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> |
52d4b9ac | 54 | #include <linux/page_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
79bd9814 | 58 | #include <linux/file.h> |
08e552c6 | 59 | #include "internal.h" |
d1a4c0b3 | 60 | #include <net/sock.h> |
4bd2c1ee | 61 | #include <net/ip.h> |
d1a4c0b3 | 62 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 63 | #include "slab.h" |
8cdea7c0 | 64 | |
8697d331 BS |
65 | #include <asm/uaccess.h> |
66 | ||
cc8e970c KM |
67 | #include <trace/events/vmscan.h> |
68 | ||
073219e9 TH |
69 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
70 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 71 | |
a181b0e8 | 72 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 73 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 74 | |
c255a458 | 75 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 76 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 77 | int do_swap_account __read_mostly; |
a42c390c MH |
78 | |
79 | /* for remember boot option*/ | |
c255a458 | 80 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
81 | static int really_do_swap_account __initdata = 1; |
82 | #else | |
ada4ba59 | 83 | static int really_do_swap_account __initdata; |
a42c390c MH |
84 | #endif |
85 | ||
c077719b | 86 | #else |
a0db00fc | 87 | #define do_swap_account 0 |
c077719b KH |
88 | #endif |
89 | ||
90 | ||
af7c4b0e JW |
91 | static const char * const mem_cgroup_stat_names[] = { |
92 | "cache", | |
93 | "rss", | |
b070e65c | 94 | "rss_huge", |
af7c4b0e | 95 | "mapped_file", |
3ea67d06 | 96 | "writeback", |
af7c4b0e JW |
97 | "swap", |
98 | }; | |
99 | ||
e9f8974f JW |
100 | enum mem_cgroup_events_index { |
101 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
102 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
103 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
104 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
105 | MEM_CGROUP_EVENTS_NSTATS, |
106 | }; | |
af7c4b0e JW |
107 | |
108 | static const char * const mem_cgroup_events_names[] = { | |
109 | "pgpgin", | |
110 | "pgpgout", | |
111 | "pgfault", | |
112 | "pgmajfault", | |
113 | }; | |
114 | ||
58cf188e SZ |
115 | static const char * const mem_cgroup_lru_names[] = { |
116 | "inactive_anon", | |
117 | "active_anon", | |
118 | "inactive_file", | |
119 | "active_file", | |
120 | "unevictable", | |
121 | }; | |
122 | ||
7a159cc9 JW |
123 | /* |
124 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
125 | * it will be incremated by the number of pages. This counter is used for | |
126 | * for trigger some periodic events. This is straightforward and better | |
127 | * than using jiffies etc. to handle periodic memcg event. | |
128 | */ | |
129 | enum mem_cgroup_events_target { | |
130 | MEM_CGROUP_TARGET_THRESH, | |
bb4cc1a8 | 131 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 132 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
133 | MEM_CGROUP_NTARGETS, |
134 | }; | |
a0db00fc KS |
135 | #define THRESHOLDS_EVENTS_TARGET 128 |
136 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
137 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 138 | |
d52aa412 | 139 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 140 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 141 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 142 | unsigned long nr_page_events; |
7a159cc9 | 143 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
144 | }; |
145 | ||
5ac8fb31 JW |
146 | struct reclaim_iter { |
147 | struct mem_cgroup *position; | |
527a5ec9 JW |
148 | /* scan generation, increased every round-trip */ |
149 | unsigned int generation; | |
150 | }; | |
151 | ||
6d12e2d8 KH |
152 | /* |
153 | * per-zone information in memory controller. | |
154 | */ | |
6d12e2d8 | 155 | struct mem_cgroup_per_zone { |
6290df54 | 156 | struct lruvec lruvec; |
1eb49272 | 157 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 158 | |
5ac8fb31 | 159 | struct reclaim_iter iter[DEF_PRIORITY + 1]; |
527a5ec9 | 160 | |
bb4cc1a8 | 161 | struct rb_node tree_node; /* RB tree node */ |
3e32cb2e | 162 | unsigned long usage_in_excess;/* Set to the value by which */ |
bb4cc1a8 AM |
163 | /* the soft limit is exceeded*/ |
164 | bool on_tree; | |
d79154bb | 165 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 166 | /* use container_of */ |
6d12e2d8 | 167 | }; |
6d12e2d8 KH |
168 | |
169 | struct mem_cgroup_per_node { | |
170 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
171 | }; | |
172 | ||
bb4cc1a8 AM |
173 | /* |
174 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
175 | * their hierarchy representation | |
176 | */ | |
177 | ||
178 | struct mem_cgroup_tree_per_zone { | |
179 | struct rb_root rb_root; | |
180 | spinlock_t lock; | |
181 | }; | |
182 | ||
183 | struct mem_cgroup_tree_per_node { | |
184 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
185 | }; | |
186 | ||
187 | struct mem_cgroup_tree { | |
188 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
189 | }; | |
190 | ||
191 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
192 | ||
2e72b634 KS |
193 | struct mem_cgroup_threshold { |
194 | struct eventfd_ctx *eventfd; | |
3e32cb2e | 195 | unsigned long threshold; |
2e72b634 KS |
196 | }; |
197 | ||
9490ff27 | 198 | /* For threshold */ |
2e72b634 | 199 | struct mem_cgroup_threshold_ary { |
748dad36 | 200 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 201 | int current_threshold; |
2e72b634 KS |
202 | /* Size of entries[] */ |
203 | unsigned int size; | |
204 | /* Array of thresholds */ | |
205 | struct mem_cgroup_threshold entries[0]; | |
206 | }; | |
2c488db2 KS |
207 | |
208 | struct mem_cgroup_thresholds { | |
209 | /* Primary thresholds array */ | |
210 | struct mem_cgroup_threshold_ary *primary; | |
211 | /* | |
212 | * Spare threshold array. | |
213 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
214 | * It must be able to store at least primary->size - 1 entries. | |
215 | */ | |
216 | struct mem_cgroup_threshold_ary *spare; | |
217 | }; | |
218 | ||
9490ff27 KH |
219 | /* for OOM */ |
220 | struct mem_cgroup_eventfd_list { | |
221 | struct list_head list; | |
222 | struct eventfd_ctx *eventfd; | |
223 | }; | |
2e72b634 | 224 | |
79bd9814 TH |
225 | /* |
226 | * cgroup_event represents events which userspace want to receive. | |
227 | */ | |
3bc942f3 | 228 | struct mem_cgroup_event { |
79bd9814 | 229 | /* |
59b6f873 | 230 | * memcg which the event belongs to. |
79bd9814 | 231 | */ |
59b6f873 | 232 | struct mem_cgroup *memcg; |
79bd9814 TH |
233 | /* |
234 | * eventfd to signal userspace about the event. | |
235 | */ | |
236 | struct eventfd_ctx *eventfd; | |
237 | /* | |
238 | * Each of these stored in a list by the cgroup. | |
239 | */ | |
240 | struct list_head list; | |
fba94807 TH |
241 | /* |
242 | * register_event() callback will be used to add new userspace | |
243 | * waiter for changes related to this event. Use eventfd_signal() | |
244 | * on eventfd to send notification to userspace. | |
245 | */ | |
59b6f873 | 246 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 247 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
248 | /* |
249 | * unregister_event() callback will be called when userspace closes | |
250 | * the eventfd or on cgroup removing. This callback must be set, | |
251 | * if you want provide notification functionality. | |
252 | */ | |
59b6f873 | 253 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 254 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
255 | /* |
256 | * All fields below needed to unregister event when | |
257 | * userspace closes eventfd. | |
258 | */ | |
259 | poll_table pt; | |
260 | wait_queue_head_t *wqh; | |
261 | wait_queue_t wait; | |
262 | struct work_struct remove; | |
263 | }; | |
264 | ||
c0ff4b85 R |
265 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
266 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 267 | |
8cdea7c0 BS |
268 | /* |
269 | * The memory controller data structure. The memory controller controls both | |
270 | * page cache and RSS per cgroup. We would eventually like to provide | |
271 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
272 | * to help the administrator determine what knobs to tune. | |
273 | * | |
274 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
275 | * we hit the water mark. May be even add a low water mark, such that |
276 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
277 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
278 | */ |
279 | struct mem_cgroup { | |
280 | struct cgroup_subsys_state css; | |
3e32cb2e JW |
281 | |
282 | /* Accounted resources */ | |
283 | struct page_counter memory; | |
284 | struct page_counter memsw; | |
285 | struct page_counter kmem; | |
286 | ||
287 | unsigned long soft_limit; | |
59927fb9 | 288 | |
70ddf637 AV |
289 | /* vmpressure notifications */ |
290 | struct vmpressure vmpressure; | |
291 | ||
2f7dd7a4 JW |
292 | /* css_online() has been completed */ |
293 | int initialized; | |
294 | ||
18f59ea7 BS |
295 | /* |
296 | * Should the accounting and control be hierarchical, per subtree? | |
297 | */ | |
298 | bool use_hierarchy; | |
510fc4e1 | 299 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
300 | |
301 | bool oom_lock; | |
302 | atomic_t under_oom; | |
3812c8c8 | 303 | atomic_t oom_wakeups; |
79dfdacc | 304 | |
1f4c025b | 305 | int swappiness; |
3c11ecf4 KH |
306 | /* OOM-Killer disable */ |
307 | int oom_kill_disable; | |
a7885eb8 | 308 | |
2e72b634 KS |
309 | /* protect arrays of thresholds */ |
310 | struct mutex thresholds_lock; | |
311 | ||
312 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 313 | struct mem_cgroup_thresholds thresholds; |
907860ed | 314 | |
2e72b634 | 315 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 316 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 317 | |
9490ff27 KH |
318 | /* For oom notifier event fd */ |
319 | struct list_head oom_notify; | |
185efc0f | 320 | |
7dc74be0 DN |
321 | /* |
322 | * Should we move charges of a task when a task is moved into this | |
323 | * mem_cgroup ? And what type of charges should we move ? | |
324 | */ | |
f894ffa8 | 325 | unsigned long move_charge_at_immigrate; |
619d094b KH |
326 | /* |
327 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
328 | */ | |
329 | atomic_t moving_account; | |
312734c0 KH |
330 | /* taken only while moving_account > 0 */ |
331 | spinlock_t move_lock; | |
d52aa412 | 332 | /* |
c62b1a3b | 333 | * percpu counter. |
d52aa412 | 334 | */ |
3a7951b4 | 335 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
336 | /* |
337 | * used when a cpu is offlined or other synchronizations | |
338 | * See mem_cgroup_read_stat(). | |
339 | */ | |
340 | struct mem_cgroup_stat_cpu nocpu_base; | |
341 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 342 | |
4bd2c1ee | 343 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
2e685cad | 344 | struct cg_proto tcp_mem; |
d1a4c0b3 | 345 | #endif |
2633d7a0 | 346 | #if defined(CONFIG_MEMCG_KMEM) |
bd673145 VD |
347 | /* analogous to slab_common's slab_caches list, but per-memcg; |
348 | * protected by memcg_slab_mutex */ | |
2633d7a0 | 349 | struct list_head memcg_slab_caches; |
2633d7a0 GC |
350 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ |
351 | int kmemcg_id; | |
352 | #endif | |
45cf7ebd GC |
353 | |
354 | int last_scanned_node; | |
355 | #if MAX_NUMNODES > 1 | |
356 | nodemask_t scan_nodes; | |
357 | atomic_t numainfo_events; | |
358 | atomic_t numainfo_updating; | |
359 | #endif | |
70ddf637 | 360 | |
fba94807 TH |
361 | /* List of events which userspace want to receive */ |
362 | struct list_head event_list; | |
363 | spinlock_t event_list_lock; | |
364 | ||
54f72fe0 JW |
365 | struct mem_cgroup_per_node *nodeinfo[0]; |
366 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
367 | }; |
368 | ||
510fc4e1 GC |
369 | /* internal only representation about the status of kmem accounting. */ |
370 | enum { | |
6de64beb | 371 | KMEM_ACCOUNTED_ACTIVE, /* accounted by this cgroup itself */ |
7de37682 | 372 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
373 | }; |
374 | ||
510fc4e1 GC |
375 | #ifdef CONFIG_MEMCG_KMEM |
376 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
377 | { | |
378 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
379 | } | |
7de37682 GC |
380 | |
381 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
382 | { | |
383 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
384 | } | |
385 | ||
386 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) | |
387 | { | |
10d5ebf4 LZ |
388 | /* |
389 | * Our caller must use css_get() first, because memcg_uncharge_kmem() | |
390 | * will call css_put() if it sees the memcg is dead. | |
391 | */ | |
392 | smp_wmb(); | |
7de37682 GC |
393 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
394 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
395 | } | |
396 | ||
397 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
398 | { | |
399 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
400 | &memcg->kmem_account_flags); | |
401 | } | |
510fc4e1 GC |
402 | #endif |
403 | ||
7dc74be0 DN |
404 | /* Stuffs for move charges at task migration. */ |
405 | /* | |
ee5e8472 GC |
406 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
407 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
408 | */ |
409 | enum move_type { | |
4ffef5fe | 410 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 411 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
412 | NR_MOVE_TYPE, |
413 | }; | |
414 | ||
4ffef5fe DN |
415 | /* "mc" and its members are protected by cgroup_mutex */ |
416 | static struct move_charge_struct { | |
b1dd693e | 417 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
418 | struct mem_cgroup *from; |
419 | struct mem_cgroup *to; | |
ee5e8472 | 420 | unsigned long immigrate_flags; |
4ffef5fe | 421 | unsigned long precharge; |
854ffa8d | 422 | unsigned long moved_charge; |
483c30b5 | 423 | unsigned long moved_swap; |
8033b97c DN |
424 | struct task_struct *moving_task; /* a task moving charges */ |
425 | wait_queue_head_t waitq; /* a waitq for other context */ | |
426 | } mc = { | |
2bd9bb20 | 427 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
428 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
429 | }; | |
4ffef5fe | 430 | |
90254a65 DN |
431 | static bool move_anon(void) |
432 | { | |
ee5e8472 | 433 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
434 | } |
435 | ||
87946a72 DN |
436 | static bool move_file(void) |
437 | { | |
ee5e8472 | 438 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
439 | } |
440 | ||
4e416953 BS |
441 | /* |
442 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
443 | * limit reclaim to prevent infinite loops, if they ever occur. | |
444 | */ | |
a0db00fc | 445 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 446 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 447 | |
217bc319 KH |
448 | enum charge_type { |
449 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 450 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 451 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 452 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
453 | NR_CHARGE_TYPE, |
454 | }; | |
455 | ||
8c7c6e34 | 456 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
457 | enum res_type { |
458 | _MEM, | |
459 | _MEMSWAP, | |
460 | _OOM_TYPE, | |
510fc4e1 | 461 | _KMEM, |
86ae53e1 GC |
462 | }; |
463 | ||
a0db00fc KS |
464 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
465 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 466 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
467 | /* Used for OOM nofiier */ |
468 | #define OOM_CONTROL (0) | |
8c7c6e34 | 469 | |
0999821b GC |
470 | /* |
471 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
472 | * As a consequence, any change that needs to protect against new child cgroups | |
473 | * appearing has to hold it as well. | |
474 | */ | |
475 | static DEFINE_MUTEX(memcg_create_mutex); | |
476 | ||
b2145145 WL |
477 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
478 | { | |
a7c6d554 | 479 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
480 | } |
481 | ||
70ddf637 AV |
482 | /* Some nice accessors for the vmpressure. */ |
483 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
484 | { | |
485 | if (!memcg) | |
486 | memcg = root_mem_cgroup; | |
487 | return &memcg->vmpressure; | |
488 | } | |
489 | ||
490 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
491 | { | |
492 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
493 | } | |
494 | ||
7ffc0edc MH |
495 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
496 | { | |
497 | return (memcg == root_mem_cgroup); | |
498 | } | |
499 | ||
4219b2da LZ |
500 | /* |
501 | * We restrict the id in the range of [1, 65535], so it can fit into | |
502 | * an unsigned short. | |
503 | */ | |
504 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
505 | ||
34c00c31 LZ |
506 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
507 | { | |
15a4c835 | 508 | return memcg->css.id; |
34c00c31 LZ |
509 | } |
510 | ||
511 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
512 | { | |
513 | struct cgroup_subsys_state *css; | |
514 | ||
7d699ddb | 515 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
516 | return mem_cgroup_from_css(css); |
517 | } | |
518 | ||
e1aab161 | 519 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 520 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 521 | |
e1aab161 GC |
522 | void sock_update_memcg(struct sock *sk) |
523 | { | |
376be5ff | 524 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 525 | struct mem_cgroup *memcg; |
3f134619 | 526 | struct cg_proto *cg_proto; |
e1aab161 GC |
527 | |
528 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
529 | ||
f3f511e1 GC |
530 | /* Socket cloning can throw us here with sk_cgrp already |
531 | * filled. It won't however, necessarily happen from | |
532 | * process context. So the test for root memcg given | |
533 | * the current task's memcg won't help us in this case. | |
534 | * | |
535 | * Respecting the original socket's memcg is a better | |
536 | * decision in this case. | |
537 | */ | |
538 | if (sk->sk_cgrp) { | |
539 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 540 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
541 | return; |
542 | } | |
543 | ||
e1aab161 GC |
544 | rcu_read_lock(); |
545 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 546 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 547 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
548 | memcg_proto_active(cg_proto) && |
549 | css_tryget_online(&memcg->css)) { | |
3f134619 | 550 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
551 | } |
552 | rcu_read_unlock(); | |
553 | } | |
554 | } | |
555 | EXPORT_SYMBOL(sock_update_memcg); | |
556 | ||
557 | void sock_release_memcg(struct sock *sk) | |
558 | { | |
376be5ff | 559 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
560 | struct mem_cgroup *memcg; |
561 | WARN_ON(!sk->sk_cgrp->memcg); | |
562 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 563 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
564 | } |
565 | } | |
d1a4c0b3 GC |
566 | |
567 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
568 | { | |
569 | if (!memcg || mem_cgroup_is_root(memcg)) | |
570 | return NULL; | |
571 | ||
2e685cad | 572 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
573 | } |
574 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 575 | |
3f134619 GC |
576 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
577 | { | |
2e685cad | 578 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
579 | return; |
580 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
581 | } | |
582 | #else | |
583 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
584 | { | |
585 | } | |
586 | #endif | |
587 | ||
a8964b9b | 588 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
589 | /* |
590 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
b8627835 LZ |
591 | * The main reason for not using cgroup id for this: |
592 | * this works better in sparse environments, where we have a lot of memcgs, | |
593 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
594 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
595 | * 200 entry array for that. | |
55007d84 GC |
596 | * |
597 | * The current size of the caches array is stored in | |
598 | * memcg_limited_groups_array_size. It will double each time we have to | |
599 | * increase it. | |
600 | */ | |
601 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
602 | int memcg_limited_groups_array_size; |
603 | ||
55007d84 GC |
604 | /* |
605 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
606 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
607 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
608 | * tunable, but that is strictly not necessary. | |
609 | * | |
b8627835 | 610 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
611 | * this constant directly from cgroup, but it is understandable that this is |
612 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 613 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
614 | * increase ours as well if it increases. |
615 | */ | |
616 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 617 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 618 | |
d7f25f8a GC |
619 | /* |
620 | * A lot of the calls to the cache allocation functions are expected to be | |
621 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
622 | * conditional to this static branch, we'll have to allow modules that does | |
623 | * kmem_cache_alloc and the such to see this symbol as well | |
624 | */ | |
a8964b9b | 625 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 626 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 627 | |
f3bb3043 VD |
628 | static void memcg_free_cache_id(int id); |
629 | ||
a8964b9b GC |
630 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
631 | { | |
55007d84 | 632 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 633 | static_key_slow_dec(&memcg_kmem_enabled_key); |
f3bb3043 | 634 | memcg_free_cache_id(memcg->kmemcg_id); |
55007d84 | 635 | } |
bea207c8 GC |
636 | /* |
637 | * This check can't live in kmem destruction function, | |
638 | * since the charges will outlive the cgroup | |
639 | */ | |
3e32cb2e | 640 | WARN_ON(page_counter_read(&memcg->kmem)); |
a8964b9b GC |
641 | } |
642 | #else | |
643 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
644 | { | |
645 | } | |
646 | #endif /* CONFIG_MEMCG_KMEM */ | |
647 | ||
648 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
649 | { | |
650 | disarm_sock_keys(memcg); | |
651 | disarm_kmem_keys(memcg); | |
652 | } | |
653 | ||
c0ff4b85 | 654 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 655 | |
f64c3f54 | 656 | static struct mem_cgroup_per_zone * |
e231875b | 657 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 658 | { |
e231875b JZ |
659 | int nid = zone_to_nid(zone); |
660 | int zid = zone_idx(zone); | |
661 | ||
54f72fe0 | 662 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
663 | } |
664 | ||
c0ff4b85 | 665 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 666 | { |
c0ff4b85 | 667 | return &memcg->css; |
d324236b WF |
668 | } |
669 | ||
f64c3f54 | 670 | static struct mem_cgroup_per_zone * |
e231875b | 671 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 672 | { |
97a6c37b JW |
673 | int nid = page_to_nid(page); |
674 | int zid = page_zonenum(page); | |
f64c3f54 | 675 | |
e231875b | 676 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
677 | } |
678 | ||
bb4cc1a8 AM |
679 | static struct mem_cgroup_tree_per_zone * |
680 | soft_limit_tree_node_zone(int nid, int zid) | |
681 | { | |
682 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
683 | } | |
684 | ||
685 | static struct mem_cgroup_tree_per_zone * | |
686 | soft_limit_tree_from_page(struct page *page) | |
687 | { | |
688 | int nid = page_to_nid(page); | |
689 | int zid = page_zonenum(page); | |
690 | ||
691 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
692 | } | |
693 | ||
cf2c8127 JW |
694 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
695 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 696 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
697 | { |
698 | struct rb_node **p = &mctz->rb_root.rb_node; | |
699 | struct rb_node *parent = NULL; | |
700 | struct mem_cgroup_per_zone *mz_node; | |
701 | ||
702 | if (mz->on_tree) | |
703 | return; | |
704 | ||
705 | mz->usage_in_excess = new_usage_in_excess; | |
706 | if (!mz->usage_in_excess) | |
707 | return; | |
708 | while (*p) { | |
709 | parent = *p; | |
710 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
711 | tree_node); | |
712 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
713 | p = &(*p)->rb_left; | |
714 | /* | |
715 | * We can't avoid mem cgroups that are over their soft | |
716 | * limit by the same amount | |
717 | */ | |
718 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
719 | p = &(*p)->rb_right; | |
720 | } | |
721 | rb_link_node(&mz->tree_node, parent, p); | |
722 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
723 | mz->on_tree = true; | |
724 | } | |
725 | ||
cf2c8127 JW |
726 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
727 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
728 | { |
729 | if (!mz->on_tree) | |
730 | return; | |
731 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
732 | mz->on_tree = false; | |
733 | } | |
734 | ||
cf2c8127 JW |
735 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
736 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 737 | { |
0a31bc97 JW |
738 | unsigned long flags; |
739 | ||
740 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 741 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 742 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
743 | } |
744 | ||
3e32cb2e JW |
745 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
746 | { | |
747 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
748 | unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit); | |
749 | unsigned long excess = 0; | |
750 | ||
751 | if (nr_pages > soft_limit) | |
752 | excess = nr_pages - soft_limit; | |
753 | ||
754 | return excess; | |
755 | } | |
bb4cc1a8 AM |
756 | |
757 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
758 | { | |
3e32cb2e | 759 | unsigned long excess; |
bb4cc1a8 AM |
760 | struct mem_cgroup_per_zone *mz; |
761 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 762 | |
e231875b | 763 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
764 | /* |
765 | * Necessary to update all ancestors when hierarchy is used. | |
766 | * because their event counter is not touched. | |
767 | */ | |
768 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 769 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 770 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
771 | /* |
772 | * We have to update the tree if mz is on RB-tree or | |
773 | * mem is over its softlimit. | |
774 | */ | |
775 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
776 | unsigned long flags; |
777 | ||
778 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
779 | /* if on-tree, remove it */ |
780 | if (mz->on_tree) | |
cf2c8127 | 781 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
782 | /* |
783 | * Insert again. mz->usage_in_excess will be updated. | |
784 | * If excess is 0, no tree ops. | |
785 | */ | |
cf2c8127 | 786 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 787 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
788 | } |
789 | } | |
790 | } | |
791 | ||
792 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
793 | { | |
bb4cc1a8 | 794 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
795 | struct mem_cgroup_per_zone *mz; |
796 | int nid, zid; | |
bb4cc1a8 | 797 | |
e231875b JZ |
798 | for_each_node(nid) { |
799 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
800 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
801 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 802 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
803 | } |
804 | } | |
805 | } | |
806 | ||
807 | static struct mem_cgroup_per_zone * | |
808 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
809 | { | |
810 | struct rb_node *rightmost = NULL; | |
811 | struct mem_cgroup_per_zone *mz; | |
812 | ||
813 | retry: | |
814 | mz = NULL; | |
815 | rightmost = rb_last(&mctz->rb_root); | |
816 | if (!rightmost) | |
817 | goto done; /* Nothing to reclaim from */ | |
818 | ||
819 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
820 | /* | |
821 | * Remove the node now but someone else can add it back, | |
822 | * we will to add it back at the end of reclaim to its correct | |
823 | * position in the tree. | |
824 | */ | |
cf2c8127 | 825 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 826 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 827 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
828 | goto retry; |
829 | done: | |
830 | return mz; | |
831 | } | |
832 | ||
833 | static struct mem_cgroup_per_zone * | |
834 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
835 | { | |
836 | struct mem_cgroup_per_zone *mz; | |
837 | ||
0a31bc97 | 838 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 839 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 840 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
841 | return mz; |
842 | } | |
843 | ||
711d3d2c KH |
844 | /* |
845 | * Implementation Note: reading percpu statistics for memcg. | |
846 | * | |
847 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
848 | * synchronization to implement "quick" read. There are trade-off between | |
849 | * reading cost and precision of value. Then, we may have a chance to implement | |
850 | * a periodic synchronizion of counter in memcg's counter. | |
851 | * | |
852 | * But this _read() function is used for user interface now. The user accounts | |
853 | * memory usage by memory cgroup and he _always_ requires exact value because | |
854 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
855 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
856 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
857 | * | |
858 | * If there are kernel internal actions which can make use of some not-exact | |
859 | * value, and reading all cpu value can be performance bottleneck in some | |
860 | * common workload, threashold and synchonization as vmstat[] should be | |
861 | * implemented. | |
862 | */ | |
c0ff4b85 | 863 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 864 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 865 | { |
7a159cc9 | 866 | long val = 0; |
c62b1a3b | 867 | int cpu; |
c62b1a3b | 868 | |
711d3d2c KH |
869 | get_online_cpus(); |
870 | for_each_online_cpu(cpu) | |
c0ff4b85 | 871 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 872 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
873 | spin_lock(&memcg->pcp_counter_lock); |
874 | val += memcg->nocpu_base.count[idx]; | |
875 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
876 | #endif |
877 | put_online_cpus(); | |
c62b1a3b KH |
878 | return val; |
879 | } | |
880 | ||
c0ff4b85 | 881 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
882 | enum mem_cgroup_events_index idx) |
883 | { | |
884 | unsigned long val = 0; | |
885 | int cpu; | |
886 | ||
9c567512 | 887 | get_online_cpus(); |
e9f8974f | 888 | for_each_online_cpu(cpu) |
c0ff4b85 | 889 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 890 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
891 | spin_lock(&memcg->pcp_counter_lock); |
892 | val += memcg->nocpu_base.events[idx]; | |
893 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 894 | #endif |
9c567512 | 895 | put_online_cpus(); |
e9f8974f JW |
896 | return val; |
897 | } | |
898 | ||
c0ff4b85 | 899 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 900 | struct page *page, |
0a31bc97 | 901 | int nr_pages) |
d52aa412 | 902 | { |
b2402857 KH |
903 | /* |
904 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
905 | * counted as CACHE even if it's on ANON LRU. | |
906 | */ | |
0a31bc97 | 907 | if (PageAnon(page)) |
b2402857 | 908 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 909 | nr_pages); |
d52aa412 | 910 | else |
b2402857 | 911 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 912 | nr_pages); |
55e462b0 | 913 | |
b070e65c DR |
914 | if (PageTransHuge(page)) |
915 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
916 | nr_pages); | |
917 | ||
e401f176 KH |
918 | /* pagein of a big page is an event. So, ignore page size */ |
919 | if (nr_pages > 0) | |
c0ff4b85 | 920 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 921 | else { |
c0ff4b85 | 922 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
923 | nr_pages = -nr_pages; /* for event */ |
924 | } | |
e401f176 | 925 | |
13114716 | 926 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
927 | } |
928 | ||
e231875b | 929 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
930 | { |
931 | struct mem_cgroup_per_zone *mz; | |
932 | ||
933 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
934 | return mz->lru_size[lru]; | |
935 | } | |
936 | ||
e231875b JZ |
937 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
938 | int nid, | |
939 | unsigned int lru_mask) | |
bb2a0de9 | 940 | { |
e231875b | 941 | unsigned long nr = 0; |
889976db YH |
942 | int zid; |
943 | ||
e231875b | 944 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 945 | |
e231875b JZ |
946 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
947 | struct mem_cgroup_per_zone *mz; | |
948 | enum lru_list lru; | |
949 | ||
950 | for_each_lru(lru) { | |
951 | if (!(BIT(lru) & lru_mask)) | |
952 | continue; | |
953 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
954 | nr += mz->lru_size[lru]; | |
955 | } | |
956 | } | |
957 | return nr; | |
889976db | 958 | } |
bb2a0de9 | 959 | |
c0ff4b85 | 960 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 961 | unsigned int lru_mask) |
6d12e2d8 | 962 | { |
e231875b | 963 | unsigned long nr = 0; |
889976db | 964 | int nid; |
6d12e2d8 | 965 | |
31aaea4a | 966 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
967 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
968 | return nr; | |
d52aa412 KH |
969 | } |
970 | ||
f53d7ce3 JW |
971 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
972 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
973 | { |
974 | unsigned long val, next; | |
975 | ||
13114716 | 976 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 977 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 978 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
979 | if ((long)next - (long)val < 0) { |
980 | switch (target) { | |
981 | case MEM_CGROUP_TARGET_THRESH: | |
982 | next = val + THRESHOLDS_EVENTS_TARGET; | |
983 | break; | |
bb4cc1a8 AM |
984 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
985 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
986 | break; | |
f53d7ce3 JW |
987 | case MEM_CGROUP_TARGET_NUMAINFO: |
988 | next = val + NUMAINFO_EVENTS_TARGET; | |
989 | break; | |
990 | default: | |
991 | break; | |
992 | } | |
993 | __this_cpu_write(memcg->stat->targets[target], next); | |
994 | return true; | |
7a159cc9 | 995 | } |
f53d7ce3 | 996 | return false; |
d2265e6f KH |
997 | } |
998 | ||
999 | /* | |
1000 | * Check events in order. | |
1001 | * | |
1002 | */ | |
c0ff4b85 | 1003 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
1004 | { |
1005 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
1006 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1007 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 1008 | bool do_softlimit; |
82b3f2a7 | 1009 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 1010 | |
bb4cc1a8 AM |
1011 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
1012 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
1013 | #if MAX_NUMNODES > 1 |
1014 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1015 | MEM_CGROUP_TARGET_NUMAINFO); | |
1016 | #endif | |
c0ff4b85 | 1017 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
1018 | if (unlikely(do_softlimit)) |
1019 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1020 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1021 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1022 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1023 | #endif |
0a31bc97 | 1024 | } |
d2265e6f KH |
1025 | } |
1026 | ||
cf475ad2 | 1027 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1028 | { |
31a78f23 BS |
1029 | /* |
1030 | * mm_update_next_owner() may clear mm->owner to NULL | |
1031 | * if it races with swapoff, page migration, etc. | |
1032 | * So this can be called with p == NULL. | |
1033 | */ | |
1034 | if (unlikely(!p)) | |
1035 | return NULL; | |
1036 | ||
073219e9 | 1037 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1038 | } |
1039 | ||
df381975 | 1040 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1041 | { |
c0ff4b85 | 1042 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 1043 | |
54595fe2 KH |
1044 | rcu_read_lock(); |
1045 | do { | |
6f6acb00 MH |
1046 | /* |
1047 | * Page cache insertions can happen withou an | |
1048 | * actual mm context, e.g. during disk probing | |
1049 | * on boot, loopback IO, acct() writes etc. | |
1050 | */ | |
1051 | if (unlikely(!mm)) | |
df381975 | 1052 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1053 | else { |
1054 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1055 | if (unlikely(!memcg)) | |
1056 | memcg = root_mem_cgroup; | |
1057 | } | |
ec903c0c | 1058 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1059 | rcu_read_unlock(); |
c0ff4b85 | 1060 | return memcg; |
54595fe2 KH |
1061 | } |
1062 | ||
5660048c JW |
1063 | /** |
1064 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1065 | * @root: hierarchy root | |
1066 | * @prev: previously returned memcg, NULL on first invocation | |
1067 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1068 | * | |
1069 | * Returns references to children of the hierarchy below @root, or | |
1070 | * @root itself, or %NULL after a full round-trip. | |
1071 | * | |
1072 | * Caller must pass the return value in @prev on subsequent | |
1073 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1074 | * to cancel a hierarchy walk before the round-trip is complete. | |
1075 | * | |
1076 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1077 | * divide up the memcgs in the hierarchy among all concurrent | |
1078 | * reclaimers operating on the same zone and priority. | |
1079 | */ | |
694fbc0f | 1080 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1081 | struct mem_cgroup *prev, |
694fbc0f | 1082 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1083 | { |
5ac8fb31 JW |
1084 | struct reclaim_iter *uninitialized_var(iter); |
1085 | struct cgroup_subsys_state *css = NULL; | |
9f3a0d09 | 1086 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1087 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1088 | |
694fbc0f AM |
1089 | if (mem_cgroup_disabled()) |
1090 | return NULL; | |
5660048c | 1091 | |
9f3a0d09 JW |
1092 | if (!root) |
1093 | root = root_mem_cgroup; | |
7d74b06f | 1094 | |
9f3a0d09 | 1095 | if (prev && !reclaim) |
5ac8fb31 | 1096 | pos = prev; |
14067bb3 | 1097 | |
9f3a0d09 JW |
1098 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1099 | if (prev) | |
5ac8fb31 | 1100 | goto out; |
694fbc0f | 1101 | return root; |
9f3a0d09 | 1102 | } |
14067bb3 | 1103 | |
542f85f9 | 1104 | rcu_read_lock(); |
5f578161 | 1105 | |
5ac8fb31 JW |
1106 | if (reclaim) { |
1107 | struct mem_cgroup_per_zone *mz; | |
1108 | ||
1109 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
1110 | iter = &mz->iter[reclaim->priority]; | |
1111 | ||
1112 | if (prev && reclaim->generation != iter->generation) | |
1113 | goto out_unlock; | |
1114 | ||
1115 | do { | |
1116 | pos = ACCESS_ONCE(iter->position); | |
1117 | /* | |
1118 | * A racing update may change the position and | |
1119 | * put the last reference, hence css_tryget(), | |
1120 | * or retry to see the updated position. | |
1121 | */ | |
1122 | } while (pos && !css_tryget(&pos->css)); | |
1123 | } | |
1124 | ||
1125 | if (pos) | |
1126 | css = &pos->css; | |
1127 | ||
1128 | for (;;) { | |
1129 | css = css_next_descendant_pre(css, &root->css); | |
1130 | if (!css) { | |
1131 | /* | |
1132 | * Reclaimers share the hierarchy walk, and a | |
1133 | * new one might jump in right at the end of | |
1134 | * the hierarchy - make sure they see at least | |
1135 | * one group and restart from the beginning. | |
1136 | */ | |
1137 | if (!prev) | |
1138 | continue; | |
1139 | break; | |
527a5ec9 | 1140 | } |
7d74b06f | 1141 | |
5ac8fb31 JW |
1142 | /* |
1143 | * Verify the css and acquire a reference. The root | |
1144 | * is provided by the caller, so we know it's alive | |
1145 | * and kicking, and don't take an extra reference. | |
1146 | */ | |
1147 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1148 | |
5ac8fb31 JW |
1149 | if (css == &root->css) |
1150 | break; | |
542f85f9 | 1151 | |
5ac8fb31 JW |
1152 | if (css_tryget_online(css)) { |
1153 | /* | |
1154 | * Make sure the memcg is initialized: | |
1155 | * mem_cgroup_css_online() orders the the | |
1156 | * initialization against setting the flag. | |
1157 | */ | |
1158 | if (smp_load_acquire(&memcg->initialized)) | |
1159 | break; | |
1160 | ||
1161 | css_put(css); | |
527a5ec9 | 1162 | } |
9f3a0d09 | 1163 | |
5ac8fb31 JW |
1164 | memcg = NULL; |
1165 | } | |
1166 | ||
1167 | if (reclaim) { | |
1168 | if (cmpxchg(&iter->position, pos, memcg) == pos) { | |
1169 | if (memcg) | |
1170 | css_get(&memcg->css); | |
1171 | if (pos) | |
1172 | css_put(&pos->css); | |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * pairs with css_tryget when dereferencing iter->position | |
1177 | * above. | |
1178 | */ | |
1179 | if (pos) | |
1180 | css_put(&pos->css); | |
1181 | ||
1182 | if (!memcg) | |
1183 | iter->generation++; | |
1184 | else if (!prev) | |
1185 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1186 | } |
5ac8fb31 | 1187 | |
542f85f9 MH |
1188 | out_unlock: |
1189 | rcu_read_unlock(); | |
5ac8fb31 | 1190 | out: |
c40046f3 MH |
1191 | if (prev && prev != root) |
1192 | css_put(&prev->css); | |
1193 | ||
9f3a0d09 | 1194 | return memcg; |
14067bb3 | 1195 | } |
7d74b06f | 1196 | |
5660048c JW |
1197 | /** |
1198 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1199 | * @root: hierarchy root | |
1200 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1201 | */ | |
1202 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1203 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1204 | { |
1205 | if (!root) | |
1206 | root = root_mem_cgroup; | |
1207 | if (prev && prev != root) | |
1208 | css_put(&prev->css); | |
1209 | } | |
7d74b06f | 1210 | |
9f3a0d09 JW |
1211 | /* |
1212 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1213 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1214 | * be used for reference counting. | |
1215 | */ | |
1216 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1217 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1218 | iter != NULL; \ |
527a5ec9 | 1219 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1220 | |
9f3a0d09 | 1221 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1222 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1223 | iter != NULL; \ |
527a5ec9 | 1224 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1225 | |
68ae564b | 1226 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1227 | { |
c0ff4b85 | 1228 | struct mem_cgroup *memcg; |
456f998e | 1229 | |
456f998e | 1230 | rcu_read_lock(); |
c0ff4b85 R |
1231 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1232 | if (unlikely(!memcg)) | |
456f998e YH |
1233 | goto out; |
1234 | ||
1235 | switch (idx) { | |
456f998e | 1236 | case PGFAULT: |
0e574a93 JW |
1237 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1238 | break; | |
1239 | case PGMAJFAULT: | |
1240 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1241 | break; |
1242 | default: | |
1243 | BUG(); | |
1244 | } | |
1245 | out: | |
1246 | rcu_read_unlock(); | |
1247 | } | |
68ae564b | 1248 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1249 | |
925b7673 JW |
1250 | /** |
1251 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1252 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1253 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1254 | * |
1255 | * Returns the lru list vector holding pages for the given @zone and | |
1256 | * @mem. This can be the global zone lruvec, if the memory controller | |
1257 | * is disabled. | |
1258 | */ | |
1259 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1260 | struct mem_cgroup *memcg) | |
1261 | { | |
1262 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1263 | struct lruvec *lruvec; |
925b7673 | 1264 | |
bea8c150 HD |
1265 | if (mem_cgroup_disabled()) { |
1266 | lruvec = &zone->lruvec; | |
1267 | goto out; | |
1268 | } | |
925b7673 | 1269 | |
e231875b | 1270 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1271 | lruvec = &mz->lruvec; |
1272 | out: | |
1273 | /* | |
1274 | * Since a node can be onlined after the mem_cgroup was created, | |
1275 | * we have to be prepared to initialize lruvec->zone here; | |
1276 | * and if offlined then reonlined, we need to reinitialize it. | |
1277 | */ | |
1278 | if (unlikely(lruvec->zone != zone)) | |
1279 | lruvec->zone = zone; | |
1280 | return lruvec; | |
925b7673 JW |
1281 | } |
1282 | ||
925b7673 | 1283 | /** |
fa9add64 | 1284 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1285 | * @page: the page |
fa9add64 | 1286 | * @zone: zone of the page |
925b7673 | 1287 | */ |
fa9add64 | 1288 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1289 | { |
08e552c6 | 1290 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1291 | struct mem_cgroup *memcg; |
1292 | struct page_cgroup *pc; | |
bea8c150 | 1293 | struct lruvec *lruvec; |
6d12e2d8 | 1294 | |
bea8c150 HD |
1295 | if (mem_cgroup_disabled()) { |
1296 | lruvec = &zone->lruvec; | |
1297 | goto out; | |
1298 | } | |
925b7673 | 1299 | |
08e552c6 | 1300 | pc = lookup_page_cgroup(page); |
38c5d72f | 1301 | memcg = pc->mem_cgroup; |
7512102c HD |
1302 | |
1303 | /* | |
fa9add64 | 1304 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1305 | * an uncharged page off lru does nothing to secure |
1306 | * its former mem_cgroup from sudden removal. | |
1307 | * | |
1308 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1309 | * under page_cgroup lock: between them, they make all uses | |
1310 | * of pc->mem_cgroup safe. | |
1311 | */ | |
fa9add64 | 1312 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1313 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1314 | ||
e231875b | 1315 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1316 | lruvec = &mz->lruvec; |
1317 | out: | |
1318 | /* | |
1319 | * Since a node can be onlined after the mem_cgroup was created, | |
1320 | * we have to be prepared to initialize lruvec->zone here; | |
1321 | * and if offlined then reonlined, we need to reinitialize it. | |
1322 | */ | |
1323 | if (unlikely(lruvec->zone != zone)) | |
1324 | lruvec->zone = zone; | |
1325 | return lruvec; | |
08e552c6 | 1326 | } |
b69408e8 | 1327 | |
925b7673 | 1328 | /** |
fa9add64 HD |
1329 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1330 | * @lruvec: mem_cgroup per zone lru vector | |
1331 | * @lru: index of lru list the page is sitting on | |
1332 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1333 | * |
fa9add64 HD |
1334 | * This function must be called when a page is added to or removed from an |
1335 | * lru list. | |
3f58a829 | 1336 | */ |
fa9add64 HD |
1337 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1338 | int nr_pages) | |
3f58a829 MK |
1339 | { |
1340 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1341 | unsigned long *lru_size; |
3f58a829 MK |
1342 | |
1343 | if (mem_cgroup_disabled()) | |
1344 | return; | |
1345 | ||
fa9add64 HD |
1346 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1347 | lru_size = mz->lru_size + lru; | |
1348 | *lru_size += nr_pages; | |
1349 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1350 | } |
544122e5 | 1351 | |
3e92041d | 1352 | /* |
c0ff4b85 | 1353 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1354 | * hierarchy subtree |
1355 | */ | |
c3ac9a8a JW |
1356 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1357 | struct mem_cgroup *memcg) | |
3e92041d | 1358 | { |
91c63734 JW |
1359 | if (root_memcg == memcg) |
1360 | return true; | |
3a981f48 | 1361 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1362 | return false; |
b47f77b5 | 1363 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1364 | } |
1365 | ||
1366 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1367 | struct mem_cgroup *memcg) | |
1368 | { | |
1369 | bool ret; | |
1370 | ||
91c63734 | 1371 | rcu_read_lock(); |
c3ac9a8a | 1372 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1373 | rcu_read_unlock(); |
1374 | return ret; | |
3e92041d MH |
1375 | } |
1376 | ||
ffbdccf5 DR |
1377 | bool task_in_mem_cgroup(struct task_struct *task, |
1378 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1379 | { |
0b7f569e | 1380 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1381 | struct task_struct *p; |
ffbdccf5 | 1382 | bool ret; |
4c4a2214 | 1383 | |
158e0a2d | 1384 | p = find_lock_task_mm(task); |
de077d22 | 1385 | if (p) { |
df381975 | 1386 | curr = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1387 | task_unlock(p); |
1388 | } else { | |
1389 | /* | |
1390 | * All threads may have already detached their mm's, but the oom | |
1391 | * killer still needs to detect if they have already been oom | |
1392 | * killed to prevent needlessly killing additional tasks. | |
1393 | */ | |
ffbdccf5 | 1394 | rcu_read_lock(); |
de077d22 DR |
1395 | curr = mem_cgroup_from_task(task); |
1396 | if (curr) | |
1397 | css_get(&curr->css); | |
ffbdccf5 | 1398 | rcu_read_unlock(); |
de077d22 | 1399 | } |
d31f56db | 1400 | /* |
c0ff4b85 | 1401 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1402 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1403 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1404 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1405 | */ |
c0ff4b85 | 1406 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1407 | css_put(&curr->css); |
4c4a2214 DR |
1408 | return ret; |
1409 | } | |
1410 | ||
c56d5c7d | 1411 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1412 | { |
9b272977 | 1413 | unsigned long inactive_ratio; |
14797e23 | 1414 | unsigned long inactive; |
9b272977 | 1415 | unsigned long active; |
c772be93 | 1416 | unsigned long gb; |
14797e23 | 1417 | |
4d7dcca2 HD |
1418 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1419 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1420 | |
c772be93 KM |
1421 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1422 | if (gb) | |
1423 | inactive_ratio = int_sqrt(10 * gb); | |
1424 | else | |
1425 | inactive_ratio = 1; | |
1426 | ||
9b272977 | 1427 | return inactive * inactive_ratio < active; |
14797e23 KM |
1428 | } |
1429 | ||
3e32cb2e | 1430 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1431 | container_of(counter, struct mem_cgroup, member) |
1432 | ||
19942822 | 1433 | /** |
9d11ea9f | 1434 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1435 | * @memcg: the memory cgroup |
19942822 | 1436 | * |
9d11ea9f | 1437 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1438 | * pages. |
19942822 | 1439 | */ |
c0ff4b85 | 1440 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1441 | { |
3e32cb2e JW |
1442 | unsigned long margin = 0; |
1443 | unsigned long count; | |
1444 | unsigned long limit; | |
9d11ea9f | 1445 | |
3e32cb2e JW |
1446 | count = page_counter_read(&memcg->memory); |
1447 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1448 | if (count < limit) | |
1449 | margin = limit - count; | |
1450 | ||
1451 | if (do_swap_account) { | |
1452 | count = page_counter_read(&memcg->memsw); | |
1453 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1454 | if (count <= limit) | |
1455 | margin = min(margin, limit - count); | |
1456 | } | |
1457 | ||
1458 | return margin; | |
19942822 JW |
1459 | } |
1460 | ||
1f4c025b | 1461 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1462 | { |
a7885eb8 | 1463 | /* root ? */ |
14208b0e | 1464 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1465 | return vm_swappiness; |
1466 | ||
bf1ff263 | 1467 | return memcg->swappiness; |
a7885eb8 KM |
1468 | } |
1469 | ||
619d094b KH |
1470 | /* |
1471 | * memcg->moving_account is used for checking possibility that some thread is | |
1472 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1473 | * a memcg, other threads should check memcg->moving_account under | |
1474 | * rcu_read_lock(), like this: | |
1475 | * | |
1476 | * CPU-A CPU-B | |
1477 | * rcu_read_lock() | |
1478 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1479 | * take heavy locks. | |
1480 | * synchronize_rcu() update something. | |
1481 | * rcu_read_unlock() | |
1482 | * start move here. | |
1483 | */ | |
4331f7d3 | 1484 | |
c0ff4b85 | 1485 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1486 | { |
619d094b | 1487 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1488 | synchronize_rcu(); |
1489 | } | |
1490 | ||
c0ff4b85 | 1491 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1492 | { |
619d094b KH |
1493 | /* |
1494 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1495 | * We check NULL in callee rather than caller. | |
1496 | */ | |
d7365e78 | 1497 | if (memcg) |
619d094b | 1498 | atomic_dec(&memcg->moving_account); |
32047e2a | 1499 | } |
619d094b | 1500 | |
32047e2a | 1501 | /* |
bdcbb659 | 1502 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1503 | * |
bdcbb659 QH |
1504 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1505 | * moving cgroups. This is for waiting at high-memory pressure | |
1506 | * caused by "move". | |
32047e2a | 1507 | */ |
c0ff4b85 | 1508 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1509 | { |
2bd9bb20 KH |
1510 | struct mem_cgroup *from; |
1511 | struct mem_cgroup *to; | |
4b534334 | 1512 | bool ret = false; |
2bd9bb20 KH |
1513 | /* |
1514 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1515 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1516 | */ | |
1517 | spin_lock(&mc.lock); | |
1518 | from = mc.from; | |
1519 | to = mc.to; | |
1520 | if (!from) | |
1521 | goto unlock; | |
3e92041d | 1522 | |
c0ff4b85 R |
1523 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1524 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1525 | unlock: |
1526 | spin_unlock(&mc.lock); | |
4b534334 KH |
1527 | return ret; |
1528 | } | |
1529 | ||
c0ff4b85 | 1530 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1531 | { |
1532 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1533 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1534 | DEFINE_WAIT(wait); |
1535 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1536 | /* moving charge context might have finished. */ | |
1537 | if (mc.moving_task) | |
1538 | schedule(); | |
1539 | finish_wait(&mc.waitq, &wait); | |
1540 | return true; | |
1541 | } | |
1542 | } | |
1543 | return false; | |
1544 | } | |
1545 | ||
312734c0 KH |
1546 | /* |
1547 | * Take this lock when | |
1548 | * - a code tries to modify page's memcg while it's USED. | |
1549 | * - a code tries to modify page state accounting in a memcg. | |
312734c0 KH |
1550 | */ |
1551 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1552 | unsigned long *flags) | |
1553 | { | |
1554 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1555 | } | |
1556 | ||
1557 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1558 | unsigned long *flags) | |
1559 | { | |
1560 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1561 | } | |
1562 | ||
58cf188e | 1563 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1564 | /** |
58cf188e | 1565 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1566 | * @memcg: The memory cgroup that went over limit |
1567 | * @p: Task that is going to be killed | |
1568 | * | |
1569 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1570 | * enabled | |
1571 | */ | |
1572 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1573 | { | |
e61734c5 | 1574 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1575 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1576 | struct mem_cgroup *iter; |
1577 | unsigned int i; | |
e222432b | 1578 | |
58cf188e | 1579 | if (!p) |
e222432b BS |
1580 | return; |
1581 | ||
08088cb9 | 1582 | mutex_lock(&oom_info_lock); |
e222432b BS |
1583 | rcu_read_lock(); |
1584 | ||
e61734c5 TH |
1585 | pr_info("Task in "); |
1586 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1587 | pr_info(" killed as a result of limit of "); | |
1588 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1589 | pr_info("\n"); | |
e222432b | 1590 | |
e222432b BS |
1591 | rcu_read_unlock(); |
1592 | ||
3e32cb2e JW |
1593 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1594 | K((u64)page_counter_read(&memcg->memory)), | |
1595 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1596 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1597 | K((u64)page_counter_read(&memcg->memsw)), | |
1598 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1599 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1600 | K((u64)page_counter_read(&memcg->kmem)), | |
1601 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1602 | |
1603 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1604 | pr_info("Memory cgroup stats for "); |
1605 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1606 | pr_cont(":"); |
1607 | ||
1608 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1609 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1610 | continue; | |
1611 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1612 | K(mem_cgroup_read_stat(iter, i))); | |
1613 | } | |
1614 | ||
1615 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1616 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1617 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1618 | ||
1619 | pr_cont("\n"); | |
1620 | } | |
08088cb9 | 1621 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1622 | } |
1623 | ||
81d39c20 KH |
1624 | /* |
1625 | * This function returns the number of memcg under hierarchy tree. Returns | |
1626 | * 1(self count) if no children. | |
1627 | */ | |
c0ff4b85 | 1628 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1629 | { |
1630 | int num = 0; | |
7d74b06f KH |
1631 | struct mem_cgroup *iter; |
1632 | ||
c0ff4b85 | 1633 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1634 | num++; |
81d39c20 KH |
1635 | return num; |
1636 | } | |
1637 | ||
a63d83f4 DR |
1638 | /* |
1639 | * Return the memory (and swap, if configured) limit for a memcg. | |
1640 | */ | |
3e32cb2e | 1641 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1642 | { |
3e32cb2e | 1643 | unsigned long limit; |
a63d83f4 | 1644 | |
3e32cb2e | 1645 | limit = memcg->memory.limit; |
9a5a8f19 | 1646 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1647 | unsigned long memsw_limit; |
9a5a8f19 | 1648 | |
3e32cb2e JW |
1649 | memsw_limit = memcg->memsw.limit; |
1650 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1651 | } |
9a5a8f19 | 1652 | return limit; |
a63d83f4 DR |
1653 | } |
1654 | ||
19965460 DR |
1655 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1656 | int order) | |
9cbb78bb DR |
1657 | { |
1658 | struct mem_cgroup *iter; | |
1659 | unsigned long chosen_points = 0; | |
1660 | unsigned long totalpages; | |
1661 | unsigned int points = 0; | |
1662 | struct task_struct *chosen = NULL; | |
1663 | ||
876aafbf | 1664 | /* |
465adcf1 DR |
1665 | * If current has a pending SIGKILL or is exiting, then automatically |
1666 | * select it. The goal is to allow it to allocate so that it may | |
1667 | * quickly exit and free its memory. | |
876aafbf | 1668 | */ |
465adcf1 | 1669 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1670 | set_thread_flag(TIF_MEMDIE); |
1671 | return; | |
1672 | } | |
1673 | ||
1674 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1675 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1676 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1677 | struct css_task_iter it; |
9cbb78bb DR |
1678 | struct task_struct *task; |
1679 | ||
72ec7029 TH |
1680 | css_task_iter_start(&iter->css, &it); |
1681 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1682 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1683 | false)) { | |
1684 | case OOM_SCAN_SELECT: | |
1685 | if (chosen) | |
1686 | put_task_struct(chosen); | |
1687 | chosen = task; | |
1688 | chosen_points = ULONG_MAX; | |
1689 | get_task_struct(chosen); | |
1690 | /* fall through */ | |
1691 | case OOM_SCAN_CONTINUE: | |
1692 | continue; | |
1693 | case OOM_SCAN_ABORT: | |
72ec7029 | 1694 | css_task_iter_end(&it); |
9cbb78bb DR |
1695 | mem_cgroup_iter_break(memcg, iter); |
1696 | if (chosen) | |
1697 | put_task_struct(chosen); | |
1698 | return; | |
1699 | case OOM_SCAN_OK: | |
1700 | break; | |
1701 | }; | |
1702 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1703 | if (!points || points < chosen_points) |
1704 | continue; | |
1705 | /* Prefer thread group leaders for display purposes */ | |
1706 | if (points == chosen_points && | |
1707 | thread_group_leader(chosen)) | |
1708 | continue; | |
1709 | ||
1710 | if (chosen) | |
1711 | put_task_struct(chosen); | |
1712 | chosen = task; | |
1713 | chosen_points = points; | |
1714 | get_task_struct(chosen); | |
9cbb78bb | 1715 | } |
72ec7029 | 1716 | css_task_iter_end(&it); |
9cbb78bb DR |
1717 | } |
1718 | ||
1719 | if (!chosen) | |
1720 | return; | |
1721 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1722 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1723 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1724 | } |
1725 | ||
4d0c066d KH |
1726 | /** |
1727 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1728 | * @memcg: the target memcg |
4d0c066d KH |
1729 | * @nid: the node ID to be checked. |
1730 | * @noswap : specify true here if the user wants flle only information. | |
1731 | * | |
1732 | * This function returns whether the specified memcg contains any | |
1733 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1734 | * pages in the node. | |
1735 | */ | |
c0ff4b85 | 1736 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1737 | int nid, bool noswap) |
1738 | { | |
c0ff4b85 | 1739 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1740 | return true; |
1741 | if (noswap || !total_swap_pages) | |
1742 | return false; | |
c0ff4b85 | 1743 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1744 | return true; |
1745 | return false; | |
1746 | ||
1747 | } | |
bb4cc1a8 | 1748 | #if MAX_NUMNODES > 1 |
889976db YH |
1749 | |
1750 | /* | |
1751 | * Always updating the nodemask is not very good - even if we have an empty | |
1752 | * list or the wrong list here, we can start from some node and traverse all | |
1753 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1754 | * | |
1755 | */ | |
c0ff4b85 | 1756 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1757 | { |
1758 | int nid; | |
453a9bf3 KH |
1759 | /* |
1760 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1761 | * pagein/pageout changes since the last update. | |
1762 | */ | |
c0ff4b85 | 1763 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1764 | return; |
c0ff4b85 | 1765 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1766 | return; |
1767 | ||
889976db | 1768 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1769 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1770 | |
31aaea4a | 1771 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1772 | |
c0ff4b85 R |
1773 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1774 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1775 | } |
453a9bf3 | 1776 | |
c0ff4b85 R |
1777 | atomic_set(&memcg->numainfo_events, 0); |
1778 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1779 | } |
1780 | ||
1781 | /* | |
1782 | * Selecting a node where we start reclaim from. Because what we need is just | |
1783 | * reducing usage counter, start from anywhere is O,K. Considering | |
1784 | * memory reclaim from current node, there are pros. and cons. | |
1785 | * | |
1786 | * Freeing memory from current node means freeing memory from a node which | |
1787 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1788 | * hit limits, it will see a contention on a node. But freeing from remote | |
1789 | * node means more costs for memory reclaim because of memory latency. | |
1790 | * | |
1791 | * Now, we use round-robin. Better algorithm is welcomed. | |
1792 | */ | |
c0ff4b85 | 1793 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1794 | { |
1795 | int node; | |
1796 | ||
c0ff4b85 R |
1797 | mem_cgroup_may_update_nodemask(memcg); |
1798 | node = memcg->last_scanned_node; | |
889976db | 1799 | |
c0ff4b85 | 1800 | node = next_node(node, memcg->scan_nodes); |
889976db | 1801 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1802 | node = first_node(memcg->scan_nodes); |
889976db YH |
1803 | /* |
1804 | * We call this when we hit limit, not when pages are added to LRU. | |
1805 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1806 | * memcg is too small and all pages are not on LRU. In that case, | |
1807 | * we use curret node. | |
1808 | */ | |
1809 | if (unlikely(node == MAX_NUMNODES)) | |
1810 | node = numa_node_id(); | |
1811 | ||
c0ff4b85 | 1812 | memcg->last_scanned_node = node; |
889976db YH |
1813 | return node; |
1814 | } | |
1815 | ||
bb4cc1a8 AM |
1816 | /* |
1817 | * Check all nodes whether it contains reclaimable pages or not. | |
1818 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1819 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1820 | * enough new information. We need to do double check. | |
1821 | */ | |
1822 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1823 | { | |
1824 | int nid; | |
1825 | ||
1826 | /* | |
1827 | * quick check...making use of scan_node. | |
1828 | * We can skip unused nodes. | |
1829 | */ | |
1830 | if (!nodes_empty(memcg->scan_nodes)) { | |
1831 | for (nid = first_node(memcg->scan_nodes); | |
1832 | nid < MAX_NUMNODES; | |
1833 | nid = next_node(nid, memcg->scan_nodes)) { | |
1834 | ||
1835 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1836 | return true; | |
1837 | } | |
1838 | } | |
1839 | /* | |
1840 | * Check rest of nodes. | |
1841 | */ | |
1842 | for_each_node_state(nid, N_MEMORY) { | |
1843 | if (node_isset(nid, memcg->scan_nodes)) | |
1844 | continue; | |
1845 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1846 | return true; | |
1847 | } | |
1848 | return false; | |
1849 | } | |
1850 | ||
889976db | 1851 | #else |
c0ff4b85 | 1852 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1853 | { |
1854 | return 0; | |
1855 | } | |
4d0c066d | 1856 | |
bb4cc1a8 AM |
1857 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
1858 | { | |
1859 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
1860 | } | |
889976db YH |
1861 | #endif |
1862 | ||
0608f43d AM |
1863 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1864 | struct zone *zone, | |
1865 | gfp_t gfp_mask, | |
1866 | unsigned long *total_scanned) | |
1867 | { | |
1868 | struct mem_cgroup *victim = NULL; | |
1869 | int total = 0; | |
1870 | int loop = 0; | |
1871 | unsigned long excess; | |
1872 | unsigned long nr_scanned; | |
1873 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1874 | .zone = zone, | |
1875 | .priority = 0, | |
1876 | }; | |
1877 | ||
3e32cb2e | 1878 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1879 | |
1880 | while (1) { | |
1881 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1882 | if (!victim) { | |
1883 | loop++; | |
1884 | if (loop >= 2) { | |
1885 | /* | |
1886 | * If we have not been able to reclaim | |
1887 | * anything, it might because there are | |
1888 | * no reclaimable pages under this hierarchy | |
1889 | */ | |
1890 | if (!total) | |
1891 | break; | |
1892 | /* | |
1893 | * We want to do more targeted reclaim. | |
1894 | * excess >> 2 is not to excessive so as to | |
1895 | * reclaim too much, nor too less that we keep | |
1896 | * coming back to reclaim from this cgroup | |
1897 | */ | |
1898 | if (total >= (excess >> 2) || | |
1899 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1900 | break; | |
1901 | } | |
1902 | continue; | |
1903 | } | |
1904 | if (!mem_cgroup_reclaimable(victim, false)) | |
1905 | continue; | |
1906 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
1907 | zone, &nr_scanned); | |
1908 | *total_scanned += nr_scanned; | |
3e32cb2e | 1909 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1910 | break; |
6d61ef40 | 1911 | } |
0608f43d AM |
1912 | mem_cgroup_iter_break(root_memcg, victim); |
1913 | return total; | |
6d61ef40 BS |
1914 | } |
1915 | ||
0056f4e6 JW |
1916 | #ifdef CONFIG_LOCKDEP |
1917 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1918 | .name = "memcg_oom_lock", | |
1919 | }; | |
1920 | #endif | |
1921 | ||
fb2a6fc5 JW |
1922 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1923 | ||
867578cb KH |
1924 | /* |
1925 | * Check OOM-Killer is already running under our hierarchy. | |
1926 | * If someone is running, return false. | |
1927 | */ | |
fb2a6fc5 | 1928 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1929 | { |
79dfdacc | 1930 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1931 | |
fb2a6fc5 JW |
1932 | spin_lock(&memcg_oom_lock); |
1933 | ||
9f3a0d09 | 1934 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1935 | if (iter->oom_lock) { |
79dfdacc MH |
1936 | /* |
1937 | * this subtree of our hierarchy is already locked | |
1938 | * so we cannot give a lock. | |
1939 | */ | |
79dfdacc | 1940 | failed = iter; |
9f3a0d09 JW |
1941 | mem_cgroup_iter_break(memcg, iter); |
1942 | break; | |
23751be0 JW |
1943 | } else |
1944 | iter->oom_lock = true; | |
7d74b06f | 1945 | } |
867578cb | 1946 | |
fb2a6fc5 JW |
1947 | if (failed) { |
1948 | /* | |
1949 | * OK, we failed to lock the whole subtree so we have | |
1950 | * to clean up what we set up to the failing subtree | |
1951 | */ | |
1952 | for_each_mem_cgroup_tree(iter, memcg) { | |
1953 | if (iter == failed) { | |
1954 | mem_cgroup_iter_break(memcg, iter); | |
1955 | break; | |
1956 | } | |
1957 | iter->oom_lock = false; | |
79dfdacc | 1958 | } |
0056f4e6 JW |
1959 | } else |
1960 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1961 | |
1962 | spin_unlock(&memcg_oom_lock); | |
1963 | ||
1964 | return !failed; | |
a636b327 | 1965 | } |
0b7f569e | 1966 | |
fb2a6fc5 | 1967 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1968 | { |
7d74b06f KH |
1969 | struct mem_cgroup *iter; |
1970 | ||
fb2a6fc5 | 1971 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1972 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1973 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1974 | iter->oom_lock = false; |
fb2a6fc5 | 1975 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1976 | } |
1977 | ||
c0ff4b85 | 1978 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1979 | { |
1980 | struct mem_cgroup *iter; | |
1981 | ||
c0ff4b85 | 1982 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1983 | atomic_inc(&iter->under_oom); |
1984 | } | |
1985 | ||
c0ff4b85 | 1986 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1987 | { |
1988 | struct mem_cgroup *iter; | |
1989 | ||
867578cb KH |
1990 | /* |
1991 | * When a new child is created while the hierarchy is under oom, | |
1992 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1993 | * atomic_add_unless() here. | |
1994 | */ | |
c0ff4b85 | 1995 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1996 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1997 | } |
1998 | ||
867578cb KH |
1999 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2000 | ||
dc98df5a | 2001 | struct oom_wait_info { |
d79154bb | 2002 | struct mem_cgroup *memcg; |
dc98df5a KH |
2003 | wait_queue_t wait; |
2004 | }; | |
2005 | ||
2006 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2007 | unsigned mode, int sync, void *arg) | |
2008 | { | |
d79154bb HD |
2009 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2010 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2011 | struct oom_wait_info *oom_wait_info; |
2012 | ||
2013 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2014 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2015 | |
dc98df5a | 2016 | /* |
d79154bb | 2017 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2018 | * Then we can use css_is_ancestor without taking care of RCU. |
2019 | */ | |
c0ff4b85 R |
2020 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2021 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2022 | return 0; |
dc98df5a KH |
2023 | return autoremove_wake_function(wait, mode, sync, arg); |
2024 | } | |
2025 | ||
c0ff4b85 | 2026 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2027 | { |
3812c8c8 | 2028 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2029 | /* for filtering, pass "memcg" as argument. */ |
2030 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2031 | } |
2032 | ||
c0ff4b85 | 2033 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2034 | { |
c0ff4b85 R |
2035 | if (memcg && atomic_read(&memcg->under_oom)) |
2036 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2037 | } |
2038 | ||
3812c8c8 | 2039 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2040 | { |
3812c8c8 JW |
2041 | if (!current->memcg_oom.may_oom) |
2042 | return; | |
867578cb | 2043 | /* |
49426420 JW |
2044 | * We are in the middle of the charge context here, so we |
2045 | * don't want to block when potentially sitting on a callstack | |
2046 | * that holds all kinds of filesystem and mm locks. | |
2047 | * | |
2048 | * Also, the caller may handle a failed allocation gracefully | |
2049 | * (like optional page cache readahead) and so an OOM killer | |
2050 | * invocation might not even be necessary. | |
2051 | * | |
2052 | * That's why we don't do anything here except remember the | |
2053 | * OOM context and then deal with it at the end of the page | |
2054 | * fault when the stack is unwound, the locks are released, | |
2055 | * and when we know whether the fault was overall successful. | |
867578cb | 2056 | */ |
49426420 JW |
2057 | css_get(&memcg->css); |
2058 | current->memcg_oom.memcg = memcg; | |
2059 | current->memcg_oom.gfp_mask = mask; | |
2060 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2061 | } |
2062 | ||
2063 | /** | |
2064 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2065 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2066 | * |
49426420 JW |
2067 | * This has to be called at the end of a page fault if the memcg OOM |
2068 | * handler was enabled. | |
3812c8c8 | 2069 | * |
49426420 | 2070 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2071 | * sleep on a waitqueue until the userspace task resolves the |
2072 | * situation. Sleeping directly in the charge context with all kinds | |
2073 | * of locks held is not a good idea, instead we remember an OOM state | |
2074 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2075 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2076 | * |
2077 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2078 | * completed, %false otherwise. |
3812c8c8 | 2079 | */ |
49426420 | 2080 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2081 | { |
49426420 | 2082 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2083 | struct oom_wait_info owait; |
49426420 | 2084 | bool locked; |
3812c8c8 JW |
2085 | |
2086 | /* OOM is global, do not handle */ | |
3812c8c8 | 2087 | if (!memcg) |
49426420 | 2088 | return false; |
3812c8c8 | 2089 | |
49426420 JW |
2090 | if (!handle) |
2091 | goto cleanup; | |
3812c8c8 JW |
2092 | |
2093 | owait.memcg = memcg; | |
2094 | owait.wait.flags = 0; | |
2095 | owait.wait.func = memcg_oom_wake_function; | |
2096 | owait.wait.private = current; | |
2097 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2098 | |
3812c8c8 | 2099 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2100 | mem_cgroup_mark_under_oom(memcg); |
2101 | ||
2102 | locked = mem_cgroup_oom_trylock(memcg); | |
2103 | ||
2104 | if (locked) | |
2105 | mem_cgroup_oom_notify(memcg); | |
2106 | ||
2107 | if (locked && !memcg->oom_kill_disable) { | |
2108 | mem_cgroup_unmark_under_oom(memcg); | |
2109 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2110 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2111 | current->memcg_oom.order); | |
2112 | } else { | |
3812c8c8 | 2113 | schedule(); |
49426420 JW |
2114 | mem_cgroup_unmark_under_oom(memcg); |
2115 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2116 | } | |
2117 | ||
2118 | if (locked) { | |
fb2a6fc5 JW |
2119 | mem_cgroup_oom_unlock(memcg); |
2120 | /* | |
2121 | * There is no guarantee that an OOM-lock contender | |
2122 | * sees the wakeups triggered by the OOM kill | |
2123 | * uncharges. Wake any sleepers explicitely. | |
2124 | */ | |
2125 | memcg_oom_recover(memcg); | |
2126 | } | |
49426420 JW |
2127 | cleanup: |
2128 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2129 | css_put(&memcg->css); |
867578cb | 2130 | return true; |
0b7f569e KH |
2131 | } |
2132 | ||
d7365e78 JW |
2133 | /** |
2134 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
2135 | * @page: page that is going to change accounted state | |
2136 | * @locked: &memcg->move_lock slowpath was taken | |
2137 | * @flags: IRQ-state flags for &memcg->move_lock | |
32047e2a | 2138 | * |
d7365e78 JW |
2139 | * This function must mark the beginning of an accounted page state |
2140 | * change to prevent double accounting when the page is concurrently | |
2141 | * being moved to another memcg: | |
32047e2a | 2142 | * |
d7365e78 JW |
2143 | * memcg = mem_cgroup_begin_page_stat(page, &locked, &flags); |
2144 | * if (TestClearPageState(page)) | |
2145 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
2146 | * mem_cgroup_end_page_stat(memcg, locked, flags); | |
32047e2a | 2147 | * |
d7365e78 JW |
2148 | * The RCU lock is held throughout the transaction. The fast path can |
2149 | * get away without acquiring the memcg->move_lock (@locked is false) | |
2150 | * because page moving starts with an RCU grace period. | |
32047e2a | 2151 | * |
d7365e78 JW |
2152 | * The RCU lock also protects the memcg from being freed when the page |
2153 | * state that is going to change is the only thing preventing the page | |
2154 | * from being uncharged. E.g. end-writeback clearing PageWriteback(), | |
2155 | * which allows migration to go ahead and uncharge the page before the | |
2156 | * account transaction might be complete. | |
d69b042f | 2157 | */ |
d7365e78 JW |
2158 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page, |
2159 | bool *locked, | |
2160 | unsigned long *flags) | |
89c06bd5 KH |
2161 | { |
2162 | struct mem_cgroup *memcg; | |
2163 | struct page_cgroup *pc; | |
2164 | ||
d7365e78 JW |
2165 | rcu_read_lock(); |
2166 | ||
2167 | if (mem_cgroup_disabled()) | |
2168 | return NULL; | |
2169 | ||
89c06bd5 KH |
2170 | pc = lookup_page_cgroup(page); |
2171 | again: | |
2172 | memcg = pc->mem_cgroup; | |
2173 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
d7365e78 JW |
2174 | return NULL; |
2175 | ||
2176 | *locked = false; | |
bdcbb659 | 2177 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2178 | return memcg; |
89c06bd5 KH |
2179 | |
2180 | move_lock_mem_cgroup(memcg, flags); | |
2181 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2182 | move_unlock_mem_cgroup(memcg, flags); | |
2183 | goto again; | |
2184 | } | |
2185 | *locked = true; | |
d7365e78 JW |
2186 | |
2187 | return memcg; | |
89c06bd5 KH |
2188 | } |
2189 | ||
d7365e78 JW |
2190 | /** |
2191 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2192 | * @memcg: the memcg that was accounted against | |
2193 | * @locked: value received from mem_cgroup_begin_page_stat() | |
2194 | * @flags: value received from mem_cgroup_begin_page_stat() | |
2195 | */ | |
2196 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg, bool locked, | |
2197 | unsigned long flags) | |
89c06bd5 | 2198 | { |
d7365e78 JW |
2199 | if (memcg && locked) |
2200 | move_unlock_mem_cgroup(memcg, &flags); | |
89c06bd5 | 2201 | |
d7365e78 | 2202 | rcu_read_unlock(); |
89c06bd5 KH |
2203 | } |
2204 | ||
d7365e78 JW |
2205 | /** |
2206 | * mem_cgroup_update_page_stat - update page state statistics | |
2207 | * @memcg: memcg to account against | |
2208 | * @idx: page state item to account | |
2209 | * @val: number of pages (positive or negative) | |
2210 | * | |
2211 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2212 | */ | |
2213 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2214 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2215 | { |
658b72c5 | 2216 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2217 | |
d7365e78 JW |
2218 | if (memcg) |
2219 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2220 | } |
26174efd | 2221 | |
cdec2e42 KH |
2222 | /* |
2223 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2224 | * TODO: maybe necessary to use big numbers in big irons. | |
2225 | */ | |
7ec99d62 | 2226 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2227 | struct memcg_stock_pcp { |
2228 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2229 | unsigned int nr_pages; |
cdec2e42 | 2230 | struct work_struct work; |
26fe6168 | 2231 | unsigned long flags; |
a0db00fc | 2232 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2233 | }; |
2234 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2235 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2236 | |
a0956d54 SS |
2237 | /** |
2238 | * consume_stock: Try to consume stocked charge on this cpu. | |
2239 | * @memcg: memcg to consume from. | |
2240 | * @nr_pages: how many pages to charge. | |
2241 | * | |
2242 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2243 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2244 | * service an allocation will refill the stock. | |
2245 | * | |
2246 | * returns true if successful, false otherwise. | |
cdec2e42 | 2247 | */ |
a0956d54 | 2248 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2249 | { |
2250 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2251 | bool ret = false; |
cdec2e42 | 2252 | |
a0956d54 | 2253 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2254 | return ret; |
a0956d54 | 2255 | |
cdec2e42 | 2256 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2257 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2258 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2259 | ret = true; |
2260 | } | |
cdec2e42 KH |
2261 | put_cpu_var(memcg_stock); |
2262 | return ret; | |
2263 | } | |
2264 | ||
2265 | /* | |
3e32cb2e | 2266 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2267 | */ |
2268 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2269 | { | |
2270 | struct mem_cgroup *old = stock->cached; | |
2271 | ||
11c9ea4e | 2272 | if (stock->nr_pages) { |
3e32cb2e | 2273 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2274 | if (do_swap_account) |
3e32cb2e | 2275 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2276 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2277 | stock->nr_pages = 0; |
cdec2e42 KH |
2278 | } |
2279 | stock->cached = NULL; | |
cdec2e42 KH |
2280 | } |
2281 | ||
2282 | /* | |
2283 | * This must be called under preempt disabled or must be called by | |
2284 | * a thread which is pinned to local cpu. | |
2285 | */ | |
2286 | static void drain_local_stock(struct work_struct *dummy) | |
2287 | { | |
7c8e0181 | 2288 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2289 | drain_stock(stock); |
26fe6168 | 2290 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2291 | } |
2292 | ||
e4777496 MH |
2293 | static void __init memcg_stock_init(void) |
2294 | { | |
2295 | int cpu; | |
2296 | ||
2297 | for_each_possible_cpu(cpu) { | |
2298 | struct memcg_stock_pcp *stock = | |
2299 | &per_cpu(memcg_stock, cpu); | |
2300 | INIT_WORK(&stock->work, drain_local_stock); | |
2301 | } | |
2302 | } | |
2303 | ||
cdec2e42 | 2304 | /* |
3e32cb2e | 2305 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2306 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2307 | */ |
c0ff4b85 | 2308 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2309 | { |
2310 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2311 | ||
c0ff4b85 | 2312 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2313 | drain_stock(stock); |
c0ff4b85 | 2314 | stock->cached = memcg; |
cdec2e42 | 2315 | } |
11c9ea4e | 2316 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2317 | put_cpu_var(memcg_stock); |
2318 | } | |
2319 | ||
2320 | /* | |
c0ff4b85 | 2321 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2322 | * of the hierarchy under it. sync flag says whether we should block |
2323 | * until the work is done. | |
cdec2e42 | 2324 | */ |
c0ff4b85 | 2325 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2326 | { |
26fe6168 | 2327 | int cpu, curcpu; |
d38144b7 | 2328 | |
cdec2e42 | 2329 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2330 | get_online_cpus(); |
5af12d0e | 2331 | curcpu = get_cpu(); |
cdec2e42 KH |
2332 | for_each_online_cpu(cpu) { |
2333 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2334 | struct mem_cgroup *memcg; |
26fe6168 | 2335 | |
c0ff4b85 R |
2336 | memcg = stock->cached; |
2337 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2338 | continue; |
c0ff4b85 | 2339 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2340 | continue; |
d1a05b69 MH |
2341 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2342 | if (cpu == curcpu) | |
2343 | drain_local_stock(&stock->work); | |
2344 | else | |
2345 | schedule_work_on(cpu, &stock->work); | |
2346 | } | |
cdec2e42 | 2347 | } |
5af12d0e | 2348 | put_cpu(); |
d38144b7 MH |
2349 | |
2350 | if (!sync) | |
2351 | goto out; | |
2352 | ||
2353 | for_each_online_cpu(cpu) { | |
2354 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2355 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2356 | flush_work(&stock->work); |
2357 | } | |
2358 | out: | |
f894ffa8 | 2359 | put_online_cpus(); |
d38144b7 MH |
2360 | } |
2361 | ||
2362 | /* | |
2363 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2364 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
3e32cb2e | 2365 | * expects some charges will be back later but cannot wait for it. |
d38144b7 | 2366 | */ |
c0ff4b85 | 2367 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2368 | { |
9f50fad6 MH |
2369 | /* |
2370 | * If someone calls draining, avoid adding more kworker runs. | |
2371 | */ | |
2372 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2373 | return; | |
c0ff4b85 | 2374 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2375 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2376 | } |
2377 | ||
2378 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2379 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2380 | { |
2381 | /* called when force_empty is called */ | |
9f50fad6 | 2382 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2383 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2384 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2385 | } |
2386 | ||
711d3d2c KH |
2387 | /* |
2388 | * This function drains percpu counter value from DEAD cpu and | |
2389 | * move it to local cpu. Note that this function can be preempted. | |
2390 | */ | |
c0ff4b85 | 2391 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2392 | { |
2393 | int i; | |
2394 | ||
c0ff4b85 | 2395 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2396 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2397 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2398 | |
c0ff4b85 R |
2399 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2400 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2401 | } |
e9f8974f | 2402 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2403 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2404 | |
c0ff4b85 R |
2405 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2406 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2407 | } |
c0ff4b85 | 2408 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2409 | } |
2410 | ||
0db0628d | 2411 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2412 | unsigned long action, |
2413 | void *hcpu) | |
2414 | { | |
2415 | int cpu = (unsigned long)hcpu; | |
2416 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2417 | struct mem_cgroup *iter; |
cdec2e42 | 2418 | |
619d094b | 2419 | if (action == CPU_ONLINE) |
1489ebad | 2420 | return NOTIFY_OK; |
1489ebad | 2421 | |
d833049b | 2422 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2423 | return NOTIFY_OK; |
711d3d2c | 2424 | |
9f3a0d09 | 2425 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2426 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2427 | ||
cdec2e42 KH |
2428 | stock = &per_cpu(memcg_stock, cpu); |
2429 | drain_stock(stock); | |
2430 | return NOTIFY_OK; | |
2431 | } | |
2432 | ||
00501b53 JW |
2433 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2434 | unsigned int nr_pages) | |
8a9f3ccd | 2435 | { |
7ec99d62 | 2436 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2437 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2438 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2439 | struct page_counter *counter; |
6539cc05 | 2440 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2441 | bool may_swap = true; |
2442 | bool drained = false; | |
05b84301 | 2443 | int ret = 0; |
a636b327 | 2444 | |
ce00a967 JW |
2445 | if (mem_cgroup_is_root(memcg)) |
2446 | goto done; | |
6539cc05 | 2447 | retry: |
b6b6cc72 MH |
2448 | if (consume_stock(memcg, nr_pages)) |
2449 | goto done; | |
8a9f3ccd | 2450 | |
3fbe7244 | 2451 | if (!do_swap_account || |
3e32cb2e JW |
2452 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2453 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2454 | goto done_restock; |
3fbe7244 | 2455 | if (do_swap_account) |
3e32cb2e JW |
2456 | page_counter_uncharge(&memcg->memsw, batch); |
2457 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2458 | } else { |
3e32cb2e | 2459 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2460 | may_swap = false; |
3fbe7244 | 2461 | } |
7a81b88c | 2462 | |
6539cc05 JW |
2463 | if (batch > nr_pages) { |
2464 | batch = nr_pages; | |
2465 | goto retry; | |
2466 | } | |
6d61ef40 | 2467 | |
06b078fc JW |
2468 | /* |
2469 | * Unlike in global OOM situations, memcg is not in a physical | |
2470 | * memory shortage. Allow dying and OOM-killed tasks to | |
2471 | * bypass the last charges so that they can exit quickly and | |
2472 | * free their memory. | |
2473 | */ | |
2474 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2475 | fatal_signal_pending(current) || | |
2476 | current->flags & PF_EXITING)) | |
2477 | goto bypass; | |
2478 | ||
2479 | if (unlikely(task_in_memcg_oom(current))) | |
2480 | goto nomem; | |
2481 | ||
6539cc05 JW |
2482 | if (!(gfp_mask & __GFP_WAIT)) |
2483 | goto nomem; | |
4b534334 | 2484 | |
b70a2a21 JW |
2485 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2486 | gfp_mask, may_swap); | |
6539cc05 | 2487 | |
61e02c74 | 2488 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2489 | goto retry; |
28c34c29 | 2490 | |
b70a2a21 JW |
2491 | if (!drained) { |
2492 | drain_all_stock_async(mem_over_limit); | |
2493 | drained = true; | |
2494 | goto retry; | |
2495 | } | |
2496 | ||
28c34c29 JW |
2497 | if (gfp_mask & __GFP_NORETRY) |
2498 | goto nomem; | |
6539cc05 JW |
2499 | /* |
2500 | * Even though the limit is exceeded at this point, reclaim | |
2501 | * may have been able to free some pages. Retry the charge | |
2502 | * before killing the task. | |
2503 | * | |
2504 | * Only for regular pages, though: huge pages are rather | |
2505 | * unlikely to succeed so close to the limit, and we fall back | |
2506 | * to regular pages anyway in case of failure. | |
2507 | */ | |
61e02c74 | 2508 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2509 | goto retry; |
2510 | /* | |
2511 | * At task move, charge accounts can be doubly counted. So, it's | |
2512 | * better to wait until the end of task_move if something is going on. | |
2513 | */ | |
2514 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2515 | goto retry; | |
2516 | ||
9b130619 JW |
2517 | if (nr_retries--) |
2518 | goto retry; | |
2519 | ||
06b078fc JW |
2520 | if (gfp_mask & __GFP_NOFAIL) |
2521 | goto bypass; | |
2522 | ||
6539cc05 JW |
2523 | if (fatal_signal_pending(current)) |
2524 | goto bypass; | |
2525 | ||
61e02c74 | 2526 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2527 | nomem: |
6d1fdc48 | 2528 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2529 | return -ENOMEM; |
867578cb | 2530 | bypass: |
ce00a967 | 2531 | return -EINTR; |
6539cc05 JW |
2532 | |
2533 | done_restock: | |
e8ea14cc | 2534 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2535 | if (batch > nr_pages) |
2536 | refill_stock(memcg, batch - nr_pages); | |
2537 | done: | |
05b84301 | 2538 | return ret; |
7a81b88c | 2539 | } |
8a9f3ccd | 2540 | |
00501b53 | 2541 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2542 | { |
ce00a967 JW |
2543 | if (mem_cgroup_is_root(memcg)) |
2544 | return; | |
2545 | ||
3e32cb2e | 2546 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2547 | if (do_swap_account) |
3e32cb2e | 2548 | page_counter_uncharge(&memcg->memsw, nr_pages); |
e8ea14cc JW |
2549 | |
2550 | css_put_many(&memcg->css, nr_pages); | |
d01dd17f KH |
2551 | } |
2552 | ||
a3b2d692 KH |
2553 | /* |
2554 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2555 | * rcu_read_lock(). The caller is responsible for calling |
2556 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2557 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2558 | */ |
2559 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2560 | { | |
a3b2d692 KH |
2561 | /* ID 0 is unused ID */ |
2562 | if (!id) | |
2563 | return NULL; | |
34c00c31 | 2564 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2565 | } |
2566 | ||
0a31bc97 JW |
2567 | /* |
2568 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2569 | * @page: the page | |
2570 | * | |
2571 | * Look up, get a css reference, and return the memcg that owns @page. | |
2572 | * | |
2573 | * The page must be locked to prevent racing with swap-in and page | |
2574 | * cache charges. If coming from an unlocked page table, the caller | |
2575 | * must ensure the page is on the LRU or this can race with charging. | |
2576 | */ | |
e42d9d5d | 2577 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2578 | { |
c0ff4b85 | 2579 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2580 | struct page_cgroup *pc; |
a3b2d692 | 2581 | unsigned short id; |
b5a84319 KH |
2582 | swp_entry_t ent; |
2583 | ||
309381fe | 2584 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2585 | |
3c776e64 | 2586 | pc = lookup_page_cgroup(page); |
a3b2d692 | 2587 | if (PageCgroupUsed(pc)) { |
c0ff4b85 | 2588 | memcg = pc->mem_cgroup; |
ec903c0c | 2589 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2590 | memcg = NULL; |
e42d9d5d | 2591 | } else if (PageSwapCache(page)) { |
3c776e64 | 2592 | ent.val = page_private(page); |
9fb4b7cc | 2593 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2594 | rcu_read_lock(); |
c0ff4b85 | 2595 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2596 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2597 | memcg = NULL; |
a3b2d692 | 2598 | rcu_read_unlock(); |
3c776e64 | 2599 | } |
c0ff4b85 | 2600 | return memcg; |
b5a84319 KH |
2601 | } |
2602 | ||
0a31bc97 JW |
2603 | static void lock_page_lru(struct page *page, int *isolated) |
2604 | { | |
2605 | struct zone *zone = page_zone(page); | |
2606 | ||
2607 | spin_lock_irq(&zone->lru_lock); | |
2608 | if (PageLRU(page)) { | |
2609 | struct lruvec *lruvec; | |
2610 | ||
2611 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2612 | ClearPageLRU(page); | |
2613 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2614 | *isolated = 1; | |
2615 | } else | |
2616 | *isolated = 0; | |
2617 | } | |
2618 | ||
2619 | static void unlock_page_lru(struct page *page, int isolated) | |
2620 | { | |
2621 | struct zone *zone = page_zone(page); | |
2622 | ||
2623 | if (isolated) { | |
2624 | struct lruvec *lruvec; | |
2625 | ||
2626 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2627 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2628 | SetPageLRU(page); | |
2629 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2630 | } | |
2631 | spin_unlock_irq(&zone->lru_lock); | |
2632 | } | |
2633 | ||
00501b53 | 2634 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2635 | bool lrucare) |
7a81b88c | 2636 | { |
ce587e65 | 2637 | struct page_cgroup *pc = lookup_page_cgroup(page); |
0a31bc97 | 2638 | int isolated; |
9ce70c02 | 2639 | |
309381fe | 2640 | VM_BUG_ON_PAGE(PageCgroupUsed(pc), page); |
ca3e0214 KH |
2641 | /* |
2642 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2643 | * accessed by any other context at this point. | |
2644 | */ | |
9ce70c02 HD |
2645 | |
2646 | /* | |
2647 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2648 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2649 | */ | |
0a31bc97 JW |
2650 | if (lrucare) |
2651 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2652 | |
0a31bc97 JW |
2653 | /* |
2654 | * Nobody should be changing or seriously looking at | |
2655 | * pc->mem_cgroup and pc->flags at this point: | |
2656 | * | |
2657 | * - the page is uncharged | |
2658 | * | |
2659 | * - the page is off-LRU | |
2660 | * | |
2661 | * - an anonymous fault has exclusive page access, except for | |
2662 | * a locked page table | |
2663 | * | |
2664 | * - a page cache insertion, a swapin fault, or a migration | |
2665 | * have the page locked | |
2666 | */ | |
c0ff4b85 | 2667 | pc->mem_cgroup = memcg; |
0a31bc97 | 2668 | pc->flags = PCG_USED | PCG_MEM | (do_swap_account ? PCG_MEMSW : 0); |
9ce70c02 | 2669 | |
0a31bc97 JW |
2670 | if (lrucare) |
2671 | unlock_page_lru(page, isolated); | |
7a81b88c | 2672 | } |
66e1707b | 2673 | |
7ae1e1d0 | 2674 | #ifdef CONFIG_MEMCG_KMEM |
bd673145 VD |
2675 | /* |
2676 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or | |
2677 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. | |
2678 | */ | |
2679 | static DEFINE_MUTEX(memcg_slab_mutex); | |
2680 | ||
d6441637 VD |
2681 | static DEFINE_MUTEX(activate_kmem_mutex); |
2682 | ||
1f458cbf GC |
2683 | /* |
2684 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2685 | * in the memcg_cache_params struct. | |
2686 | */ | |
2687 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2688 | { | |
2689 | struct kmem_cache *cachep; | |
2690 | ||
2691 | VM_BUG_ON(p->is_root_cache); | |
2692 | cachep = p->root_cache; | |
7a67d7ab | 2693 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2694 | } |
2695 | ||
749c5415 | 2696 | #ifdef CONFIG_SLABINFO |
2da8ca82 | 2697 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
749c5415 | 2698 | { |
2da8ca82 | 2699 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
749c5415 GC |
2700 | struct memcg_cache_params *params; |
2701 | ||
cf2b8fbf | 2702 | if (!memcg_kmem_is_active(memcg)) |
749c5415 GC |
2703 | return -EIO; |
2704 | ||
2705 | print_slabinfo_header(m); | |
2706 | ||
bd673145 | 2707 | mutex_lock(&memcg_slab_mutex); |
749c5415 GC |
2708 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) |
2709 | cache_show(memcg_params_to_cache(params), m); | |
bd673145 | 2710 | mutex_unlock(&memcg_slab_mutex); |
749c5415 GC |
2711 | |
2712 | return 0; | |
2713 | } | |
2714 | #endif | |
2715 | ||
3e32cb2e JW |
2716 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2717 | unsigned long nr_pages) | |
7ae1e1d0 | 2718 | { |
3e32cb2e | 2719 | struct page_counter *counter; |
7ae1e1d0 | 2720 | int ret = 0; |
7ae1e1d0 | 2721 | |
3e32cb2e JW |
2722 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2723 | if (ret < 0) | |
7ae1e1d0 GC |
2724 | return ret; |
2725 | ||
3e32cb2e | 2726 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2727 | if (ret == -EINTR) { |
2728 | /* | |
00501b53 JW |
2729 | * try_charge() chose to bypass to root due to OOM kill or |
2730 | * fatal signal. Since our only options are to either fail | |
2731 | * the allocation or charge it to this cgroup, do it as a | |
2732 | * temporary condition. But we can't fail. From a kmem/slab | |
2733 | * perspective, the cache has already been selected, by | |
2734 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2735 | * our minds. |
2736 | * | |
2737 | * This condition will only trigger if the task entered | |
00501b53 JW |
2738 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2739 | * during try_charge() above. Tasks that were already dying | |
2740 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2741 | * directed to the root cgroup in memcontrol.h |
2742 | */ | |
3e32cb2e | 2743 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2744 | if (do_swap_account) |
3e32cb2e | 2745 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2746 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2747 | ret = 0; |
2748 | } else if (ret) | |
3e32cb2e | 2749 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2750 | |
2751 | return ret; | |
2752 | } | |
2753 | ||
3e32cb2e JW |
2754 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, |
2755 | unsigned long nr_pages) | |
7ae1e1d0 | 2756 | { |
3e32cb2e | 2757 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2758 | if (do_swap_account) |
3e32cb2e | 2759 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 GC |
2760 | |
2761 | /* Not down to 0 */ | |
e8ea14cc JW |
2762 | if (page_counter_uncharge(&memcg->kmem, nr_pages)) { |
2763 | css_put_many(&memcg->css, nr_pages); | |
7de37682 | 2764 | return; |
e8ea14cc | 2765 | } |
7de37682 | 2766 | |
10d5ebf4 LZ |
2767 | /* |
2768 | * Releases a reference taken in kmem_cgroup_css_offline in case | |
2769 | * this last uncharge is racing with the offlining code or it is | |
2770 | * outliving the memcg existence. | |
2771 | * | |
2772 | * The memory barrier imposed by test&clear is paired with the | |
2773 | * explicit one in memcg_kmem_mark_dead(). | |
2774 | */ | |
7de37682 | 2775 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 2776 | css_put(&memcg->css); |
e8ea14cc JW |
2777 | |
2778 | css_put_many(&memcg->css, nr_pages); | |
7ae1e1d0 GC |
2779 | } |
2780 | ||
2633d7a0 GC |
2781 | /* |
2782 | * helper for acessing a memcg's index. It will be used as an index in the | |
2783 | * child cache array in kmem_cache, and also to derive its name. This function | |
2784 | * will return -1 when this is not a kmem-limited memcg. | |
2785 | */ | |
2786 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2787 | { | |
2788 | return memcg ? memcg->kmemcg_id : -1; | |
2789 | } | |
2790 | ||
f3bb3043 | 2791 | static int memcg_alloc_cache_id(void) |
55007d84 | 2792 | { |
f3bb3043 VD |
2793 | int id, size; |
2794 | int err; | |
2795 | ||
2796 | id = ida_simple_get(&kmem_limited_groups, | |
2797 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2798 | if (id < 0) | |
2799 | return id; | |
55007d84 | 2800 | |
f3bb3043 VD |
2801 | if (id < memcg_limited_groups_array_size) |
2802 | return id; | |
2803 | ||
2804 | /* | |
2805 | * There's no space for the new id in memcg_caches arrays, | |
2806 | * so we have to grow them. | |
2807 | */ | |
2808 | ||
2809 | size = 2 * (id + 1); | |
55007d84 GC |
2810 | if (size < MEMCG_CACHES_MIN_SIZE) |
2811 | size = MEMCG_CACHES_MIN_SIZE; | |
2812 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2813 | size = MEMCG_CACHES_MAX_SIZE; | |
2814 | ||
f3bb3043 VD |
2815 | mutex_lock(&memcg_slab_mutex); |
2816 | err = memcg_update_all_caches(size); | |
2817 | mutex_unlock(&memcg_slab_mutex); | |
2818 | ||
2819 | if (err) { | |
2820 | ida_simple_remove(&kmem_limited_groups, id); | |
2821 | return err; | |
2822 | } | |
2823 | return id; | |
2824 | } | |
2825 | ||
2826 | static void memcg_free_cache_id(int id) | |
2827 | { | |
2828 | ida_simple_remove(&kmem_limited_groups, id); | |
55007d84 GC |
2829 | } |
2830 | ||
2831 | /* | |
2832 | * We should update the current array size iff all caches updates succeed. This | |
2833 | * can only be done from the slab side. The slab mutex needs to be held when | |
2834 | * calling this. | |
2835 | */ | |
2836 | void memcg_update_array_size(int num) | |
2837 | { | |
f3bb3043 | 2838 | memcg_limited_groups_array_size = num; |
55007d84 GC |
2839 | } |
2840 | ||
776ed0f0 VD |
2841 | static void memcg_register_cache(struct mem_cgroup *memcg, |
2842 | struct kmem_cache *root_cache) | |
2633d7a0 | 2843 | { |
93f39eea VD |
2844 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by |
2845 | memcg_slab_mutex */ | |
bd673145 | 2846 | struct kmem_cache *cachep; |
d7f25f8a GC |
2847 | int id; |
2848 | ||
bd673145 VD |
2849 | lockdep_assert_held(&memcg_slab_mutex); |
2850 | ||
2851 | id = memcg_cache_id(memcg); | |
2852 | ||
2853 | /* | |
2854 | * Since per-memcg caches are created asynchronously on first | |
2855 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
2856 | * create the same cache, but only one of them may succeed. | |
2857 | */ | |
2858 | if (cache_from_memcg_idx(root_cache, id)) | |
1aa13254 VD |
2859 | return; |
2860 | ||
073ee1c6 | 2861 | cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1); |
776ed0f0 | 2862 | cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf); |
2edefe11 | 2863 | /* |
bd673145 VD |
2864 | * If we could not create a memcg cache, do not complain, because |
2865 | * that's not critical at all as we can always proceed with the root | |
2866 | * cache. | |
2edefe11 | 2867 | */ |
bd673145 VD |
2868 | if (!cachep) |
2869 | return; | |
2edefe11 | 2870 | |
33a690c4 | 2871 | css_get(&memcg->css); |
bd673145 | 2872 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
1aa13254 | 2873 | |
d7f25f8a | 2874 | /* |
959c8963 VD |
2875 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
2876 | * barrier here to ensure nobody will see the kmem_cache partially | |
2877 | * initialized. | |
d7f25f8a | 2878 | */ |
959c8963 VD |
2879 | smp_wmb(); |
2880 | ||
bd673145 VD |
2881 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
2882 | root_cache->memcg_params->memcg_caches[id] = cachep; | |
1aa13254 | 2883 | } |
d7f25f8a | 2884 | |
776ed0f0 | 2885 | static void memcg_unregister_cache(struct kmem_cache *cachep) |
1aa13254 | 2886 | { |
bd673145 | 2887 | struct kmem_cache *root_cache; |
1aa13254 VD |
2888 | struct mem_cgroup *memcg; |
2889 | int id; | |
2890 | ||
bd673145 | 2891 | lockdep_assert_held(&memcg_slab_mutex); |
d7f25f8a | 2892 | |
bd673145 | 2893 | BUG_ON(is_root_cache(cachep)); |
2edefe11 | 2894 | |
bd673145 VD |
2895 | root_cache = cachep->memcg_params->root_cache; |
2896 | memcg = cachep->memcg_params->memcg; | |
96403da2 | 2897 | id = memcg_cache_id(memcg); |
d7f25f8a | 2898 | |
bd673145 VD |
2899 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
2900 | root_cache->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a | 2901 | |
bd673145 VD |
2902 | list_del(&cachep->memcg_params->list); |
2903 | ||
2904 | kmem_cache_destroy(cachep); | |
33a690c4 VD |
2905 | |
2906 | /* drop the reference taken in memcg_register_cache */ | |
2907 | css_put(&memcg->css); | |
2633d7a0 GC |
2908 | } |
2909 | ||
0e9d92f2 GC |
2910 | /* |
2911 | * During the creation a new cache, we need to disable our accounting mechanism | |
2912 | * altogether. This is true even if we are not creating, but rather just | |
2913 | * enqueing new caches to be created. | |
2914 | * | |
2915 | * This is because that process will trigger allocations; some visible, like | |
2916 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
2917 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
2918 | * objects during debug. | |
2919 | * | |
2920 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
2921 | * to it. This may not be a bounded recursion: since the first cache creation | |
2922 | * failed to complete (waiting on the allocation), we'll just try to create the | |
2923 | * cache again, failing at the same point. | |
2924 | * | |
2925 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
2926 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
2927 | * inside the following two functions. | |
2928 | */ | |
2929 | static inline void memcg_stop_kmem_account(void) | |
2930 | { | |
2931 | VM_BUG_ON(!current->mm); | |
2932 | current->memcg_kmem_skip_account++; | |
2933 | } | |
2934 | ||
2935 | static inline void memcg_resume_kmem_account(void) | |
2936 | { | |
2937 | VM_BUG_ON(!current->mm); | |
2938 | current->memcg_kmem_skip_account--; | |
2939 | } | |
2940 | ||
776ed0f0 | 2941 | int __memcg_cleanup_cache_params(struct kmem_cache *s) |
7cf27982 GC |
2942 | { |
2943 | struct kmem_cache *c; | |
b8529907 | 2944 | int i, failed = 0; |
7cf27982 | 2945 | |
bd673145 | 2946 | mutex_lock(&memcg_slab_mutex); |
7a67d7ab QH |
2947 | for_each_memcg_cache_index(i) { |
2948 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
2949 | if (!c) |
2950 | continue; | |
2951 | ||
776ed0f0 | 2952 | memcg_unregister_cache(c); |
b8529907 VD |
2953 | |
2954 | if (cache_from_memcg_idx(s, i)) | |
2955 | failed++; | |
7cf27982 | 2956 | } |
bd673145 | 2957 | mutex_unlock(&memcg_slab_mutex); |
b8529907 | 2958 | return failed; |
7cf27982 GC |
2959 | } |
2960 | ||
776ed0f0 | 2961 | static void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
2962 | { |
2963 | struct kmem_cache *cachep; | |
bd673145 | 2964 | struct memcg_cache_params *params, *tmp; |
1f458cbf GC |
2965 | |
2966 | if (!memcg_kmem_is_active(memcg)) | |
2967 | return; | |
2968 | ||
bd673145 VD |
2969 | mutex_lock(&memcg_slab_mutex); |
2970 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { | |
1f458cbf | 2971 | cachep = memcg_params_to_cache(params); |
bd673145 VD |
2972 | kmem_cache_shrink(cachep); |
2973 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
776ed0f0 | 2974 | memcg_unregister_cache(cachep); |
1f458cbf | 2975 | } |
bd673145 | 2976 | mutex_unlock(&memcg_slab_mutex); |
1f458cbf GC |
2977 | } |
2978 | ||
776ed0f0 | 2979 | struct memcg_register_cache_work { |
5722d094 VD |
2980 | struct mem_cgroup *memcg; |
2981 | struct kmem_cache *cachep; | |
2982 | struct work_struct work; | |
2983 | }; | |
2984 | ||
776ed0f0 | 2985 | static void memcg_register_cache_func(struct work_struct *w) |
d7f25f8a | 2986 | { |
776ed0f0 VD |
2987 | struct memcg_register_cache_work *cw = |
2988 | container_of(w, struct memcg_register_cache_work, work); | |
5722d094 VD |
2989 | struct mem_cgroup *memcg = cw->memcg; |
2990 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2991 | |
bd673145 | 2992 | mutex_lock(&memcg_slab_mutex); |
776ed0f0 | 2993 | memcg_register_cache(memcg, cachep); |
bd673145 VD |
2994 | mutex_unlock(&memcg_slab_mutex); |
2995 | ||
5722d094 | 2996 | css_put(&memcg->css); |
d7f25f8a GC |
2997 | kfree(cw); |
2998 | } | |
2999 | ||
3000 | /* | |
3001 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3002 | */ |
776ed0f0 VD |
3003 | static void __memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3004 | struct kmem_cache *cachep) | |
d7f25f8a | 3005 | { |
776ed0f0 | 3006 | struct memcg_register_cache_work *cw; |
d7f25f8a | 3007 | |
776ed0f0 | 3008 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
ca0dde97 LZ |
3009 | if (cw == NULL) { |
3010 | css_put(&memcg->css); | |
d7f25f8a GC |
3011 | return; |
3012 | } | |
3013 | ||
3014 | cw->memcg = memcg; | |
3015 | cw->cachep = cachep; | |
3016 | ||
776ed0f0 | 3017 | INIT_WORK(&cw->work, memcg_register_cache_func); |
d7f25f8a GC |
3018 | schedule_work(&cw->work); |
3019 | } | |
3020 | ||
776ed0f0 VD |
3021 | static void memcg_schedule_register_cache(struct mem_cgroup *memcg, |
3022 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
3023 | { |
3024 | /* | |
3025 | * We need to stop accounting when we kmalloc, because if the | |
3026 | * corresponding kmalloc cache is not yet created, the first allocation | |
776ed0f0 | 3027 | * in __memcg_schedule_register_cache will recurse. |
0e9d92f2 GC |
3028 | * |
3029 | * However, it is better to enclose the whole function. Depending on | |
3030 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3031 | * trigger an allocation. This too, will make us recurse. Because at | |
3032 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3033 | * the safest choice is to do it like this, wrapping the whole function. | |
3034 | */ | |
3035 | memcg_stop_kmem_account(); | |
776ed0f0 | 3036 | __memcg_schedule_register_cache(memcg, cachep); |
0e9d92f2 GC |
3037 | memcg_resume_kmem_account(); |
3038 | } | |
c67a8a68 VD |
3039 | |
3040 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) | |
3041 | { | |
3e32cb2e | 3042 | unsigned int nr_pages = 1 << order; |
c67a8a68 VD |
3043 | int res; |
3044 | ||
3e32cb2e | 3045 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, nr_pages); |
c67a8a68 | 3046 | if (!res) |
3e32cb2e | 3047 | atomic_add(nr_pages, &cachep->memcg_params->nr_pages); |
c67a8a68 VD |
3048 | return res; |
3049 | } | |
3050 | ||
3051 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) | |
3052 | { | |
3e32cb2e JW |
3053 | unsigned int nr_pages = 1 << order; |
3054 | ||
3055 | memcg_uncharge_kmem(cachep->memcg_params->memcg, nr_pages); | |
3056 | atomic_sub(nr_pages, &cachep->memcg_params->nr_pages); | |
c67a8a68 VD |
3057 | } |
3058 | ||
d7f25f8a GC |
3059 | /* |
3060 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3061 | * We try to use the current memcg's version of the cache. | |
3062 | * | |
3063 | * If the cache does not exist yet, if we are the first user of it, | |
3064 | * we either create it immediately, if possible, or create it asynchronously | |
3065 | * in a workqueue. | |
3066 | * In the latter case, we will let the current allocation go through with | |
3067 | * the original cache. | |
3068 | * | |
3069 | * Can't be called in interrupt context or from kernel threads. | |
3070 | * This function needs to be called with rcu_read_lock() held. | |
3071 | */ | |
3072 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3073 | gfp_t gfp) | |
3074 | { | |
3075 | struct mem_cgroup *memcg; | |
959c8963 | 3076 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
3077 | |
3078 | VM_BUG_ON(!cachep->memcg_params); | |
3079 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3080 | ||
0e9d92f2 GC |
3081 | if (!current->mm || current->memcg_kmem_skip_account) |
3082 | return cachep; | |
3083 | ||
d7f25f8a GC |
3084 | rcu_read_lock(); |
3085 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a | 3086 | |
cf2b8fbf | 3087 | if (!memcg_kmem_is_active(memcg)) |
ca0dde97 | 3088 | goto out; |
d7f25f8a | 3089 | |
959c8963 VD |
3090 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
3091 | if (likely(memcg_cachep)) { | |
3092 | cachep = memcg_cachep; | |
ca0dde97 | 3093 | goto out; |
d7f25f8a GC |
3094 | } |
3095 | ||
ca0dde97 | 3096 | /* The corresponding put will be done in the workqueue. */ |
ec903c0c | 3097 | if (!css_tryget_online(&memcg->css)) |
ca0dde97 LZ |
3098 | goto out; |
3099 | rcu_read_unlock(); | |
3100 | ||
3101 | /* | |
3102 | * If we are in a safe context (can wait, and not in interrupt | |
3103 | * context), we could be be predictable and return right away. | |
3104 | * This would guarantee that the allocation being performed | |
3105 | * already belongs in the new cache. | |
3106 | * | |
3107 | * However, there are some clashes that can arrive from locking. | |
3108 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
3109 | * memcg_create_kmem_cache, this means no further allocation |
3110 | * could happen with the slab_mutex held. So it's better to | |
3111 | * defer everything. | |
ca0dde97 | 3112 | */ |
776ed0f0 | 3113 | memcg_schedule_register_cache(memcg, cachep); |
ca0dde97 LZ |
3114 | return cachep; |
3115 | out: | |
3116 | rcu_read_unlock(); | |
3117 | return cachep; | |
d7f25f8a | 3118 | } |
d7f25f8a | 3119 | |
7ae1e1d0 GC |
3120 | /* |
3121 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3122 | * possible for the given order. But the page is not allocated yet, so we'll | |
3123 | * need a further commit step to do the final arrangements. | |
3124 | * | |
3125 | * It is possible for the task to switch cgroups in this mean time, so at | |
3126 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3127 | * the handle argument to return to the caller which cgroup we should commit | |
3128 | * against. We could also return the memcg directly and avoid the pointer | |
3129 | * passing, but a boolean return value gives better semantics considering | |
3130 | * the compiled-out case as well. | |
3131 | * | |
3132 | * Returning true means the allocation is possible. | |
3133 | */ | |
3134 | bool | |
3135 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3136 | { | |
3137 | struct mem_cgroup *memcg; | |
3138 | int ret; | |
3139 | ||
3140 | *_memcg = NULL; | |
6d42c232 GC |
3141 | |
3142 | /* | |
3143 | * Disabling accounting is only relevant for some specific memcg | |
3144 | * internal allocations. Therefore we would initially not have such | |
52383431 VD |
3145 | * check here, since direct calls to the page allocator that are |
3146 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen | |
3147 | * outside memcg core. We are mostly concerned with cache allocations, | |
3148 | * and by having this test at memcg_kmem_get_cache, we are already able | |
3149 | * to relay the allocation to the root cache and bypass the memcg cache | |
3150 | * altogether. | |
6d42c232 GC |
3151 | * |
3152 | * There is one exception, though: the SLUB allocator does not create | |
3153 | * large order caches, but rather service large kmallocs directly from | |
3154 | * the page allocator. Therefore, the following sequence when backed by | |
3155 | * the SLUB allocator: | |
3156 | * | |
f894ffa8 AM |
3157 | * memcg_stop_kmem_account(); |
3158 | * kmalloc(<large_number>) | |
3159 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3160 | * |
3161 | * would effectively ignore the fact that we should skip accounting, | |
3162 | * since it will drive us directly to this function without passing | |
3163 | * through the cache selector memcg_kmem_get_cache. Such large | |
3164 | * allocations are extremely rare but can happen, for instance, for the | |
3165 | * cache arrays. We bring this test here. | |
3166 | */ | |
3167 | if (!current->mm || current->memcg_kmem_skip_account) | |
3168 | return true; | |
3169 | ||
df381975 | 3170 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 3171 | |
cf2b8fbf | 3172 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
3173 | css_put(&memcg->css); |
3174 | return true; | |
3175 | } | |
3176 | ||
3e32cb2e | 3177 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
3178 | if (!ret) |
3179 | *_memcg = memcg; | |
7ae1e1d0 GC |
3180 | |
3181 | css_put(&memcg->css); | |
3182 | return (ret == 0); | |
3183 | } | |
3184 | ||
3185 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3186 | int order) | |
3187 | { | |
3188 | struct page_cgroup *pc; | |
3189 | ||
3190 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3191 | ||
3192 | /* The page allocation failed. Revert */ | |
3193 | if (!page) { | |
3e32cb2e | 3194 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
3195 | return; |
3196 | } | |
a840cda6 JW |
3197 | /* |
3198 | * The page is freshly allocated and not visible to any | |
3199 | * outside callers yet. Set up pc non-atomically. | |
3200 | */ | |
7ae1e1d0 | 3201 | pc = lookup_page_cgroup(page); |
7ae1e1d0 | 3202 | pc->mem_cgroup = memcg; |
a840cda6 | 3203 | pc->flags = PCG_USED; |
7ae1e1d0 GC |
3204 | } |
3205 | ||
3206 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3207 | { | |
3208 | struct mem_cgroup *memcg = NULL; | |
3209 | struct page_cgroup *pc; | |
3210 | ||
3211 | ||
3212 | pc = lookup_page_cgroup(page); | |
7ae1e1d0 GC |
3213 | if (!PageCgroupUsed(pc)) |
3214 | return; | |
3215 | ||
a840cda6 JW |
3216 | memcg = pc->mem_cgroup; |
3217 | pc->flags = 0; | |
7ae1e1d0 GC |
3218 | |
3219 | /* | |
3220 | * We trust that only if there is a memcg associated with the page, it | |
3221 | * is a valid allocation | |
3222 | */ | |
3223 | if (!memcg) | |
3224 | return; | |
3225 | ||
309381fe | 3226 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
3e32cb2e | 3227 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 | 3228 | } |
1f458cbf | 3229 | #else |
776ed0f0 | 3230 | static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3231 | { |
3232 | } | |
7ae1e1d0 GC |
3233 | #endif /* CONFIG_MEMCG_KMEM */ |
3234 | ||
ca3e0214 KH |
3235 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3236 | ||
ca3e0214 KH |
3237 | /* |
3238 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3239 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3240 | * charge/uncharge will be never happen and move_account() is done under | |
3241 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3242 | */ |
e94c8a9c | 3243 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3244 | { |
3245 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c | 3246 | struct page_cgroup *pc; |
b070e65c | 3247 | struct mem_cgroup *memcg; |
e94c8a9c | 3248 | int i; |
ca3e0214 | 3249 | |
3d37c4a9 KH |
3250 | if (mem_cgroup_disabled()) |
3251 | return; | |
b070e65c DR |
3252 | |
3253 | memcg = head_pc->mem_cgroup; | |
e94c8a9c KH |
3254 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3255 | pc = head_pc + i; | |
b070e65c | 3256 | pc->mem_cgroup = memcg; |
0a31bc97 | 3257 | pc->flags = head_pc->flags; |
e94c8a9c | 3258 | } |
b070e65c DR |
3259 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3260 | HPAGE_PMD_NR); | |
ca3e0214 | 3261 | } |
12d27107 | 3262 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3263 | |
f817ed48 | 3264 | /** |
de3638d9 | 3265 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3266 | * @page: the page |
7ec99d62 | 3267 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3268 | * @pc: page_cgroup of the page. |
3269 | * @from: mem_cgroup which the page is moved from. | |
3270 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3271 | * | |
3272 | * The caller must confirm following. | |
08e552c6 | 3273 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3274 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3275 | * |
2f3479b1 KH |
3276 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3277 | * from old cgroup. | |
f817ed48 | 3278 | */ |
7ec99d62 JW |
3279 | static int mem_cgroup_move_account(struct page *page, |
3280 | unsigned int nr_pages, | |
3281 | struct page_cgroup *pc, | |
3282 | struct mem_cgroup *from, | |
2f3479b1 | 3283 | struct mem_cgroup *to) |
f817ed48 | 3284 | { |
de3638d9 JW |
3285 | unsigned long flags; |
3286 | int ret; | |
987eba66 | 3287 | |
f817ed48 | 3288 | VM_BUG_ON(from == to); |
309381fe | 3289 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3290 | /* |
3291 | * The page is isolated from LRU. So, collapse function | |
3292 | * will not handle this page. But page splitting can happen. | |
3293 | * Do this check under compound_page_lock(). The caller should | |
3294 | * hold it. | |
3295 | */ | |
3296 | ret = -EBUSY; | |
7ec99d62 | 3297 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3298 | goto out; |
3299 | ||
0a31bc97 JW |
3300 | /* |
3301 | * Prevent mem_cgroup_migrate() from looking at pc->mem_cgroup | |
3302 | * of its source page while we change it: page migration takes | |
3303 | * both pages off the LRU, but page cache replacement doesn't. | |
3304 | */ | |
3305 | if (!trylock_page(page)) | |
3306 | goto out; | |
de3638d9 JW |
3307 | |
3308 | ret = -EINVAL; | |
3309 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
0a31bc97 | 3310 | goto out_unlock; |
de3638d9 | 3311 | |
312734c0 | 3312 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3313 | |
0a31bc97 | 3314 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
3315 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
3316 | nr_pages); | |
3317 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3318 | nr_pages); | |
3319 | } | |
3ea67d06 | 3320 | |
59d1d256 JW |
3321 | if (PageWriteback(page)) { |
3322 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3323 | nr_pages); | |
3324 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3325 | nr_pages); | |
3326 | } | |
3ea67d06 | 3327 | |
0a31bc97 JW |
3328 | /* |
3329 | * It is safe to change pc->mem_cgroup here because the page | |
3330 | * is referenced, charged, and isolated - we can't race with | |
3331 | * uncharging, charging, migration, or LRU putback. | |
3332 | */ | |
d69b042f | 3333 | |
854ffa8d | 3334 | /* caller should have done css_get */ |
08e552c6 | 3335 | pc->mem_cgroup = to; |
312734c0 | 3336 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 | 3337 | ret = 0; |
0a31bc97 JW |
3338 | |
3339 | local_irq_disable(); | |
3340 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 3341 | memcg_check_events(to, page); |
0a31bc97 | 3342 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 3343 | memcg_check_events(from, page); |
0a31bc97 JW |
3344 | local_irq_enable(); |
3345 | out_unlock: | |
3346 | unlock_page(page); | |
de3638d9 | 3347 | out: |
f817ed48 KH |
3348 | return ret; |
3349 | } | |
3350 | ||
2ef37d3f MH |
3351 | /** |
3352 | * mem_cgroup_move_parent - moves page to the parent group | |
3353 | * @page: the page to move | |
3354 | * @pc: page_cgroup of the page | |
3355 | * @child: page's cgroup | |
3356 | * | |
3357 | * move charges to its parent or the root cgroup if the group has no | |
3358 | * parent (aka use_hierarchy==0). | |
3359 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3360 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3361 | * it signals a race with a page removal/uncharge or migration. In the | |
3362 | * first case the page is on the way out and it will vanish from the LRU | |
3363 | * on the next attempt and the call should be retried later. | |
3364 | * Isolation from the LRU fails only if page has been isolated from | |
3365 | * the LRU since we looked at it and that usually means either global | |
3366 | * reclaim or migration going on. The page will either get back to the | |
3367 | * LRU or vanish. | |
3368 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3369 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3370 | * disappear in the next attempt. | |
f817ed48 | 3371 | */ |
5564e88b JW |
3372 | static int mem_cgroup_move_parent(struct page *page, |
3373 | struct page_cgroup *pc, | |
6068bf01 | 3374 | struct mem_cgroup *child) |
f817ed48 | 3375 | { |
f817ed48 | 3376 | struct mem_cgroup *parent; |
7ec99d62 | 3377 | unsigned int nr_pages; |
4be4489f | 3378 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3379 | int ret; |
3380 | ||
d8423011 | 3381 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3382 | |
57f9fd7d DN |
3383 | ret = -EBUSY; |
3384 | if (!get_page_unless_zero(page)) | |
3385 | goto out; | |
3386 | if (isolate_lru_page(page)) | |
3387 | goto put; | |
52dbb905 | 3388 | |
7ec99d62 | 3389 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3390 | |
cc926f78 KH |
3391 | parent = parent_mem_cgroup(child); |
3392 | /* | |
3393 | * If no parent, move charges to root cgroup. | |
3394 | */ | |
3395 | if (!parent) | |
3396 | parent = root_mem_cgroup; | |
f817ed48 | 3397 | |
2ef37d3f | 3398 | if (nr_pages > 1) { |
309381fe | 3399 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
987eba66 | 3400 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3401 | } |
987eba66 | 3402 | |
cc926f78 | 3403 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3404 | pc, child, parent); |
3e32cb2e | 3405 | if (!ret) { |
e8ea14cc JW |
3406 | if (!mem_cgroup_is_root(parent)) |
3407 | css_get_many(&parent->css, nr_pages); | |
3e32cb2e JW |
3408 | /* Take charge off the local counters */ |
3409 | page_counter_cancel(&child->memory, nr_pages); | |
3410 | if (do_swap_account) | |
3411 | page_counter_cancel(&child->memsw, nr_pages); | |
e8ea14cc | 3412 | css_put_many(&child->css, nr_pages); |
3e32cb2e | 3413 | } |
8dba474f | 3414 | |
7ec99d62 | 3415 | if (nr_pages > 1) |
987eba66 | 3416 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3417 | putback_lru_page(page); |
57f9fd7d | 3418 | put: |
40d58138 | 3419 | put_page(page); |
57f9fd7d | 3420 | out: |
f817ed48 KH |
3421 | return ret; |
3422 | } | |
3423 | ||
c255a458 | 3424 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
3425 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
3426 | bool charge) | |
d13d1443 | 3427 | { |
0a31bc97 JW |
3428 | int val = (charge) ? 1 : -1; |
3429 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 3430 | } |
02491447 DN |
3431 | |
3432 | /** | |
3433 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3434 | * @entry: swap entry to be moved | |
3435 | * @from: mem_cgroup which the entry is moved from | |
3436 | * @to: mem_cgroup which the entry is moved to | |
3437 | * | |
3438 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3439 | * as the mem_cgroup's id of @from. | |
3440 | * | |
3441 | * Returns 0 on success, -EINVAL on failure. | |
3442 | * | |
3e32cb2e | 3443 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3444 | * both res and memsw, and called css_get(). |
3445 | */ | |
3446 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3447 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3448 | { |
3449 | unsigned short old_id, new_id; | |
3450 | ||
34c00c31 LZ |
3451 | old_id = mem_cgroup_id(from); |
3452 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3453 | |
3454 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3455 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3456 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3457 | /* |
483c30b5 | 3458 | * This function is only called from task migration context now. |
3e32cb2e | 3459 | * It postpones page_counter and refcount handling till the end |
483c30b5 | 3460 | * of task migration(mem_cgroup_clear_mc()) for performance |
4050377b LZ |
3461 | * improvement. But we cannot postpone css_get(to) because if |
3462 | * the process that has been moved to @to does swap-in, the | |
3463 | * refcount of @to might be decreased to 0. | |
3464 | * | |
3465 | * We are in attach() phase, so the cgroup is guaranteed to be | |
3466 | * alive, so we can just call css_get(). | |
02491447 | 3467 | */ |
4050377b | 3468 | css_get(&to->css); |
02491447 DN |
3469 | return 0; |
3470 | } | |
3471 | return -EINVAL; | |
3472 | } | |
3473 | #else | |
3474 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3475 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3476 | { |
3477 | return -EINVAL; | |
3478 | } | |
8c7c6e34 | 3479 | #endif |
d13d1443 | 3480 | |
f212ad7c DN |
3481 | #ifdef CONFIG_DEBUG_VM |
3482 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3483 | { | |
3484 | struct page_cgroup *pc; | |
3485 | ||
3486 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3487 | /* |
3488 | * Can be NULL while feeding pages into the page allocator for | |
3489 | * the first time, i.e. during boot or memory hotplug; | |
3490 | * or when mem_cgroup_disabled(). | |
3491 | */ | |
f212ad7c DN |
3492 | if (likely(pc) && PageCgroupUsed(pc)) |
3493 | return pc; | |
3494 | return NULL; | |
3495 | } | |
3496 | ||
3497 | bool mem_cgroup_bad_page_check(struct page *page) | |
3498 | { | |
3499 | if (mem_cgroup_disabled()) | |
3500 | return false; | |
3501 | ||
3502 | return lookup_page_cgroup_used(page) != NULL; | |
3503 | } | |
3504 | ||
3505 | void mem_cgroup_print_bad_page(struct page *page) | |
3506 | { | |
3507 | struct page_cgroup *pc; | |
3508 | ||
3509 | pc = lookup_page_cgroup_used(page); | |
3510 | if (pc) { | |
d045197f AM |
3511 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
3512 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
3513 | } |
3514 | } | |
3515 | #endif | |
3516 | ||
3e32cb2e JW |
3517 | static DEFINE_MUTEX(memcg_limit_mutex); |
3518 | ||
d38d2a75 | 3519 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3520 | unsigned long limit) |
628f4235 | 3521 | { |
3e32cb2e JW |
3522 | unsigned long curusage; |
3523 | unsigned long oldusage; | |
3524 | bool enlarge = false; | |
81d39c20 | 3525 | int retry_count; |
3e32cb2e | 3526 | int ret; |
81d39c20 KH |
3527 | |
3528 | /* | |
3529 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3530 | * is depends on callers. We set our retry-count to be function | |
3531 | * of # of children which we should visit in this loop. | |
3532 | */ | |
3e32cb2e JW |
3533 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3534 | mem_cgroup_count_children(memcg); | |
81d39c20 | 3535 | |
3e32cb2e | 3536 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 3537 | |
3e32cb2e | 3538 | do { |
628f4235 KH |
3539 | if (signal_pending(current)) { |
3540 | ret = -EINTR; | |
3541 | break; | |
3542 | } | |
3e32cb2e JW |
3543 | |
3544 | mutex_lock(&memcg_limit_mutex); | |
3545 | if (limit > memcg->memsw.limit) { | |
3546 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3547 | ret = -EINVAL; |
628f4235 KH |
3548 | break; |
3549 | } | |
3e32cb2e JW |
3550 | if (limit > memcg->memory.limit) |
3551 | enlarge = true; | |
3552 | ret = page_counter_limit(&memcg->memory, limit); | |
3553 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3554 | |
3555 | if (!ret) | |
3556 | break; | |
3557 | ||
b70a2a21 JW |
3558 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
3559 | ||
3e32cb2e | 3560 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 3561 | /* Usage is reduced ? */ |
f894ffa8 | 3562 | if (curusage >= oldusage) |
81d39c20 KH |
3563 | retry_count--; |
3564 | else | |
3565 | oldusage = curusage; | |
3e32cb2e JW |
3566 | } while (retry_count); |
3567 | ||
3c11ecf4 KH |
3568 | if (!ret && enlarge) |
3569 | memcg_oom_recover(memcg); | |
14797e23 | 3570 | |
8c7c6e34 KH |
3571 | return ret; |
3572 | } | |
3573 | ||
338c8431 | 3574 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3575 | unsigned long limit) |
8c7c6e34 | 3576 | { |
3e32cb2e JW |
3577 | unsigned long curusage; |
3578 | unsigned long oldusage; | |
3579 | bool enlarge = false; | |
81d39c20 | 3580 | int retry_count; |
3e32cb2e | 3581 | int ret; |
8c7c6e34 | 3582 | |
81d39c20 | 3583 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
3584 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3585 | mem_cgroup_count_children(memcg); | |
3586 | ||
3587 | oldusage = page_counter_read(&memcg->memsw); | |
3588 | ||
3589 | do { | |
8c7c6e34 KH |
3590 | if (signal_pending(current)) { |
3591 | ret = -EINTR; | |
3592 | break; | |
3593 | } | |
3e32cb2e JW |
3594 | |
3595 | mutex_lock(&memcg_limit_mutex); | |
3596 | if (limit < memcg->memory.limit) { | |
3597 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3598 | ret = -EINVAL; |
8c7c6e34 KH |
3599 | break; |
3600 | } | |
3e32cb2e JW |
3601 | if (limit > memcg->memsw.limit) |
3602 | enlarge = true; | |
3603 | ret = page_counter_limit(&memcg->memsw, limit); | |
3604 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3605 | |
3606 | if (!ret) | |
3607 | break; | |
3608 | ||
b70a2a21 JW |
3609 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3610 | ||
3e32cb2e | 3611 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3612 | /* Usage is reduced ? */ |
8c7c6e34 | 3613 | if (curusage >= oldusage) |
628f4235 | 3614 | retry_count--; |
81d39c20 KH |
3615 | else |
3616 | oldusage = curusage; | |
3e32cb2e JW |
3617 | } while (retry_count); |
3618 | ||
3c11ecf4 KH |
3619 | if (!ret && enlarge) |
3620 | memcg_oom_recover(memcg); | |
3e32cb2e | 3621 | |
628f4235 KH |
3622 | return ret; |
3623 | } | |
3624 | ||
0608f43d AM |
3625 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3626 | gfp_t gfp_mask, | |
3627 | unsigned long *total_scanned) | |
3628 | { | |
3629 | unsigned long nr_reclaimed = 0; | |
3630 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3631 | unsigned long reclaimed; | |
3632 | int loop = 0; | |
3633 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3634 | unsigned long excess; |
0608f43d AM |
3635 | unsigned long nr_scanned; |
3636 | ||
3637 | if (order > 0) | |
3638 | return 0; | |
3639 | ||
3640 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3641 | /* | |
3642 | * This loop can run a while, specially if mem_cgroup's continuously | |
3643 | * keep exceeding their soft limit and putting the system under | |
3644 | * pressure | |
3645 | */ | |
3646 | do { | |
3647 | if (next_mz) | |
3648 | mz = next_mz; | |
3649 | else | |
3650 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3651 | if (!mz) | |
3652 | break; | |
3653 | ||
3654 | nr_scanned = 0; | |
3655 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3656 | gfp_mask, &nr_scanned); | |
3657 | nr_reclaimed += reclaimed; | |
3658 | *total_scanned += nr_scanned; | |
0a31bc97 | 3659 | spin_lock_irq(&mctz->lock); |
0608f43d AM |
3660 | |
3661 | /* | |
3662 | * If we failed to reclaim anything from this memory cgroup | |
3663 | * it is time to move on to the next cgroup | |
3664 | */ | |
3665 | next_mz = NULL; | |
3666 | if (!reclaimed) { | |
3667 | do { | |
3668 | /* | |
3669 | * Loop until we find yet another one. | |
3670 | * | |
3671 | * By the time we get the soft_limit lock | |
3672 | * again, someone might have aded the | |
3673 | * group back on the RB tree. Iterate to | |
3674 | * make sure we get a different mem. | |
3675 | * mem_cgroup_largest_soft_limit_node returns | |
3676 | * NULL if no other cgroup is present on | |
3677 | * the tree | |
3678 | */ | |
3679 | next_mz = | |
3680 | __mem_cgroup_largest_soft_limit_node(mctz); | |
3681 | if (next_mz == mz) | |
3682 | css_put(&next_mz->memcg->css); | |
3683 | else /* next_mz == NULL or other memcg */ | |
3684 | break; | |
3685 | } while (1); | |
3686 | } | |
cf2c8127 | 3687 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 3688 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3689 | /* |
3690 | * One school of thought says that we should not add | |
3691 | * back the node to the tree if reclaim returns 0. | |
3692 | * But our reclaim could return 0, simply because due | |
3693 | * to priority we are exposing a smaller subset of | |
3694 | * memory to reclaim from. Consider this as a longer | |
3695 | * term TODO. | |
3696 | */ | |
3697 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3698 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3699 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3700 | css_put(&mz->memcg->css); |
3701 | loop++; | |
3702 | /* | |
3703 | * Could not reclaim anything and there are no more | |
3704 | * mem cgroups to try or we seem to be looping without | |
3705 | * reclaiming anything. | |
3706 | */ | |
3707 | if (!nr_reclaimed && | |
3708 | (next_mz == NULL || | |
3709 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3710 | break; | |
3711 | } while (!nr_reclaimed); | |
3712 | if (next_mz) | |
3713 | css_put(&next_mz->memcg->css); | |
3714 | return nr_reclaimed; | |
3715 | } | |
3716 | ||
2ef37d3f MH |
3717 | /** |
3718 | * mem_cgroup_force_empty_list - clears LRU of a group | |
3719 | * @memcg: group to clear | |
3720 | * @node: NUMA node | |
3721 | * @zid: zone id | |
3722 | * @lru: lru to to clear | |
3723 | * | |
3c935d18 | 3724 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
3725 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
3726 | * group. | |
cc847582 | 3727 | */ |
2ef37d3f | 3728 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3729 | int node, int zid, enum lru_list lru) |
cc847582 | 3730 | { |
bea8c150 | 3731 | struct lruvec *lruvec; |
2ef37d3f | 3732 | unsigned long flags; |
072c56c1 | 3733 | struct list_head *list; |
925b7673 JW |
3734 | struct page *busy; |
3735 | struct zone *zone; | |
072c56c1 | 3736 | |
08e552c6 | 3737 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
3738 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
3739 | list = &lruvec->lists[lru]; | |
cc847582 | 3740 | |
f817ed48 | 3741 | busy = NULL; |
2ef37d3f | 3742 | do { |
925b7673 | 3743 | struct page_cgroup *pc; |
5564e88b JW |
3744 | struct page *page; |
3745 | ||
08e552c6 | 3746 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3747 | if (list_empty(list)) { |
08e552c6 | 3748 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3749 | break; |
f817ed48 | 3750 | } |
925b7673 JW |
3751 | page = list_entry(list->prev, struct page, lru); |
3752 | if (busy == page) { | |
3753 | list_move(&page->lru, list); | |
648bcc77 | 3754 | busy = NULL; |
08e552c6 | 3755 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3756 | continue; |
3757 | } | |
08e552c6 | 3758 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3759 | |
925b7673 | 3760 | pc = lookup_page_cgroup(page); |
5564e88b | 3761 | |
3c935d18 | 3762 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 3763 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 3764 | busy = page; |
f817ed48 KH |
3765 | } else |
3766 | busy = NULL; | |
2a7a0e0f | 3767 | cond_resched(); |
2ef37d3f | 3768 | } while (!list_empty(list)); |
cc847582 KH |
3769 | } |
3770 | ||
3771 | /* | |
c26251f9 MH |
3772 | * make mem_cgroup's charge to be 0 if there is no task by moving |
3773 | * all the charges and pages to the parent. | |
cc847582 | 3774 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
3775 | * |
3776 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 3777 | */ |
ab5196c2 | 3778 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 3779 | { |
c26251f9 | 3780 | int node, zid; |
f817ed48 | 3781 | |
fce66477 | 3782 | do { |
52d4b9ac KH |
3783 | /* This is for making all *used* pages to be on LRU. */ |
3784 | lru_add_drain_all(); | |
c0ff4b85 | 3785 | drain_all_stock_sync(memcg); |
c0ff4b85 | 3786 | mem_cgroup_start_move(memcg); |
31aaea4a | 3787 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 3788 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3789 | enum lru_list lru; |
3790 | for_each_lru(lru) { | |
2ef37d3f | 3791 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3792 | node, zid, lru); |
f817ed48 | 3793 | } |
1ecaab2b | 3794 | } |
f817ed48 | 3795 | } |
c0ff4b85 R |
3796 | mem_cgroup_end_move(memcg); |
3797 | memcg_oom_recover(memcg); | |
52d4b9ac | 3798 | cond_resched(); |
f817ed48 | 3799 | |
2ef37d3f | 3800 | /* |
bea207c8 GC |
3801 | * Kernel memory may not necessarily be trackable to a specific |
3802 | * process. So they are not migrated, and therefore we can't | |
3803 | * expect their value to drop to 0 here. | |
3804 | * Having res filled up with kmem only is enough. | |
3805 | * | |
2ef37d3f MH |
3806 | * This is a safety check because mem_cgroup_force_empty_list |
3807 | * could have raced with mem_cgroup_replace_page_cache callers | |
3808 | * so the lru seemed empty but the page could have been added | |
3809 | * right after the check. RES_USAGE should be safe as we always | |
3810 | * charge before adding to the LRU. | |
3811 | */ | |
3e32cb2e JW |
3812 | } while (page_counter_read(&memcg->memory) - |
3813 | page_counter_read(&memcg->kmem) > 0); | |
c26251f9 MH |
3814 | } |
3815 | ||
ea280e7b TH |
3816 | /* |
3817 | * Test whether @memcg has children, dead or alive. Note that this | |
3818 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3819 | * returns %true if there are child csses according to the cgroup | |
3820 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3821 | */ | |
b5f99b53 GC |
3822 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3823 | { | |
ea280e7b TH |
3824 | bool ret; |
3825 | ||
696ac172 | 3826 | /* |
ea280e7b TH |
3827 | * The lock does not prevent addition or deletion of children, but |
3828 | * it prevents a new child from being initialized based on this | |
3829 | * parent in css_online(), so it's enough to decide whether | |
3830 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3831 | */ |
ea280e7b TH |
3832 | lockdep_assert_held(&memcg_create_mutex); |
3833 | ||
3834 | rcu_read_lock(); | |
3835 | ret = css_next_child(NULL, &memcg->css); | |
3836 | rcu_read_unlock(); | |
3837 | return ret; | |
b5f99b53 GC |
3838 | } |
3839 | ||
c26251f9 MH |
3840 | /* |
3841 | * Reclaims as many pages from the given memcg as possible and moves | |
3842 | * the rest to the parent. | |
3843 | * | |
3844 | * Caller is responsible for holding css reference for memcg. | |
3845 | */ | |
3846 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3847 | { | |
3848 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3849 | |
c1e862c1 KH |
3850 | /* we call try-to-free pages for make this cgroup empty */ |
3851 | lru_add_drain_all(); | |
f817ed48 | 3852 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3853 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3854 | int progress; |
c1e862c1 | 3855 | |
c26251f9 MH |
3856 | if (signal_pending(current)) |
3857 | return -EINTR; | |
3858 | ||
b70a2a21 JW |
3859 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3860 | GFP_KERNEL, true); | |
c1e862c1 | 3861 | if (!progress) { |
f817ed48 | 3862 | nr_retries--; |
c1e862c1 | 3863 | /* maybe some writeback is necessary */ |
8aa7e847 | 3864 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3865 | } |
f817ed48 KH |
3866 | |
3867 | } | |
ab5196c2 MH |
3868 | |
3869 | return 0; | |
cc847582 KH |
3870 | } |
3871 | ||
6770c64e TH |
3872 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3873 | char *buf, size_t nbytes, | |
3874 | loff_t off) | |
c1e862c1 | 3875 | { |
6770c64e | 3876 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3877 | |
d8423011 MH |
3878 | if (mem_cgroup_is_root(memcg)) |
3879 | return -EINVAL; | |
6770c64e | 3880 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3881 | } |
3882 | ||
182446d0 TH |
3883 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3884 | struct cftype *cft) | |
18f59ea7 | 3885 | { |
182446d0 | 3886 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3887 | } |
3888 | ||
182446d0 TH |
3889 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3890 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3891 | { |
3892 | int retval = 0; | |
182446d0 | 3893 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3894 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3895 | |
0999821b | 3896 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3897 | |
3898 | if (memcg->use_hierarchy == val) | |
3899 | goto out; | |
3900 | ||
18f59ea7 | 3901 | /* |
af901ca1 | 3902 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3903 | * in the child subtrees. If it is unset, then the change can |
3904 | * occur, provided the current cgroup has no children. | |
3905 | * | |
3906 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3907 | * set if there are no children. | |
3908 | */ | |
c0ff4b85 | 3909 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3910 | (val == 1 || val == 0)) { |
ea280e7b | 3911 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3912 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3913 | else |
3914 | retval = -EBUSY; | |
3915 | } else | |
3916 | retval = -EINVAL; | |
567fb435 GC |
3917 | |
3918 | out: | |
0999821b | 3919 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3920 | |
3921 | return retval; | |
3922 | } | |
3923 | ||
3e32cb2e JW |
3924 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3925 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3926 | { |
3927 | struct mem_cgroup *iter; | |
3928 | long val = 0; | |
3929 | ||
3930 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3931 | for_each_mem_cgroup_tree(iter, memcg) | |
3932 | val += mem_cgroup_read_stat(iter, idx); | |
3933 | ||
3934 | if (val < 0) /* race ? */ | |
3935 | val = 0; | |
3936 | return val; | |
3937 | } | |
3938 | ||
3939 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3940 | { | |
3941 | u64 val; | |
3942 | ||
3e32cb2e JW |
3943 | if (mem_cgroup_is_root(memcg)) { |
3944 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3945 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3946 | if (swap) | |
3947 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3948 | } else { | |
ce00a967 | 3949 | if (!swap) |
3e32cb2e | 3950 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3951 | else |
3e32cb2e | 3952 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3953 | } |
ce00a967 JW |
3954 | return val << PAGE_SHIFT; |
3955 | } | |
3956 | ||
3e32cb2e JW |
3957 | enum { |
3958 | RES_USAGE, | |
3959 | RES_LIMIT, | |
3960 | RES_MAX_USAGE, | |
3961 | RES_FAILCNT, | |
3962 | RES_SOFT_LIMIT, | |
3963 | }; | |
ce00a967 | 3964 | |
791badbd | 3965 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3966 | struct cftype *cft) |
8cdea7c0 | 3967 | { |
182446d0 | 3968 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3969 | struct page_counter *counter; |
af36f906 | 3970 | |
3e32cb2e | 3971 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3972 | case _MEM: |
3e32cb2e JW |
3973 | counter = &memcg->memory; |
3974 | break; | |
8c7c6e34 | 3975 | case _MEMSWAP: |
3e32cb2e JW |
3976 | counter = &memcg->memsw; |
3977 | break; | |
510fc4e1 | 3978 | case _KMEM: |
3e32cb2e | 3979 | counter = &memcg->kmem; |
510fc4e1 | 3980 | break; |
8c7c6e34 KH |
3981 | default: |
3982 | BUG(); | |
8c7c6e34 | 3983 | } |
3e32cb2e JW |
3984 | |
3985 | switch (MEMFILE_ATTR(cft->private)) { | |
3986 | case RES_USAGE: | |
3987 | if (counter == &memcg->memory) | |
3988 | return mem_cgroup_usage(memcg, false); | |
3989 | if (counter == &memcg->memsw) | |
3990 | return mem_cgroup_usage(memcg, true); | |
3991 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3992 | case RES_LIMIT: | |
3993 | return (u64)counter->limit * PAGE_SIZE; | |
3994 | case RES_MAX_USAGE: | |
3995 | return (u64)counter->watermark * PAGE_SIZE; | |
3996 | case RES_FAILCNT: | |
3997 | return counter->failcnt; | |
3998 | case RES_SOFT_LIMIT: | |
3999 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
4000 | default: | |
4001 | BUG(); | |
4002 | } | |
8cdea7c0 | 4003 | } |
510fc4e1 | 4004 | |
510fc4e1 | 4005 | #ifdef CONFIG_MEMCG_KMEM |
d6441637 VD |
4006 | /* should be called with activate_kmem_mutex held */ |
4007 | static int __memcg_activate_kmem(struct mem_cgroup *memcg, | |
3e32cb2e | 4008 | unsigned long nr_pages) |
d6441637 VD |
4009 | { |
4010 | int err = 0; | |
4011 | int memcg_id; | |
4012 | ||
4013 | if (memcg_kmem_is_active(memcg)) | |
4014 | return 0; | |
4015 | ||
4016 | /* | |
4017 | * We are going to allocate memory for data shared by all memory | |
4018 | * cgroups so let's stop accounting here. | |
4019 | */ | |
4020 | memcg_stop_kmem_account(); | |
4021 | ||
510fc4e1 GC |
4022 | /* |
4023 | * For simplicity, we won't allow this to be disabled. It also can't | |
4024 | * be changed if the cgroup has children already, or if tasks had | |
4025 | * already joined. | |
4026 | * | |
4027 | * If tasks join before we set the limit, a person looking at | |
4028 | * kmem.usage_in_bytes will have no way to determine when it took | |
4029 | * place, which makes the value quite meaningless. | |
4030 | * | |
4031 | * After it first became limited, changes in the value of the limit are | |
4032 | * of course permitted. | |
510fc4e1 | 4033 | */ |
0999821b | 4034 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
4035 | if (cgroup_has_tasks(memcg->css.cgroup) || |
4036 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
4037 | err = -EBUSY; |
4038 | mutex_unlock(&memcg_create_mutex); | |
4039 | if (err) | |
4040 | goto out; | |
510fc4e1 | 4041 | |
f3bb3043 | 4042 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
4043 | if (memcg_id < 0) { |
4044 | err = memcg_id; | |
4045 | goto out; | |
4046 | } | |
4047 | ||
d6441637 VD |
4048 | memcg->kmemcg_id = memcg_id; |
4049 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
d6441637 VD |
4050 | |
4051 | /* | |
4052 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
4053 | * active bit set yet, so this should succeed. | |
4054 | */ | |
3e32cb2e | 4055 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
4056 | VM_BUG_ON(err); |
4057 | ||
4058 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
4059 | /* | |
4060 | * Setting the active bit after enabling static branching will | |
4061 | * guarantee no one starts accounting before all call sites are | |
4062 | * patched. | |
4063 | */ | |
4064 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 4065 | out: |
d6441637 VD |
4066 | memcg_resume_kmem_account(); |
4067 | return err; | |
d6441637 VD |
4068 | } |
4069 | ||
4070 | static int memcg_activate_kmem(struct mem_cgroup *memcg, | |
3e32cb2e | 4071 | unsigned long nr_pages) |
d6441637 VD |
4072 | { |
4073 | int ret; | |
4074 | ||
4075 | mutex_lock(&activate_kmem_mutex); | |
3e32cb2e | 4076 | ret = __memcg_activate_kmem(memcg, nr_pages); |
d6441637 VD |
4077 | mutex_unlock(&activate_kmem_mutex); |
4078 | return ret; | |
4079 | } | |
4080 | ||
4081 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 4082 | unsigned long limit) |
d6441637 VD |
4083 | { |
4084 | int ret; | |
4085 | ||
3e32cb2e | 4086 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 4087 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 4088 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 4089 | else |
3e32cb2e JW |
4090 | ret = page_counter_limit(&memcg->kmem, limit); |
4091 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
4092 | return ret; |
4093 | } | |
4094 | ||
55007d84 | 4095 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 4096 | { |
55007d84 | 4097 | int ret = 0; |
510fc4e1 | 4098 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 4099 | |
d6441637 VD |
4100 | if (!parent) |
4101 | return 0; | |
55007d84 | 4102 | |
d6441637 | 4103 | mutex_lock(&activate_kmem_mutex); |
55007d84 | 4104 | /* |
d6441637 VD |
4105 | * If the parent cgroup is not kmem-active now, it cannot be activated |
4106 | * after this point, because it has at least one child already. | |
55007d84 | 4107 | */ |
d6441637 | 4108 | if (memcg_kmem_is_active(parent)) |
3e32cb2e | 4109 | ret = __memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
d6441637 | 4110 | mutex_unlock(&activate_kmem_mutex); |
55007d84 | 4111 | return ret; |
510fc4e1 | 4112 | } |
d6441637 VD |
4113 | #else |
4114 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 4115 | unsigned long limit) |
d6441637 VD |
4116 | { |
4117 | return -EINVAL; | |
4118 | } | |
6d043990 | 4119 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 4120 | |
628f4235 KH |
4121 | /* |
4122 | * The user of this function is... | |
4123 | * RES_LIMIT. | |
4124 | */ | |
451af504 TH |
4125 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
4126 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 4127 | { |
451af504 | 4128 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4129 | unsigned long nr_pages; |
628f4235 KH |
4130 | int ret; |
4131 | ||
451af504 | 4132 | buf = strstrip(buf); |
3e32cb2e JW |
4133 | ret = page_counter_memparse(buf, &nr_pages); |
4134 | if (ret) | |
4135 | return ret; | |
af36f906 | 4136 | |
3e32cb2e | 4137 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 4138 | case RES_LIMIT: |
4b3bde4c BS |
4139 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4140 | ret = -EINVAL; | |
4141 | break; | |
4142 | } | |
3e32cb2e JW |
4143 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4144 | case _MEM: | |
4145 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 4146 | break; |
3e32cb2e JW |
4147 | case _MEMSWAP: |
4148 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 4149 | break; |
3e32cb2e JW |
4150 | case _KMEM: |
4151 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
4152 | break; | |
4153 | } | |
296c81d8 | 4154 | break; |
3e32cb2e JW |
4155 | case RES_SOFT_LIMIT: |
4156 | memcg->soft_limit = nr_pages; | |
4157 | ret = 0; | |
628f4235 KH |
4158 | break; |
4159 | } | |
451af504 | 4160 | return ret ?: nbytes; |
8cdea7c0 BS |
4161 | } |
4162 | ||
6770c64e TH |
4163 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
4164 | size_t nbytes, loff_t off) | |
c84872e1 | 4165 | { |
6770c64e | 4166 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 4167 | struct page_counter *counter; |
c84872e1 | 4168 | |
3e32cb2e JW |
4169 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
4170 | case _MEM: | |
4171 | counter = &memcg->memory; | |
4172 | break; | |
4173 | case _MEMSWAP: | |
4174 | counter = &memcg->memsw; | |
4175 | break; | |
4176 | case _KMEM: | |
4177 | counter = &memcg->kmem; | |
4178 | break; | |
4179 | default: | |
4180 | BUG(); | |
4181 | } | |
af36f906 | 4182 | |
3e32cb2e | 4183 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 4184 | case RES_MAX_USAGE: |
3e32cb2e | 4185 | page_counter_reset_watermark(counter); |
29f2a4da PE |
4186 | break; |
4187 | case RES_FAILCNT: | |
3e32cb2e | 4188 | counter->failcnt = 0; |
29f2a4da | 4189 | break; |
3e32cb2e JW |
4190 | default: |
4191 | BUG(); | |
29f2a4da | 4192 | } |
f64c3f54 | 4193 | |
6770c64e | 4194 | return nbytes; |
c84872e1 PE |
4195 | } |
4196 | ||
182446d0 | 4197 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4198 | struct cftype *cft) |
4199 | { | |
182446d0 | 4200 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
4201 | } |
4202 | ||
02491447 | 4203 | #ifdef CONFIG_MMU |
182446d0 | 4204 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
4205 | struct cftype *cft, u64 val) |
4206 | { | |
182446d0 | 4207 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
4208 | |
4209 | if (val >= (1 << NR_MOVE_TYPE)) | |
4210 | return -EINVAL; | |
ee5e8472 | 4211 | |
7dc74be0 | 4212 | /* |
ee5e8472 GC |
4213 | * No kind of locking is needed in here, because ->can_attach() will |
4214 | * check this value once in the beginning of the process, and then carry | |
4215 | * on with stale data. This means that changes to this value will only | |
4216 | * affect task migrations starting after the change. | |
7dc74be0 | 4217 | */ |
c0ff4b85 | 4218 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4219 | return 0; |
4220 | } | |
02491447 | 4221 | #else |
182446d0 | 4222 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
4223 | struct cftype *cft, u64 val) |
4224 | { | |
4225 | return -ENOSYS; | |
4226 | } | |
4227 | #endif | |
7dc74be0 | 4228 | |
406eb0c9 | 4229 | #ifdef CONFIG_NUMA |
2da8ca82 | 4230 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 4231 | { |
25485de6 GT |
4232 | struct numa_stat { |
4233 | const char *name; | |
4234 | unsigned int lru_mask; | |
4235 | }; | |
4236 | ||
4237 | static const struct numa_stat stats[] = { | |
4238 | { "total", LRU_ALL }, | |
4239 | { "file", LRU_ALL_FILE }, | |
4240 | { "anon", LRU_ALL_ANON }, | |
4241 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
4242 | }; | |
4243 | const struct numa_stat *stat; | |
406eb0c9 | 4244 | int nid; |
25485de6 | 4245 | unsigned long nr; |
2da8ca82 | 4246 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 4247 | |
25485de6 GT |
4248 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
4249 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
4250 | seq_printf(m, "%s=%lu", stat->name, nr); | |
4251 | for_each_node_state(nid, N_MEMORY) { | |
4252 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
4253 | stat->lru_mask); | |
4254 | seq_printf(m, " N%d=%lu", nid, nr); | |
4255 | } | |
4256 | seq_putc(m, '\n'); | |
406eb0c9 | 4257 | } |
406eb0c9 | 4258 | |
071aee13 YH |
4259 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
4260 | struct mem_cgroup *iter; | |
4261 | ||
4262 | nr = 0; | |
4263 | for_each_mem_cgroup_tree(iter, memcg) | |
4264 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
4265 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
4266 | for_each_node_state(nid, N_MEMORY) { | |
4267 | nr = 0; | |
4268 | for_each_mem_cgroup_tree(iter, memcg) | |
4269 | nr += mem_cgroup_node_nr_lru_pages( | |
4270 | iter, nid, stat->lru_mask); | |
4271 | seq_printf(m, " N%d=%lu", nid, nr); | |
4272 | } | |
4273 | seq_putc(m, '\n'); | |
406eb0c9 | 4274 | } |
406eb0c9 | 4275 | |
406eb0c9 YH |
4276 | return 0; |
4277 | } | |
4278 | #endif /* CONFIG_NUMA */ | |
4279 | ||
af7c4b0e JW |
4280 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
4281 | { | |
4282 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4283 | } | |
4284 | ||
2da8ca82 | 4285 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4286 | { |
2da8ca82 | 4287 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 4288 | unsigned long memory, memsw; |
af7c4b0e JW |
4289 | struct mem_cgroup *mi; |
4290 | unsigned int i; | |
406eb0c9 | 4291 | |
af7c4b0e | 4292 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4293 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4294 | continue; |
af7c4b0e JW |
4295 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4296 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4297 | } |
7b854121 | 4298 | |
af7c4b0e JW |
4299 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4300 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4301 | mem_cgroup_read_events(memcg, i)); | |
4302 | ||
4303 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4304 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4305 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4306 | ||
14067bb3 | 4307 | /* Hierarchical information */ |
3e32cb2e JW |
4308 | memory = memsw = PAGE_COUNTER_MAX; |
4309 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
4310 | memory = min(memory, mi->memory.limit); | |
4311 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 4312 | } |
3e32cb2e JW |
4313 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4314 | (u64)memory * PAGE_SIZE); | |
4315 | if (do_swap_account) | |
4316 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
4317 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4318 | |
af7c4b0e JW |
4319 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4320 | long long val = 0; | |
4321 | ||
bff6bb83 | 4322 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4323 | continue; |
af7c4b0e JW |
4324 | for_each_mem_cgroup_tree(mi, memcg) |
4325 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4326 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4327 | } | |
4328 | ||
4329 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4330 | unsigned long long val = 0; | |
4331 | ||
4332 | for_each_mem_cgroup_tree(mi, memcg) | |
4333 | val += mem_cgroup_read_events(mi, i); | |
4334 | seq_printf(m, "total_%s %llu\n", | |
4335 | mem_cgroup_events_names[i], val); | |
4336 | } | |
4337 | ||
4338 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4339 | unsigned long long val = 0; | |
4340 | ||
4341 | for_each_mem_cgroup_tree(mi, memcg) | |
4342 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4343 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4344 | } |
14067bb3 | 4345 | |
7f016ee8 | 4346 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4347 | { |
4348 | int nid, zid; | |
4349 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4350 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4351 | unsigned long recent_rotated[2] = {0, 0}; |
4352 | unsigned long recent_scanned[2] = {0, 0}; | |
4353 | ||
4354 | for_each_online_node(nid) | |
4355 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 4356 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 4357 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4358 | |
89abfab1 HD |
4359 | recent_rotated[0] += rstat->recent_rotated[0]; |
4360 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4361 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4362 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4363 | } |
78ccf5b5 JW |
4364 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4365 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4366 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4367 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4368 | } |
4369 | #endif | |
4370 | ||
d2ceb9b7 KH |
4371 | return 0; |
4372 | } | |
4373 | ||
182446d0 TH |
4374 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4375 | struct cftype *cft) | |
a7885eb8 | 4376 | { |
182446d0 | 4377 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4378 | |
1f4c025b | 4379 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4380 | } |
4381 | ||
182446d0 TH |
4382 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4383 | struct cftype *cft, u64 val) | |
a7885eb8 | 4384 | { |
182446d0 | 4385 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4386 | |
3dae7fec | 4387 | if (val > 100) |
a7885eb8 KM |
4388 | return -EINVAL; |
4389 | ||
14208b0e | 4390 | if (css->parent) |
3dae7fec JW |
4391 | memcg->swappiness = val; |
4392 | else | |
4393 | vm_swappiness = val; | |
068b38c1 | 4394 | |
a7885eb8 KM |
4395 | return 0; |
4396 | } | |
4397 | ||
2e72b634 KS |
4398 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4399 | { | |
4400 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4401 | unsigned long usage; |
2e72b634 KS |
4402 | int i; |
4403 | ||
4404 | rcu_read_lock(); | |
4405 | if (!swap) | |
2c488db2 | 4406 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4407 | else |
2c488db2 | 4408 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4409 | |
4410 | if (!t) | |
4411 | goto unlock; | |
4412 | ||
ce00a967 | 4413 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4414 | |
4415 | /* | |
748dad36 | 4416 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4417 | * If it's not true, a threshold was crossed after last |
4418 | * call of __mem_cgroup_threshold(). | |
4419 | */ | |
5407a562 | 4420 | i = t->current_threshold; |
2e72b634 KS |
4421 | |
4422 | /* | |
4423 | * Iterate backward over array of thresholds starting from | |
4424 | * current_threshold and check if a threshold is crossed. | |
4425 | * If none of thresholds below usage is crossed, we read | |
4426 | * only one element of the array here. | |
4427 | */ | |
4428 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4429 | eventfd_signal(t->entries[i].eventfd, 1); | |
4430 | ||
4431 | /* i = current_threshold + 1 */ | |
4432 | i++; | |
4433 | ||
4434 | /* | |
4435 | * Iterate forward over array of thresholds starting from | |
4436 | * current_threshold+1 and check if a threshold is crossed. | |
4437 | * If none of thresholds above usage is crossed, we read | |
4438 | * only one element of the array here. | |
4439 | */ | |
4440 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4441 | eventfd_signal(t->entries[i].eventfd, 1); | |
4442 | ||
4443 | /* Update current_threshold */ | |
5407a562 | 4444 | t->current_threshold = i - 1; |
2e72b634 KS |
4445 | unlock: |
4446 | rcu_read_unlock(); | |
4447 | } | |
4448 | ||
4449 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4450 | { | |
ad4ca5f4 KS |
4451 | while (memcg) { |
4452 | __mem_cgroup_threshold(memcg, false); | |
4453 | if (do_swap_account) | |
4454 | __mem_cgroup_threshold(memcg, true); | |
4455 | ||
4456 | memcg = parent_mem_cgroup(memcg); | |
4457 | } | |
2e72b634 KS |
4458 | } |
4459 | ||
4460 | static int compare_thresholds(const void *a, const void *b) | |
4461 | { | |
4462 | const struct mem_cgroup_threshold *_a = a; | |
4463 | const struct mem_cgroup_threshold *_b = b; | |
4464 | ||
2bff24a3 GT |
4465 | if (_a->threshold > _b->threshold) |
4466 | return 1; | |
4467 | ||
4468 | if (_a->threshold < _b->threshold) | |
4469 | return -1; | |
4470 | ||
4471 | return 0; | |
2e72b634 KS |
4472 | } |
4473 | ||
c0ff4b85 | 4474 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4475 | { |
4476 | struct mem_cgroup_eventfd_list *ev; | |
4477 | ||
2bcf2e92 MH |
4478 | spin_lock(&memcg_oom_lock); |
4479 | ||
c0ff4b85 | 4480 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4481 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4482 | |
4483 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4484 | return 0; |
4485 | } | |
4486 | ||
c0ff4b85 | 4487 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4488 | { |
7d74b06f KH |
4489 | struct mem_cgroup *iter; |
4490 | ||
c0ff4b85 | 4491 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4492 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4493 | } |
4494 | ||
59b6f873 | 4495 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4496 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4497 | { |
2c488db2 KS |
4498 | struct mem_cgroup_thresholds *thresholds; |
4499 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4500 | unsigned long threshold; |
4501 | unsigned long usage; | |
2c488db2 | 4502 | int i, size, ret; |
2e72b634 | 4503 | |
3e32cb2e | 4504 | ret = page_counter_memparse(args, &threshold); |
2e72b634 KS |
4505 | if (ret) |
4506 | return ret; | |
4507 | ||
4508 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4509 | |
05b84301 | 4510 | if (type == _MEM) { |
2c488db2 | 4511 | thresholds = &memcg->thresholds; |
ce00a967 | 4512 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4513 | } else if (type == _MEMSWAP) { |
2c488db2 | 4514 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4515 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4516 | } else |
2e72b634 KS |
4517 | BUG(); |
4518 | ||
2e72b634 | 4519 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4520 | if (thresholds->primary) |
2e72b634 KS |
4521 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4522 | ||
2c488db2 | 4523 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4524 | |
4525 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4526 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4527 | GFP_KERNEL); |
2c488db2 | 4528 | if (!new) { |
2e72b634 KS |
4529 | ret = -ENOMEM; |
4530 | goto unlock; | |
4531 | } | |
2c488db2 | 4532 | new->size = size; |
2e72b634 KS |
4533 | |
4534 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4535 | if (thresholds->primary) { |
4536 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4537 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4538 | } |
4539 | ||
2e72b634 | 4540 | /* Add new threshold */ |
2c488db2 KS |
4541 | new->entries[size - 1].eventfd = eventfd; |
4542 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4543 | |
4544 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4545 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4546 | compare_thresholds, NULL); |
4547 | ||
4548 | /* Find current threshold */ | |
2c488db2 | 4549 | new->current_threshold = -1; |
2e72b634 | 4550 | for (i = 0; i < size; i++) { |
748dad36 | 4551 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4552 | /* |
2c488db2 KS |
4553 | * new->current_threshold will not be used until |
4554 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4555 | * it here. |
4556 | */ | |
2c488db2 | 4557 | ++new->current_threshold; |
748dad36 SZ |
4558 | } else |
4559 | break; | |
2e72b634 KS |
4560 | } |
4561 | ||
2c488db2 KS |
4562 | /* Free old spare buffer and save old primary buffer as spare */ |
4563 | kfree(thresholds->spare); | |
4564 | thresholds->spare = thresholds->primary; | |
4565 | ||
4566 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4567 | |
907860ed | 4568 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4569 | synchronize_rcu(); |
4570 | ||
2e72b634 KS |
4571 | unlock: |
4572 | mutex_unlock(&memcg->thresholds_lock); | |
4573 | ||
4574 | return ret; | |
4575 | } | |
4576 | ||
59b6f873 | 4577 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4578 | struct eventfd_ctx *eventfd, const char *args) |
4579 | { | |
59b6f873 | 4580 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4581 | } |
4582 | ||
59b6f873 | 4583 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4584 | struct eventfd_ctx *eventfd, const char *args) |
4585 | { | |
59b6f873 | 4586 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4587 | } |
4588 | ||
59b6f873 | 4589 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4590 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4591 | { |
2c488db2 KS |
4592 | struct mem_cgroup_thresholds *thresholds; |
4593 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4594 | unsigned long usage; |
2c488db2 | 4595 | int i, j, size; |
2e72b634 KS |
4596 | |
4597 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4598 | |
4599 | if (type == _MEM) { | |
2c488db2 | 4600 | thresholds = &memcg->thresholds; |
ce00a967 | 4601 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4602 | } else if (type == _MEMSWAP) { |
2c488db2 | 4603 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4604 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4605 | } else |
2e72b634 KS |
4606 | BUG(); |
4607 | ||
371528ca AV |
4608 | if (!thresholds->primary) |
4609 | goto unlock; | |
4610 | ||
2e72b634 KS |
4611 | /* Check if a threshold crossed before removing */ |
4612 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4613 | ||
4614 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4615 | size = 0; |
4616 | for (i = 0; i < thresholds->primary->size; i++) { | |
4617 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4618 | size++; |
4619 | } | |
4620 | ||
2c488db2 | 4621 | new = thresholds->spare; |
907860ed | 4622 | |
2e72b634 KS |
4623 | /* Set thresholds array to NULL if we don't have thresholds */ |
4624 | if (!size) { | |
2c488db2 KS |
4625 | kfree(new); |
4626 | new = NULL; | |
907860ed | 4627 | goto swap_buffers; |
2e72b634 KS |
4628 | } |
4629 | ||
2c488db2 | 4630 | new->size = size; |
2e72b634 KS |
4631 | |
4632 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4633 | new->current_threshold = -1; |
4634 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4635 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4636 | continue; |
4637 | ||
2c488db2 | 4638 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4639 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4640 | /* |
2c488db2 | 4641 | * new->current_threshold will not be used |
2e72b634 KS |
4642 | * until rcu_assign_pointer(), so it's safe to increment |
4643 | * it here. | |
4644 | */ | |
2c488db2 | 4645 | ++new->current_threshold; |
2e72b634 KS |
4646 | } |
4647 | j++; | |
4648 | } | |
4649 | ||
907860ed | 4650 | swap_buffers: |
2c488db2 KS |
4651 | /* Swap primary and spare array */ |
4652 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4653 | /* If all events are unregistered, free the spare array */ |
4654 | if (!new) { | |
4655 | kfree(thresholds->spare); | |
4656 | thresholds->spare = NULL; | |
4657 | } | |
4658 | ||
2c488db2 | 4659 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4660 | |
907860ed | 4661 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4662 | synchronize_rcu(); |
371528ca | 4663 | unlock: |
2e72b634 | 4664 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4665 | } |
c1e862c1 | 4666 | |
59b6f873 | 4667 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4668 | struct eventfd_ctx *eventfd) |
4669 | { | |
59b6f873 | 4670 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4671 | } |
4672 | ||
59b6f873 | 4673 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4674 | struct eventfd_ctx *eventfd) |
4675 | { | |
59b6f873 | 4676 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4677 | } |
4678 | ||
59b6f873 | 4679 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4680 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4681 | { |
9490ff27 | 4682 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4683 | |
9490ff27 KH |
4684 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4685 | if (!event) | |
4686 | return -ENOMEM; | |
4687 | ||
1af8efe9 | 4688 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4689 | |
4690 | event->eventfd = eventfd; | |
4691 | list_add(&event->list, &memcg->oom_notify); | |
4692 | ||
4693 | /* already in OOM ? */ | |
79dfdacc | 4694 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4695 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4696 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4697 | |
4698 | return 0; | |
4699 | } | |
4700 | ||
59b6f873 | 4701 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4702 | struct eventfd_ctx *eventfd) |
9490ff27 | 4703 | { |
9490ff27 | 4704 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4705 | |
1af8efe9 | 4706 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4707 | |
c0ff4b85 | 4708 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4709 | if (ev->eventfd == eventfd) { |
4710 | list_del(&ev->list); | |
4711 | kfree(ev); | |
4712 | } | |
4713 | } | |
4714 | ||
1af8efe9 | 4715 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4716 | } |
4717 | ||
2da8ca82 | 4718 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4719 | { |
2da8ca82 | 4720 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4721 | |
791badbd TH |
4722 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4723 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4724 | return 0; |
4725 | } | |
4726 | ||
182446d0 | 4727 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4728 | struct cftype *cft, u64 val) |
4729 | { | |
182446d0 | 4730 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4731 | |
4732 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4733 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4734 | return -EINVAL; |
4735 | ||
c0ff4b85 | 4736 | memcg->oom_kill_disable = val; |
4d845ebf | 4737 | if (!val) |
c0ff4b85 | 4738 | memcg_oom_recover(memcg); |
3dae7fec | 4739 | |
3c11ecf4 KH |
4740 | return 0; |
4741 | } | |
4742 | ||
c255a458 | 4743 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4744 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4745 | { |
55007d84 GC |
4746 | int ret; |
4747 | ||
2633d7a0 | 4748 | memcg->kmemcg_id = -1; |
55007d84 GC |
4749 | ret = memcg_propagate_kmem(memcg); |
4750 | if (ret) | |
4751 | return ret; | |
2633d7a0 | 4752 | |
1d62e436 | 4753 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4754 | } |
e5671dfa | 4755 | |
10d5ebf4 | 4756 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4757 | { |
1d62e436 | 4758 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 LZ |
4759 | } |
4760 | ||
4761 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
4762 | { | |
4763 | if (!memcg_kmem_is_active(memcg)) | |
4764 | return; | |
4765 | ||
4766 | /* | |
4767 | * kmem charges can outlive the cgroup. In the case of slab | |
4768 | * pages, for instance, a page contain objects from various | |
4769 | * processes. As we prevent from taking a reference for every | |
4770 | * such allocation we have to be careful when doing uncharge | |
4771 | * (see memcg_uncharge_kmem) and here during offlining. | |
4772 | * | |
4773 | * The idea is that that only the _last_ uncharge which sees | |
4774 | * the dead memcg will drop the last reference. An additional | |
4775 | * reference is taken here before the group is marked dead | |
4776 | * which is then paired with css_put during uncharge resp. here. | |
4777 | * | |
4778 | * Although this might sound strange as this path is called from | |
ec903c0c TH |
4779 | * css_offline() when the referencemight have dropped down to 0 and |
4780 | * shouldn't be incremented anymore (css_tryget_online() would | |
4781 | * fail) we do not have other options because of the kmem | |
4782 | * allocations lifetime. | |
10d5ebf4 LZ |
4783 | */ |
4784 | css_get(&memcg->css); | |
7de37682 GC |
4785 | |
4786 | memcg_kmem_mark_dead(memcg); | |
4787 | ||
3e32cb2e | 4788 | if (page_counter_read(&memcg->kmem)) |
7de37682 GC |
4789 | return; |
4790 | ||
7de37682 | 4791 | if (memcg_kmem_test_and_clear_dead(memcg)) |
10d5ebf4 | 4792 | css_put(&memcg->css); |
d1a4c0b3 | 4793 | } |
e5671dfa | 4794 | #else |
cbe128e3 | 4795 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4796 | { |
4797 | return 0; | |
4798 | } | |
d1a4c0b3 | 4799 | |
10d5ebf4 LZ |
4800 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4801 | { | |
4802 | } | |
4803 | ||
4804 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) | |
d1a4c0b3 GC |
4805 | { |
4806 | } | |
e5671dfa GC |
4807 | #endif |
4808 | ||
3bc942f3 TH |
4809 | /* |
4810 | * DO NOT USE IN NEW FILES. | |
4811 | * | |
4812 | * "cgroup.event_control" implementation. | |
4813 | * | |
4814 | * This is way over-engineered. It tries to support fully configurable | |
4815 | * events for each user. Such level of flexibility is completely | |
4816 | * unnecessary especially in the light of the planned unified hierarchy. | |
4817 | * | |
4818 | * Please deprecate this and replace with something simpler if at all | |
4819 | * possible. | |
4820 | */ | |
4821 | ||
79bd9814 TH |
4822 | /* |
4823 | * Unregister event and free resources. | |
4824 | * | |
4825 | * Gets called from workqueue. | |
4826 | */ | |
3bc942f3 | 4827 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4828 | { |
3bc942f3 TH |
4829 | struct mem_cgroup_event *event = |
4830 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4831 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4832 | |
4833 | remove_wait_queue(event->wqh, &event->wait); | |
4834 | ||
59b6f873 | 4835 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4836 | |
4837 | /* Notify userspace the event is going away. */ | |
4838 | eventfd_signal(event->eventfd, 1); | |
4839 | ||
4840 | eventfd_ctx_put(event->eventfd); | |
4841 | kfree(event); | |
59b6f873 | 4842 | css_put(&memcg->css); |
79bd9814 TH |
4843 | } |
4844 | ||
4845 | /* | |
4846 | * Gets called on POLLHUP on eventfd when user closes it. | |
4847 | * | |
4848 | * Called with wqh->lock held and interrupts disabled. | |
4849 | */ | |
3bc942f3 TH |
4850 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4851 | int sync, void *key) | |
79bd9814 | 4852 | { |
3bc942f3 TH |
4853 | struct mem_cgroup_event *event = |
4854 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4855 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4856 | unsigned long flags = (unsigned long)key; |
4857 | ||
4858 | if (flags & POLLHUP) { | |
4859 | /* | |
4860 | * If the event has been detached at cgroup removal, we | |
4861 | * can simply return knowing the other side will cleanup | |
4862 | * for us. | |
4863 | * | |
4864 | * We can't race against event freeing since the other | |
4865 | * side will require wqh->lock via remove_wait_queue(), | |
4866 | * which we hold. | |
4867 | */ | |
fba94807 | 4868 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4869 | if (!list_empty(&event->list)) { |
4870 | list_del_init(&event->list); | |
4871 | /* | |
4872 | * We are in atomic context, but cgroup_event_remove() | |
4873 | * may sleep, so we have to call it in workqueue. | |
4874 | */ | |
4875 | schedule_work(&event->remove); | |
4876 | } | |
fba94807 | 4877 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4878 | } |
4879 | ||
4880 | return 0; | |
4881 | } | |
4882 | ||
3bc942f3 | 4883 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4884 | wait_queue_head_t *wqh, poll_table *pt) |
4885 | { | |
3bc942f3 TH |
4886 | struct mem_cgroup_event *event = |
4887 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4888 | |
4889 | event->wqh = wqh; | |
4890 | add_wait_queue(wqh, &event->wait); | |
4891 | } | |
4892 | ||
4893 | /* | |
3bc942f3 TH |
4894 | * DO NOT USE IN NEW FILES. |
4895 | * | |
79bd9814 TH |
4896 | * Parse input and register new cgroup event handler. |
4897 | * | |
4898 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4899 | * Interpretation of args is defined by control file implementation. | |
4900 | */ | |
451af504 TH |
4901 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4902 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4903 | { |
451af504 | 4904 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4905 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4906 | struct mem_cgroup_event *event; |
79bd9814 TH |
4907 | struct cgroup_subsys_state *cfile_css; |
4908 | unsigned int efd, cfd; | |
4909 | struct fd efile; | |
4910 | struct fd cfile; | |
fba94807 | 4911 | const char *name; |
79bd9814 TH |
4912 | char *endp; |
4913 | int ret; | |
4914 | ||
451af504 TH |
4915 | buf = strstrip(buf); |
4916 | ||
4917 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4918 | if (*endp != ' ') |
4919 | return -EINVAL; | |
451af504 | 4920 | buf = endp + 1; |
79bd9814 | 4921 | |
451af504 | 4922 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4923 | if ((*endp != ' ') && (*endp != '\0')) |
4924 | return -EINVAL; | |
451af504 | 4925 | buf = endp + 1; |
79bd9814 TH |
4926 | |
4927 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4928 | if (!event) | |
4929 | return -ENOMEM; | |
4930 | ||
59b6f873 | 4931 | event->memcg = memcg; |
79bd9814 | 4932 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4933 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4934 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4935 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4936 | |
4937 | efile = fdget(efd); | |
4938 | if (!efile.file) { | |
4939 | ret = -EBADF; | |
4940 | goto out_kfree; | |
4941 | } | |
4942 | ||
4943 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4944 | if (IS_ERR(event->eventfd)) { | |
4945 | ret = PTR_ERR(event->eventfd); | |
4946 | goto out_put_efile; | |
4947 | } | |
4948 | ||
4949 | cfile = fdget(cfd); | |
4950 | if (!cfile.file) { | |
4951 | ret = -EBADF; | |
4952 | goto out_put_eventfd; | |
4953 | } | |
4954 | ||
4955 | /* the process need read permission on control file */ | |
4956 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4957 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4958 | if (ret < 0) | |
4959 | goto out_put_cfile; | |
4960 | ||
fba94807 TH |
4961 | /* |
4962 | * Determine the event callbacks and set them in @event. This used | |
4963 | * to be done via struct cftype but cgroup core no longer knows | |
4964 | * about these events. The following is crude but the whole thing | |
4965 | * is for compatibility anyway. | |
3bc942f3 TH |
4966 | * |
4967 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
4968 | */ |
4969 | name = cfile.file->f_dentry->d_name.name; | |
4970 | ||
4971 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4972 | event->register_event = mem_cgroup_usage_register_event; | |
4973 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4974 | } else if (!strcmp(name, "memory.oom_control")) { | |
4975 | event->register_event = mem_cgroup_oom_register_event; | |
4976 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4977 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4978 | event->register_event = vmpressure_register_event; | |
4979 | event->unregister_event = vmpressure_unregister_event; | |
4980 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4981 | event->register_event = memsw_cgroup_usage_register_event; |
4982 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4983 | } else { |
4984 | ret = -EINVAL; | |
4985 | goto out_put_cfile; | |
4986 | } | |
4987 | ||
79bd9814 | 4988 | /* |
b5557c4c TH |
4989 | * Verify @cfile should belong to @css. Also, remaining events are |
4990 | * automatically removed on cgroup destruction but the removal is | |
4991 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4992 | */ |
ec903c0c TH |
4993 | cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent, |
4994 | &memory_cgrp_subsys); | |
79bd9814 | 4995 | ret = -EINVAL; |
5a17f543 | 4996 | if (IS_ERR(cfile_css)) |
79bd9814 | 4997 | goto out_put_cfile; |
5a17f543 TH |
4998 | if (cfile_css != css) { |
4999 | css_put(cfile_css); | |
79bd9814 | 5000 | goto out_put_cfile; |
5a17f543 | 5001 | } |
79bd9814 | 5002 | |
451af504 | 5003 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
5004 | if (ret) |
5005 | goto out_put_css; | |
5006 | ||
5007 | efile.file->f_op->poll(efile.file, &event->pt); | |
5008 | ||
fba94807 TH |
5009 | spin_lock(&memcg->event_list_lock); |
5010 | list_add(&event->list, &memcg->event_list); | |
5011 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
5012 | |
5013 | fdput(cfile); | |
5014 | fdput(efile); | |
5015 | ||
451af504 | 5016 | return nbytes; |
79bd9814 TH |
5017 | |
5018 | out_put_css: | |
b5557c4c | 5019 | css_put(css); |
79bd9814 TH |
5020 | out_put_cfile: |
5021 | fdput(cfile); | |
5022 | out_put_eventfd: | |
5023 | eventfd_ctx_put(event->eventfd); | |
5024 | out_put_efile: | |
5025 | fdput(efile); | |
5026 | out_kfree: | |
5027 | kfree(event); | |
5028 | ||
5029 | return ret; | |
5030 | } | |
5031 | ||
8cdea7c0 BS |
5032 | static struct cftype mem_cgroup_files[] = { |
5033 | { | |
0eea1030 | 5034 | .name = "usage_in_bytes", |
8c7c6e34 | 5035 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 5036 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5037 | }, |
c84872e1 PE |
5038 | { |
5039 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5040 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 5041 | .write = mem_cgroup_reset, |
791badbd | 5042 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 5043 | }, |
8cdea7c0 | 5044 | { |
0eea1030 | 5045 | .name = "limit_in_bytes", |
8c7c6e34 | 5046 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 5047 | .write = mem_cgroup_write, |
791badbd | 5048 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5049 | }, |
296c81d8 BS |
5050 | { |
5051 | .name = "soft_limit_in_bytes", | |
5052 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 5053 | .write = mem_cgroup_write, |
791badbd | 5054 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 5055 | }, |
8cdea7c0 BS |
5056 | { |
5057 | .name = "failcnt", | |
8c7c6e34 | 5058 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 5059 | .write = mem_cgroup_reset, |
791badbd | 5060 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 5061 | }, |
d2ceb9b7 KH |
5062 | { |
5063 | .name = "stat", | |
2da8ca82 | 5064 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 5065 | }, |
c1e862c1 KH |
5066 | { |
5067 | .name = "force_empty", | |
6770c64e | 5068 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 5069 | }, |
18f59ea7 BS |
5070 | { |
5071 | .name = "use_hierarchy", | |
5072 | .write_u64 = mem_cgroup_hierarchy_write, | |
5073 | .read_u64 = mem_cgroup_hierarchy_read, | |
5074 | }, | |
79bd9814 | 5075 | { |
3bc942f3 | 5076 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 5077 | .write = memcg_write_event_control, |
79bd9814 TH |
5078 | .flags = CFTYPE_NO_PREFIX, |
5079 | .mode = S_IWUGO, | |
5080 | }, | |
a7885eb8 KM |
5081 | { |
5082 | .name = "swappiness", | |
5083 | .read_u64 = mem_cgroup_swappiness_read, | |
5084 | .write_u64 = mem_cgroup_swappiness_write, | |
5085 | }, | |
7dc74be0 DN |
5086 | { |
5087 | .name = "move_charge_at_immigrate", | |
5088 | .read_u64 = mem_cgroup_move_charge_read, | |
5089 | .write_u64 = mem_cgroup_move_charge_write, | |
5090 | }, | |
9490ff27 KH |
5091 | { |
5092 | .name = "oom_control", | |
2da8ca82 | 5093 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 5094 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
5095 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
5096 | }, | |
70ddf637 AV |
5097 | { |
5098 | .name = "pressure_level", | |
70ddf637 | 5099 | }, |
406eb0c9 YH |
5100 | #ifdef CONFIG_NUMA |
5101 | { | |
5102 | .name = "numa_stat", | |
2da8ca82 | 5103 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
5104 | }, |
5105 | #endif | |
510fc4e1 GC |
5106 | #ifdef CONFIG_MEMCG_KMEM |
5107 | { | |
5108 | .name = "kmem.limit_in_bytes", | |
5109 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 5110 | .write = mem_cgroup_write, |
791badbd | 5111 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5112 | }, |
5113 | { | |
5114 | .name = "kmem.usage_in_bytes", | |
5115 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 5116 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5117 | }, |
5118 | { | |
5119 | .name = "kmem.failcnt", | |
5120 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 5121 | .write = mem_cgroup_reset, |
791badbd | 5122 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
5123 | }, |
5124 | { | |
5125 | .name = "kmem.max_usage_in_bytes", | |
5126 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 5127 | .write = mem_cgroup_reset, |
791badbd | 5128 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 5129 | }, |
749c5415 GC |
5130 | #ifdef CONFIG_SLABINFO |
5131 | { | |
5132 | .name = "kmem.slabinfo", | |
2da8ca82 | 5133 | .seq_show = mem_cgroup_slabinfo_read, |
749c5415 GC |
5134 | }, |
5135 | #endif | |
8c7c6e34 | 5136 | #endif |
6bc10349 | 5137 | { }, /* terminate */ |
af36f906 | 5138 | }; |
8c7c6e34 | 5139 | |
2d11085e MH |
5140 | #ifdef CONFIG_MEMCG_SWAP |
5141 | static struct cftype memsw_cgroup_files[] = { | |
5142 | { | |
5143 | .name = "memsw.usage_in_bytes", | |
5144 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 5145 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5146 | }, |
5147 | { | |
5148 | .name = "memsw.max_usage_in_bytes", | |
5149 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 5150 | .write = mem_cgroup_reset, |
791badbd | 5151 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5152 | }, |
5153 | { | |
5154 | .name = "memsw.limit_in_bytes", | |
5155 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 5156 | .write = mem_cgroup_write, |
791badbd | 5157 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5158 | }, |
5159 | { | |
5160 | .name = "memsw.failcnt", | |
5161 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 5162 | .write = mem_cgroup_reset, |
791badbd | 5163 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
5164 | }, |
5165 | { }, /* terminate */ | |
5166 | }; | |
5167 | #endif | |
c0ff4b85 | 5168 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5169 | { |
5170 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 5171 | struct mem_cgroup_per_zone *mz; |
41e3355d | 5172 | int zone, tmp = node; |
1ecaab2b KH |
5173 | /* |
5174 | * This routine is called against possible nodes. | |
5175 | * But it's BUG to call kmalloc() against offline node. | |
5176 | * | |
5177 | * TODO: this routine can waste much memory for nodes which will | |
5178 | * never be onlined. It's better to use memory hotplug callback | |
5179 | * function. | |
5180 | */ | |
41e3355d KH |
5181 | if (!node_state(node, N_NORMAL_MEMORY)) |
5182 | tmp = -1; | |
17295c88 | 5183 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5184 | if (!pn) |
5185 | return 1; | |
1ecaab2b | 5186 | |
1ecaab2b KH |
5187 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
5188 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 5189 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
5190 | mz->usage_in_excess = 0; |
5191 | mz->on_tree = false; | |
d79154bb | 5192 | mz->memcg = memcg; |
1ecaab2b | 5193 | } |
54f72fe0 | 5194 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5195 | return 0; |
5196 | } | |
5197 | ||
c0ff4b85 | 5198 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5199 | { |
54f72fe0 | 5200 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
5201 | } |
5202 | ||
33327948 KH |
5203 | static struct mem_cgroup *mem_cgroup_alloc(void) |
5204 | { | |
d79154bb | 5205 | struct mem_cgroup *memcg; |
8ff69e2c | 5206 | size_t size; |
33327948 | 5207 | |
8ff69e2c VD |
5208 | size = sizeof(struct mem_cgroup); |
5209 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 5210 | |
8ff69e2c | 5211 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 5212 | if (!memcg) |
e7bbcdf3 DC |
5213 | return NULL; |
5214 | ||
d79154bb HD |
5215 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
5216 | if (!memcg->stat) | |
d2e61b8d | 5217 | goto out_free; |
d79154bb HD |
5218 | spin_lock_init(&memcg->pcp_counter_lock); |
5219 | return memcg; | |
d2e61b8d DC |
5220 | |
5221 | out_free: | |
8ff69e2c | 5222 | kfree(memcg); |
d2e61b8d | 5223 | return NULL; |
33327948 KH |
5224 | } |
5225 | ||
59927fb9 | 5226 | /* |
c8b2a36f GC |
5227 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
5228 | * (scanning all at force_empty is too costly...) | |
5229 | * | |
5230 | * Instead of clearing all references at force_empty, we remember | |
5231 | * the number of reference from swap_cgroup and free mem_cgroup when | |
5232 | * it goes down to 0. | |
5233 | * | |
5234 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 5235 | */ |
c8b2a36f GC |
5236 | |
5237 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 5238 | { |
c8b2a36f | 5239 | int node; |
59927fb9 | 5240 | |
bb4cc1a8 | 5241 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
5242 | |
5243 | for_each_node(node) | |
5244 | free_mem_cgroup_per_zone_info(memcg, node); | |
5245 | ||
5246 | free_percpu(memcg->stat); | |
5247 | ||
3f134619 GC |
5248 | /* |
5249 | * We need to make sure that (at least for now), the jump label | |
5250 | * destruction code runs outside of the cgroup lock. This is because | |
5251 | * get_online_cpus(), which is called from the static_branch update, | |
5252 | * can't be called inside the cgroup_lock. cpusets are the ones | |
5253 | * enforcing this dependency, so if they ever change, we might as well. | |
5254 | * | |
5255 | * schedule_work() will guarantee this happens. Be careful if you need | |
5256 | * to move this code around, and make sure it is outside | |
5257 | * the cgroup_lock. | |
5258 | */ | |
a8964b9b | 5259 | disarm_static_keys(memcg); |
8ff69e2c | 5260 | kfree(memcg); |
59927fb9 | 5261 | } |
3afe36b1 | 5262 | |
7bcc1bb1 DN |
5263 | /* |
5264 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
5265 | */ | |
e1aab161 | 5266 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 5267 | { |
3e32cb2e | 5268 | if (!memcg->memory.parent) |
7bcc1bb1 | 5269 | return NULL; |
3e32cb2e | 5270 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 5271 | } |
e1aab161 | 5272 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 5273 | |
bb4cc1a8 AM |
5274 | static void __init mem_cgroup_soft_limit_tree_init(void) |
5275 | { | |
5276 | struct mem_cgroup_tree_per_node *rtpn; | |
5277 | struct mem_cgroup_tree_per_zone *rtpz; | |
5278 | int tmp, node, zone; | |
5279 | ||
5280 | for_each_node(node) { | |
5281 | tmp = node; | |
5282 | if (!node_state(node, N_NORMAL_MEMORY)) | |
5283 | tmp = -1; | |
5284 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
5285 | BUG_ON(!rtpn); | |
5286 | ||
5287 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5288 | ||
5289 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5290 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5291 | rtpz->rb_root = RB_ROOT; | |
5292 | spin_lock_init(&rtpz->lock); | |
5293 | } | |
5294 | } | |
5295 | } | |
5296 | ||
0eb253e2 | 5297 | static struct cgroup_subsys_state * __ref |
eb95419b | 5298 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 5299 | { |
d142e3e6 | 5300 | struct mem_cgroup *memcg; |
04046e1a | 5301 | long error = -ENOMEM; |
6d12e2d8 | 5302 | int node; |
8cdea7c0 | 5303 | |
c0ff4b85 R |
5304 | memcg = mem_cgroup_alloc(); |
5305 | if (!memcg) | |
04046e1a | 5306 | return ERR_PTR(error); |
78fb7466 | 5307 | |
3ed28fa1 | 5308 | for_each_node(node) |
c0ff4b85 | 5309 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5310 | goto free_out; |
f64c3f54 | 5311 | |
c077719b | 5312 | /* root ? */ |
eb95419b | 5313 | if (parent_css == NULL) { |
a41c58a6 | 5314 | root_mem_cgroup = memcg; |
3e32cb2e JW |
5315 | page_counter_init(&memcg->memory, NULL); |
5316 | page_counter_init(&memcg->memsw, NULL); | |
5317 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 5318 | } |
28dbc4b6 | 5319 | |
d142e3e6 GC |
5320 | memcg->last_scanned_node = MAX_NUMNODES; |
5321 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
5322 | memcg->move_charge_at_immigrate = 0; |
5323 | mutex_init(&memcg->thresholds_lock); | |
5324 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5325 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5326 | INIT_LIST_HEAD(&memcg->event_list); |
5327 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
5328 | |
5329 | return &memcg->css; | |
5330 | ||
5331 | free_out: | |
5332 | __mem_cgroup_free(memcg); | |
5333 | return ERR_PTR(error); | |
5334 | } | |
5335 | ||
5336 | static int | |
eb95419b | 5337 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 5338 | { |
eb95419b | 5339 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 5340 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 5341 | int ret; |
d142e3e6 | 5342 | |
15a4c835 | 5343 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
5344 | return -ENOSPC; |
5345 | ||
63876986 | 5346 | if (!parent) |
d142e3e6 GC |
5347 | return 0; |
5348 | ||
0999821b | 5349 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
5350 | |
5351 | memcg->use_hierarchy = parent->use_hierarchy; | |
5352 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5353 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5354 | ||
5355 | if (parent->use_hierarchy) { | |
3e32cb2e JW |
5356 | page_counter_init(&memcg->memory, &parent->memory); |
5357 | page_counter_init(&memcg->memsw, &parent->memsw); | |
5358 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 5359 | |
7bcc1bb1 | 5360 | /* |
8d76a979 LZ |
5361 | * No need to take a reference to the parent because cgroup |
5362 | * core guarantees its existence. | |
7bcc1bb1 | 5363 | */ |
18f59ea7 | 5364 | } else { |
3e32cb2e JW |
5365 | page_counter_init(&memcg->memory, NULL); |
5366 | page_counter_init(&memcg->memsw, NULL); | |
5367 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
5368 | /* |
5369 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5370 | * much sense so let cgroup subsystem know about this | |
5371 | * unfortunate state in our controller. | |
5372 | */ | |
d142e3e6 | 5373 | if (parent != root_mem_cgroup) |
073219e9 | 5374 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5375 | } |
0999821b | 5376 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 5377 | |
2f7dd7a4 JW |
5378 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
5379 | if (ret) | |
5380 | return ret; | |
5381 | ||
5382 | /* | |
5383 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
5384 | * orders reading memcg->initialized against its callers | |
5385 | * reading the memcg members. | |
5386 | */ | |
5387 | smp_store_release(&memcg->initialized, 1); | |
5388 | ||
5389 | return 0; | |
8cdea7c0 BS |
5390 | } |
5391 | ||
eb95419b | 5392 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5393 | { |
eb95419b | 5394 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5395 | struct mem_cgroup_event *event, *tmp; |
4fb1a86f | 5396 | struct cgroup_subsys_state *iter; |
79bd9814 TH |
5397 | |
5398 | /* | |
5399 | * Unregister events and notify userspace. | |
5400 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5401 | * directory to avoid race between userspace and kernelspace. | |
5402 | */ | |
fba94807 TH |
5403 | spin_lock(&memcg->event_list_lock); |
5404 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5405 | list_del_init(&event->list); |
5406 | schedule_work(&event->remove); | |
5407 | } | |
fba94807 | 5408 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5409 | |
10d5ebf4 LZ |
5410 | kmem_cgroup_css_offline(memcg); |
5411 | ||
4fb1a86f FB |
5412 | /* |
5413 | * This requires that offlining is serialized. Right now that is | |
5414 | * guaranteed because css_killed_work_fn() holds the cgroup_mutex. | |
5415 | */ | |
5416 | css_for_each_descendant_post(iter, css) | |
5417 | mem_cgroup_reparent_charges(mem_cgroup_from_css(iter)); | |
5418 | ||
776ed0f0 | 5419 | memcg_unregister_all_caches(memcg); |
33cb876e | 5420 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
5421 | } |
5422 | ||
eb95419b | 5423 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5424 | { |
eb95419b | 5425 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
96f1c58d JW |
5426 | /* |
5427 | * XXX: css_offline() would be where we should reparent all | |
5428 | * memory to prepare the cgroup for destruction. However, | |
3e32cb2e | 5429 | * memcg does not do css_tryget_online() and page_counter charging |
96f1c58d JW |
5430 | * under the same RCU lock region, which means that charging |
5431 | * could race with offlining. Offlining only happens to | |
5432 | * cgroups with no tasks in them but charges can show up | |
5433 | * without any tasks from the swapin path when the target | |
5434 | * memcg is looked up from the swapout record and not from the | |
5435 | * current task as it usually is. A race like this can leak | |
5436 | * charges and put pages with stale cgroup pointers into | |
5437 | * circulation: | |
5438 | * | |
5439 | * #0 #1 | |
5440 | * lookup_swap_cgroup_id() | |
5441 | * rcu_read_lock() | |
5442 | * mem_cgroup_lookup() | |
ec903c0c | 5443 | * css_tryget_online() |
96f1c58d | 5444 | * rcu_read_unlock() |
ec903c0c | 5445 | * disable css_tryget_online() |
96f1c58d JW |
5446 | * call_rcu() |
5447 | * offline_css() | |
5448 | * reparent_charges() | |
3e32cb2e | 5449 | * page_counter_try_charge() |
96f1c58d JW |
5450 | * css_put() |
5451 | * css_free() | |
5452 | * pc->mem_cgroup = dead memcg | |
5453 | * add page to lru | |
5454 | * | |
5455 | * The bulk of the charges are still moved in offline_css() to | |
5456 | * avoid pinning a lot of pages in case a long-term reference | |
5457 | * like a swapout record is deferring the css_free() to long | |
5458 | * after offlining. But this makes sure we catch any charges | |
5459 | * made after offlining: | |
5460 | */ | |
5461 | mem_cgroup_reparent_charges(memcg); | |
c268e994 | 5462 | |
10d5ebf4 | 5463 | memcg_destroy_kmem(memcg); |
465939a1 | 5464 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
5465 | } |
5466 | ||
1ced953b TH |
5467 | /** |
5468 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5469 | * @css: the target css | |
5470 | * | |
5471 | * Reset the states of the mem_cgroup associated with @css. This is | |
5472 | * invoked when the userland requests disabling on the default hierarchy | |
5473 | * but the memcg is pinned through dependency. The memcg should stop | |
5474 | * applying policies and should revert to the vanilla state as it may be | |
5475 | * made visible again. | |
5476 | * | |
5477 | * The current implementation only resets the essential configurations. | |
5478 | * This needs to be expanded to cover all the visible parts. | |
5479 | */ | |
5480 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5481 | { | |
5482 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5483 | ||
3e32cb2e JW |
5484 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
5485 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
5486 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
5487 | memcg->soft_limit = 0; | |
1ced953b TH |
5488 | } |
5489 | ||
02491447 | 5490 | #ifdef CONFIG_MMU |
7dc74be0 | 5491 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5492 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5493 | { |
05b84301 | 5494 | int ret; |
9476db97 JW |
5495 | |
5496 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 5497 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 5498 | if (!ret) { |
854ffa8d | 5499 | mc.precharge += count; |
854ffa8d DN |
5500 | return ret; |
5501 | } | |
692e7c45 | 5502 | if (ret == -EINTR) { |
00501b53 | 5503 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
5504 | return ret; |
5505 | } | |
9476db97 JW |
5506 | |
5507 | /* Try charges one by one with reclaim */ | |
854ffa8d | 5508 | while (count--) { |
00501b53 | 5509 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
5510 | /* |
5511 | * In case of failure, any residual charges against | |
5512 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
5513 | * later on. However, cancel any charges that are |
5514 | * bypassed to root right away or they'll be lost. | |
9476db97 | 5515 | */ |
692e7c45 | 5516 | if (ret == -EINTR) |
00501b53 | 5517 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 5518 | if (ret) |
38c5d72f | 5519 | return ret; |
854ffa8d | 5520 | mc.precharge++; |
9476db97 | 5521 | cond_resched(); |
854ffa8d | 5522 | } |
9476db97 | 5523 | return 0; |
4ffef5fe DN |
5524 | } |
5525 | ||
5526 | /** | |
8d32ff84 | 5527 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5528 | * @vma: the vma the pte to be checked belongs |
5529 | * @addr: the address corresponding to the pte to be checked | |
5530 | * @ptent: the pte to be checked | |
02491447 | 5531 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5532 | * |
5533 | * Returns | |
5534 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5535 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5536 | * move charge. if @target is not NULL, the page is stored in target->page | |
5537 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5538 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5539 | * target for charge migration. if @target is not NULL, the entry is stored | |
5540 | * in target->ent. | |
4ffef5fe DN |
5541 | * |
5542 | * Called with pte lock held. | |
5543 | */ | |
4ffef5fe DN |
5544 | union mc_target { |
5545 | struct page *page; | |
02491447 | 5546 | swp_entry_t ent; |
4ffef5fe DN |
5547 | }; |
5548 | ||
4ffef5fe | 5549 | enum mc_target_type { |
8d32ff84 | 5550 | MC_TARGET_NONE = 0, |
4ffef5fe | 5551 | MC_TARGET_PAGE, |
02491447 | 5552 | MC_TARGET_SWAP, |
4ffef5fe DN |
5553 | }; |
5554 | ||
90254a65 DN |
5555 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5556 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5557 | { |
90254a65 | 5558 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5559 | |
90254a65 DN |
5560 | if (!page || !page_mapped(page)) |
5561 | return NULL; | |
5562 | if (PageAnon(page)) { | |
5563 | /* we don't move shared anon */ | |
4b91355e | 5564 | if (!move_anon()) |
90254a65 | 5565 | return NULL; |
87946a72 DN |
5566 | } else if (!move_file()) |
5567 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5568 | return NULL; |
5569 | if (!get_page_unless_zero(page)) | |
5570 | return NULL; | |
5571 | ||
5572 | return page; | |
5573 | } | |
5574 | ||
4b91355e | 5575 | #ifdef CONFIG_SWAP |
90254a65 DN |
5576 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5577 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5578 | { | |
90254a65 DN |
5579 | struct page *page = NULL; |
5580 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5581 | ||
5582 | if (!move_anon() || non_swap_entry(ent)) | |
5583 | return NULL; | |
4b91355e KH |
5584 | /* |
5585 | * Because lookup_swap_cache() updates some statistics counter, | |
5586 | * we call find_get_page() with swapper_space directly. | |
5587 | */ | |
33806f06 | 5588 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
5589 | if (do_swap_account) |
5590 | entry->val = ent.val; | |
5591 | ||
5592 | return page; | |
5593 | } | |
4b91355e KH |
5594 | #else |
5595 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5596 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5597 | { | |
5598 | return NULL; | |
5599 | } | |
5600 | #endif | |
90254a65 | 5601 | |
87946a72 DN |
5602 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5603 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5604 | { | |
5605 | struct page *page = NULL; | |
87946a72 DN |
5606 | struct address_space *mapping; |
5607 | pgoff_t pgoff; | |
5608 | ||
5609 | if (!vma->vm_file) /* anonymous vma */ | |
5610 | return NULL; | |
5611 | if (!move_file()) | |
5612 | return NULL; | |
5613 | ||
87946a72 DN |
5614 | mapping = vma->vm_file->f_mapping; |
5615 | if (pte_none(ptent)) | |
5616 | pgoff = linear_page_index(vma, addr); | |
5617 | else /* pte_file(ptent) is true */ | |
5618 | pgoff = pte_to_pgoff(ptent); | |
5619 | ||
5620 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5621 | #ifdef CONFIG_SWAP |
5622 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5623 | if (shmem_mapping(mapping)) { |
5624 | page = find_get_entry(mapping, pgoff); | |
5625 | if (radix_tree_exceptional_entry(page)) { | |
5626 | swp_entry_t swp = radix_to_swp_entry(page); | |
5627 | if (do_swap_account) | |
5628 | *entry = swp; | |
5629 | page = find_get_page(swap_address_space(swp), swp.val); | |
5630 | } | |
5631 | } else | |
5632 | page = find_get_page(mapping, pgoff); | |
5633 | #else | |
5634 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5635 | #endif |
87946a72 DN |
5636 | return page; |
5637 | } | |
5638 | ||
8d32ff84 | 5639 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5640 | unsigned long addr, pte_t ptent, union mc_target *target) |
5641 | { | |
5642 | struct page *page = NULL; | |
5643 | struct page_cgroup *pc; | |
8d32ff84 | 5644 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5645 | swp_entry_t ent = { .val = 0 }; |
5646 | ||
5647 | if (pte_present(ptent)) | |
5648 | page = mc_handle_present_pte(vma, addr, ptent); | |
5649 | else if (is_swap_pte(ptent)) | |
5650 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5651 | else if (pte_none(ptent) || pte_file(ptent)) |
5652 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5653 | |
5654 | if (!page && !ent.val) | |
8d32ff84 | 5655 | return ret; |
02491447 DN |
5656 | if (page) { |
5657 | pc = lookup_page_cgroup(page); | |
5658 | /* | |
0a31bc97 JW |
5659 | * Do only loose check w/o serialization. |
5660 | * mem_cgroup_move_account() checks the pc is valid or | |
5661 | * not under LRU exclusion. | |
02491447 DN |
5662 | */ |
5663 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5664 | ret = MC_TARGET_PAGE; | |
5665 | if (target) | |
5666 | target->page = page; | |
5667 | } | |
5668 | if (!ret || !target) | |
5669 | put_page(page); | |
5670 | } | |
90254a65 DN |
5671 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5672 | if (ent.val && !ret && | |
34c00c31 | 5673 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5674 | ret = MC_TARGET_SWAP; |
5675 | if (target) | |
5676 | target->ent = ent; | |
4ffef5fe | 5677 | } |
4ffef5fe DN |
5678 | return ret; |
5679 | } | |
5680 | ||
12724850 NH |
5681 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5682 | /* | |
5683 | * We don't consider swapping or file mapped pages because THP does not | |
5684 | * support them for now. | |
5685 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5686 | */ | |
5687 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5688 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5689 | { | |
5690 | struct page *page = NULL; | |
5691 | struct page_cgroup *pc; | |
5692 | enum mc_target_type ret = MC_TARGET_NONE; | |
5693 | ||
5694 | page = pmd_page(pmd); | |
309381fe | 5695 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
5696 | if (!move_anon()) |
5697 | return ret; | |
5698 | pc = lookup_page_cgroup(page); | |
5699 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5700 | ret = MC_TARGET_PAGE; | |
5701 | if (target) { | |
5702 | get_page(page); | |
5703 | target->page = page; | |
5704 | } | |
5705 | } | |
5706 | return ret; | |
5707 | } | |
5708 | #else | |
5709 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5710 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5711 | { | |
5712 | return MC_TARGET_NONE; | |
5713 | } | |
5714 | #endif | |
5715 | ||
4ffef5fe DN |
5716 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5717 | unsigned long addr, unsigned long end, | |
5718 | struct mm_walk *walk) | |
5719 | { | |
5720 | struct vm_area_struct *vma = walk->private; | |
5721 | pte_t *pte; | |
5722 | spinlock_t *ptl; | |
5723 | ||
bf929152 | 5724 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
5725 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5726 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5727 | spin_unlock(ptl); |
1a5a9906 | 5728 | return 0; |
12724850 | 5729 | } |
03319327 | 5730 | |
45f83cef AA |
5731 | if (pmd_trans_unstable(pmd)) |
5732 | return 0; | |
4ffef5fe DN |
5733 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5734 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5735 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5736 | mc.precharge++; /* increment precharge temporarily */ |
5737 | pte_unmap_unlock(pte - 1, ptl); | |
5738 | cond_resched(); | |
5739 | ||
7dc74be0 DN |
5740 | return 0; |
5741 | } | |
5742 | ||
4ffef5fe DN |
5743 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5744 | { | |
5745 | unsigned long precharge; | |
5746 | struct vm_area_struct *vma; | |
5747 | ||
dfe076b0 | 5748 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5749 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5750 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5751 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5752 | .mm = mm, | |
5753 | .private = vma, | |
5754 | }; | |
5755 | if (is_vm_hugetlb_page(vma)) | |
5756 | continue; | |
4ffef5fe DN |
5757 | walk_page_range(vma->vm_start, vma->vm_end, |
5758 | &mem_cgroup_count_precharge_walk); | |
5759 | } | |
dfe076b0 | 5760 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5761 | |
5762 | precharge = mc.precharge; | |
5763 | mc.precharge = 0; | |
5764 | ||
5765 | return precharge; | |
5766 | } | |
5767 | ||
4ffef5fe DN |
5768 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5769 | { | |
dfe076b0 DN |
5770 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5771 | ||
5772 | VM_BUG_ON(mc.moving_task); | |
5773 | mc.moving_task = current; | |
5774 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5775 | } |
5776 | ||
dfe076b0 DN |
5777 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5778 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5779 | { |
2bd9bb20 KH |
5780 | struct mem_cgroup *from = mc.from; |
5781 | struct mem_cgroup *to = mc.to; | |
5782 | ||
4ffef5fe | 5783 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5784 | if (mc.precharge) { |
00501b53 | 5785 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5786 | mc.precharge = 0; |
5787 | } | |
5788 | /* | |
5789 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5790 | * we must uncharge here. | |
5791 | */ | |
5792 | if (mc.moved_charge) { | |
00501b53 | 5793 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5794 | mc.moved_charge = 0; |
4ffef5fe | 5795 | } |
483c30b5 DN |
5796 | /* we must fixup refcnts and charges */ |
5797 | if (mc.moved_swap) { | |
483c30b5 | 5798 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5799 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5800 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5801 | |
05b84301 | 5802 | /* |
3e32cb2e JW |
5803 | * we charged both to->memory and to->memsw, so we |
5804 | * should uncharge to->memory. | |
05b84301 | 5805 | */ |
ce00a967 | 5806 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5807 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5808 | ||
e8ea14cc | 5809 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 5810 | |
4050377b | 5811 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
5812 | mc.moved_swap = 0; |
5813 | } | |
dfe076b0 DN |
5814 | memcg_oom_recover(from); |
5815 | memcg_oom_recover(to); | |
5816 | wake_up_all(&mc.waitq); | |
5817 | } | |
5818 | ||
5819 | static void mem_cgroup_clear_mc(void) | |
5820 | { | |
5821 | struct mem_cgroup *from = mc.from; | |
5822 | ||
5823 | /* | |
5824 | * we must clear moving_task before waking up waiters at the end of | |
5825 | * task migration. | |
5826 | */ | |
5827 | mc.moving_task = NULL; | |
5828 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5829 | spin_lock(&mc.lock); |
4ffef5fe DN |
5830 | mc.from = NULL; |
5831 | mc.to = NULL; | |
2bd9bb20 | 5832 | spin_unlock(&mc.lock); |
32047e2a | 5833 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5834 | } |
5835 | ||
eb95419b | 5836 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5837 | struct cgroup_taskset *tset) |
7dc74be0 | 5838 | { |
2f7ee569 | 5839 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5840 | int ret = 0; |
eb95419b | 5841 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 5842 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 5843 | |
ee5e8472 GC |
5844 | /* |
5845 | * We are now commited to this value whatever it is. Changes in this | |
5846 | * tunable will only affect upcoming migrations, not the current one. | |
5847 | * So we need to save it, and keep it going. | |
5848 | */ | |
5849 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
5850 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
5851 | struct mm_struct *mm; |
5852 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5853 | ||
c0ff4b85 | 5854 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5855 | |
5856 | mm = get_task_mm(p); | |
5857 | if (!mm) | |
5858 | return 0; | |
7dc74be0 | 5859 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5860 | if (mm->owner == p) { |
5861 | VM_BUG_ON(mc.from); | |
5862 | VM_BUG_ON(mc.to); | |
5863 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5864 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5865 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5866 | mem_cgroup_start_move(from); |
2bd9bb20 | 5867 | spin_lock(&mc.lock); |
4ffef5fe | 5868 | mc.from = from; |
c0ff4b85 | 5869 | mc.to = memcg; |
ee5e8472 | 5870 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 5871 | spin_unlock(&mc.lock); |
dfe076b0 | 5872 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5873 | |
5874 | ret = mem_cgroup_precharge_mc(mm); | |
5875 | if (ret) | |
5876 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5877 | } |
5878 | mmput(mm); | |
7dc74be0 DN |
5879 | } |
5880 | return ret; | |
5881 | } | |
5882 | ||
eb95419b | 5883 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5884 | struct cgroup_taskset *tset) |
7dc74be0 | 5885 | { |
4ffef5fe | 5886 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5887 | } |
5888 | ||
4ffef5fe DN |
5889 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5890 | unsigned long addr, unsigned long end, | |
5891 | struct mm_walk *walk) | |
7dc74be0 | 5892 | { |
4ffef5fe DN |
5893 | int ret = 0; |
5894 | struct vm_area_struct *vma = walk->private; | |
5895 | pte_t *pte; | |
5896 | spinlock_t *ptl; | |
12724850 NH |
5897 | enum mc_target_type target_type; |
5898 | union mc_target target; | |
5899 | struct page *page; | |
5900 | struct page_cgroup *pc; | |
4ffef5fe | 5901 | |
12724850 NH |
5902 | /* |
5903 | * We don't take compound_lock() here but no race with splitting thp | |
5904 | * happens because: | |
5905 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5906 | * under splitting, which means there's no concurrent thp split, | |
5907 | * - if another thread runs into split_huge_page() just after we | |
5908 | * entered this if-block, the thread must wait for page table lock | |
5909 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5910 | * part of thp split is not executed yet. | |
5911 | */ | |
bf929152 | 5912 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5913 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5914 | spin_unlock(ptl); |
12724850 NH |
5915 | return 0; |
5916 | } | |
5917 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5918 | if (target_type == MC_TARGET_PAGE) { | |
5919 | page = target.page; | |
5920 | if (!isolate_lru_page(page)) { | |
5921 | pc = lookup_page_cgroup(page); | |
5922 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5923 | pc, mc.from, mc.to)) { |
12724850 NH |
5924 | mc.precharge -= HPAGE_PMD_NR; |
5925 | mc.moved_charge += HPAGE_PMD_NR; | |
5926 | } | |
5927 | putback_lru_page(page); | |
5928 | } | |
5929 | put_page(page); | |
5930 | } | |
bf929152 | 5931 | spin_unlock(ptl); |
1a5a9906 | 5932 | return 0; |
12724850 NH |
5933 | } |
5934 | ||
45f83cef AA |
5935 | if (pmd_trans_unstable(pmd)) |
5936 | return 0; | |
4ffef5fe DN |
5937 | retry: |
5938 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5939 | for (; addr != end; addr += PAGE_SIZE) { | |
5940 | pte_t ptent = *(pte++); | |
02491447 | 5941 | swp_entry_t ent; |
4ffef5fe DN |
5942 | |
5943 | if (!mc.precharge) | |
5944 | break; | |
5945 | ||
8d32ff84 | 5946 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5947 | case MC_TARGET_PAGE: |
5948 | page = target.page; | |
5949 | if (isolate_lru_page(page)) | |
5950 | goto put; | |
5951 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5952 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5953 | mc.from, mc.to)) { |
4ffef5fe | 5954 | mc.precharge--; |
854ffa8d DN |
5955 | /* we uncharge from mc.from later. */ |
5956 | mc.moved_charge++; | |
4ffef5fe DN |
5957 | } |
5958 | putback_lru_page(page); | |
8d32ff84 | 5959 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5960 | put_page(page); |
5961 | break; | |
02491447 DN |
5962 | case MC_TARGET_SWAP: |
5963 | ent = target.ent; | |
e91cbb42 | 5964 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5965 | mc.precharge--; |
483c30b5 DN |
5966 | /* we fixup refcnts and charges later. */ |
5967 | mc.moved_swap++; | |
5968 | } | |
02491447 | 5969 | break; |
4ffef5fe DN |
5970 | default: |
5971 | break; | |
5972 | } | |
5973 | } | |
5974 | pte_unmap_unlock(pte - 1, ptl); | |
5975 | cond_resched(); | |
5976 | ||
5977 | if (addr != end) { | |
5978 | /* | |
5979 | * We have consumed all precharges we got in can_attach(). | |
5980 | * We try charge one by one, but don't do any additional | |
5981 | * charges to mc.to if we have failed in charge once in attach() | |
5982 | * phase. | |
5983 | */ | |
854ffa8d | 5984 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5985 | if (!ret) |
5986 | goto retry; | |
5987 | } | |
5988 | ||
5989 | return ret; | |
5990 | } | |
5991 | ||
5992 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5993 | { | |
5994 | struct vm_area_struct *vma; | |
5995 | ||
5996 | lru_add_drain_all(); | |
dfe076b0 DN |
5997 | retry: |
5998 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5999 | /* | |
6000 | * Someone who are holding the mmap_sem might be waiting in | |
6001 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6002 | * and retry. Because we cancel precharges, we might not be able | |
6003 | * to move enough charges, but moving charge is a best-effort | |
6004 | * feature anyway, so it wouldn't be a big problem. | |
6005 | */ | |
6006 | __mem_cgroup_clear_mc(); | |
6007 | cond_resched(); | |
6008 | goto retry; | |
6009 | } | |
4ffef5fe DN |
6010 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6011 | int ret; | |
6012 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6013 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6014 | .mm = mm, | |
6015 | .private = vma, | |
6016 | }; | |
6017 | if (is_vm_hugetlb_page(vma)) | |
6018 | continue; | |
4ffef5fe DN |
6019 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6020 | &mem_cgroup_move_charge_walk); | |
6021 | if (ret) | |
6022 | /* | |
6023 | * means we have consumed all precharges and failed in | |
6024 | * doing additional charge. Just abandon here. | |
6025 | */ | |
6026 | break; | |
6027 | } | |
dfe076b0 | 6028 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6029 | } |
6030 | ||
eb95419b | 6031 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6032 | struct cgroup_taskset *tset) |
67e465a7 | 6033 | { |
2f7ee569 | 6034 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6035 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6036 | |
dfe076b0 | 6037 | if (mm) { |
a433658c KM |
6038 | if (mc.to) |
6039 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6040 | mmput(mm); |
6041 | } | |
a433658c KM |
6042 | if (mc.to) |
6043 | mem_cgroup_clear_mc(); | |
67e465a7 | 6044 | } |
5cfb80a7 | 6045 | #else /* !CONFIG_MMU */ |
eb95419b | 6046 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6047 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6048 | { |
6049 | return 0; | |
6050 | } | |
eb95419b | 6051 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 6052 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6053 | { |
6054 | } | |
eb95419b | 6055 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 6056 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
6057 | { |
6058 | } | |
6059 | #endif | |
67e465a7 | 6060 | |
f00baae7 TH |
6061 | /* |
6062 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6063 | * to verify whether we're attached to the default hierarchy on each mount |
6064 | * attempt. | |
f00baae7 | 6065 | */ |
eb95419b | 6066 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6067 | { |
6068 | /* | |
aa6ec29b | 6069 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6070 | * guarantees that @root doesn't have any children, so turning it |
6071 | * on for the root memcg is enough. | |
6072 | */ | |
aa6ec29b | 6073 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 6074 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
6075 | } |
6076 | ||
073219e9 | 6077 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6078 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6079 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6080 | .css_offline = mem_cgroup_css_offline, |
6081 | .css_free = mem_cgroup_css_free, | |
1ced953b | 6082 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6083 | .can_attach = mem_cgroup_can_attach, |
6084 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6085 | .attach = mem_cgroup_move_task, |
f00baae7 | 6086 | .bind = mem_cgroup_bind, |
5577964e | 6087 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 6088 | .early_init = 0, |
8cdea7c0 | 6089 | }; |
c077719b | 6090 | |
c255a458 | 6091 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6092 | static int __init enable_swap_account(char *s) |
6093 | { | |
a2c8990a | 6094 | if (!strcmp(s, "1")) |
a42c390c | 6095 | really_do_swap_account = 1; |
a2c8990a | 6096 | else if (!strcmp(s, "0")) |
a42c390c MH |
6097 | really_do_swap_account = 0; |
6098 | return 1; | |
6099 | } | |
a2c8990a | 6100 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6101 | |
2d11085e MH |
6102 | static void __init memsw_file_init(void) |
6103 | { | |
2cf669a5 TH |
6104 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6105 | memsw_cgroup_files)); | |
6acc8b02 MH |
6106 | } |
6107 | ||
6108 | static void __init enable_swap_cgroup(void) | |
6109 | { | |
6110 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6111 | do_swap_account = 1; | |
6112 | memsw_file_init(); | |
6113 | } | |
2d11085e | 6114 | } |
6acc8b02 | 6115 | |
2d11085e | 6116 | #else |
6acc8b02 | 6117 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6118 | { |
6119 | } | |
c077719b | 6120 | #endif |
2d11085e | 6121 | |
0a31bc97 JW |
6122 | #ifdef CONFIG_MEMCG_SWAP |
6123 | /** | |
6124 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6125 | * @page: page whose memsw charge to transfer | |
6126 | * @entry: swap entry to move the charge to | |
6127 | * | |
6128 | * Transfer the memsw charge of @page to @entry. | |
6129 | */ | |
6130 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6131 | { | |
6132 | struct page_cgroup *pc; | |
6133 | unsigned short oldid; | |
6134 | ||
6135 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6136 | VM_BUG_ON_PAGE(page_count(page), page); | |
6137 | ||
6138 | if (!do_swap_account) | |
6139 | return; | |
6140 | ||
6141 | pc = lookup_page_cgroup(page); | |
6142 | ||
6143 | /* Readahead page, never charged */ | |
6144 | if (!PageCgroupUsed(pc)) | |
6145 | return; | |
6146 | ||
6147 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEMSW), page); | |
6148 | ||
6149 | oldid = swap_cgroup_record(entry, mem_cgroup_id(pc->mem_cgroup)); | |
6150 | VM_BUG_ON_PAGE(oldid, page); | |
6151 | ||
6152 | pc->flags &= ~PCG_MEMSW; | |
6153 | css_get(&pc->mem_cgroup->css); | |
6154 | mem_cgroup_swap_statistics(pc->mem_cgroup, true); | |
6155 | } | |
6156 | ||
6157 | /** | |
6158 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
6159 | * @entry: swap entry to uncharge | |
6160 | * | |
6161 | * Drop the memsw charge associated with @entry. | |
6162 | */ | |
6163 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
6164 | { | |
6165 | struct mem_cgroup *memcg; | |
6166 | unsigned short id; | |
6167 | ||
6168 | if (!do_swap_account) | |
6169 | return; | |
6170 | ||
6171 | id = swap_cgroup_record(entry, 0); | |
6172 | rcu_read_lock(); | |
6173 | memcg = mem_cgroup_lookup(id); | |
6174 | if (memcg) { | |
ce00a967 | 6175 | if (!mem_cgroup_is_root(memcg)) |
3e32cb2e | 6176 | page_counter_uncharge(&memcg->memsw, 1); |
0a31bc97 JW |
6177 | mem_cgroup_swap_statistics(memcg, false); |
6178 | css_put(&memcg->css); | |
6179 | } | |
6180 | rcu_read_unlock(); | |
6181 | } | |
6182 | #endif | |
6183 | ||
00501b53 JW |
6184 | /** |
6185 | * mem_cgroup_try_charge - try charging a page | |
6186 | * @page: page to charge | |
6187 | * @mm: mm context of the victim | |
6188 | * @gfp_mask: reclaim mode | |
6189 | * @memcgp: charged memcg return | |
6190 | * | |
6191 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6192 | * pages according to @gfp_mask if necessary. | |
6193 | * | |
6194 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
6195 | * Otherwise, an error code is returned. | |
6196 | * | |
6197 | * After page->mapping has been set up, the caller must finalize the | |
6198 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
6199 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
6200 | */ | |
6201 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
6202 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
6203 | { | |
6204 | struct mem_cgroup *memcg = NULL; | |
6205 | unsigned int nr_pages = 1; | |
6206 | int ret = 0; | |
6207 | ||
6208 | if (mem_cgroup_disabled()) | |
6209 | goto out; | |
6210 | ||
6211 | if (PageSwapCache(page)) { | |
6212 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
6213 | /* | |
6214 | * Every swap fault against a single page tries to charge the | |
6215 | * page, bail as early as possible. shmem_unuse() encounters | |
6216 | * already charged pages, too. The USED bit is protected by | |
6217 | * the page lock, which serializes swap cache removal, which | |
6218 | * in turn serializes uncharging. | |
6219 | */ | |
6220 | if (PageCgroupUsed(pc)) | |
6221 | goto out; | |
6222 | } | |
6223 | ||
6224 | if (PageTransHuge(page)) { | |
6225 | nr_pages <<= compound_order(page); | |
6226 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6227 | } | |
6228 | ||
6229 | if (do_swap_account && PageSwapCache(page)) | |
6230 | memcg = try_get_mem_cgroup_from_page(page); | |
6231 | if (!memcg) | |
6232 | memcg = get_mem_cgroup_from_mm(mm); | |
6233 | ||
6234 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6235 | ||
6236 | css_put(&memcg->css); | |
6237 | ||
6238 | if (ret == -EINTR) { | |
6239 | memcg = root_mem_cgroup; | |
6240 | ret = 0; | |
6241 | } | |
6242 | out: | |
6243 | *memcgp = memcg; | |
6244 | return ret; | |
6245 | } | |
6246 | ||
6247 | /** | |
6248 | * mem_cgroup_commit_charge - commit a page charge | |
6249 | * @page: page to charge | |
6250 | * @memcg: memcg to charge the page to | |
6251 | * @lrucare: page might be on LRU already | |
6252 | * | |
6253 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6254 | * after page->mapping has been set up. This must happen atomically | |
6255 | * as part of the page instantiation, i.e. under the page table lock | |
6256 | * for anonymous pages, under the page lock for page and swap cache. | |
6257 | * | |
6258 | * In addition, the page must not be on the LRU during the commit, to | |
6259 | * prevent racing with task migration. If it might be, use @lrucare. | |
6260 | * | |
6261 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6262 | */ | |
6263 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
6264 | bool lrucare) | |
6265 | { | |
6266 | unsigned int nr_pages = 1; | |
6267 | ||
6268 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6269 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6270 | ||
6271 | if (mem_cgroup_disabled()) | |
6272 | return; | |
6273 | /* | |
6274 | * Swap faults will attempt to charge the same page multiple | |
6275 | * times. But reuse_swap_page() might have removed the page | |
6276 | * from swapcache already, so we can't check PageSwapCache(). | |
6277 | */ | |
6278 | if (!memcg) | |
6279 | return; | |
6280 | ||
6abb5a86 JW |
6281 | commit_charge(page, memcg, lrucare); |
6282 | ||
00501b53 JW |
6283 | if (PageTransHuge(page)) { |
6284 | nr_pages <<= compound_order(page); | |
6285 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6286 | } | |
6287 | ||
6abb5a86 JW |
6288 | local_irq_disable(); |
6289 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
6290 | memcg_check_events(memcg, page); | |
6291 | local_irq_enable(); | |
00501b53 JW |
6292 | |
6293 | if (do_swap_account && PageSwapCache(page)) { | |
6294 | swp_entry_t entry = { .val = page_private(page) }; | |
6295 | /* | |
6296 | * The swap entry might not get freed for a long time, | |
6297 | * let's not wait for it. The page already received a | |
6298 | * memory+swap charge, drop the swap entry duplicate. | |
6299 | */ | |
6300 | mem_cgroup_uncharge_swap(entry); | |
6301 | } | |
6302 | } | |
6303 | ||
6304 | /** | |
6305 | * mem_cgroup_cancel_charge - cancel a page charge | |
6306 | * @page: page to charge | |
6307 | * @memcg: memcg to charge the page to | |
6308 | * | |
6309 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6310 | */ | |
6311 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
6312 | { | |
6313 | unsigned int nr_pages = 1; | |
6314 | ||
6315 | if (mem_cgroup_disabled()) | |
6316 | return; | |
6317 | /* | |
6318 | * Swap faults will attempt to charge the same page multiple | |
6319 | * times. But reuse_swap_page() might have removed the page | |
6320 | * from swapcache already, so we can't check PageSwapCache(). | |
6321 | */ | |
6322 | if (!memcg) | |
6323 | return; | |
6324 | ||
6325 | if (PageTransHuge(page)) { | |
6326 | nr_pages <<= compound_order(page); | |
6327 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6328 | } | |
6329 | ||
6330 | cancel_charge(memcg, nr_pages); | |
6331 | } | |
6332 | ||
747db954 JW |
6333 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
6334 | unsigned long nr_mem, unsigned long nr_memsw, | |
6335 | unsigned long nr_anon, unsigned long nr_file, | |
6336 | unsigned long nr_huge, struct page *dummy_page) | |
6337 | { | |
6338 | unsigned long flags; | |
6339 | ||
ce00a967 JW |
6340 | if (!mem_cgroup_is_root(memcg)) { |
6341 | if (nr_mem) | |
3e32cb2e | 6342 | page_counter_uncharge(&memcg->memory, nr_mem); |
ce00a967 | 6343 | if (nr_memsw) |
3e32cb2e | 6344 | page_counter_uncharge(&memcg->memsw, nr_memsw); |
ce00a967 JW |
6345 | memcg_oom_recover(memcg); |
6346 | } | |
747db954 JW |
6347 | |
6348 | local_irq_save(flags); | |
6349 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
6350 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
6351 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
6352 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
6353 | __this_cpu_add(memcg->stat->nr_page_events, nr_anon + nr_file); | |
6354 | memcg_check_events(memcg, dummy_page); | |
6355 | local_irq_restore(flags); | |
e8ea14cc JW |
6356 | |
6357 | if (!mem_cgroup_is_root(memcg)) | |
6358 | css_put_many(&memcg->css, max(nr_mem, nr_memsw)); | |
747db954 JW |
6359 | } |
6360 | ||
6361 | static void uncharge_list(struct list_head *page_list) | |
6362 | { | |
6363 | struct mem_cgroup *memcg = NULL; | |
6364 | unsigned long nr_memsw = 0; | |
6365 | unsigned long nr_anon = 0; | |
6366 | unsigned long nr_file = 0; | |
6367 | unsigned long nr_huge = 0; | |
6368 | unsigned long pgpgout = 0; | |
6369 | unsigned long nr_mem = 0; | |
6370 | struct list_head *next; | |
6371 | struct page *page; | |
6372 | ||
6373 | next = page_list->next; | |
6374 | do { | |
6375 | unsigned int nr_pages = 1; | |
6376 | struct page_cgroup *pc; | |
6377 | ||
6378 | page = list_entry(next, struct page, lru); | |
6379 | next = page->lru.next; | |
6380 | ||
6381 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6382 | VM_BUG_ON_PAGE(page_count(page), page); | |
6383 | ||
6384 | pc = lookup_page_cgroup(page); | |
6385 | if (!PageCgroupUsed(pc)) | |
6386 | continue; | |
6387 | ||
6388 | /* | |
6389 | * Nobody should be changing or seriously looking at | |
6390 | * pc->mem_cgroup and pc->flags at this point, we have | |
6391 | * fully exclusive access to the page. | |
6392 | */ | |
6393 | ||
6394 | if (memcg != pc->mem_cgroup) { | |
6395 | if (memcg) { | |
6396 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6397 | nr_anon, nr_file, nr_huge, page); | |
6398 | pgpgout = nr_mem = nr_memsw = 0; | |
6399 | nr_anon = nr_file = nr_huge = 0; | |
6400 | } | |
6401 | memcg = pc->mem_cgroup; | |
6402 | } | |
6403 | ||
6404 | if (PageTransHuge(page)) { | |
6405 | nr_pages <<= compound_order(page); | |
6406 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6407 | nr_huge += nr_pages; | |
6408 | } | |
6409 | ||
6410 | if (PageAnon(page)) | |
6411 | nr_anon += nr_pages; | |
6412 | else | |
6413 | nr_file += nr_pages; | |
6414 | ||
6415 | if (pc->flags & PCG_MEM) | |
6416 | nr_mem += nr_pages; | |
6417 | if (pc->flags & PCG_MEMSW) | |
6418 | nr_memsw += nr_pages; | |
6419 | pc->flags = 0; | |
6420 | ||
6421 | pgpgout++; | |
6422 | } while (next != page_list); | |
6423 | ||
6424 | if (memcg) | |
6425 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6426 | nr_anon, nr_file, nr_huge, page); | |
6427 | } | |
6428 | ||
0a31bc97 JW |
6429 | /** |
6430 | * mem_cgroup_uncharge - uncharge a page | |
6431 | * @page: page to uncharge | |
6432 | * | |
6433 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6434 | * mem_cgroup_commit_charge(). | |
6435 | */ | |
6436 | void mem_cgroup_uncharge(struct page *page) | |
6437 | { | |
0a31bc97 | 6438 | struct page_cgroup *pc; |
0a31bc97 JW |
6439 | |
6440 | if (mem_cgroup_disabled()) | |
6441 | return; | |
6442 | ||
747db954 | 6443 | /* Don't touch page->lru of any random page, pre-check: */ |
0a31bc97 | 6444 | pc = lookup_page_cgroup(page); |
0a31bc97 JW |
6445 | if (!PageCgroupUsed(pc)) |
6446 | return; | |
6447 | ||
747db954 JW |
6448 | INIT_LIST_HEAD(&page->lru); |
6449 | uncharge_list(&page->lru); | |
6450 | } | |
0a31bc97 | 6451 | |
747db954 JW |
6452 | /** |
6453 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6454 | * @page_list: list of pages to uncharge | |
6455 | * | |
6456 | * Uncharge a list of pages previously charged with | |
6457 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6458 | */ | |
6459 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6460 | { | |
6461 | if (mem_cgroup_disabled()) | |
6462 | return; | |
0a31bc97 | 6463 | |
747db954 JW |
6464 | if (!list_empty(page_list)) |
6465 | uncharge_list(page_list); | |
0a31bc97 JW |
6466 | } |
6467 | ||
6468 | /** | |
6469 | * mem_cgroup_migrate - migrate a charge to another page | |
6470 | * @oldpage: currently charged page | |
6471 | * @newpage: page to transfer the charge to | |
6472 | * @lrucare: both pages might be on the LRU already | |
6473 | * | |
6474 | * Migrate the charge from @oldpage to @newpage. | |
6475 | * | |
6476 | * Both pages must be locked, @newpage->mapping must be set up. | |
6477 | */ | |
6478 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
6479 | bool lrucare) | |
6480 | { | |
0a31bc97 JW |
6481 | struct page_cgroup *pc; |
6482 | int isolated; | |
6483 | ||
6484 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6485 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
6486 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
6487 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
6488 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
6489 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6490 | newpage); | |
0a31bc97 JW |
6491 | |
6492 | if (mem_cgroup_disabled()) | |
6493 | return; | |
6494 | ||
6495 | /* Page cache replacement: new page already charged? */ | |
6496 | pc = lookup_page_cgroup(newpage); | |
6497 | if (PageCgroupUsed(pc)) | |
6498 | return; | |
6499 | ||
6500 | /* Re-entrant migration: old page already uncharged? */ | |
6501 | pc = lookup_page_cgroup(oldpage); | |
6502 | if (!PageCgroupUsed(pc)) | |
6503 | return; | |
6504 | ||
6505 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEM), oldpage); | |
6506 | VM_BUG_ON_PAGE(do_swap_account && !(pc->flags & PCG_MEMSW), oldpage); | |
6507 | ||
0a31bc97 JW |
6508 | if (lrucare) |
6509 | lock_page_lru(oldpage, &isolated); | |
6510 | ||
6511 | pc->flags = 0; | |
6512 | ||
6513 | if (lrucare) | |
6514 | unlock_page_lru(oldpage, isolated); | |
6515 | ||
6abb5a86 | 6516 | commit_charge(newpage, pc->mem_cgroup, lrucare); |
0a31bc97 JW |
6517 | } |
6518 | ||
2d11085e | 6519 | /* |
1081312f MH |
6520 | * subsys_initcall() for memory controller. |
6521 | * | |
6522 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6523 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6524 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6525 | * be initialized from here. | |
2d11085e MH |
6526 | */ |
6527 | static int __init mem_cgroup_init(void) | |
6528 | { | |
6529 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6530 | enable_swap_cgroup(); |
bb4cc1a8 | 6531 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6532 | memcg_stock_init(); |
2d11085e MH |
6533 | return 0; |
6534 | } | |
6535 | subsys_initcall(mem_cgroup_init); |