { | |
5817 | struct mem_cgroup *memcg; | |
5818 | unsigned short oldid; | |
5819 | ||
5820 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5821 | VM_BUG_ON_PAGE(page_count(page), page); | |
5822 | ||
5823 | if (!do_swap_account) | |
5824 | return; | |
5825 | ||
5826 | memcg = page->mem_cgroup; | |
5827 | ||
5828 | /* Readahead page, never charged */ | |
5829 | if (!memcg) | |
5830 | return; | |
5831 | ||
5832 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5833 | VM_BUG_ON_PAGE(oldid, page); | |
5834 | mem_cgroup_swap_statistics(memcg, true); | |
5835 | ||
5836 | page->mem_cgroup = NULL; | |
5837 | ||
5838 | if (!mem_cgroup_is_root(memcg)) | |
5839 | page_counter_uncharge(&memcg->memory, 1); | |
5840 | ||
5841 | /* XXX: caller holds IRQ-safe mapping->tree_lock */ | |
5842 | VM_BUG_ON(!irqs_disabled()); | |
5843 | ||
5844 | mem_cgroup_charge_statistics(memcg, page, -1); | |
5845 | memcg_check_events(memcg, page); | |
5846 | } | |
5847 | ||
5848 | /** | |
5849 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5850 | * @entry: swap entry to uncharge | |
5851 | * | |
5852 | * Drop the memsw charge associated with @entry. | |
5853 | */ | |
5854 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5855 | { | |
5856 | struct mem_cgroup *memcg; | |
5857 | unsigned short id; | |
5858 | ||
5859 | if (!do_swap_account) | |
5860 | return; | |
5861 | ||
5862 | id = swap_cgroup_record(entry, 0); | |
5863 | rcu_read_lock(); | |
5864 | memcg = mem_cgroup_lookup(id); | |
5865 | if (memcg) { | |
5866 | if (!mem_cgroup_is_root(memcg)) | |
5867 | page_counter_uncharge(&memcg->memsw, 1); | |
5868 | mem_cgroup_swap_statistics(memcg, false); | |
5869 | css_put(&memcg->css); | |
5870 | } | |
5871 | rcu_read_unlock(); | |
5872 | } | |
5873 | ||
5874 | /* for remember boot option*/ | |
5875 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5876 | static int really_do_swap_account __initdata = 1; | |
5877 | #else | |
5878 | static int really_do_swap_account __initdata; | |
5879 | #endif | |
5880 | ||
5881 | static int __init enable_swap_account(char *s) | |
5882 | { | |
5883 | if (!strcmp(s, "1")) | |
5884 | really_do_swap_account = 1; | |
5885 | else if (!strcmp(s, "0")) | |
5886 | really_do_swap_account = 0; | |
5887 | return 1; | |
5888 | } | |
5889 | __setup("swapaccount=", enable_swap_account); | |
5890 | ||
5891 | static struct cftype memsw_cgroup_files[] = { | |
5892 | { | |
5893 | .name = "memsw.usage_in_bytes", | |
5894 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5895 | .read_u64 = mem_cgroup_read_u64, | |
5896 | }, | |
5897 | { | |
5898 | .name = "memsw.max_usage_in_bytes", | |
5899 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5900 | .write = mem_cgroup_reset, | |
5901 | .read_u64 = mem_cgroup_read_u64, | |
5902 | }, | |
5903 | { | |
5904 | .name = "memsw.limit_in_bytes", | |
5905 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5906 | .write = mem_cgroup_write, | |
5907 | .read_u64 = mem_cgroup_read_u64, | |
5908 | }, | |
5909 | { | |
5910 | .name = "memsw.failcnt", | |
5911 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5912 | .write = mem_cgroup_reset, | |
5913 | .read_u64 = mem_cgroup_read_u64, | |
5914 | }, | |
5915 | { }, /* terminate */ | |
5916 | }; | |
5917 | ||
5918 | static int __init mem_cgroup_swap_init(void) | |
5919 | { | |
5920 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5921 | do_swap_account = 1; | |
5922 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, | |
5923 | memsw_cgroup_files)); | |
5924 | } | |
5925 | return 0; | |
5926 | } | |
5927 | subsys_initcall(mem_cgroup_swap_init); | |
5928 | ||
5929 | #endif /* CONFIG_MEMCG_SWAP */ |