]>
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 */ |
510fc4e1 GC |
372 | }; |
373 | ||
510fc4e1 GC |
374 | #ifdef CONFIG_MEMCG_KMEM |
375 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
376 | { | |
377 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
378 | } | |
7de37682 GC |
379 | |
380 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
381 | { | |
382 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
383 | } | |
384 | ||
510fc4e1 GC |
385 | #endif |
386 | ||
7dc74be0 DN |
387 | /* Stuffs for move charges at task migration. */ |
388 | /* | |
ee5e8472 GC |
389 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
390 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
391 | */ |
392 | enum move_type { | |
4ffef5fe | 393 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 394 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
395 | NR_MOVE_TYPE, |
396 | }; | |
397 | ||
4ffef5fe DN |
398 | /* "mc" and its members are protected by cgroup_mutex */ |
399 | static struct move_charge_struct { | |
b1dd693e | 400 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
401 | struct mem_cgroup *from; |
402 | struct mem_cgroup *to; | |
ee5e8472 | 403 | unsigned long immigrate_flags; |
4ffef5fe | 404 | unsigned long precharge; |
854ffa8d | 405 | unsigned long moved_charge; |
483c30b5 | 406 | unsigned long moved_swap; |
8033b97c DN |
407 | struct task_struct *moving_task; /* a task moving charges */ |
408 | wait_queue_head_t waitq; /* a waitq for other context */ | |
409 | } mc = { | |
2bd9bb20 | 410 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
411 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
412 | }; | |
4ffef5fe | 413 | |
90254a65 DN |
414 | static bool move_anon(void) |
415 | { | |
ee5e8472 | 416 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
417 | } |
418 | ||
87946a72 DN |
419 | static bool move_file(void) |
420 | { | |
ee5e8472 | 421 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
422 | } |
423 | ||
4e416953 BS |
424 | /* |
425 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
426 | * limit reclaim to prevent infinite loops, if they ever occur. | |
427 | */ | |
a0db00fc | 428 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 429 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 430 | |
217bc319 KH |
431 | enum charge_type { |
432 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 433 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 434 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 435 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
436 | NR_CHARGE_TYPE, |
437 | }; | |
438 | ||
8c7c6e34 | 439 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
440 | enum res_type { |
441 | _MEM, | |
442 | _MEMSWAP, | |
443 | _OOM_TYPE, | |
510fc4e1 | 444 | _KMEM, |
86ae53e1 GC |
445 | }; |
446 | ||
a0db00fc KS |
447 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
448 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 449 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
450 | /* Used for OOM nofiier */ |
451 | #define OOM_CONTROL (0) | |
8c7c6e34 | 452 | |
0999821b GC |
453 | /* |
454 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
455 | * As a consequence, any change that needs to protect against new child cgroups | |
456 | * appearing has to hold it as well. | |
457 | */ | |
458 | static DEFINE_MUTEX(memcg_create_mutex); | |
459 | ||
b2145145 WL |
460 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
461 | { | |
a7c6d554 | 462 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
463 | } |
464 | ||
70ddf637 AV |
465 | /* Some nice accessors for the vmpressure. */ |
466 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
467 | { | |
468 | if (!memcg) | |
469 | memcg = root_mem_cgroup; | |
470 | return &memcg->vmpressure; | |
471 | } | |
472 | ||
473 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
474 | { | |
475 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
476 | } | |
477 | ||
7ffc0edc MH |
478 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
479 | { | |
480 | return (memcg == root_mem_cgroup); | |
481 | } | |
482 | ||
4219b2da LZ |
483 | /* |
484 | * We restrict the id in the range of [1, 65535], so it can fit into | |
485 | * an unsigned short. | |
486 | */ | |
487 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
488 | ||
34c00c31 LZ |
489 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
490 | { | |
15a4c835 | 491 | return memcg->css.id; |
34c00c31 LZ |
492 | } |
493 | ||
494 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
495 | { | |
496 | struct cgroup_subsys_state *css; | |
497 | ||
7d699ddb | 498 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
499 | return mem_cgroup_from_css(css); |
500 | } | |
501 | ||
e1aab161 | 502 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 503 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 504 | |
e1aab161 GC |
505 | void sock_update_memcg(struct sock *sk) |
506 | { | |
376be5ff | 507 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 508 | struct mem_cgroup *memcg; |
3f134619 | 509 | struct cg_proto *cg_proto; |
e1aab161 GC |
510 | |
511 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
512 | ||
f3f511e1 GC |
513 | /* Socket cloning can throw us here with sk_cgrp already |
514 | * filled. It won't however, necessarily happen from | |
515 | * process context. So the test for root memcg given | |
516 | * the current task's memcg won't help us in this case. | |
517 | * | |
518 | * Respecting the original socket's memcg is a better | |
519 | * decision in this case. | |
520 | */ | |
521 | if (sk->sk_cgrp) { | |
522 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 523 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
524 | return; |
525 | } | |
526 | ||
e1aab161 GC |
527 | rcu_read_lock(); |
528 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 529 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 530 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
531 | memcg_proto_active(cg_proto) && |
532 | css_tryget_online(&memcg->css)) { | |
3f134619 | 533 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
534 | } |
535 | rcu_read_unlock(); | |
536 | } | |
537 | } | |
538 | EXPORT_SYMBOL(sock_update_memcg); | |
539 | ||
540 | void sock_release_memcg(struct sock *sk) | |
541 | { | |
376be5ff | 542 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
543 | struct mem_cgroup *memcg; |
544 | WARN_ON(!sk->sk_cgrp->memcg); | |
545 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 546 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
547 | } |
548 | } | |
d1a4c0b3 GC |
549 | |
550 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
551 | { | |
552 | if (!memcg || mem_cgroup_is_root(memcg)) | |
553 | return NULL; | |
554 | ||
2e685cad | 555 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
556 | } |
557 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 558 | |
3f134619 GC |
559 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
560 | { | |
2e685cad | 561 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
562 | return; |
563 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
564 | } | |
565 | #else | |
566 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
567 | { | |
568 | } | |
569 | #endif | |
570 | ||
a8964b9b | 571 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
572 | /* |
573 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
b8627835 LZ |
574 | * The main reason for not using cgroup id for this: |
575 | * this works better in sparse environments, where we have a lot of memcgs, | |
576 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
577 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
578 | * 200 entry array for that. | |
55007d84 GC |
579 | * |
580 | * The current size of the caches array is stored in | |
581 | * memcg_limited_groups_array_size. It will double each time we have to | |
582 | * increase it. | |
583 | */ | |
584 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
585 | int memcg_limited_groups_array_size; |
586 | ||
55007d84 GC |
587 | /* |
588 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
589 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
590 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
591 | * tunable, but that is strictly not necessary. | |
592 | * | |
b8627835 | 593 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
594 | * this constant directly from cgroup, but it is understandable that this is |
595 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 596 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
597 | * increase ours as well if it increases. |
598 | */ | |
599 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 600 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 601 | |
d7f25f8a GC |
602 | /* |
603 | * A lot of the calls to the cache allocation functions are expected to be | |
604 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
605 | * conditional to this static branch, we'll have to allow modules that does | |
606 | * kmem_cache_alloc and the such to see this symbol as well | |
607 | */ | |
a8964b9b | 608 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 609 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 610 | |
f3bb3043 VD |
611 | static void memcg_free_cache_id(int id); |
612 | ||
a8964b9b GC |
613 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
614 | { | |
55007d84 | 615 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 616 | static_key_slow_dec(&memcg_kmem_enabled_key); |
f3bb3043 | 617 | memcg_free_cache_id(memcg->kmemcg_id); |
55007d84 | 618 | } |
bea207c8 GC |
619 | /* |
620 | * This check can't live in kmem destruction function, | |
621 | * since the charges will outlive the cgroup | |
622 | */ | |
3e32cb2e | 623 | WARN_ON(page_counter_read(&memcg->kmem)); |
a8964b9b GC |
624 | } |
625 | #else | |
626 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
627 | { | |
628 | } | |
629 | #endif /* CONFIG_MEMCG_KMEM */ | |
630 | ||
631 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
632 | { | |
633 | disarm_sock_keys(memcg); | |
634 | disarm_kmem_keys(memcg); | |
635 | } | |
636 | ||
f64c3f54 | 637 | static struct mem_cgroup_per_zone * |
e231875b | 638 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 639 | { |
e231875b JZ |
640 | int nid = zone_to_nid(zone); |
641 | int zid = zone_idx(zone); | |
642 | ||
54f72fe0 | 643 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
644 | } |
645 | ||
c0ff4b85 | 646 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 647 | { |
c0ff4b85 | 648 | return &memcg->css; |
d324236b WF |
649 | } |
650 | ||
f64c3f54 | 651 | static struct mem_cgroup_per_zone * |
e231875b | 652 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 653 | { |
97a6c37b JW |
654 | int nid = page_to_nid(page); |
655 | int zid = page_zonenum(page); | |
f64c3f54 | 656 | |
e231875b | 657 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
658 | } |
659 | ||
bb4cc1a8 AM |
660 | static struct mem_cgroup_tree_per_zone * |
661 | soft_limit_tree_node_zone(int nid, int zid) | |
662 | { | |
663 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
664 | } | |
665 | ||
666 | static struct mem_cgroup_tree_per_zone * | |
667 | soft_limit_tree_from_page(struct page *page) | |
668 | { | |
669 | int nid = page_to_nid(page); | |
670 | int zid = page_zonenum(page); | |
671 | ||
672 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
673 | } | |
674 | ||
cf2c8127 JW |
675 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
676 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 677 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
678 | { |
679 | struct rb_node **p = &mctz->rb_root.rb_node; | |
680 | struct rb_node *parent = NULL; | |
681 | struct mem_cgroup_per_zone *mz_node; | |
682 | ||
683 | if (mz->on_tree) | |
684 | return; | |
685 | ||
686 | mz->usage_in_excess = new_usage_in_excess; | |
687 | if (!mz->usage_in_excess) | |
688 | return; | |
689 | while (*p) { | |
690 | parent = *p; | |
691 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
692 | tree_node); | |
693 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
694 | p = &(*p)->rb_left; | |
695 | /* | |
696 | * We can't avoid mem cgroups that are over their soft | |
697 | * limit by the same amount | |
698 | */ | |
699 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
700 | p = &(*p)->rb_right; | |
701 | } | |
702 | rb_link_node(&mz->tree_node, parent, p); | |
703 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
704 | mz->on_tree = true; | |
705 | } | |
706 | ||
cf2c8127 JW |
707 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
708 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
709 | { |
710 | if (!mz->on_tree) | |
711 | return; | |
712 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
713 | mz->on_tree = false; | |
714 | } | |
715 | ||
cf2c8127 JW |
716 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
717 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 718 | { |
0a31bc97 JW |
719 | unsigned long flags; |
720 | ||
721 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 722 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 723 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
724 | } |
725 | ||
3e32cb2e JW |
726 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
727 | { | |
728 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
729 | unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit); | |
730 | unsigned long excess = 0; | |
731 | ||
732 | if (nr_pages > soft_limit) | |
733 | excess = nr_pages - soft_limit; | |
734 | ||
735 | return excess; | |
736 | } | |
bb4cc1a8 AM |
737 | |
738 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
739 | { | |
3e32cb2e | 740 | unsigned long excess; |
bb4cc1a8 AM |
741 | struct mem_cgroup_per_zone *mz; |
742 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 743 | |
e231875b | 744 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
745 | /* |
746 | * Necessary to update all ancestors when hierarchy is used. | |
747 | * because their event counter is not touched. | |
748 | */ | |
749 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 750 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 751 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
752 | /* |
753 | * We have to update the tree if mz is on RB-tree or | |
754 | * mem is over its softlimit. | |
755 | */ | |
756 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
757 | unsigned long flags; |
758 | ||
759 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
760 | /* if on-tree, remove it */ |
761 | if (mz->on_tree) | |
cf2c8127 | 762 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
763 | /* |
764 | * Insert again. mz->usage_in_excess will be updated. | |
765 | * If excess is 0, no tree ops. | |
766 | */ | |
cf2c8127 | 767 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 768 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
769 | } |
770 | } | |
771 | } | |
772 | ||
773 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
774 | { | |
bb4cc1a8 | 775 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
776 | struct mem_cgroup_per_zone *mz; |
777 | int nid, zid; | |
bb4cc1a8 | 778 | |
e231875b JZ |
779 | for_each_node(nid) { |
780 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
781 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
782 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 783 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
784 | } |
785 | } | |
786 | } | |
787 | ||
788 | static struct mem_cgroup_per_zone * | |
789 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
790 | { | |
791 | struct rb_node *rightmost = NULL; | |
792 | struct mem_cgroup_per_zone *mz; | |
793 | ||
794 | retry: | |
795 | mz = NULL; | |
796 | rightmost = rb_last(&mctz->rb_root); | |
797 | if (!rightmost) | |
798 | goto done; /* Nothing to reclaim from */ | |
799 | ||
800 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
801 | /* | |
802 | * Remove the node now but someone else can add it back, | |
803 | * we will to add it back at the end of reclaim to its correct | |
804 | * position in the tree. | |
805 | */ | |
cf2c8127 | 806 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 807 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 808 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
809 | goto retry; |
810 | done: | |
811 | return mz; | |
812 | } | |
813 | ||
814 | static struct mem_cgroup_per_zone * | |
815 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
816 | { | |
817 | struct mem_cgroup_per_zone *mz; | |
818 | ||
0a31bc97 | 819 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 820 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 821 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
822 | return mz; |
823 | } | |
824 | ||
711d3d2c KH |
825 | /* |
826 | * Implementation Note: reading percpu statistics for memcg. | |
827 | * | |
828 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
829 | * synchronization to implement "quick" read. There are trade-off between | |
830 | * reading cost and precision of value. Then, we may have a chance to implement | |
831 | * a periodic synchronizion of counter in memcg's counter. | |
832 | * | |
833 | * But this _read() function is used for user interface now. The user accounts | |
834 | * memory usage by memory cgroup and he _always_ requires exact value because | |
835 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
836 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
837 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
838 | * | |
839 | * If there are kernel internal actions which can make use of some not-exact | |
840 | * value, and reading all cpu value can be performance bottleneck in some | |
841 | * common workload, threashold and synchonization as vmstat[] should be | |
842 | * implemented. | |
843 | */ | |
c0ff4b85 | 844 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 845 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 846 | { |
7a159cc9 | 847 | long val = 0; |
c62b1a3b | 848 | int cpu; |
c62b1a3b | 849 | |
711d3d2c KH |
850 | get_online_cpus(); |
851 | for_each_online_cpu(cpu) | |
c0ff4b85 | 852 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 853 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
854 | spin_lock(&memcg->pcp_counter_lock); |
855 | val += memcg->nocpu_base.count[idx]; | |
856 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
857 | #endif |
858 | put_online_cpus(); | |
c62b1a3b KH |
859 | return val; |
860 | } | |
861 | ||
c0ff4b85 | 862 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
863 | enum mem_cgroup_events_index idx) |
864 | { | |
865 | unsigned long val = 0; | |
866 | int cpu; | |
867 | ||
9c567512 | 868 | get_online_cpus(); |
e9f8974f | 869 | for_each_online_cpu(cpu) |
c0ff4b85 | 870 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 871 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
872 | spin_lock(&memcg->pcp_counter_lock); |
873 | val += memcg->nocpu_base.events[idx]; | |
874 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 875 | #endif |
9c567512 | 876 | put_online_cpus(); |
e9f8974f JW |
877 | return val; |
878 | } | |
879 | ||
c0ff4b85 | 880 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 881 | struct page *page, |
0a31bc97 | 882 | int nr_pages) |
d52aa412 | 883 | { |
b2402857 KH |
884 | /* |
885 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
886 | * counted as CACHE even if it's on ANON LRU. | |
887 | */ | |
0a31bc97 | 888 | if (PageAnon(page)) |
b2402857 | 889 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 890 | nr_pages); |
d52aa412 | 891 | else |
b2402857 | 892 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 893 | nr_pages); |
55e462b0 | 894 | |
b070e65c DR |
895 | if (PageTransHuge(page)) |
896 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
897 | nr_pages); | |
898 | ||
e401f176 KH |
899 | /* pagein of a big page is an event. So, ignore page size */ |
900 | if (nr_pages > 0) | |
c0ff4b85 | 901 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 902 | else { |
c0ff4b85 | 903 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
904 | nr_pages = -nr_pages; /* for event */ |
905 | } | |
e401f176 | 906 | |
13114716 | 907 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
908 | } |
909 | ||
e231875b | 910 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
911 | { |
912 | struct mem_cgroup_per_zone *mz; | |
913 | ||
914 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
915 | return mz->lru_size[lru]; | |
916 | } | |
917 | ||
e231875b JZ |
918 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
919 | int nid, | |
920 | unsigned int lru_mask) | |
bb2a0de9 | 921 | { |
e231875b | 922 | unsigned long nr = 0; |
889976db YH |
923 | int zid; |
924 | ||
e231875b | 925 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 926 | |
e231875b JZ |
927 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
928 | struct mem_cgroup_per_zone *mz; | |
929 | enum lru_list lru; | |
930 | ||
931 | for_each_lru(lru) { | |
932 | if (!(BIT(lru) & lru_mask)) | |
933 | continue; | |
934 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
935 | nr += mz->lru_size[lru]; | |
936 | } | |
937 | } | |
938 | return nr; | |
889976db | 939 | } |
bb2a0de9 | 940 | |
c0ff4b85 | 941 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 942 | unsigned int lru_mask) |
6d12e2d8 | 943 | { |
e231875b | 944 | unsigned long nr = 0; |
889976db | 945 | int nid; |
6d12e2d8 | 946 | |
31aaea4a | 947 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
948 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
949 | return nr; | |
d52aa412 KH |
950 | } |
951 | ||
f53d7ce3 JW |
952 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
953 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
954 | { |
955 | unsigned long val, next; | |
956 | ||
13114716 | 957 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 958 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 959 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
960 | if ((long)next - (long)val < 0) { |
961 | switch (target) { | |
962 | case MEM_CGROUP_TARGET_THRESH: | |
963 | next = val + THRESHOLDS_EVENTS_TARGET; | |
964 | break; | |
bb4cc1a8 AM |
965 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
966 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
967 | break; | |
f53d7ce3 JW |
968 | case MEM_CGROUP_TARGET_NUMAINFO: |
969 | next = val + NUMAINFO_EVENTS_TARGET; | |
970 | break; | |
971 | default: | |
972 | break; | |
973 | } | |
974 | __this_cpu_write(memcg->stat->targets[target], next); | |
975 | return true; | |
7a159cc9 | 976 | } |
f53d7ce3 | 977 | return false; |
d2265e6f KH |
978 | } |
979 | ||
980 | /* | |
981 | * Check events in order. | |
982 | * | |
983 | */ | |
c0ff4b85 | 984 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
985 | { |
986 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
987 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
988 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 989 | bool do_softlimit; |
82b3f2a7 | 990 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 991 | |
bb4cc1a8 AM |
992 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
993 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
994 | #if MAX_NUMNODES > 1 |
995 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
996 | MEM_CGROUP_TARGET_NUMAINFO); | |
997 | #endif | |
c0ff4b85 | 998 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
999 | if (unlikely(do_softlimit)) |
1000 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1001 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1002 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1003 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1004 | #endif |
0a31bc97 | 1005 | } |
d2265e6f KH |
1006 | } |
1007 | ||
cf475ad2 | 1008 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1009 | { |
31a78f23 BS |
1010 | /* |
1011 | * mm_update_next_owner() may clear mm->owner to NULL | |
1012 | * if it races with swapoff, page migration, etc. | |
1013 | * So this can be called with p == NULL. | |
1014 | */ | |
1015 | if (unlikely(!p)) | |
1016 | return NULL; | |
1017 | ||
073219e9 | 1018 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1019 | } |
1020 | ||
df381975 | 1021 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1022 | { |
c0ff4b85 | 1023 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 1024 | |
54595fe2 KH |
1025 | rcu_read_lock(); |
1026 | do { | |
6f6acb00 MH |
1027 | /* |
1028 | * Page cache insertions can happen withou an | |
1029 | * actual mm context, e.g. during disk probing | |
1030 | * on boot, loopback IO, acct() writes etc. | |
1031 | */ | |
1032 | if (unlikely(!mm)) | |
df381975 | 1033 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1034 | else { |
1035 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1036 | if (unlikely(!memcg)) | |
1037 | memcg = root_mem_cgroup; | |
1038 | } | |
ec903c0c | 1039 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1040 | rcu_read_unlock(); |
c0ff4b85 | 1041 | return memcg; |
54595fe2 KH |
1042 | } |
1043 | ||
5660048c JW |
1044 | /** |
1045 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1046 | * @root: hierarchy root | |
1047 | * @prev: previously returned memcg, NULL on first invocation | |
1048 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1049 | * | |
1050 | * Returns references to children of the hierarchy below @root, or | |
1051 | * @root itself, or %NULL after a full round-trip. | |
1052 | * | |
1053 | * Caller must pass the return value in @prev on subsequent | |
1054 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1055 | * to cancel a hierarchy walk before the round-trip is complete. | |
1056 | * | |
1057 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1058 | * divide up the memcgs in the hierarchy among all concurrent | |
1059 | * reclaimers operating on the same zone and priority. | |
1060 | */ | |
694fbc0f | 1061 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1062 | struct mem_cgroup *prev, |
694fbc0f | 1063 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1064 | { |
5ac8fb31 JW |
1065 | struct reclaim_iter *uninitialized_var(iter); |
1066 | struct cgroup_subsys_state *css = NULL; | |
9f3a0d09 | 1067 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1068 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1069 | |
694fbc0f AM |
1070 | if (mem_cgroup_disabled()) |
1071 | return NULL; | |
5660048c | 1072 | |
9f3a0d09 JW |
1073 | if (!root) |
1074 | root = root_mem_cgroup; | |
7d74b06f | 1075 | |
9f3a0d09 | 1076 | if (prev && !reclaim) |
5ac8fb31 | 1077 | pos = prev; |
14067bb3 | 1078 | |
9f3a0d09 JW |
1079 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1080 | if (prev) | |
5ac8fb31 | 1081 | goto out; |
694fbc0f | 1082 | return root; |
9f3a0d09 | 1083 | } |
14067bb3 | 1084 | |
542f85f9 | 1085 | rcu_read_lock(); |
5f578161 | 1086 | |
5ac8fb31 JW |
1087 | if (reclaim) { |
1088 | struct mem_cgroup_per_zone *mz; | |
1089 | ||
1090 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
1091 | iter = &mz->iter[reclaim->priority]; | |
1092 | ||
1093 | if (prev && reclaim->generation != iter->generation) | |
1094 | goto out_unlock; | |
1095 | ||
1096 | do { | |
1097 | pos = ACCESS_ONCE(iter->position); | |
1098 | /* | |
1099 | * A racing update may change the position and | |
1100 | * put the last reference, hence css_tryget(), | |
1101 | * or retry to see the updated position. | |
1102 | */ | |
1103 | } while (pos && !css_tryget(&pos->css)); | |
1104 | } | |
1105 | ||
1106 | if (pos) | |
1107 | css = &pos->css; | |
1108 | ||
1109 | for (;;) { | |
1110 | css = css_next_descendant_pre(css, &root->css); | |
1111 | if (!css) { | |
1112 | /* | |
1113 | * Reclaimers share the hierarchy walk, and a | |
1114 | * new one might jump in right at the end of | |
1115 | * the hierarchy - make sure they see at least | |
1116 | * one group and restart from the beginning. | |
1117 | */ | |
1118 | if (!prev) | |
1119 | continue; | |
1120 | break; | |
527a5ec9 | 1121 | } |
7d74b06f | 1122 | |
5ac8fb31 JW |
1123 | /* |
1124 | * Verify the css and acquire a reference. The root | |
1125 | * is provided by the caller, so we know it's alive | |
1126 | * and kicking, and don't take an extra reference. | |
1127 | */ | |
1128 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1129 | |
5ac8fb31 JW |
1130 | if (css == &root->css) |
1131 | break; | |
542f85f9 | 1132 | |
b2052564 | 1133 | if (css_tryget(css)) { |
5ac8fb31 JW |
1134 | /* |
1135 | * Make sure the memcg is initialized: | |
1136 | * mem_cgroup_css_online() orders the the | |
1137 | * initialization against setting the flag. | |
1138 | */ | |
1139 | if (smp_load_acquire(&memcg->initialized)) | |
1140 | break; | |
1141 | ||
1142 | css_put(css); | |
527a5ec9 | 1143 | } |
9f3a0d09 | 1144 | |
5ac8fb31 JW |
1145 | memcg = NULL; |
1146 | } | |
1147 | ||
1148 | if (reclaim) { | |
1149 | if (cmpxchg(&iter->position, pos, memcg) == pos) { | |
1150 | if (memcg) | |
1151 | css_get(&memcg->css); | |
1152 | if (pos) | |
1153 | css_put(&pos->css); | |
1154 | } | |
1155 | ||
1156 | /* | |
1157 | * pairs with css_tryget when dereferencing iter->position | |
1158 | * above. | |
1159 | */ | |
1160 | if (pos) | |
1161 | css_put(&pos->css); | |
1162 | ||
1163 | if (!memcg) | |
1164 | iter->generation++; | |
1165 | else if (!prev) | |
1166 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1167 | } |
5ac8fb31 | 1168 | |
542f85f9 MH |
1169 | out_unlock: |
1170 | rcu_read_unlock(); | |
5ac8fb31 | 1171 | out: |
c40046f3 MH |
1172 | if (prev && prev != root) |
1173 | css_put(&prev->css); | |
1174 | ||
9f3a0d09 | 1175 | return memcg; |
14067bb3 | 1176 | } |
7d74b06f | 1177 | |
5660048c JW |
1178 | /** |
1179 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1180 | * @root: hierarchy root | |
1181 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1182 | */ | |
1183 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1184 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1185 | { |
1186 | if (!root) | |
1187 | root = root_mem_cgroup; | |
1188 | if (prev && prev != root) | |
1189 | css_put(&prev->css); | |
1190 | } | |
7d74b06f | 1191 | |
9f3a0d09 JW |
1192 | /* |
1193 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1194 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1195 | * be used for reference counting. | |
1196 | */ | |
1197 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1198 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1199 | iter != NULL; \ |
527a5ec9 | 1200 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1201 | |
9f3a0d09 | 1202 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1203 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1204 | iter != NULL; \ |
527a5ec9 | 1205 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1206 | |
68ae564b | 1207 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1208 | { |
c0ff4b85 | 1209 | struct mem_cgroup *memcg; |
456f998e | 1210 | |
456f998e | 1211 | rcu_read_lock(); |
c0ff4b85 R |
1212 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1213 | if (unlikely(!memcg)) | |
456f998e YH |
1214 | goto out; |
1215 | ||
1216 | switch (idx) { | |
456f998e | 1217 | case PGFAULT: |
0e574a93 JW |
1218 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1219 | break; | |
1220 | case PGMAJFAULT: | |
1221 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1222 | break; |
1223 | default: | |
1224 | BUG(); | |
1225 | } | |
1226 | out: | |
1227 | rcu_read_unlock(); | |
1228 | } | |
68ae564b | 1229 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1230 | |
925b7673 JW |
1231 | /** |
1232 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1233 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1234 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1235 | * |
1236 | * Returns the lru list vector holding pages for the given @zone and | |
1237 | * @mem. This can be the global zone lruvec, if the memory controller | |
1238 | * is disabled. | |
1239 | */ | |
1240 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1241 | struct mem_cgroup *memcg) | |
1242 | { | |
1243 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1244 | struct lruvec *lruvec; |
925b7673 | 1245 | |
bea8c150 HD |
1246 | if (mem_cgroup_disabled()) { |
1247 | lruvec = &zone->lruvec; | |
1248 | goto out; | |
1249 | } | |
925b7673 | 1250 | |
e231875b | 1251 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1252 | lruvec = &mz->lruvec; |
1253 | out: | |
1254 | /* | |
1255 | * Since a node can be onlined after the mem_cgroup was created, | |
1256 | * we have to be prepared to initialize lruvec->zone here; | |
1257 | * and if offlined then reonlined, we need to reinitialize it. | |
1258 | */ | |
1259 | if (unlikely(lruvec->zone != zone)) | |
1260 | lruvec->zone = zone; | |
1261 | return lruvec; | |
925b7673 JW |
1262 | } |
1263 | ||
925b7673 | 1264 | /** |
dfe0e773 | 1265 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1266 | * @page: the page |
fa9add64 | 1267 | * @zone: zone of the page |
dfe0e773 JW |
1268 | * |
1269 | * This function is only safe when following the LRU page isolation | |
1270 | * and putback protocol: the LRU lock must be held, and the page must | |
1271 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1272 | */ |
fa9add64 | 1273 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1274 | { |
08e552c6 | 1275 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1276 | struct mem_cgroup *memcg; |
1277 | struct page_cgroup *pc; | |
bea8c150 | 1278 | struct lruvec *lruvec; |
6d12e2d8 | 1279 | |
bea8c150 HD |
1280 | if (mem_cgroup_disabled()) { |
1281 | lruvec = &zone->lruvec; | |
1282 | goto out; | |
1283 | } | |
925b7673 | 1284 | |
08e552c6 | 1285 | pc = lookup_page_cgroup(page); |
38c5d72f | 1286 | memcg = pc->mem_cgroup; |
7512102c HD |
1287 | |
1288 | /* | |
dfe0e773 JW |
1289 | * Swapcache readahead pages are added to the LRU - and |
1290 | * possibly migrated - before they are charged. Ensure | |
1291 | * pc->mem_cgroup is sane. | |
7512102c | 1292 | */ |
fa9add64 | 1293 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1294 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1295 | ||
e231875b | 1296 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1297 | lruvec = &mz->lruvec; |
1298 | out: | |
1299 | /* | |
1300 | * Since a node can be onlined after the mem_cgroup was created, | |
1301 | * we have to be prepared to initialize lruvec->zone here; | |
1302 | * and if offlined then reonlined, we need to reinitialize it. | |
1303 | */ | |
1304 | if (unlikely(lruvec->zone != zone)) | |
1305 | lruvec->zone = zone; | |
1306 | return lruvec; | |
08e552c6 | 1307 | } |
b69408e8 | 1308 | |
925b7673 | 1309 | /** |
fa9add64 HD |
1310 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1311 | * @lruvec: mem_cgroup per zone lru vector | |
1312 | * @lru: index of lru list the page is sitting on | |
1313 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1314 | * |
fa9add64 HD |
1315 | * This function must be called when a page is added to or removed from an |
1316 | * lru list. | |
3f58a829 | 1317 | */ |
fa9add64 HD |
1318 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1319 | int nr_pages) | |
3f58a829 MK |
1320 | { |
1321 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1322 | unsigned long *lru_size; |
3f58a829 MK |
1323 | |
1324 | if (mem_cgroup_disabled()) | |
1325 | return; | |
1326 | ||
fa9add64 HD |
1327 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1328 | lru_size = mz->lru_size + lru; | |
1329 | *lru_size += nr_pages; | |
1330 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1331 | } |
544122e5 | 1332 | |
3e92041d | 1333 | /* |
c0ff4b85 | 1334 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1335 | * hierarchy subtree |
1336 | */ | |
c3ac9a8a JW |
1337 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1338 | struct mem_cgroup *memcg) | |
3e92041d | 1339 | { |
91c63734 JW |
1340 | if (root_memcg == memcg) |
1341 | return true; | |
3a981f48 | 1342 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1343 | return false; |
b47f77b5 | 1344 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1345 | } |
1346 | ||
1347 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1348 | struct mem_cgroup *memcg) | |
1349 | { | |
1350 | bool ret; | |
1351 | ||
91c63734 | 1352 | rcu_read_lock(); |
c3ac9a8a | 1353 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1354 | rcu_read_unlock(); |
1355 | return ret; | |
3e92041d MH |
1356 | } |
1357 | ||
ffbdccf5 DR |
1358 | bool task_in_mem_cgroup(struct task_struct *task, |
1359 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1360 | { |
0b7f569e | 1361 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1362 | struct task_struct *p; |
ffbdccf5 | 1363 | bool ret; |
4c4a2214 | 1364 | |
158e0a2d | 1365 | p = find_lock_task_mm(task); |
de077d22 | 1366 | if (p) { |
df381975 | 1367 | curr = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1368 | task_unlock(p); |
1369 | } else { | |
1370 | /* | |
1371 | * All threads may have already detached their mm's, but the oom | |
1372 | * killer still needs to detect if they have already been oom | |
1373 | * killed to prevent needlessly killing additional tasks. | |
1374 | */ | |
ffbdccf5 | 1375 | rcu_read_lock(); |
de077d22 DR |
1376 | curr = mem_cgroup_from_task(task); |
1377 | if (curr) | |
1378 | css_get(&curr->css); | |
ffbdccf5 | 1379 | rcu_read_unlock(); |
de077d22 | 1380 | } |
d31f56db | 1381 | /* |
c0ff4b85 | 1382 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1383 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1384 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1385 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1386 | */ |
c0ff4b85 | 1387 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1388 | css_put(&curr->css); |
4c4a2214 DR |
1389 | return ret; |
1390 | } | |
1391 | ||
c56d5c7d | 1392 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1393 | { |
9b272977 | 1394 | unsigned long inactive_ratio; |
14797e23 | 1395 | unsigned long inactive; |
9b272977 | 1396 | unsigned long active; |
c772be93 | 1397 | unsigned long gb; |
14797e23 | 1398 | |
4d7dcca2 HD |
1399 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1400 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1401 | |
c772be93 KM |
1402 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1403 | if (gb) | |
1404 | inactive_ratio = int_sqrt(10 * gb); | |
1405 | else | |
1406 | inactive_ratio = 1; | |
1407 | ||
9b272977 | 1408 | return inactive * inactive_ratio < active; |
14797e23 KM |
1409 | } |
1410 | ||
3e32cb2e | 1411 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1412 | container_of(counter, struct mem_cgroup, member) |
1413 | ||
19942822 | 1414 | /** |
9d11ea9f | 1415 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1416 | * @memcg: the memory cgroup |
19942822 | 1417 | * |
9d11ea9f | 1418 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1419 | * pages. |
19942822 | 1420 | */ |
c0ff4b85 | 1421 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1422 | { |
3e32cb2e JW |
1423 | unsigned long margin = 0; |
1424 | unsigned long count; | |
1425 | unsigned long limit; | |
9d11ea9f | 1426 | |
3e32cb2e JW |
1427 | count = page_counter_read(&memcg->memory); |
1428 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1429 | if (count < limit) | |
1430 | margin = limit - count; | |
1431 | ||
1432 | if (do_swap_account) { | |
1433 | count = page_counter_read(&memcg->memsw); | |
1434 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1435 | if (count <= limit) | |
1436 | margin = min(margin, limit - count); | |
1437 | } | |
1438 | ||
1439 | return margin; | |
19942822 JW |
1440 | } |
1441 | ||
1f4c025b | 1442 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1443 | { |
a7885eb8 | 1444 | /* root ? */ |
14208b0e | 1445 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1446 | return vm_swappiness; |
1447 | ||
bf1ff263 | 1448 | return memcg->swappiness; |
a7885eb8 KM |
1449 | } |
1450 | ||
619d094b KH |
1451 | /* |
1452 | * memcg->moving_account is used for checking possibility that some thread is | |
1453 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1454 | * a memcg, other threads should check memcg->moving_account under | |
1455 | * rcu_read_lock(), like this: | |
1456 | * | |
1457 | * CPU-A CPU-B | |
1458 | * rcu_read_lock() | |
1459 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1460 | * take heavy locks. | |
1461 | * synchronize_rcu() update something. | |
1462 | * rcu_read_unlock() | |
1463 | * start move here. | |
1464 | */ | |
4331f7d3 | 1465 | |
c0ff4b85 | 1466 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1467 | { |
619d094b | 1468 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1469 | synchronize_rcu(); |
1470 | } | |
1471 | ||
c0ff4b85 | 1472 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1473 | { |
619d094b KH |
1474 | /* |
1475 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1476 | * We check NULL in callee rather than caller. | |
1477 | */ | |
d7365e78 | 1478 | if (memcg) |
619d094b | 1479 | atomic_dec(&memcg->moving_account); |
32047e2a | 1480 | } |
619d094b | 1481 | |
32047e2a | 1482 | /* |
bdcbb659 | 1483 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1484 | * |
bdcbb659 QH |
1485 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1486 | * moving cgroups. This is for waiting at high-memory pressure | |
1487 | * caused by "move". | |
32047e2a | 1488 | */ |
c0ff4b85 | 1489 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1490 | { |
2bd9bb20 KH |
1491 | struct mem_cgroup *from; |
1492 | struct mem_cgroup *to; | |
4b534334 | 1493 | bool ret = false; |
2bd9bb20 KH |
1494 | /* |
1495 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1496 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1497 | */ | |
1498 | spin_lock(&mc.lock); | |
1499 | from = mc.from; | |
1500 | to = mc.to; | |
1501 | if (!from) | |
1502 | goto unlock; | |
3e92041d | 1503 | |
c0ff4b85 R |
1504 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1505 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1506 | unlock: |
1507 | spin_unlock(&mc.lock); | |
4b534334 KH |
1508 | return ret; |
1509 | } | |
1510 | ||
c0ff4b85 | 1511 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1512 | { |
1513 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1514 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1515 | DEFINE_WAIT(wait); |
1516 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1517 | /* moving charge context might have finished. */ | |
1518 | if (mc.moving_task) | |
1519 | schedule(); | |
1520 | finish_wait(&mc.waitq, &wait); | |
1521 | return true; | |
1522 | } | |
1523 | } | |
1524 | return false; | |
1525 | } | |
1526 | ||
312734c0 KH |
1527 | /* |
1528 | * Take this lock when | |
1529 | * - a code tries to modify page's memcg while it's USED. | |
1530 | * - a code tries to modify page state accounting in a memcg. | |
312734c0 KH |
1531 | */ |
1532 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1533 | unsigned long *flags) | |
1534 | { | |
1535 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1536 | } | |
1537 | ||
1538 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1539 | unsigned long *flags) | |
1540 | { | |
1541 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1542 | } | |
1543 | ||
58cf188e | 1544 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1545 | /** |
58cf188e | 1546 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1547 | * @memcg: The memory cgroup that went over limit |
1548 | * @p: Task that is going to be killed | |
1549 | * | |
1550 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1551 | * enabled | |
1552 | */ | |
1553 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1554 | { | |
e61734c5 | 1555 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1556 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1557 | struct mem_cgroup *iter; |
1558 | unsigned int i; | |
e222432b | 1559 | |
58cf188e | 1560 | if (!p) |
e222432b BS |
1561 | return; |
1562 | ||
08088cb9 | 1563 | mutex_lock(&oom_info_lock); |
e222432b BS |
1564 | rcu_read_lock(); |
1565 | ||
e61734c5 TH |
1566 | pr_info("Task in "); |
1567 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1568 | pr_info(" killed as a result of limit of "); | |
1569 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1570 | pr_info("\n"); | |
e222432b | 1571 | |
e222432b BS |
1572 | rcu_read_unlock(); |
1573 | ||
3e32cb2e JW |
1574 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1575 | K((u64)page_counter_read(&memcg->memory)), | |
1576 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1577 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1578 | K((u64)page_counter_read(&memcg->memsw)), | |
1579 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1580 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1581 | K((u64)page_counter_read(&memcg->kmem)), | |
1582 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1583 | |
1584 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1585 | pr_info("Memory cgroup stats for "); |
1586 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1587 | pr_cont(":"); |
1588 | ||
1589 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1590 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1591 | continue; | |
1592 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1593 | K(mem_cgroup_read_stat(iter, i))); | |
1594 | } | |
1595 | ||
1596 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1597 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1598 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1599 | ||
1600 | pr_cont("\n"); | |
1601 | } | |
08088cb9 | 1602 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1603 | } |
1604 | ||
81d39c20 KH |
1605 | /* |
1606 | * This function returns the number of memcg under hierarchy tree. Returns | |
1607 | * 1(self count) if no children. | |
1608 | */ | |
c0ff4b85 | 1609 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1610 | { |
1611 | int num = 0; | |
7d74b06f KH |
1612 | struct mem_cgroup *iter; |
1613 | ||
c0ff4b85 | 1614 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1615 | num++; |
81d39c20 KH |
1616 | return num; |
1617 | } | |
1618 | ||
a63d83f4 DR |
1619 | /* |
1620 | * Return the memory (and swap, if configured) limit for a memcg. | |
1621 | */ | |
3e32cb2e | 1622 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1623 | { |
3e32cb2e | 1624 | unsigned long limit; |
a63d83f4 | 1625 | |
3e32cb2e | 1626 | limit = memcg->memory.limit; |
9a5a8f19 | 1627 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1628 | unsigned long memsw_limit; |
9a5a8f19 | 1629 | |
3e32cb2e JW |
1630 | memsw_limit = memcg->memsw.limit; |
1631 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1632 | } |
9a5a8f19 | 1633 | return limit; |
a63d83f4 DR |
1634 | } |
1635 | ||
19965460 DR |
1636 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1637 | int order) | |
9cbb78bb DR |
1638 | { |
1639 | struct mem_cgroup *iter; | |
1640 | unsigned long chosen_points = 0; | |
1641 | unsigned long totalpages; | |
1642 | unsigned int points = 0; | |
1643 | struct task_struct *chosen = NULL; | |
1644 | ||
876aafbf | 1645 | /* |
465adcf1 DR |
1646 | * If current has a pending SIGKILL or is exiting, then automatically |
1647 | * select it. The goal is to allow it to allocate so that it may | |
1648 | * quickly exit and free its memory. | |
876aafbf | 1649 | */ |
465adcf1 | 1650 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1651 | set_thread_flag(TIF_MEMDIE); |
1652 | return; | |
1653 | } | |
1654 | ||
1655 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1656 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1657 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1658 | struct css_task_iter it; |
9cbb78bb DR |
1659 | struct task_struct *task; |
1660 | ||
72ec7029 TH |
1661 | css_task_iter_start(&iter->css, &it); |
1662 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1663 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1664 | false)) { | |
1665 | case OOM_SCAN_SELECT: | |
1666 | if (chosen) | |
1667 | put_task_struct(chosen); | |
1668 | chosen = task; | |
1669 | chosen_points = ULONG_MAX; | |
1670 | get_task_struct(chosen); | |
1671 | /* fall through */ | |
1672 | case OOM_SCAN_CONTINUE: | |
1673 | continue; | |
1674 | case OOM_SCAN_ABORT: | |
72ec7029 | 1675 | css_task_iter_end(&it); |
9cbb78bb DR |
1676 | mem_cgroup_iter_break(memcg, iter); |
1677 | if (chosen) | |
1678 | put_task_struct(chosen); | |
1679 | return; | |
1680 | case OOM_SCAN_OK: | |
1681 | break; | |
1682 | }; | |
1683 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1684 | if (!points || points < chosen_points) |
1685 | continue; | |
1686 | /* Prefer thread group leaders for display purposes */ | |
1687 | if (points == chosen_points && | |
1688 | thread_group_leader(chosen)) | |
1689 | continue; | |
1690 | ||
1691 | if (chosen) | |
1692 | put_task_struct(chosen); | |
1693 | chosen = task; | |
1694 | chosen_points = points; | |
1695 | get_task_struct(chosen); | |
9cbb78bb | 1696 | } |
72ec7029 | 1697 | css_task_iter_end(&it); |
9cbb78bb DR |
1698 | } |
1699 | ||
1700 | if (!chosen) | |
1701 | return; | |
1702 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1703 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1704 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1705 | } |
1706 | ||
4d0c066d KH |
1707 | /** |
1708 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1709 | * @memcg: the target memcg |
4d0c066d KH |
1710 | * @nid: the node ID to be checked. |
1711 | * @noswap : specify true here if the user wants flle only information. | |
1712 | * | |
1713 | * This function returns whether the specified memcg contains any | |
1714 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1715 | * pages in the node. | |
1716 | */ | |
c0ff4b85 | 1717 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1718 | int nid, bool noswap) |
1719 | { | |
c0ff4b85 | 1720 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1721 | return true; |
1722 | if (noswap || !total_swap_pages) | |
1723 | return false; | |
c0ff4b85 | 1724 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1725 | return true; |
1726 | return false; | |
1727 | ||
1728 | } | |
bb4cc1a8 | 1729 | #if MAX_NUMNODES > 1 |
889976db YH |
1730 | |
1731 | /* | |
1732 | * Always updating the nodemask is not very good - even if we have an empty | |
1733 | * list or the wrong list here, we can start from some node and traverse all | |
1734 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1735 | * | |
1736 | */ | |
c0ff4b85 | 1737 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1738 | { |
1739 | int nid; | |
453a9bf3 KH |
1740 | /* |
1741 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1742 | * pagein/pageout changes since the last update. | |
1743 | */ | |
c0ff4b85 | 1744 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1745 | return; |
c0ff4b85 | 1746 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1747 | return; |
1748 | ||
889976db | 1749 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1750 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1751 | |
31aaea4a | 1752 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1753 | |
c0ff4b85 R |
1754 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1755 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1756 | } |
453a9bf3 | 1757 | |
c0ff4b85 R |
1758 | atomic_set(&memcg->numainfo_events, 0); |
1759 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1760 | } |
1761 | ||
1762 | /* | |
1763 | * Selecting a node where we start reclaim from. Because what we need is just | |
1764 | * reducing usage counter, start from anywhere is O,K. Considering | |
1765 | * memory reclaim from current node, there are pros. and cons. | |
1766 | * | |
1767 | * Freeing memory from current node means freeing memory from a node which | |
1768 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1769 | * hit limits, it will see a contention on a node. But freeing from remote | |
1770 | * node means more costs for memory reclaim because of memory latency. | |
1771 | * | |
1772 | * Now, we use round-robin. Better algorithm is welcomed. | |
1773 | */ | |
c0ff4b85 | 1774 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1775 | { |
1776 | int node; | |
1777 | ||
c0ff4b85 R |
1778 | mem_cgroup_may_update_nodemask(memcg); |
1779 | node = memcg->last_scanned_node; | |
889976db | 1780 | |
c0ff4b85 | 1781 | node = next_node(node, memcg->scan_nodes); |
889976db | 1782 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1783 | node = first_node(memcg->scan_nodes); |
889976db YH |
1784 | /* |
1785 | * We call this when we hit limit, not when pages are added to LRU. | |
1786 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1787 | * memcg is too small and all pages are not on LRU. In that case, | |
1788 | * we use curret node. | |
1789 | */ | |
1790 | if (unlikely(node == MAX_NUMNODES)) | |
1791 | node = numa_node_id(); | |
1792 | ||
c0ff4b85 | 1793 | memcg->last_scanned_node = node; |
889976db YH |
1794 | return node; |
1795 | } | |
1796 | ||
bb4cc1a8 AM |
1797 | /* |
1798 | * Check all nodes whether it contains reclaimable pages or not. | |
1799 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1800 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1801 | * enough new information. We need to do double check. | |
1802 | */ | |
1803 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) | |
1804 | { | |
1805 | int nid; | |
1806 | ||
1807 | /* | |
1808 | * quick check...making use of scan_node. | |
1809 | * We can skip unused nodes. | |
1810 | */ | |
1811 | if (!nodes_empty(memcg->scan_nodes)) { | |
1812 | for (nid = first_node(memcg->scan_nodes); | |
1813 | nid < MAX_NUMNODES; | |
1814 | nid = next_node(nid, memcg->scan_nodes)) { | |
1815 | ||
1816 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1817 | return true; | |
1818 | } | |
1819 | } | |
1820 | /* | |
1821 | * Check rest of nodes. | |
1822 | */ | |
1823 | for_each_node_state(nid, N_MEMORY) { | |
1824 | if (node_isset(nid, memcg->scan_nodes)) | |
1825 | continue; | |
1826 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) | |
1827 | return true; | |
1828 | } | |
1829 | return false; | |
1830 | } | |
1831 | ||
889976db | 1832 | #else |
c0ff4b85 | 1833 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1834 | { |
1835 | return 0; | |
1836 | } | |
4d0c066d | 1837 | |
bb4cc1a8 AM |
1838 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
1839 | { | |
1840 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); | |
1841 | } | |
889976db YH |
1842 | #endif |
1843 | ||
0608f43d AM |
1844 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1845 | struct zone *zone, | |
1846 | gfp_t gfp_mask, | |
1847 | unsigned long *total_scanned) | |
1848 | { | |
1849 | struct mem_cgroup *victim = NULL; | |
1850 | int total = 0; | |
1851 | int loop = 0; | |
1852 | unsigned long excess; | |
1853 | unsigned long nr_scanned; | |
1854 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1855 | .zone = zone, | |
1856 | .priority = 0, | |
1857 | }; | |
1858 | ||
3e32cb2e | 1859 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1860 | |
1861 | while (1) { | |
1862 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1863 | if (!victim) { | |
1864 | loop++; | |
1865 | if (loop >= 2) { | |
1866 | /* | |
1867 | * If we have not been able to reclaim | |
1868 | * anything, it might because there are | |
1869 | * no reclaimable pages under this hierarchy | |
1870 | */ | |
1871 | if (!total) | |
1872 | break; | |
1873 | /* | |
1874 | * We want to do more targeted reclaim. | |
1875 | * excess >> 2 is not to excessive so as to | |
1876 | * reclaim too much, nor too less that we keep | |
1877 | * coming back to reclaim from this cgroup | |
1878 | */ | |
1879 | if (total >= (excess >> 2) || | |
1880 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1881 | break; | |
1882 | } | |
1883 | continue; | |
1884 | } | |
1885 | if (!mem_cgroup_reclaimable(victim, false)) | |
1886 | continue; | |
1887 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, | |
1888 | zone, &nr_scanned); | |
1889 | *total_scanned += nr_scanned; | |
3e32cb2e | 1890 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1891 | break; |
6d61ef40 | 1892 | } |
0608f43d AM |
1893 | mem_cgroup_iter_break(root_memcg, victim); |
1894 | return total; | |
6d61ef40 BS |
1895 | } |
1896 | ||
0056f4e6 JW |
1897 | #ifdef CONFIG_LOCKDEP |
1898 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1899 | .name = "memcg_oom_lock", | |
1900 | }; | |
1901 | #endif | |
1902 | ||
fb2a6fc5 JW |
1903 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1904 | ||
867578cb KH |
1905 | /* |
1906 | * Check OOM-Killer is already running under our hierarchy. | |
1907 | * If someone is running, return false. | |
1908 | */ | |
fb2a6fc5 | 1909 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1910 | { |
79dfdacc | 1911 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1912 | |
fb2a6fc5 JW |
1913 | spin_lock(&memcg_oom_lock); |
1914 | ||
9f3a0d09 | 1915 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1916 | if (iter->oom_lock) { |
79dfdacc MH |
1917 | /* |
1918 | * this subtree of our hierarchy is already locked | |
1919 | * so we cannot give a lock. | |
1920 | */ | |
79dfdacc | 1921 | failed = iter; |
9f3a0d09 JW |
1922 | mem_cgroup_iter_break(memcg, iter); |
1923 | break; | |
23751be0 JW |
1924 | } else |
1925 | iter->oom_lock = true; | |
7d74b06f | 1926 | } |
867578cb | 1927 | |
fb2a6fc5 JW |
1928 | if (failed) { |
1929 | /* | |
1930 | * OK, we failed to lock the whole subtree so we have | |
1931 | * to clean up what we set up to the failing subtree | |
1932 | */ | |
1933 | for_each_mem_cgroup_tree(iter, memcg) { | |
1934 | if (iter == failed) { | |
1935 | mem_cgroup_iter_break(memcg, iter); | |
1936 | break; | |
1937 | } | |
1938 | iter->oom_lock = false; | |
79dfdacc | 1939 | } |
0056f4e6 JW |
1940 | } else |
1941 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1942 | |
1943 | spin_unlock(&memcg_oom_lock); | |
1944 | ||
1945 | return !failed; | |
a636b327 | 1946 | } |
0b7f569e | 1947 | |
fb2a6fc5 | 1948 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1949 | { |
7d74b06f KH |
1950 | struct mem_cgroup *iter; |
1951 | ||
fb2a6fc5 | 1952 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1953 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1954 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1955 | iter->oom_lock = false; |
fb2a6fc5 | 1956 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1957 | } |
1958 | ||
c0ff4b85 | 1959 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1960 | { |
1961 | struct mem_cgroup *iter; | |
1962 | ||
c0ff4b85 | 1963 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1964 | atomic_inc(&iter->under_oom); |
1965 | } | |
1966 | ||
c0ff4b85 | 1967 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1968 | { |
1969 | struct mem_cgroup *iter; | |
1970 | ||
867578cb KH |
1971 | /* |
1972 | * When a new child is created while the hierarchy is under oom, | |
1973 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1974 | * atomic_add_unless() here. | |
1975 | */ | |
c0ff4b85 | 1976 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1977 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1978 | } |
1979 | ||
867578cb KH |
1980 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1981 | ||
dc98df5a | 1982 | struct oom_wait_info { |
d79154bb | 1983 | struct mem_cgroup *memcg; |
dc98df5a KH |
1984 | wait_queue_t wait; |
1985 | }; | |
1986 | ||
1987 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1988 | unsigned mode, int sync, void *arg) | |
1989 | { | |
d79154bb HD |
1990 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1991 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1992 | struct oom_wait_info *oom_wait_info; |
1993 | ||
1994 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1995 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1996 | |
dc98df5a | 1997 | /* |
d79154bb | 1998 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1999 | * Then we can use css_is_ancestor without taking care of RCU. |
2000 | */ | |
c0ff4b85 R |
2001 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2002 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2003 | return 0; |
dc98df5a KH |
2004 | return autoremove_wake_function(wait, mode, sync, arg); |
2005 | } | |
2006 | ||
c0ff4b85 | 2007 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2008 | { |
3812c8c8 | 2009 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
2010 | /* for filtering, pass "memcg" as argument. */ |
2011 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2012 | } |
2013 | ||
c0ff4b85 | 2014 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2015 | { |
c0ff4b85 R |
2016 | if (memcg && atomic_read(&memcg->under_oom)) |
2017 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2018 | } |
2019 | ||
3812c8c8 | 2020 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 2021 | { |
3812c8c8 JW |
2022 | if (!current->memcg_oom.may_oom) |
2023 | return; | |
867578cb | 2024 | /* |
49426420 JW |
2025 | * We are in the middle of the charge context here, so we |
2026 | * don't want to block when potentially sitting on a callstack | |
2027 | * that holds all kinds of filesystem and mm locks. | |
2028 | * | |
2029 | * Also, the caller may handle a failed allocation gracefully | |
2030 | * (like optional page cache readahead) and so an OOM killer | |
2031 | * invocation might not even be necessary. | |
2032 | * | |
2033 | * That's why we don't do anything here except remember the | |
2034 | * OOM context and then deal with it at the end of the page | |
2035 | * fault when the stack is unwound, the locks are released, | |
2036 | * and when we know whether the fault was overall successful. | |
867578cb | 2037 | */ |
49426420 JW |
2038 | css_get(&memcg->css); |
2039 | current->memcg_oom.memcg = memcg; | |
2040 | current->memcg_oom.gfp_mask = mask; | |
2041 | current->memcg_oom.order = order; | |
3812c8c8 JW |
2042 | } |
2043 | ||
2044 | /** | |
2045 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 2046 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 2047 | * |
49426420 JW |
2048 | * This has to be called at the end of a page fault if the memcg OOM |
2049 | * handler was enabled. | |
3812c8c8 | 2050 | * |
49426420 | 2051 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
2052 | * sleep on a waitqueue until the userspace task resolves the |
2053 | * situation. Sleeping directly in the charge context with all kinds | |
2054 | * of locks held is not a good idea, instead we remember an OOM state | |
2055 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 2056 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
2057 | * |
2058 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 2059 | * completed, %false otherwise. |
3812c8c8 | 2060 | */ |
49426420 | 2061 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 2062 | { |
49426420 | 2063 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 2064 | struct oom_wait_info owait; |
49426420 | 2065 | bool locked; |
3812c8c8 JW |
2066 | |
2067 | /* OOM is global, do not handle */ | |
3812c8c8 | 2068 | if (!memcg) |
49426420 | 2069 | return false; |
3812c8c8 | 2070 | |
49426420 JW |
2071 | if (!handle) |
2072 | goto cleanup; | |
3812c8c8 JW |
2073 | |
2074 | owait.memcg = memcg; | |
2075 | owait.wait.flags = 0; | |
2076 | owait.wait.func = memcg_oom_wake_function; | |
2077 | owait.wait.private = current; | |
2078 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 2079 | |
3812c8c8 | 2080 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
2081 | mem_cgroup_mark_under_oom(memcg); |
2082 | ||
2083 | locked = mem_cgroup_oom_trylock(memcg); | |
2084 | ||
2085 | if (locked) | |
2086 | mem_cgroup_oom_notify(memcg); | |
2087 | ||
2088 | if (locked && !memcg->oom_kill_disable) { | |
2089 | mem_cgroup_unmark_under_oom(memcg); | |
2090 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2091 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
2092 | current->memcg_oom.order); | |
2093 | } else { | |
3812c8c8 | 2094 | schedule(); |
49426420 JW |
2095 | mem_cgroup_unmark_under_oom(memcg); |
2096 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2097 | } | |
2098 | ||
2099 | if (locked) { | |
fb2a6fc5 JW |
2100 | mem_cgroup_oom_unlock(memcg); |
2101 | /* | |
2102 | * There is no guarantee that an OOM-lock contender | |
2103 | * sees the wakeups triggered by the OOM kill | |
2104 | * uncharges. Wake any sleepers explicitely. | |
2105 | */ | |
2106 | memcg_oom_recover(memcg); | |
2107 | } | |
49426420 JW |
2108 | cleanup: |
2109 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2110 | css_put(&memcg->css); |
867578cb | 2111 | return true; |
0b7f569e KH |
2112 | } |
2113 | ||
d7365e78 JW |
2114 | /** |
2115 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
2116 | * @page: page that is going to change accounted state | |
2117 | * @locked: &memcg->move_lock slowpath was taken | |
2118 | * @flags: IRQ-state flags for &memcg->move_lock | |
32047e2a | 2119 | * |
d7365e78 JW |
2120 | * This function must mark the beginning of an accounted page state |
2121 | * change to prevent double accounting when the page is concurrently | |
2122 | * being moved to another memcg: | |
32047e2a | 2123 | * |
d7365e78 JW |
2124 | * memcg = mem_cgroup_begin_page_stat(page, &locked, &flags); |
2125 | * if (TestClearPageState(page)) | |
2126 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
2127 | * mem_cgroup_end_page_stat(memcg, locked, flags); | |
32047e2a | 2128 | * |
d7365e78 JW |
2129 | * The RCU lock is held throughout the transaction. The fast path can |
2130 | * get away without acquiring the memcg->move_lock (@locked is false) | |
2131 | * because page moving starts with an RCU grace period. | |
32047e2a | 2132 | * |
d7365e78 JW |
2133 | * The RCU lock also protects the memcg from being freed when the page |
2134 | * state that is going to change is the only thing preventing the page | |
2135 | * from being uncharged. E.g. end-writeback clearing PageWriteback(), | |
2136 | * which allows migration to go ahead and uncharge the page before the | |
2137 | * account transaction might be complete. | |
d69b042f | 2138 | */ |
d7365e78 JW |
2139 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page, |
2140 | bool *locked, | |
2141 | unsigned long *flags) | |
89c06bd5 KH |
2142 | { |
2143 | struct mem_cgroup *memcg; | |
2144 | struct page_cgroup *pc; | |
2145 | ||
d7365e78 JW |
2146 | rcu_read_lock(); |
2147 | ||
2148 | if (mem_cgroup_disabled()) | |
2149 | return NULL; | |
2150 | ||
89c06bd5 KH |
2151 | pc = lookup_page_cgroup(page); |
2152 | again: | |
2153 | memcg = pc->mem_cgroup; | |
2154 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
d7365e78 JW |
2155 | return NULL; |
2156 | ||
2157 | *locked = false; | |
bdcbb659 | 2158 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2159 | return memcg; |
89c06bd5 KH |
2160 | |
2161 | move_lock_mem_cgroup(memcg, flags); | |
2162 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2163 | move_unlock_mem_cgroup(memcg, flags); | |
2164 | goto again; | |
2165 | } | |
2166 | *locked = true; | |
d7365e78 JW |
2167 | |
2168 | return memcg; | |
89c06bd5 KH |
2169 | } |
2170 | ||
d7365e78 JW |
2171 | /** |
2172 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2173 | * @memcg: the memcg that was accounted against | |
2174 | * @locked: value received from mem_cgroup_begin_page_stat() | |
2175 | * @flags: value received from mem_cgroup_begin_page_stat() | |
2176 | */ | |
2177 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg, bool locked, | |
2178 | unsigned long flags) | |
89c06bd5 | 2179 | { |
d7365e78 JW |
2180 | if (memcg && locked) |
2181 | move_unlock_mem_cgroup(memcg, &flags); | |
89c06bd5 | 2182 | |
d7365e78 | 2183 | rcu_read_unlock(); |
89c06bd5 KH |
2184 | } |
2185 | ||
d7365e78 JW |
2186 | /** |
2187 | * mem_cgroup_update_page_stat - update page state statistics | |
2188 | * @memcg: memcg to account against | |
2189 | * @idx: page state item to account | |
2190 | * @val: number of pages (positive or negative) | |
2191 | * | |
2192 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2193 | */ | |
2194 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2195 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2196 | { |
658b72c5 | 2197 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2198 | |
d7365e78 JW |
2199 | if (memcg) |
2200 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2201 | } |
26174efd | 2202 | |
cdec2e42 KH |
2203 | /* |
2204 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2205 | * TODO: maybe necessary to use big numbers in big irons. | |
2206 | */ | |
7ec99d62 | 2207 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2208 | struct memcg_stock_pcp { |
2209 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2210 | unsigned int nr_pages; |
cdec2e42 | 2211 | struct work_struct work; |
26fe6168 | 2212 | unsigned long flags; |
a0db00fc | 2213 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2214 | }; |
2215 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2216 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2217 | |
a0956d54 SS |
2218 | /** |
2219 | * consume_stock: Try to consume stocked charge on this cpu. | |
2220 | * @memcg: memcg to consume from. | |
2221 | * @nr_pages: how many pages to charge. | |
2222 | * | |
2223 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2224 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2225 | * service an allocation will refill the stock. | |
2226 | * | |
2227 | * returns true if successful, false otherwise. | |
cdec2e42 | 2228 | */ |
a0956d54 | 2229 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2230 | { |
2231 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2232 | bool ret = false; |
cdec2e42 | 2233 | |
a0956d54 | 2234 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2235 | return ret; |
a0956d54 | 2236 | |
cdec2e42 | 2237 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2238 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2239 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2240 | ret = true; |
2241 | } | |
cdec2e42 KH |
2242 | put_cpu_var(memcg_stock); |
2243 | return ret; | |
2244 | } | |
2245 | ||
2246 | /* | |
3e32cb2e | 2247 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2248 | */ |
2249 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2250 | { | |
2251 | struct mem_cgroup *old = stock->cached; | |
2252 | ||
11c9ea4e | 2253 | if (stock->nr_pages) { |
3e32cb2e | 2254 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2255 | if (do_swap_account) |
3e32cb2e | 2256 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2257 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2258 | stock->nr_pages = 0; |
cdec2e42 KH |
2259 | } |
2260 | stock->cached = NULL; | |
cdec2e42 KH |
2261 | } |
2262 | ||
2263 | /* | |
2264 | * This must be called under preempt disabled or must be called by | |
2265 | * a thread which is pinned to local cpu. | |
2266 | */ | |
2267 | static void drain_local_stock(struct work_struct *dummy) | |
2268 | { | |
7c8e0181 | 2269 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2270 | drain_stock(stock); |
26fe6168 | 2271 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2272 | } |
2273 | ||
e4777496 MH |
2274 | static void __init memcg_stock_init(void) |
2275 | { | |
2276 | int cpu; | |
2277 | ||
2278 | for_each_possible_cpu(cpu) { | |
2279 | struct memcg_stock_pcp *stock = | |
2280 | &per_cpu(memcg_stock, cpu); | |
2281 | INIT_WORK(&stock->work, drain_local_stock); | |
2282 | } | |
2283 | } | |
2284 | ||
cdec2e42 | 2285 | /* |
3e32cb2e | 2286 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2287 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2288 | */ |
c0ff4b85 | 2289 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2290 | { |
2291 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2292 | ||
c0ff4b85 | 2293 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2294 | drain_stock(stock); |
c0ff4b85 | 2295 | stock->cached = memcg; |
cdec2e42 | 2296 | } |
11c9ea4e | 2297 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2298 | put_cpu_var(memcg_stock); |
2299 | } | |
2300 | ||
2301 | /* | |
c0ff4b85 | 2302 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2303 | * of the hierarchy under it. |
cdec2e42 | 2304 | */ |
6d3d6aa2 | 2305 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2306 | { |
26fe6168 | 2307 | int cpu, curcpu; |
d38144b7 | 2308 | |
6d3d6aa2 JW |
2309 | /* If someone's already draining, avoid adding running more workers. */ |
2310 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2311 | return; | |
cdec2e42 | 2312 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2313 | get_online_cpus(); |
5af12d0e | 2314 | curcpu = get_cpu(); |
cdec2e42 KH |
2315 | for_each_online_cpu(cpu) { |
2316 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2317 | struct mem_cgroup *memcg; |
26fe6168 | 2318 | |
c0ff4b85 R |
2319 | memcg = stock->cached; |
2320 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2321 | continue; |
c0ff4b85 | 2322 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2323 | continue; |
d1a05b69 MH |
2324 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2325 | if (cpu == curcpu) | |
2326 | drain_local_stock(&stock->work); | |
2327 | else | |
2328 | schedule_work_on(cpu, &stock->work); | |
2329 | } | |
cdec2e42 | 2330 | } |
5af12d0e | 2331 | put_cpu(); |
f894ffa8 | 2332 | put_online_cpus(); |
9f50fad6 | 2333 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2334 | } |
2335 | ||
711d3d2c KH |
2336 | /* |
2337 | * This function drains percpu counter value from DEAD cpu and | |
2338 | * move it to local cpu. Note that this function can be preempted. | |
2339 | */ | |
c0ff4b85 | 2340 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2341 | { |
2342 | int i; | |
2343 | ||
c0ff4b85 | 2344 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2345 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2346 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2347 | |
c0ff4b85 R |
2348 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2349 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2350 | } |
e9f8974f | 2351 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2352 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2353 | |
c0ff4b85 R |
2354 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2355 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2356 | } |
c0ff4b85 | 2357 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2358 | } |
2359 | ||
0db0628d | 2360 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2361 | unsigned long action, |
2362 | void *hcpu) | |
2363 | { | |
2364 | int cpu = (unsigned long)hcpu; | |
2365 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2366 | struct mem_cgroup *iter; |
cdec2e42 | 2367 | |
619d094b | 2368 | if (action == CPU_ONLINE) |
1489ebad | 2369 | return NOTIFY_OK; |
1489ebad | 2370 | |
d833049b | 2371 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2372 | return NOTIFY_OK; |
711d3d2c | 2373 | |
9f3a0d09 | 2374 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2375 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2376 | ||
cdec2e42 KH |
2377 | stock = &per_cpu(memcg_stock, cpu); |
2378 | drain_stock(stock); | |
2379 | return NOTIFY_OK; | |
2380 | } | |
2381 | ||
00501b53 JW |
2382 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2383 | unsigned int nr_pages) | |
8a9f3ccd | 2384 | { |
7ec99d62 | 2385 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2386 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2387 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2388 | struct page_counter *counter; |
6539cc05 | 2389 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2390 | bool may_swap = true; |
2391 | bool drained = false; | |
05b84301 | 2392 | int ret = 0; |
a636b327 | 2393 | |
ce00a967 JW |
2394 | if (mem_cgroup_is_root(memcg)) |
2395 | goto done; | |
6539cc05 | 2396 | retry: |
b6b6cc72 MH |
2397 | if (consume_stock(memcg, nr_pages)) |
2398 | goto done; | |
8a9f3ccd | 2399 | |
3fbe7244 | 2400 | if (!do_swap_account || |
3e32cb2e JW |
2401 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2402 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2403 | goto done_restock; |
3fbe7244 | 2404 | if (do_swap_account) |
3e32cb2e JW |
2405 | page_counter_uncharge(&memcg->memsw, batch); |
2406 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2407 | } else { |
3e32cb2e | 2408 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2409 | may_swap = false; |
3fbe7244 | 2410 | } |
7a81b88c | 2411 | |
6539cc05 JW |
2412 | if (batch > nr_pages) { |
2413 | batch = nr_pages; | |
2414 | goto retry; | |
2415 | } | |
6d61ef40 | 2416 | |
06b078fc JW |
2417 | /* |
2418 | * Unlike in global OOM situations, memcg is not in a physical | |
2419 | * memory shortage. Allow dying and OOM-killed tasks to | |
2420 | * bypass the last charges so that they can exit quickly and | |
2421 | * free their memory. | |
2422 | */ | |
2423 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2424 | fatal_signal_pending(current) || | |
2425 | current->flags & PF_EXITING)) | |
2426 | goto bypass; | |
2427 | ||
2428 | if (unlikely(task_in_memcg_oom(current))) | |
2429 | goto nomem; | |
2430 | ||
6539cc05 JW |
2431 | if (!(gfp_mask & __GFP_WAIT)) |
2432 | goto nomem; | |
4b534334 | 2433 | |
b70a2a21 JW |
2434 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2435 | gfp_mask, may_swap); | |
6539cc05 | 2436 | |
61e02c74 | 2437 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2438 | goto retry; |
28c34c29 | 2439 | |
b70a2a21 | 2440 | if (!drained) { |
6d3d6aa2 | 2441 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2442 | drained = true; |
2443 | goto retry; | |
2444 | } | |
2445 | ||
28c34c29 JW |
2446 | if (gfp_mask & __GFP_NORETRY) |
2447 | goto nomem; | |
6539cc05 JW |
2448 | /* |
2449 | * Even though the limit is exceeded at this point, reclaim | |
2450 | * may have been able to free some pages. Retry the charge | |
2451 | * before killing the task. | |
2452 | * | |
2453 | * Only for regular pages, though: huge pages are rather | |
2454 | * unlikely to succeed so close to the limit, and we fall back | |
2455 | * to regular pages anyway in case of failure. | |
2456 | */ | |
61e02c74 | 2457 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2458 | goto retry; |
2459 | /* | |
2460 | * At task move, charge accounts can be doubly counted. So, it's | |
2461 | * better to wait until the end of task_move if something is going on. | |
2462 | */ | |
2463 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2464 | goto retry; | |
2465 | ||
9b130619 JW |
2466 | if (nr_retries--) |
2467 | goto retry; | |
2468 | ||
06b078fc JW |
2469 | if (gfp_mask & __GFP_NOFAIL) |
2470 | goto bypass; | |
2471 | ||
6539cc05 JW |
2472 | if (fatal_signal_pending(current)) |
2473 | goto bypass; | |
2474 | ||
61e02c74 | 2475 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2476 | nomem: |
6d1fdc48 | 2477 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2478 | return -ENOMEM; |
867578cb | 2479 | bypass: |
ce00a967 | 2480 | return -EINTR; |
6539cc05 JW |
2481 | |
2482 | done_restock: | |
e8ea14cc | 2483 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2484 | if (batch > nr_pages) |
2485 | refill_stock(memcg, batch - nr_pages); | |
2486 | done: | |
05b84301 | 2487 | return ret; |
7a81b88c | 2488 | } |
8a9f3ccd | 2489 | |
00501b53 | 2490 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2491 | { |
ce00a967 JW |
2492 | if (mem_cgroup_is_root(memcg)) |
2493 | return; | |
2494 | ||
3e32cb2e | 2495 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2496 | if (do_swap_account) |
3e32cb2e | 2497 | page_counter_uncharge(&memcg->memsw, nr_pages); |
e8ea14cc JW |
2498 | |
2499 | css_put_many(&memcg->css, nr_pages); | |
d01dd17f KH |
2500 | } |
2501 | ||
a3b2d692 KH |
2502 | /* |
2503 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2504 | * rcu_read_lock(). The caller is responsible for calling |
2505 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2506 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2507 | */ |
2508 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2509 | { | |
a3b2d692 KH |
2510 | /* ID 0 is unused ID */ |
2511 | if (!id) | |
2512 | return NULL; | |
34c00c31 | 2513 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2514 | } |
2515 | ||
0a31bc97 JW |
2516 | /* |
2517 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2518 | * @page: the page | |
2519 | * | |
2520 | * Look up, get a css reference, and return the memcg that owns @page. | |
2521 | * | |
2522 | * The page must be locked to prevent racing with swap-in and page | |
2523 | * cache charges. If coming from an unlocked page table, the caller | |
2524 | * must ensure the page is on the LRU or this can race with charging. | |
2525 | */ | |
e42d9d5d | 2526 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2527 | { |
c0ff4b85 | 2528 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2529 | struct page_cgroup *pc; |
a3b2d692 | 2530 | unsigned short id; |
b5a84319 KH |
2531 | swp_entry_t ent; |
2532 | ||
309381fe | 2533 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2534 | |
3c776e64 | 2535 | pc = lookup_page_cgroup(page); |
a3b2d692 | 2536 | if (PageCgroupUsed(pc)) { |
c0ff4b85 | 2537 | memcg = pc->mem_cgroup; |
ec903c0c | 2538 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2539 | memcg = NULL; |
e42d9d5d | 2540 | } else if (PageSwapCache(page)) { |
3c776e64 | 2541 | ent.val = page_private(page); |
9fb4b7cc | 2542 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2543 | rcu_read_lock(); |
c0ff4b85 | 2544 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2545 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2546 | memcg = NULL; |
a3b2d692 | 2547 | rcu_read_unlock(); |
3c776e64 | 2548 | } |
c0ff4b85 | 2549 | return memcg; |
b5a84319 KH |
2550 | } |
2551 | ||
0a31bc97 JW |
2552 | static void lock_page_lru(struct page *page, int *isolated) |
2553 | { | |
2554 | struct zone *zone = page_zone(page); | |
2555 | ||
2556 | spin_lock_irq(&zone->lru_lock); | |
2557 | if (PageLRU(page)) { | |
2558 | struct lruvec *lruvec; | |
2559 | ||
2560 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2561 | ClearPageLRU(page); | |
2562 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2563 | *isolated = 1; | |
2564 | } else | |
2565 | *isolated = 0; | |
2566 | } | |
2567 | ||
2568 | static void unlock_page_lru(struct page *page, int isolated) | |
2569 | { | |
2570 | struct zone *zone = page_zone(page); | |
2571 | ||
2572 | if (isolated) { | |
2573 | struct lruvec *lruvec; | |
2574 | ||
2575 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2576 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2577 | SetPageLRU(page); | |
2578 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2579 | } | |
2580 | spin_unlock_irq(&zone->lru_lock); | |
2581 | } | |
2582 | ||
00501b53 | 2583 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2584 | bool lrucare) |
7a81b88c | 2585 | { |
ce587e65 | 2586 | struct page_cgroup *pc = lookup_page_cgroup(page); |
0a31bc97 | 2587 | int isolated; |
9ce70c02 | 2588 | |
309381fe | 2589 | VM_BUG_ON_PAGE(PageCgroupUsed(pc), page); |
ca3e0214 KH |
2590 | /* |
2591 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2592 | * accessed by any other context at this point. | |
2593 | */ | |
9ce70c02 HD |
2594 | |
2595 | /* | |
2596 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2597 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2598 | */ | |
0a31bc97 JW |
2599 | if (lrucare) |
2600 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2601 | |
0a31bc97 JW |
2602 | /* |
2603 | * Nobody should be changing or seriously looking at | |
2604 | * pc->mem_cgroup and pc->flags at this point: | |
2605 | * | |
2606 | * - the page is uncharged | |
2607 | * | |
2608 | * - the page is off-LRU | |
2609 | * | |
2610 | * - an anonymous fault has exclusive page access, except for | |
2611 | * a locked page table | |
2612 | * | |
2613 | * - a page cache insertion, a swapin fault, or a migration | |
2614 | * have the page locked | |
2615 | */ | |
c0ff4b85 | 2616 | pc->mem_cgroup = memcg; |
0a31bc97 | 2617 | pc->flags = PCG_USED | PCG_MEM | (do_swap_account ? PCG_MEMSW : 0); |
9ce70c02 | 2618 | |
0a31bc97 JW |
2619 | if (lrucare) |
2620 | unlock_page_lru(page, isolated); | |
7a81b88c | 2621 | } |
66e1707b | 2622 | |
7ae1e1d0 | 2623 | #ifdef CONFIG_MEMCG_KMEM |
bd673145 VD |
2624 | /* |
2625 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or | |
2626 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. | |
2627 | */ | |
2628 | static DEFINE_MUTEX(memcg_slab_mutex); | |
2629 | ||
1f458cbf GC |
2630 | /* |
2631 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2632 | * in the memcg_cache_params struct. | |
2633 | */ | |
2634 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2635 | { | |
2636 | struct kmem_cache *cachep; | |
2637 | ||
2638 | VM_BUG_ON(p->is_root_cache); | |
2639 | cachep = p->root_cache; | |
7a67d7ab | 2640 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2641 | } |
2642 | ||
749c5415 | 2643 | #ifdef CONFIG_SLABINFO |
2da8ca82 | 2644 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
749c5415 | 2645 | { |
2da8ca82 | 2646 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
749c5415 GC |
2647 | struct memcg_cache_params *params; |
2648 | ||
cf2b8fbf | 2649 | if (!memcg_kmem_is_active(memcg)) |
749c5415 GC |
2650 | return -EIO; |
2651 | ||
2652 | print_slabinfo_header(m); | |
2653 | ||
bd673145 | 2654 | mutex_lock(&memcg_slab_mutex); |
749c5415 GC |
2655 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) |
2656 | cache_show(memcg_params_to_cache(params), m); | |
bd673145 | 2657 | mutex_unlock(&memcg_slab_mutex); |
749c5415 GC |
2658 | |
2659 | return 0; | |
2660 | } | |
2661 | #endif | |
2662 | ||
3e32cb2e JW |
2663 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2664 | unsigned long nr_pages) | |
7ae1e1d0 | 2665 | { |
3e32cb2e | 2666 | struct page_counter *counter; |
7ae1e1d0 | 2667 | int ret = 0; |
7ae1e1d0 | 2668 | |
3e32cb2e JW |
2669 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2670 | if (ret < 0) | |
7ae1e1d0 GC |
2671 | return ret; |
2672 | ||
3e32cb2e | 2673 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2674 | if (ret == -EINTR) { |
2675 | /* | |
00501b53 JW |
2676 | * try_charge() chose to bypass to root due to OOM kill or |
2677 | * fatal signal. Since our only options are to either fail | |
2678 | * the allocation or charge it to this cgroup, do it as a | |
2679 | * temporary condition. But we can't fail. From a kmem/slab | |
2680 | * perspective, the cache has already been selected, by | |
2681 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2682 | * our minds. |
2683 | * | |
2684 | * This condition will only trigger if the task entered | |
00501b53 JW |
2685 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2686 | * during try_charge() above. Tasks that were already dying | |
2687 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2688 | * directed to the root cgroup in memcontrol.h |
2689 | */ | |
3e32cb2e | 2690 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2691 | if (do_swap_account) |
3e32cb2e | 2692 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2693 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2694 | ret = 0; |
2695 | } else if (ret) | |
3e32cb2e | 2696 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2697 | |
2698 | return ret; | |
2699 | } | |
2700 | ||
3e32cb2e JW |
2701 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, |
2702 | unsigned long nr_pages) | |
7ae1e1d0 | 2703 | { |
3e32cb2e | 2704 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2705 | if (do_swap_account) |
3e32cb2e | 2706 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2707 | |
64f21993 | 2708 | page_counter_uncharge(&memcg->kmem, nr_pages); |
e8ea14cc JW |
2709 | |
2710 | css_put_many(&memcg->css, nr_pages); | |
7ae1e1d0 GC |
2711 | } |
2712 | ||
2633d7a0 GC |
2713 | /* |
2714 | * helper for acessing a memcg's index. It will be used as an index in the | |
2715 | * child cache array in kmem_cache, and also to derive its name. This function | |
2716 | * will return -1 when this is not a kmem-limited memcg. | |
2717 | */ | |
2718 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2719 | { | |
2720 | return memcg ? memcg->kmemcg_id : -1; | |
2721 | } | |
2722 | ||
f3bb3043 | 2723 | static int memcg_alloc_cache_id(void) |
55007d84 | 2724 | { |
f3bb3043 VD |
2725 | int id, size; |
2726 | int err; | |
2727 | ||
2728 | id = ida_simple_get(&kmem_limited_groups, | |
2729 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2730 | if (id < 0) | |
2731 | return id; | |
55007d84 | 2732 | |
f3bb3043 VD |
2733 | if (id < memcg_limited_groups_array_size) |
2734 | return id; | |
2735 | ||
2736 | /* | |
2737 | * There's no space for the new id in memcg_caches arrays, | |
2738 | * so we have to grow them. | |
2739 | */ | |
2740 | ||
2741 | size = 2 * (id + 1); | |
55007d84 GC |
2742 | if (size < MEMCG_CACHES_MIN_SIZE) |
2743 | size = MEMCG_CACHES_MIN_SIZE; | |
2744 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2745 | size = MEMCG_CACHES_MAX_SIZE; | |
2746 | ||
f3bb3043 VD |
2747 | mutex_lock(&memcg_slab_mutex); |
2748 | err = memcg_update_all_caches(size); | |
2749 | mutex_unlock(&memcg_slab_mutex); | |
2750 | ||
2751 | if (err) { | |
2752 | ida_simple_remove(&kmem_limited_groups, id); | |
2753 | return err; | |
2754 | } | |
2755 | return id; | |
2756 | } | |
2757 | ||
2758 | static void memcg_free_cache_id(int id) | |
2759 | { | |
2760 | ida_simple_remove(&kmem_limited_groups, id); | |
55007d84 GC |
2761 | } |
2762 | ||
2763 | /* | |
2764 | * We should update the current array size iff all caches updates succeed. This | |
2765 | * can only be done from the slab side. The slab mutex needs to be held when | |
2766 | * calling this. | |
2767 | */ | |
2768 | void memcg_update_array_size(int num) | |
2769 | { | |
f3bb3043 | 2770 | memcg_limited_groups_array_size = num; |
55007d84 GC |
2771 | } |
2772 | ||
776ed0f0 VD |
2773 | static void memcg_register_cache(struct mem_cgroup *memcg, |
2774 | struct kmem_cache *root_cache) | |
2633d7a0 | 2775 | { |
93f39eea VD |
2776 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by |
2777 | memcg_slab_mutex */ | |
bd673145 | 2778 | struct kmem_cache *cachep; |
d7f25f8a GC |
2779 | int id; |
2780 | ||
bd673145 VD |
2781 | lockdep_assert_held(&memcg_slab_mutex); |
2782 | ||
2783 | id = memcg_cache_id(memcg); | |
2784 | ||
2785 | /* | |
2786 | * Since per-memcg caches are created asynchronously on first | |
2787 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
2788 | * create the same cache, but only one of them may succeed. | |
2789 | */ | |
2790 | if (cache_from_memcg_idx(root_cache, id)) | |
1aa13254 VD |
2791 | return; |
2792 | ||
073ee1c6 | 2793 | cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1); |
776ed0f0 | 2794 | cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf); |
2edefe11 | 2795 | /* |
bd673145 VD |
2796 | * If we could not create a memcg cache, do not complain, because |
2797 | * that's not critical at all as we can always proceed with the root | |
2798 | * cache. | |
2edefe11 | 2799 | */ |
bd673145 VD |
2800 | if (!cachep) |
2801 | return; | |
2edefe11 | 2802 | |
33a690c4 | 2803 | css_get(&memcg->css); |
bd673145 | 2804 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
1aa13254 | 2805 | |
d7f25f8a | 2806 | /* |
959c8963 VD |
2807 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
2808 | * barrier here to ensure nobody will see the kmem_cache partially | |
2809 | * initialized. | |
d7f25f8a | 2810 | */ |
959c8963 VD |
2811 | smp_wmb(); |
2812 | ||
bd673145 VD |
2813 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
2814 | root_cache->memcg_params->memcg_caches[id] = cachep; | |
1aa13254 | 2815 | } |
d7f25f8a | 2816 | |
776ed0f0 | 2817 | static void memcg_unregister_cache(struct kmem_cache *cachep) |
1aa13254 | 2818 | { |
bd673145 | 2819 | struct kmem_cache *root_cache; |
1aa13254 VD |
2820 | struct mem_cgroup *memcg; |
2821 | int id; | |
2822 | ||
bd673145 | 2823 | lockdep_assert_held(&memcg_slab_mutex); |
d7f25f8a | 2824 | |
bd673145 | 2825 | BUG_ON(is_root_cache(cachep)); |
2edefe11 | 2826 | |
bd673145 VD |
2827 | root_cache = cachep->memcg_params->root_cache; |
2828 | memcg = cachep->memcg_params->memcg; | |
96403da2 | 2829 | id = memcg_cache_id(memcg); |
d7f25f8a | 2830 | |
bd673145 VD |
2831 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
2832 | root_cache->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a | 2833 | |
bd673145 VD |
2834 | list_del(&cachep->memcg_params->list); |
2835 | ||
2836 | kmem_cache_destroy(cachep); | |
33a690c4 VD |
2837 | |
2838 | /* drop the reference taken in memcg_register_cache */ | |
2839 | css_put(&memcg->css); | |
2633d7a0 GC |
2840 | } |
2841 | ||
0e9d92f2 GC |
2842 | /* |
2843 | * During the creation a new cache, we need to disable our accounting mechanism | |
2844 | * altogether. This is true even if we are not creating, but rather just | |
2845 | * enqueing new caches to be created. | |
2846 | * | |
2847 | * This is because that process will trigger allocations; some visible, like | |
2848 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
2849 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
2850 | * objects during debug. | |
2851 | * | |
2852 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
2853 | * to it. This may not be a bounded recursion: since the first cache creation | |
2854 | * failed to complete (waiting on the allocation), we'll just try to create the | |
2855 | * cache again, failing at the same point. | |
2856 | * | |
2857 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
2858 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
2859 | * inside the following two functions. | |
2860 | */ | |
2861 | static inline void memcg_stop_kmem_account(void) | |
2862 | { | |
2863 | VM_BUG_ON(!current->mm); | |
2864 | current->memcg_kmem_skip_account++; | |
2865 | } | |
2866 | ||
2867 | static inline void memcg_resume_kmem_account(void) | |
2868 | { | |
2869 | VM_BUG_ON(!current->mm); | |
2870 | current->memcg_kmem_skip_account--; | |
2871 | } | |
2872 | ||
776ed0f0 | 2873 | int __memcg_cleanup_cache_params(struct kmem_cache *s) |
7cf27982 GC |
2874 | { |
2875 | struct kmem_cache *c; | |
b8529907 | 2876 | int i, failed = 0; |
7cf27982 | 2877 | |
bd673145 | 2878 | mutex_lock(&memcg_slab_mutex); |
7a67d7ab QH |
2879 | for_each_memcg_cache_index(i) { |
2880 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
2881 | if (!c) |
2882 | continue; | |
2883 | ||
776ed0f0 | 2884 | memcg_unregister_cache(c); |
b8529907 VD |
2885 | |
2886 | if (cache_from_memcg_idx(s, i)) | |
2887 | failed++; | |
7cf27982 | 2888 | } |
bd673145 | 2889 | mutex_unlock(&memcg_slab_mutex); |
b8529907 | 2890 | return failed; |
7cf27982 GC |
2891 | } |
2892 | ||
776ed0f0 | 2893 | static void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
2894 | { |
2895 | struct kmem_cache *cachep; | |
bd673145 | 2896 | struct memcg_cache_params *params, *tmp; |
1f458cbf GC |
2897 | |
2898 | if (!memcg_kmem_is_active(memcg)) | |
2899 | return; | |
2900 | ||
bd673145 VD |
2901 | mutex_lock(&memcg_slab_mutex); |
2902 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { | |
1f458cbf | 2903 | cachep = memcg_params_to_cache(params); |
bd673145 VD |
2904 | kmem_cache_shrink(cachep); |
2905 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
776ed0f0 | 2906 | memcg_unregister_cache(cachep); |
1f458cbf | 2907 | } |
bd673145 | 2908 | mutex_unlock(&memcg_slab_mutex); |
1f458cbf GC |
2909 | } |
2910 | ||
776ed0f0 | 2911 | struct memcg_register_cache_work { |
5722d094 VD |
2912 | struct mem_cgroup *memcg; |
2913 | struct kmem_cache *cachep; | |
2914 | struct work_struct work; | |
2915 | }; | |
2916 | ||
776ed0f0 | 2917 | static void memcg_register_cache_func(struct work_struct *w) |
d7f25f8a | 2918 | { |
776ed0f0 VD |
2919 | struct memcg_register_cache_work *cw = |
2920 | container_of(w, struct memcg_register_cache_work, work); | |
5722d094 VD |
2921 | struct mem_cgroup *memcg = cw->memcg; |
2922 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2923 | |
bd673145 | 2924 | mutex_lock(&memcg_slab_mutex); |
776ed0f0 | 2925 | memcg_register_cache(memcg, cachep); |
bd673145 VD |
2926 | mutex_unlock(&memcg_slab_mutex); |
2927 | ||
5722d094 | 2928 | css_put(&memcg->css); |
d7f25f8a GC |
2929 | kfree(cw); |
2930 | } | |
2931 | ||
2932 | /* | |
2933 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2934 | */ |
776ed0f0 VD |
2935 | static void __memcg_schedule_register_cache(struct mem_cgroup *memcg, |
2936 | struct kmem_cache *cachep) | |
d7f25f8a | 2937 | { |
776ed0f0 | 2938 | struct memcg_register_cache_work *cw; |
d7f25f8a | 2939 | |
776ed0f0 | 2940 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
ca0dde97 LZ |
2941 | if (cw == NULL) { |
2942 | css_put(&memcg->css); | |
d7f25f8a GC |
2943 | return; |
2944 | } | |
2945 | ||
2946 | cw->memcg = memcg; | |
2947 | cw->cachep = cachep; | |
2948 | ||
776ed0f0 | 2949 | INIT_WORK(&cw->work, memcg_register_cache_func); |
d7f25f8a GC |
2950 | schedule_work(&cw->work); |
2951 | } | |
2952 | ||
776ed0f0 VD |
2953 | static void memcg_schedule_register_cache(struct mem_cgroup *memcg, |
2954 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2955 | { |
2956 | /* | |
2957 | * We need to stop accounting when we kmalloc, because if the | |
2958 | * corresponding kmalloc cache is not yet created, the first allocation | |
776ed0f0 | 2959 | * in __memcg_schedule_register_cache will recurse. |
0e9d92f2 GC |
2960 | * |
2961 | * However, it is better to enclose the whole function. Depending on | |
2962 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2963 | * trigger an allocation. This too, will make us recurse. Because at | |
2964 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2965 | * the safest choice is to do it like this, wrapping the whole function. | |
2966 | */ | |
2967 | memcg_stop_kmem_account(); | |
776ed0f0 | 2968 | __memcg_schedule_register_cache(memcg, cachep); |
0e9d92f2 GC |
2969 | memcg_resume_kmem_account(); |
2970 | } | |
c67a8a68 VD |
2971 | |
2972 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) | |
2973 | { | |
3e32cb2e | 2974 | unsigned int nr_pages = 1 << order; |
c67a8a68 VD |
2975 | int res; |
2976 | ||
3e32cb2e | 2977 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, nr_pages); |
c67a8a68 | 2978 | if (!res) |
3e32cb2e | 2979 | atomic_add(nr_pages, &cachep->memcg_params->nr_pages); |
c67a8a68 VD |
2980 | return res; |
2981 | } | |
2982 | ||
2983 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) | |
2984 | { | |
3e32cb2e JW |
2985 | unsigned int nr_pages = 1 << order; |
2986 | ||
2987 | memcg_uncharge_kmem(cachep->memcg_params->memcg, nr_pages); | |
2988 | atomic_sub(nr_pages, &cachep->memcg_params->nr_pages); | |
c67a8a68 VD |
2989 | } |
2990 | ||
d7f25f8a GC |
2991 | /* |
2992 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2993 | * We try to use the current memcg's version of the cache. | |
2994 | * | |
2995 | * If the cache does not exist yet, if we are the first user of it, | |
2996 | * we either create it immediately, if possible, or create it asynchronously | |
2997 | * in a workqueue. | |
2998 | * In the latter case, we will let the current allocation go through with | |
2999 | * the original cache. | |
3000 | * | |
3001 | * Can't be called in interrupt context or from kernel threads. | |
3002 | * This function needs to be called with rcu_read_lock() held. | |
3003 | */ | |
3004 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3005 | gfp_t gfp) | |
3006 | { | |
3007 | struct mem_cgroup *memcg; | |
959c8963 | 3008 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
3009 | |
3010 | VM_BUG_ON(!cachep->memcg_params); | |
3011 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3012 | ||
0e9d92f2 GC |
3013 | if (!current->mm || current->memcg_kmem_skip_account) |
3014 | return cachep; | |
3015 | ||
d7f25f8a GC |
3016 | rcu_read_lock(); |
3017 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a | 3018 | |
cf2b8fbf | 3019 | if (!memcg_kmem_is_active(memcg)) |
ca0dde97 | 3020 | goto out; |
d7f25f8a | 3021 | |
959c8963 VD |
3022 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
3023 | if (likely(memcg_cachep)) { | |
3024 | cachep = memcg_cachep; | |
ca0dde97 | 3025 | goto out; |
d7f25f8a GC |
3026 | } |
3027 | ||
ca0dde97 | 3028 | /* The corresponding put will be done in the workqueue. */ |
ec903c0c | 3029 | if (!css_tryget_online(&memcg->css)) |
ca0dde97 LZ |
3030 | goto out; |
3031 | rcu_read_unlock(); | |
3032 | ||
3033 | /* | |
3034 | * If we are in a safe context (can wait, and not in interrupt | |
3035 | * context), we could be be predictable and return right away. | |
3036 | * This would guarantee that the allocation being performed | |
3037 | * already belongs in the new cache. | |
3038 | * | |
3039 | * However, there are some clashes that can arrive from locking. | |
3040 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
3041 | * memcg_create_kmem_cache, this means no further allocation |
3042 | * could happen with the slab_mutex held. So it's better to | |
3043 | * defer everything. | |
ca0dde97 | 3044 | */ |
776ed0f0 | 3045 | memcg_schedule_register_cache(memcg, cachep); |
ca0dde97 LZ |
3046 | return cachep; |
3047 | out: | |
3048 | rcu_read_unlock(); | |
3049 | return cachep; | |
d7f25f8a | 3050 | } |
d7f25f8a | 3051 | |
7ae1e1d0 GC |
3052 | /* |
3053 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3054 | * possible for the given order. But the page is not allocated yet, so we'll | |
3055 | * need a further commit step to do the final arrangements. | |
3056 | * | |
3057 | * It is possible for the task to switch cgroups in this mean time, so at | |
3058 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3059 | * the handle argument to return to the caller which cgroup we should commit | |
3060 | * against. We could also return the memcg directly and avoid the pointer | |
3061 | * passing, but a boolean return value gives better semantics considering | |
3062 | * the compiled-out case as well. | |
3063 | * | |
3064 | * Returning true means the allocation is possible. | |
3065 | */ | |
3066 | bool | |
3067 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3068 | { | |
3069 | struct mem_cgroup *memcg; | |
3070 | int ret; | |
3071 | ||
3072 | *_memcg = NULL; | |
6d42c232 GC |
3073 | |
3074 | /* | |
3075 | * Disabling accounting is only relevant for some specific memcg | |
3076 | * internal allocations. Therefore we would initially not have such | |
52383431 VD |
3077 | * check here, since direct calls to the page allocator that are |
3078 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen | |
3079 | * outside memcg core. We are mostly concerned with cache allocations, | |
3080 | * and by having this test at memcg_kmem_get_cache, we are already able | |
3081 | * to relay the allocation to the root cache and bypass the memcg cache | |
3082 | * altogether. | |
6d42c232 GC |
3083 | * |
3084 | * There is one exception, though: the SLUB allocator does not create | |
3085 | * large order caches, but rather service large kmallocs directly from | |
3086 | * the page allocator. Therefore, the following sequence when backed by | |
3087 | * the SLUB allocator: | |
3088 | * | |
f894ffa8 AM |
3089 | * memcg_stop_kmem_account(); |
3090 | * kmalloc(<large_number>) | |
3091 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
3092 | * |
3093 | * would effectively ignore the fact that we should skip accounting, | |
3094 | * since it will drive us directly to this function without passing | |
3095 | * through the cache selector memcg_kmem_get_cache. Such large | |
3096 | * allocations are extremely rare but can happen, for instance, for the | |
3097 | * cache arrays. We bring this test here. | |
3098 | */ | |
3099 | if (!current->mm || current->memcg_kmem_skip_account) | |
3100 | return true; | |
3101 | ||
df381975 | 3102 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 3103 | |
cf2b8fbf | 3104 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
3105 | css_put(&memcg->css); |
3106 | return true; | |
3107 | } | |
3108 | ||
3e32cb2e | 3109 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
3110 | if (!ret) |
3111 | *_memcg = memcg; | |
7ae1e1d0 GC |
3112 | |
3113 | css_put(&memcg->css); | |
3114 | return (ret == 0); | |
3115 | } | |
3116 | ||
3117 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3118 | int order) | |
3119 | { | |
3120 | struct page_cgroup *pc; | |
3121 | ||
3122 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3123 | ||
3124 | /* The page allocation failed. Revert */ | |
3125 | if (!page) { | |
3e32cb2e | 3126 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
3127 | return; |
3128 | } | |
a840cda6 JW |
3129 | /* |
3130 | * The page is freshly allocated and not visible to any | |
3131 | * outside callers yet. Set up pc non-atomically. | |
3132 | */ | |
7ae1e1d0 | 3133 | pc = lookup_page_cgroup(page); |
7ae1e1d0 | 3134 | pc->mem_cgroup = memcg; |
a840cda6 | 3135 | pc->flags = PCG_USED; |
7ae1e1d0 GC |
3136 | } |
3137 | ||
3138 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3139 | { | |
3140 | struct mem_cgroup *memcg = NULL; | |
3141 | struct page_cgroup *pc; | |
3142 | ||
3143 | ||
3144 | pc = lookup_page_cgroup(page); | |
7ae1e1d0 GC |
3145 | if (!PageCgroupUsed(pc)) |
3146 | return; | |
3147 | ||
a840cda6 JW |
3148 | memcg = pc->mem_cgroup; |
3149 | pc->flags = 0; | |
7ae1e1d0 GC |
3150 | |
3151 | /* | |
3152 | * We trust that only if there is a memcg associated with the page, it | |
3153 | * is a valid allocation | |
3154 | */ | |
3155 | if (!memcg) | |
3156 | return; | |
3157 | ||
309381fe | 3158 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
3e32cb2e | 3159 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 | 3160 | } |
1f458cbf | 3161 | #else |
776ed0f0 | 3162 | static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3163 | { |
3164 | } | |
7ae1e1d0 GC |
3165 | #endif /* CONFIG_MEMCG_KMEM */ |
3166 | ||
ca3e0214 KH |
3167 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3168 | ||
ca3e0214 KH |
3169 | /* |
3170 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3171 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3172 | * charge/uncharge will be never happen and move_account() is done under | |
3173 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3174 | */ |
e94c8a9c | 3175 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3176 | { |
b9982f8d | 3177 | struct page_cgroup *head_pc; |
e94c8a9c | 3178 | struct page_cgroup *pc; |
b070e65c | 3179 | struct mem_cgroup *memcg; |
e94c8a9c | 3180 | int i; |
ca3e0214 | 3181 | |
3d37c4a9 KH |
3182 | if (mem_cgroup_disabled()) |
3183 | return; | |
b070e65c | 3184 | |
b9982f8d MH |
3185 | head_pc = lookup_page_cgroup(head); |
3186 | ||
b070e65c | 3187 | memcg = head_pc->mem_cgroup; |
e94c8a9c KH |
3188 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3189 | pc = head_pc + i; | |
b070e65c | 3190 | pc->mem_cgroup = memcg; |
0a31bc97 | 3191 | pc->flags = head_pc->flags; |
e94c8a9c | 3192 | } |
b070e65c DR |
3193 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
3194 | HPAGE_PMD_NR); | |
ca3e0214 | 3195 | } |
12d27107 | 3196 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3197 | |
f817ed48 | 3198 | /** |
de3638d9 | 3199 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3200 | * @page: the page |
7ec99d62 | 3201 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3202 | * @pc: page_cgroup of the page. |
3203 | * @from: mem_cgroup which the page is moved from. | |
3204 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3205 | * | |
3206 | * The caller must confirm following. | |
08e552c6 | 3207 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3208 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3209 | * |
2f3479b1 KH |
3210 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3211 | * from old cgroup. | |
f817ed48 | 3212 | */ |
7ec99d62 JW |
3213 | static int mem_cgroup_move_account(struct page *page, |
3214 | unsigned int nr_pages, | |
3215 | struct page_cgroup *pc, | |
3216 | struct mem_cgroup *from, | |
2f3479b1 | 3217 | struct mem_cgroup *to) |
f817ed48 | 3218 | { |
de3638d9 JW |
3219 | unsigned long flags; |
3220 | int ret; | |
987eba66 | 3221 | |
f817ed48 | 3222 | VM_BUG_ON(from == to); |
309381fe | 3223 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3224 | /* |
3225 | * The page is isolated from LRU. So, collapse function | |
3226 | * will not handle this page. But page splitting can happen. | |
3227 | * Do this check under compound_page_lock(). The caller should | |
3228 | * hold it. | |
3229 | */ | |
3230 | ret = -EBUSY; | |
7ec99d62 | 3231 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3232 | goto out; |
3233 | ||
0a31bc97 JW |
3234 | /* |
3235 | * Prevent mem_cgroup_migrate() from looking at pc->mem_cgroup | |
3236 | * of its source page while we change it: page migration takes | |
3237 | * both pages off the LRU, but page cache replacement doesn't. | |
3238 | */ | |
3239 | if (!trylock_page(page)) | |
3240 | goto out; | |
de3638d9 JW |
3241 | |
3242 | ret = -EINVAL; | |
3243 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
0a31bc97 | 3244 | goto out_unlock; |
de3638d9 | 3245 | |
312734c0 | 3246 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3247 | |
0a31bc97 | 3248 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
3249 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
3250 | nr_pages); | |
3251 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3252 | nr_pages); | |
3253 | } | |
3ea67d06 | 3254 | |
59d1d256 JW |
3255 | if (PageWriteback(page)) { |
3256 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3257 | nr_pages); | |
3258 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3259 | nr_pages); | |
3260 | } | |
3ea67d06 | 3261 | |
0a31bc97 JW |
3262 | /* |
3263 | * It is safe to change pc->mem_cgroup here because the page | |
3264 | * is referenced, charged, and isolated - we can't race with | |
3265 | * uncharging, charging, migration, or LRU putback. | |
3266 | */ | |
d69b042f | 3267 | |
854ffa8d | 3268 | /* caller should have done css_get */ |
08e552c6 | 3269 | pc->mem_cgroup = to; |
312734c0 | 3270 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 | 3271 | ret = 0; |
0a31bc97 JW |
3272 | |
3273 | local_irq_disable(); | |
3274 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 3275 | memcg_check_events(to, page); |
0a31bc97 | 3276 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 3277 | memcg_check_events(from, page); |
0a31bc97 JW |
3278 | local_irq_enable(); |
3279 | out_unlock: | |
3280 | unlock_page(page); | |
de3638d9 | 3281 | out: |
f817ed48 KH |
3282 | return ret; |
3283 | } | |
3284 | ||
c255a458 | 3285 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
3286 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
3287 | bool charge) | |
d13d1443 | 3288 | { |
0a31bc97 JW |
3289 | int val = (charge) ? 1 : -1; |
3290 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 3291 | } |
02491447 DN |
3292 | |
3293 | /** | |
3294 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3295 | * @entry: swap entry to be moved | |
3296 | * @from: mem_cgroup which the entry is moved from | |
3297 | * @to: mem_cgroup which the entry is moved to | |
3298 | * | |
3299 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3300 | * as the mem_cgroup's id of @from. | |
3301 | * | |
3302 | * Returns 0 on success, -EINVAL on failure. | |
3303 | * | |
3e32cb2e | 3304 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3305 | * both res and memsw, and called css_get(). |
3306 | */ | |
3307 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3308 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3309 | { |
3310 | unsigned short old_id, new_id; | |
3311 | ||
34c00c31 LZ |
3312 | old_id = mem_cgroup_id(from); |
3313 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3314 | |
3315 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3316 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3317 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3318 | /* |
483c30b5 | 3319 | * This function is only called from task migration context now. |
3e32cb2e | 3320 | * It postpones page_counter and refcount handling till the end |
483c30b5 | 3321 | * of task migration(mem_cgroup_clear_mc()) for performance |
4050377b LZ |
3322 | * improvement. But we cannot postpone css_get(to) because if |
3323 | * the process that has been moved to @to does swap-in, the | |
3324 | * refcount of @to might be decreased to 0. | |
3325 | * | |
3326 | * We are in attach() phase, so the cgroup is guaranteed to be | |
3327 | * alive, so we can just call css_get(). | |
02491447 | 3328 | */ |
4050377b | 3329 | css_get(&to->css); |
02491447 DN |
3330 | return 0; |
3331 | } | |
3332 | return -EINVAL; | |
3333 | } | |
3334 | #else | |
3335 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3336 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3337 | { |
3338 | return -EINVAL; | |
3339 | } | |
8c7c6e34 | 3340 | #endif |
d13d1443 | 3341 | |
f212ad7c DN |
3342 | #ifdef CONFIG_DEBUG_VM |
3343 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3344 | { | |
3345 | struct page_cgroup *pc; | |
3346 | ||
3347 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3348 | /* |
3349 | * Can be NULL while feeding pages into the page allocator for | |
3350 | * the first time, i.e. during boot or memory hotplug; | |
3351 | * or when mem_cgroup_disabled(). | |
3352 | */ | |
f212ad7c DN |
3353 | if (likely(pc) && PageCgroupUsed(pc)) |
3354 | return pc; | |
3355 | return NULL; | |
3356 | } | |
3357 | ||
3358 | bool mem_cgroup_bad_page_check(struct page *page) | |
3359 | { | |
3360 | if (mem_cgroup_disabled()) | |
3361 | return false; | |
3362 | ||
3363 | return lookup_page_cgroup_used(page) != NULL; | |
3364 | } | |
3365 | ||
3366 | void mem_cgroup_print_bad_page(struct page *page) | |
3367 | { | |
3368 | struct page_cgroup *pc; | |
3369 | ||
3370 | pc = lookup_page_cgroup_used(page); | |
3371 | if (pc) { | |
d045197f AM |
3372 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
3373 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
3374 | } |
3375 | } | |
3376 | #endif | |
3377 | ||
3e32cb2e JW |
3378 | static DEFINE_MUTEX(memcg_limit_mutex); |
3379 | ||
d38d2a75 | 3380 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3381 | unsigned long limit) |
628f4235 | 3382 | { |
3e32cb2e JW |
3383 | unsigned long curusage; |
3384 | unsigned long oldusage; | |
3385 | bool enlarge = false; | |
81d39c20 | 3386 | int retry_count; |
3e32cb2e | 3387 | int ret; |
81d39c20 KH |
3388 | |
3389 | /* | |
3390 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3391 | * is depends on callers. We set our retry-count to be function | |
3392 | * of # of children which we should visit in this loop. | |
3393 | */ | |
3e32cb2e JW |
3394 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3395 | mem_cgroup_count_children(memcg); | |
81d39c20 | 3396 | |
3e32cb2e | 3397 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 3398 | |
3e32cb2e | 3399 | do { |
628f4235 KH |
3400 | if (signal_pending(current)) { |
3401 | ret = -EINTR; | |
3402 | break; | |
3403 | } | |
3e32cb2e JW |
3404 | |
3405 | mutex_lock(&memcg_limit_mutex); | |
3406 | if (limit > memcg->memsw.limit) { | |
3407 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3408 | ret = -EINVAL; |
628f4235 KH |
3409 | break; |
3410 | } | |
3e32cb2e JW |
3411 | if (limit > memcg->memory.limit) |
3412 | enlarge = true; | |
3413 | ret = page_counter_limit(&memcg->memory, limit); | |
3414 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3415 | |
3416 | if (!ret) | |
3417 | break; | |
3418 | ||
b70a2a21 JW |
3419 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
3420 | ||
3e32cb2e | 3421 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 3422 | /* Usage is reduced ? */ |
f894ffa8 | 3423 | if (curusage >= oldusage) |
81d39c20 KH |
3424 | retry_count--; |
3425 | else | |
3426 | oldusage = curusage; | |
3e32cb2e JW |
3427 | } while (retry_count); |
3428 | ||
3c11ecf4 KH |
3429 | if (!ret && enlarge) |
3430 | memcg_oom_recover(memcg); | |
14797e23 | 3431 | |
8c7c6e34 KH |
3432 | return ret; |
3433 | } | |
3434 | ||
338c8431 | 3435 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3436 | unsigned long limit) |
8c7c6e34 | 3437 | { |
3e32cb2e JW |
3438 | unsigned long curusage; |
3439 | unsigned long oldusage; | |
3440 | bool enlarge = false; | |
81d39c20 | 3441 | int retry_count; |
3e32cb2e | 3442 | int ret; |
8c7c6e34 | 3443 | |
81d39c20 | 3444 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
3445 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3446 | mem_cgroup_count_children(memcg); | |
3447 | ||
3448 | oldusage = page_counter_read(&memcg->memsw); | |
3449 | ||
3450 | do { | |
8c7c6e34 KH |
3451 | if (signal_pending(current)) { |
3452 | ret = -EINTR; | |
3453 | break; | |
3454 | } | |
3e32cb2e JW |
3455 | |
3456 | mutex_lock(&memcg_limit_mutex); | |
3457 | if (limit < memcg->memory.limit) { | |
3458 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3459 | ret = -EINVAL; |
8c7c6e34 KH |
3460 | break; |
3461 | } | |
3e32cb2e JW |
3462 | if (limit > memcg->memsw.limit) |
3463 | enlarge = true; | |
3464 | ret = page_counter_limit(&memcg->memsw, limit); | |
3465 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3466 | |
3467 | if (!ret) | |
3468 | break; | |
3469 | ||
b70a2a21 JW |
3470 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3471 | ||
3e32cb2e | 3472 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3473 | /* Usage is reduced ? */ |
8c7c6e34 | 3474 | if (curusage >= oldusage) |
628f4235 | 3475 | retry_count--; |
81d39c20 KH |
3476 | else |
3477 | oldusage = curusage; | |
3e32cb2e JW |
3478 | } while (retry_count); |
3479 | ||
3c11ecf4 KH |
3480 | if (!ret && enlarge) |
3481 | memcg_oom_recover(memcg); | |
3e32cb2e | 3482 | |
628f4235 KH |
3483 | return ret; |
3484 | } | |
3485 | ||
0608f43d AM |
3486 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3487 | gfp_t gfp_mask, | |
3488 | unsigned long *total_scanned) | |
3489 | { | |
3490 | unsigned long nr_reclaimed = 0; | |
3491 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3492 | unsigned long reclaimed; | |
3493 | int loop = 0; | |
3494 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3495 | unsigned long excess; |
0608f43d AM |
3496 | unsigned long nr_scanned; |
3497 | ||
3498 | if (order > 0) | |
3499 | return 0; | |
3500 | ||
3501 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3502 | /* | |
3503 | * This loop can run a while, specially if mem_cgroup's continuously | |
3504 | * keep exceeding their soft limit and putting the system under | |
3505 | * pressure | |
3506 | */ | |
3507 | do { | |
3508 | if (next_mz) | |
3509 | mz = next_mz; | |
3510 | else | |
3511 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3512 | if (!mz) | |
3513 | break; | |
3514 | ||
3515 | nr_scanned = 0; | |
3516 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3517 | gfp_mask, &nr_scanned); | |
3518 | nr_reclaimed += reclaimed; | |
3519 | *total_scanned += nr_scanned; | |
0a31bc97 | 3520 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3521 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3522 | |
3523 | /* | |
3524 | * If we failed to reclaim anything from this memory cgroup | |
3525 | * it is time to move on to the next cgroup | |
3526 | */ | |
3527 | next_mz = NULL; | |
bc2f2e7f VD |
3528 | if (!reclaimed) |
3529 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3530 | ||
3e32cb2e | 3531 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3532 | /* |
3533 | * One school of thought says that we should not add | |
3534 | * back the node to the tree if reclaim returns 0. | |
3535 | * But our reclaim could return 0, simply because due | |
3536 | * to priority we are exposing a smaller subset of | |
3537 | * memory to reclaim from. Consider this as a longer | |
3538 | * term TODO. | |
3539 | */ | |
3540 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3541 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3542 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3543 | css_put(&mz->memcg->css); |
3544 | loop++; | |
3545 | /* | |
3546 | * Could not reclaim anything and there are no more | |
3547 | * mem cgroups to try or we seem to be looping without | |
3548 | * reclaiming anything. | |
3549 | */ | |
3550 | if (!nr_reclaimed && | |
3551 | (next_mz == NULL || | |
3552 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3553 | break; | |
3554 | } while (!nr_reclaimed); | |
3555 | if (next_mz) | |
3556 | css_put(&next_mz->memcg->css); | |
3557 | return nr_reclaimed; | |
3558 | } | |
3559 | ||
ea280e7b TH |
3560 | /* |
3561 | * Test whether @memcg has children, dead or alive. Note that this | |
3562 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3563 | * returns %true if there are child csses according to the cgroup | |
3564 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3565 | */ | |
b5f99b53 GC |
3566 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3567 | { | |
ea280e7b TH |
3568 | bool ret; |
3569 | ||
696ac172 | 3570 | /* |
ea280e7b TH |
3571 | * The lock does not prevent addition or deletion of children, but |
3572 | * it prevents a new child from being initialized based on this | |
3573 | * parent in css_online(), so it's enough to decide whether | |
3574 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3575 | */ |
ea280e7b TH |
3576 | lockdep_assert_held(&memcg_create_mutex); |
3577 | ||
3578 | rcu_read_lock(); | |
3579 | ret = css_next_child(NULL, &memcg->css); | |
3580 | rcu_read_unlock(); | |
3581 | return ret; | |
b5f99b53 GC |
3582 | } |
3583 | ||
c26251f9 MH |
3584 | /* |
3585 | * Reclaims as many pages from the given memcg as possible and moves | |
3586 | * the rest to the parent. | |
3587 | * | |
3588 | * Caller is responsible for holding css reference for memcg. | |
3589 | */ | |
3590 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3591 | { | |
3592 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3593 | |
c1e862c1 KH |
3594 | /* we call try-to-free pages for make this cgroup empty */ |
3595 | lru_add_drain_all(); | |
f817ed48 | 3596 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3597 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3598 | int progress; |
c1e862c1 | 3599 | |
c26251f9 MH |
3600 | if (signal_pending(current)) |
3601 | return -EINTR; | |
3602 | ||
b70a2a21 JW |
3603 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3604 | GFP_KERNEL, true); | |
c1e862c1 | 3605 | if (!progress) { |
f817ed48 | 3606 | nr_retries--; |
c1e862c1 | 3607 | /* maybe some writeback is necessary */ |
8aa7e847 | 3608 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3609 | } |
f817ed48 KH |
3610 | |
3611 | } | |
ab5196c2 MH |
3612 | |
3613 | return 0; | |
cc847582 KH |
3614 | } |
3615 | ||
6770c64e TH |
3616 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3617 | char *buf, size_t nbytes, | |
3618 | loff_t off) | |
c1e862c1 | 3619 | { |
6770c64e | 3620 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3621 | |
d8423011 MH |
3622 | if (mem_cgroup_is_root(memcg)) |
3623 | return -EINVAL; | |
6770c64e | 3624 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3625 | } |
3626 | ||
182446d0 TH |
3627 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3628 | struct cftype *cft) | |
18f59ea7 | 3629 | { |
182446d0 | 3630 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3631 | } |
3632 | ||
182446d0 TH |
3633 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3634 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3635 | { |
3636 | int retval = 0; | |
182446d0 | 3637 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3638 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3639 | |
0999821b | 3640 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3641 | |
3642 | if (memcg->use_hierarchy == val) | |
3643 | goto out; | |
3644 | ||
18f59ea7 | 3645 | /* |
af901ca1 | 3646 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3647 | * in the child subtrees. If it is unset, then the change can |
3648 | * occur, provided the current cgroup has no children. | |
3649 | * | |
3650 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3651 | * set if there are no children. | |
3652 | */ | |
c0ff4b85 | 3653 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3654 | (val == 1 || val == 0)) { |
ea280e7b | 3655 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3656 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3657 | else |
3658 | retval = -EBUSY; | |
3659 | } else | |
3660 | retval = -EINVAL; | |
567fb435 GC |
3661 | |
3662 | out: | |
0999821b | 3663 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3664 | |
3665 | return retval; | |
3666 | } | |
3667 | ||
3e32cb2e JW |
3668 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3669 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3670 | { |
3671 | struct mem_cgroup *iter; | |
3672 | long val = 0; | |
3673 | ||
3674 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3675 | for_each_mem_cgroup_tree(iter, memcg) | |
3676 | val += mem_cgroup_read_stat(iter, idx); | |
3677 | ||
3678 | if (val < 0) /* race ? */ | |
3679 | val = 0; | |
3680 | return val; | |
3681 | } | |
3682 | ||
3683 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3684 | { | |
3685 | u64 val; | |
3686 | ||
3e32cb2e JW |
3687 | if (mem_cgroup_is_root(memcg)) { |
3688 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3689 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3690 | if (swap) | |
3691 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3692 | } else { | |
ce00a967 | 3693 | if (!swap) |
3e32cb2e | 3694 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3695 | else |
3e32cb2e | 3696 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3697 | } |
ce00a967 JW |
3698 | return val << PAGE_SHIFT; |
3699 | } | |
3700 | ||
3e32cb2e JW |
3701 | enum { |
3702 | RES_USAGE, | |
3703 | RES_LIMIT, | |
3704 | RES_MAX_USAGE, | |
3705 | RES_FAILCNT, | |
3706 | RES_SOFT_LIMIT, | |
3707 | }; | |
ce00a967 | 3708 | |
791badbd | 3709 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3710 | struct cftype *cft) |
8cdea7c0 | 3711 | { |
182446d0 | 3712 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3713 | struct page_counter *counter; |
af36f906 | 3714 | |
3e32cb2e | 3715 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3716 | case _MEM: |
3e32cb2e JW |
3717 | counter = &memcg->memory; |
3718 | break; | |
8c7c6e34 | 3719 | case _MEMSWAP: |
3e32cb2e JW |
3720 | counter = &memcg->memsw; |
3721 | break; | |
510fc4e1 | 3722 | case _KMEM: |
3e32cb2e | 3723 | counter = &memcg->kmem; |
510fc4e1 | 3724 | break; |
8c7c6e34 KH |
3725 | default: |
3726 | BUG(); | |
8c7c6e34 | 3727 | } |
3e32cb2e JW |
3728 | |
3729 | switch (MEMFILE_ATTR(cft->private)) { | |
3730 | case RES_USAGE: | |
3731 | if (counter == &memcg->memory) | |
3732 | return mem_cgroup_usage(memcg, false); | |
3733 | if (counter == &memcg->memsw) | |
3734 | return mem_cgroup_usage(memcg, true); | |
3735 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3736 | case RES_LIMIT: | |
3737 | return (u64)counter->limit * PAGE_SIZE; | |
3738 | case RES_MAX_USAGE: | |
3739 | return (u64)counter->watermark * PAGE_SIZE; | |
3740 | case RES_FAILCNT: | |
3741 | return counter->failcnt; | |
3742 | case RES_SOFT_LIMIT: | |
3743 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3744 | default: | |
3745 | BUG(); | |
3746 | } | |
8cdea7c0 | 3747 | } |
510fc4e1 | 3748 | |
510fc4e1 | 3749 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
3750 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
3751 | unsigned long nr_pages) | |
d6441637 VD |
3752 | { |
3753 | int err = 0; | |
3754 | int memcg_id; | |
3755 | ||
3756 | if (memcg_kmem_is_active(memcg)) | |
3757 | return 0; | |
3758 | ||
3759 | /* | |
3760 | * We are going to allocate memory for data shared by all memory | |
3761 | * cgroups so let's stop accounting here. | |
3762 | */ | |
3763 | memcg_stop_kmem_account(); | |
3764 | ||
510fc4e1 GC |
3765 | /* |
3766 | * For simplicity, we won't allow this to be disabled. It also can't | |
3767 | * be changed if the cgroup has children already, or if tasks had | |
3768 | * already joined. | |
3769 | * | |
3770 | * If tasks join before we set the limit, a person looking at | |
3771 | * kmem.usage_in_bytes will have no way to determine when it took | |
3772 | * place, which makes the value quite meaningless. | |
3773 | * | |
3774 | * After it first became limited, changes in the value of the limit are | |
3775 | * of course permitted. | |
510fc4e1 | 3776 | */ |
0999821b | 3777 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
3778 | if (cgroup_has_tasks(memcg->css.cgroup) || |
3779 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
3780 | err = -EBUSY; |
3781 | mutex_unlock(&memcg_create_mutex); | |
3782 | if (err) | |
3783 | goto out; | |
510fc4e1 | 3784 | |
f3bb3043 | 3785 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
3786 | if (memcg_id < 0) { |
3787 | err = memcg_id; | |
3788 | goto out; | |
3789 | } | |
3790 | ||
d6441637 VD |
3791 | memcg->kmemcg_id = memcg_id; |
3792 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
d6441637 VD |
3793 | |
3794 | /* | |
3795 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
3796 | * active bit set yet, so this should succeed. | |
3797 | */ | |
3e32cb2e | 3798 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
3799 | VM_BUG_ON(err); |
3800 | ||
3801 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
3802 | /* | |
3803 | * Setting the active bit after enabling static branching will | |
3804 | * guarantee no one starts accounting before all call sites are | |
3805 | * patched. | |
3806 | */ | |
3807 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 3808 | out: |
d6441637 VD |
3809 | memcg_resume_kmem_account(); |
3810 | return err; | |
d6441637 VD |
3811 | } |
3812 | ||
d6441637 | 3813 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3814 | unsigned long limit) |
d6441637 VD |
3815 | { |
3816 | int ret; | |
3817 | ||
3e32cb2e | 3818 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 3819 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 3820 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 3821 | else |
3e32cb2e JW |
3822 | ret = page_counter_limit(&memcg->kmem, limit); |
3823 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
3824 | return ret; |
3825 | } | |
3826 | ||
55007d84 | 3827 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 3828 | { |
55007d84 | 3829 | int ret = 0; |
510fc4e1 | 3830 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 3831 | |
d6441637 VD |
3832 | if (!parent) |
3833 | return 0; | |
55007d84 | 3834 | |
8c0145b6 | 3835 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 3836 | /* |
d6441637 VD |
3837 | * If the parent cgroup is not kmem-active now, it cannot be activated |
3838 | * after this point, because it has at least one child already. | |
55007d84 | 3839 | */ |
d6441637 | 3840 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3841 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3842 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3843 | return ret; |
510fc4e1 | 3844 | } |
d6441637 VD |
3845 | #else |
3846 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3847 | unsigned long limit) |
d6441637 VD |
3848 | { |
3849 | return -EINVAL; | |
3850 | } | |
6d043990 | 3851 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3852 | |
628f4235 KH |
3853 | /* |
3854 | * The user of this function is... | |
3855 | * RES_LIMIT. | |
3856 | */ | |
451af504 TH |
3857 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3858 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3859 | { |
451af504 | 3860 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3861 | unsigned long nr_pages; |
628f4235 KH |
3862 | int ret; |
3863 | ||
451af504 | 3864 | buf = strstrip(buf); |
3e32cb2e JW |
3865 | ret = page_counter_memparse(buf, &nr_pages); |
3866 | if (ret) | |
3867 | return ret; | |
af36f906 | 3868 | |
3e32cb2e | 3869 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3870 | case RES_LIMIT: |
4b3bde4c BS |
3871 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3872 | ret = -EINVAL; | |
3873 | break; | |
3874 | } | |
3e32cb2e JW |
3875 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3876 | case _MEM: | |
3877 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3878 | break; |
3e32cb2e JW |
3879 | case _MEMSWAP: |
3880 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3881 | break; |
3e32cb2e JW |
3882 | case _KMEM: |
3883 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3884 | break; | |
3885 | } | |
296c81d8 | 3886 | break; |
3e32cb2e JW |
3887 | case RES_SOFT_LIMIT: |
3888 | memcg->soft_limit = nr_pages; | |
3889 | ret = 0; | |
628f4235 KH |
3890 | break; |
3891 | } | |
451af504 | 3892 | return ret ?: nbytes; |
8cdea7c0 BS |
3893 | } |
3894 | ||
6770c64e TH |
3895 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3896 | size_t nbytes, loff_t off) | |
c84872e1 | 3897 | { |
6770c64e | 3898 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3899 | struct page_counter *counter; |
c84872e1 | 3900 | |
3e32cb2e JW |
3901 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3902 | case _MEM: | |
3903 | counter = &memcg->memory; | |
3904 | break; | |
3905 | case _MEMSWAP: | |
3906 | counter = &memcg->memsw; | |
3907 | break; | |
3908 | case _KMEM: | |
3909 | counter = &memcg->kmem; | |
3910 | break; | |
3911 | default: | |
3912 | BUG(); | |
3913 | } | |
af36f906 | 3914 | |
3e32cb2e | 3915 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3916 | case RES_MAX_USAGE: |
3e32cb2e | 3917 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3918 | break; |
3919 | case RES_FAILCNT: | |
3e32cb2e | 3920 | counter->failcnt = 0; |
29f2a4da | 3921 | break; |
3e32cb2e JW |
3922 | default: |
3923 | BUG(); | |
29f2a4da | 3924 | } |
f64c3f54 | 3925 | |
6770c64e | 3926 | return nbytes; |
c84872e1 PE |
3927 | } |
3928 | ||
182446d0 | 3929 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3930 | struct cftype *cft) |
3931 | { | |
182446d0 | 3932 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3933 | } |
3934 | ||
02491447 | 3935 | #ifdef CONFIG_MMU |
182446d0 | 3936 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3937 | struct cftype *cft, u64 val) |
3938 | { | |
182446d0 | 3939 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
3940 | |
3941 | if (val >= (1 << NR_MOVE_TYPE)) | |
3942 | return -EINVAL; | |
ee5e8472 | 3943 | |
7dc74be0 | 3944 | /* |
ee5e8472 GC |
3945 | * No kind of locking is needed in here, because ->can_attach() will |
3946 | * check this value once in the beginning of the process, and then carry | |
3947 | * on with stale data. This means that changes to this value will only | |
3948 | * affect task migrations starting after the change. | |
7dc74be0 | 3949 | */ |
c0ff4b85 | 3950 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3951 | return 0; |
3952 | } | |
02491447 | 3953 | #else |
182446d0 | 3954 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3955 | struct cftype *cft, u64 val) |
3956 | { | |
3957 | return -ENOSYS; | |
3958 | } | |
3959 | #endif | |
7dc74be0 | 3960 | |
406eb0c9 | 3961 | #ifdef CONFIG_NUMA |
2da8ca82 | 3962 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3963 | { |
25485de6 GT |
3964 | struct numa_stat { |
3965 | const char *name; | |
3966 | unsigned int lru_mask; | |
3967 | }; | |
3968 | ||
3969 | static const struct numa_stat stats[] = { | |
3970 | { "total", LRU_ALL }, | |
3971 | { "file", LRU_ALL_FILE }, | |
3972 | { "anon", LRU_ALL_ANON }, | |
3973 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3974 | }; | |
3975 | const struct numa_stat *stat; | |
406eb0c9 | 3976 | int nid; |
25485de6 | 3977 | unsigned long nr; |
2da8ca82 | 3978 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3979 | |
25485de6 GT |
3980 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3981 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3982 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3983 | for_each_node_state(nid, N_MEMORY) { | |
3984 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3985 | stat->lru_mask); | |
3986 | seq_printf(m, " N%d=%lu", nid, nr); | |
3987 | } | |
3988 | seq_putc(m, '\n'); | |
406eb0c9 | 3989 | } |
406eb0c9 | 3990 | |
071aee13 YH |
3991 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3992 | struct mem_cgroup *iter; | |
3993 | ||
3994 | nr = 0; | |
3995 | for_each_mem_cgroup_tree(iter, memcg) | |
3996 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3997 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3998 | for_each_node_state(nid, N_MEMORY) { | |
3999 | nr = 0; | |
4000 | for_each_mem_cgroup_tree(iter, memcg) | |
4001 | nr += mem_cgroup_node_nr_lru_pages( | |
4002 | iter, nid, stat->lru_mask); | |
4003 | seq_printf(m, " N%d=%lu", nid, nr); | |
4004 | } | |
4005 | seq_putc(m, '\n'); | |
406eb0c9 | 4006 | } |
406eb0c9 | 4007 | |
406eb0c9 YH |
4008 | return 0; |
4009 | } | |
4010 | #endif /* CONFIG_NUMA */ | |
4011 | ||
af7c4b0e JW |
4012 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
4013 | { | |
4014 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4015 | } | |
4016 | ||
2da8ca82 | 4017 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 4018 | { |
2da8ca82 | 4019 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 4020 | unsigned long memory, memsw; |
af7c4b0e JW |
4021 | struct mem_cgroup *mi; |
4022 | unsigned int i; | |
406eb0c9 | 4023 | |
af7c4b0e | 4024 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4025 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4026 | continue; |
af7c4b0e JW |
4027 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4028 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4029 | } |
7b854121 | 4030 | |
af7c4b0e JW |
4031 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4032 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4033 | mem_cgroup_read_events(memcg, i)); | |
4034 | ||
4035 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4036 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4037 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4038 | ||
14067bb3 | 4039 | /* Hierarchical information */ |
3e32cb2e JW |
4040 | memory = memsw = PAGE_COUNTER_MAX; |
4041 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
4042 | memory = min(memory, mi->memory.limit); | |
4043 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 4044 | } |
3e32cb2e JW |
4045 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4046 | (u64)memory * PAGE_SIZE); | |
4047 | if (do_swap_account) | |
4048 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
4049 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4050 | |
af7c4b0e JW |
4051 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4052 | long long val = 0; | |
4053 | ||
bff6bb83 | 4054 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4055 | continue; |
af7c4b0e JW |
4056 | for_each_mem_cgroup_tree(mi, memcg) |
4057 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4058 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4059 | } | |
4060 | ||
4061 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4062 | unsigned long long val = 0; | |
4063 | ||
4064 | for_each_mem_cgroup_tree(mi, memcg) | |
4065 | val += mem_cgroup_read_events(mi, i); | |
4066 | seq_printf(m, "total_%s %llu\n", | |
4067 | mem_cgroup_events_names[i], val); | |
4068 | } | |
4069 | ||
4070 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4071 | unsigned long long val = 0; | |
4072 | ||
4073 | for_each_mem_cgroup_tree(mi, memcg) | |
4074 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4075 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4076 | } |
14067bb3 | 4077 | |
7f016ee8 | 4078 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4079 | { |
4080 | int nid, zid; | |
4081 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4082 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4083 | unsigned long recent_rotated[2] = {0, 0}; |
4084 | unsigned long recent_scanned[2] = {0, 0}; | |
4085 | ||
4086 | for_each_online_node(nid) | |
4087 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 4088 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 4089 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4090 | |
89abfab1 HD |
4091 | recent_rotated[0] += rstat->recent_rotated[0]; |
4092 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4093 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4094 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4095 | } |
78ccf5b5 JW |
4096 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4097 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4098 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4099 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4100 | } |
4101 | #endif | |
4102 | ||
d2ceb9b7 KH |
4103 | return 0; |
4104 | } | |
4105 | ||
182446d0 TH |
4106 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4107 | struct cftype *cft) | |
a7885eb8 | 4108 | { |
182446d0 | 4109 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4110 | |
1f4c025b | 4111 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4112 | } |
4113 | ||
182446d0 TH |
4114 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4115 | struct cftype *cft, u64 val) | |
a7885eb8 | 4116 | { |
182446d0 | 4117 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4118 | |
3dae7fec | 4119 | if (val > 100) |
a7885eb8 KM |
4120 | return -EINVAL; |
4121 | ||
14208b0e | 4122 | if (css->parent) |
3dae7fec JW |
4123 | memcg->swappiness = val; |
4124 | else | |
4125 | vm_swappiness = val; | |
068b38c1 | 4126 | |
a7885eb8 KM |
4127 | return 0; |
4128 | } | |
4129 | ||
2e72b634 KS |
4130 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4131 | { | |
4132 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4133 | unsigned long usage; |
2e72b634 KS |
4134 | int i; |
4135 | ||
4136 | rcu_read_lock(); | |
4137 | if (!swap) | |
2c488db2 | 4138 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4139 | else |
2c488db2 | 4140 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4141 | |
4142 | if (!t) | |
4143 | goto unlock; | |
4144 | ||
ce00a967 | 4145 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4146 | |
4147 | /* | |
748dad36 | 4148 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4149 | * If it's not true, a threshold was crossed after last |
4150 | * call of __mem_cgroup_threshold(). | |
4151 | */ | |
5407a562 | 4152 | i = t->current_threshold; |
2e72b634 KS |
4153 | |
4154 | /* | |
4155 | * Iterate backward over array of thresholds starting from | |
4156 | * current_threshold and check if a threshold is crossed. | |
4157 | * If none of thresholds below usage is crossed, we read | |
4158 | * only one element of the array here. | |
4159 | */ | |
4160 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4161 | eventfd_signal(t->entries[i].eventfd, 1); | |
4162 | ||
4163 | /* i = current_threshold + 1 */ | |
4164 | i++; | |
4165 | ||
4166 | /* | |
4167 | * Iterate forward over array of thresholds starting from | |
4168 | * current_threshold+1 and check if a threshold is crossed. | |
4169 | * If none of thresholds above usage is crossed, we read | |
4170 | * only one element of the array here. | |
4171 | */ | |
4172 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4173 | eventfd_signal(t->entries[i].eventfd, 1); | |
4174 | ||
4175 | /* Update current_threshold */ | |
5407a562 | 4176 | t->current_threshold = i - 1; |
2e72b634 KS |
4177 | unlock: |
4178 | rcu_read_unlock(); | |
4179 | } | |
4180 | ||
4181 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4182 | { | |
ad4ca5f4 KS |
4183 | while (memcg) { |
4184 | __mem_cgroup_threshold(memcg, false); | |
4185 | if (do_swap_account) | |
4186 | __mem_cgroup_threshold(memcg, true); | |
4187 | ||
4188 | memcg = parent_mem_cgroup(memcg); | |
4189 | } | |
2e72b634 KS |
4190 | } |
4191 | ||
4192 | static int compare_thresholds(const void *a, const void *b) | |
4193 | { | |
4194 | const struct mem_cgroup_threshold *_a = a; | |
4195 | const struct mem_cgroup_threshold *_b = b; | |
4196 | ||
2bff24a3 GT |
4197 | if (_a->threshold > _b->threshold) |
4198 | return 1; | |
4199 | ||
4200 | if (_a->threshold < _b->threshold) | |
4201 | return -1; | |
4202 | ||
4203 | return 0; | |
2e72b634 KS |
4204 | } |
4205 | ||
c0ff4b85 | 4206 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4207 | { |
4208 | struct mem_cgroup_eventfd_list *ev; | |
4209 | ||
2bcf2e92 MH |
4210 | spin_lock(&memcg_oom_lock); |
4211 | ||
c0ff4b85 | 4212 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4213 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4214 | |
4215 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4216 | return 0; |
4217 | } | |
4218 | ||
c0ff4b85 | 4219 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4220 | { |
7d74b06f KH |
4221 | struct mem_cgroup *iter; |
4222 | ||
c0ff4b85 | 4223 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4224 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4225 | } |
4226 | ||
59b6f873 | 4227 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4228 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4229 | { |
2c488db2 KS |
4230 | struct mem_cgroup_thresholds *thresholds; |
4231 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4232 | unsigned long threshold; |
4233 | unsigned long usage; | |
2c488db2 | 4234 | int i, size, ret; |
2e72b634 | 4235 | |
3e32cb2e | 4236 | ret = page_counter_memparse(args, &threshold); |
2e72b634 KS |
4237 | if (ret) |
4238 | return ret; | |
4239 | ||
4240 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4241 | |
05b84301 | 4242 | if (type == _MEM) { |
2c488db2 | 4243 | thresholds = &memcg->thresholds; |
ce00a967 | 4244 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4245 | } else if (type == _MEMSWAP) { |
2c488db2 | 4246 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4247 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4248 | } else |
2e72b634 KS |
4249 | BUG(); |
4250 | ||
2e72b634 | 4251 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4252 | if (thresholds->primary) |
2e72b634 KS |
4253 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4254 | ||
2c488db2 | 4255 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4256 | |
4257 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4258 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4259 | GFP_KERNEL); |
2c488db2 | 4260 | if (!new) { |
2e72b634 KS |
4261 | ret = -ENOMEM; |
4262 | goto unlock; | |
4263 | } | |
2c488db2 | 4264 | new->size = size; |
2e72b634 KS |
4265 | |
4266 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4267 | if (thresholds->primary) { |
4268 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4269 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4270 | } |
4271 | ||
2e72b634 | 4272 | /* Add new threshold */ |
2c488db2 KS |
4273 | new->entries[size - 1].eventfd = eventfd; |
4274 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4275 | |
4276 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4277 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4278 | compare_thresholds, NULL); |
4279 | ||
4280 | /* Find current threshold */ | |
2c488db2 | 4281 | new->current_threshold = -1; |
2e72b634 | 4282 | for (i = 0; i < size; i++) { |
748dad36 | 4283 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4284 | /* |
2c488db2 KS |
4285 | * new->current_threshold will not be used until |
4286 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4287 | * it here. |
4288 | */ | |
2c488db2 | 4289 | ++new->current_threshold; |
748dad36 SZ |
4290 | } else |
4291 | break; | |
2e72b634 KS |
4292 | } |
4293 | ||
2c488db2 KS |
4294 | /* Free old spare buffer and save old primary buffer as spare */ |
4295 | kfree(thresholds->spare); | |
4296 | thresholds->spare = thresholds->primary; | |
4297 | ||
4298 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4299 | |
907860ed | 4300 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4301 | synchronize_rcu(); |
4302 | ||
2e72b634 KS |
4303 | unlock: |
4304 | mutex_unlock(&memcg->thresholds_lock); | |
4305 | ||
4306 | return ret; | |
4307 | } | |
4308 | ||
59b6f873 | 4309 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4310 | struct eventfd_ctx *eventfd, const char *args) |
4311 | { | |
59b6f873 | 4312 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4313 | } |
4314 | ||
59b6f873 | 4315 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4316 | struct eventfd_ctx *eventfd, const char *args) |
4317 | { | |
59b6f873 | 4318 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4319 | } |
4320 | ||
59b6f873 | 4321 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4322 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4323 | { |
2c488db2 KS |
4324 | struct mem_cgroup_thresholds *thresholds; |
4325 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4326 | unsigned long usage; |
2c488db2 | 4327 | int i, j, size; |
2e72b634 KS |
4328 | |
4329 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4330 | |
4331 | if (type == _MEM) { | |
2c488db2 | 4332 | thresholds = &memcg->thresholds; |
ce00a967 | 4333 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4334 | } else if (type == _MEMSWAP) { |
2c488db2 | 4335 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4336 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4337 | } else |
2e72b634 KS |
4338 | BUG(); |
4339 | ||
371528ca AV |
4340 | if (!thresholds->primary) |
4341 | goto unlock; | |
4342 | ||
2e72b634 KS |
4343 | /* Check if a threshold crossed before removing */ |
4344 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4345 | ||
4346 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4347 | size = 0; |
4348 | for (i = 0; i < thresholds->primary->size; i++) { | |
4349 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4350 | size++; |
4351 | } | |
4352 | ||
2c488db2 | 4353 | new = thresholds->spare; |
907860ed | 4354 | |
2e72b634 KS |
4355 | /* Set thresholds array to NULL if we don't have thresholds */ |
4356 | if (!size) { | |
2c488db2 KS |
4357 | kfree(new); |
4358 | new = NULL; | |
907860ed | 4359 | goto swap_buffers; |
2e72b634 KS |
4360 | } |
4361 | ||
2c488db2 | 4362 | new->size = size; |
2e72b634 KS |
4363 | |
4364 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4365 | new->current_threshold = -1; |
4366 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4367 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4368 | continue; |
4369 | ||
2c488db2 | 4370 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4371 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4372 | /* |
2c488db2 | 4373 | * new->current_threshold will not be used |
2e72b634 KS |
4374 | * until rcu_assign_pointer(), so it's safe to increment |
4375 | * it here. | |
4376 | */ | |
2c488db2 | 4377 | ++new->current_threshold; |
2e72b634 KS |
4378 | } |
4379 | j++; | |
4380 | } | |
4381 | ||
907860ed | 4382 | swap_buffers: |
2c488db2 KS |
4383 | /* Swap primary and spare array */ |
4384 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4385 | /* If all events are unregistered, free the spare array */ |
4386 | if (!new) { | |
4387 | kfree(thresholds->spare); | |
4388 | thresholds->spare = NULL; | |
4389 | } | |
4390 | ||
2c488db2 | 4391 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4392 | |
907860ed | 4393 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4394 | synchronize_rcu(); |
371528ca | 4395 | unlock: |
2e72b634 | 4396 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4397 | } |
c1e862c1 | 4398 | |
59b6f873 | 4399 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4400 | struct eventfd_ctx *eventfd) |
4401 | { | |
59b6f873 | 4402 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4403 | } |
4404 | ||
59b6f873 | 4405 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4406 | struct eventfd_ctx *eventfd) |
4407 | { | |
59b6f873 | 4408 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4409 | } |
4410 | ||
59b6f873 | 4411 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4412 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4413 | { |
9490ff27 | 4414 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4415 | |
9490ff27 KH |
4416 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4417 | if (!event) | |
4418 | return -ENOMEM; | |
4419 | ||
1af8efe9 | 4420 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4421 | |
4422 | event->eventfd = eventfd; | |
4423 | list_add(&event->list, &memcg->oom_notify); | |
4424 | ||
4425 | /* already in OOM ? */ | |
79dfdacc | 4426 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4427 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4428 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4429 | |
4430 | return 0; | |
4431 | } | |
4432 | ||
59b6f873 | 4433 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4434 | struct eventfd_ctx *eventfd) |
9490ff27 | 4435 | { |
9490ff27 | 4436 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4437 | |
1af8efe9 | 4438 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4439 | |
c0ff4b85 | 4440 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4441 | if (ev->eventfd == eventfd) { |
4442 | list_del(&ev->list); | |
4443 | kfree(ev); | |
4444 | } | |
4445 | } | |
4446 | ||
1af8efe9 | 4447 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4448 | } |
4449 | ||
2da8ca82 | 4450 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4451 | { |
2da8ca82 | 4452 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4453 | |
791badbd TH |
4454 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4455 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4456 | return 0; |
4457 | } | |
4458 | ||
182446d0 | 4459 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4460 | struct cftype *cft, u64 val) |
4461 | { | |
182446d0 | 4462 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4463 | |
4464 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4465 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4466 | return -EINVAL; |
4467 | ||
c0ff4b85 | 4468 | memcg->oom_kill_disable = val; |
4d845ebf | 4469 | if (!val) |
c0ff4b85 | 4470 | memcg_oom_recover(memcg); |
3dae7fec | 4471 | |
3c11ecf4 KH |
4472 | return 0; |
4473 | } | |
4474 | ||
c255a458 | 4475 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4476 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4477 | { |
55007d84 GC |
4478 | int ret; |
4479 | ||
2633d7a0 | 4480 | memcg->kmemcg_id = -1; |
55007d84 GC |
4481 | ret = memcg_propagate_kmem(memcg); |
4482 | if (ret) | |
4483 | return ret; | |
2633d7a0 | 4484 | |
1d62e436 | 4485 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4486 | } |
e5671dfa | 4487 | |
10d5ebf4 | 4488 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4489 | { |
1d62e436 | 4490 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 4491 | } |
e5671dfa | 4492 | #else |
cbe128e3 | 4493 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4494 | { |
4495 | return 0; | |
4496 | } | |
d1a4c0b3 | 4497 | |
10d5ebf4 LZ |
4498 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4499 | { | |
4500 | } | |
e5671dfa GC |
4501 | #endif |
4502 | ||
3bc942f3 TH |
4503 | /* |
4504 | * DO NOT USE IN NEW FILES. | |
4505 | * | |
4506 | * "cgroup.event_control" implementation. | |
4507 | * | |
4508 | * This is way over-engineered. It tries to support fully configurable | |
4509 | * events for each user. Such level of flexibility is completely | |
4510 | * unnecessary especially in the light of the planned unified hierarchy. | |
4511 | * | |
4512 | * Please deprecate this and replace with something simpler if at all | |
4513 | * possible. | |
4514 | */ | |
4515 | ||
79bd9814 TH |
4516 | /* |
4517 | * Unregister event and free resources. | |
4518 | * | |
4519 | * Gets called from workqueue. | |
4520 | */ | |
3bc942f3 | 4521 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4522 | { |
3bc942f3 TH |
4523 | struct mem_cgroup_event *event = |
4524 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4525 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4526 | |
4527 | remove_wait_queue(event->wqh, &event->wait); | |
4528 | ||
59b6f873 | 4529 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4530 | |
4531 | /* Notify userspace the event is going away. */ | |
4532 | eventfd_signal(event->eventfd, 1); | |
4533 | ||
4534 | eventfd_ctx_put(event->eventfd); | |
4535 | kfree(event); | |
59b6f873 | 4536 | css_put(&memcg->css); |
79bd9814 TH |
4537 | } |
4538 | ||
4539 | /* | |
4540 | * Gets called on POLLHUP on eventfd when user closes it. | |
4541 | * | |
4542 | * Called with wqh->lock held and interrupts disabled. | |
4543 | */ | |
3bc942f3 TH |
4544 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4545 | int sync, void *key) | |
79bd9814 | 4546 | { |
3bc942f3 TH |
4547 | struct mem_cgroup_event *event = |
4548 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4549 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4550 | unsigned long flags = (unsigned long)key; |
4551 | ||
4552 | if (flags & POLLHUP) { | |
4553 | /* | |
4554 | * If the event has been detached at cgroup removal, we | |
4555 | * can simply return knowing the other side will cleanup | |
4556 | * for us. | |
4557 | * | |
4558 | * We can't race against event freeing since the other | |
4559 | * side will require wqh->lock via remove_wait_queue(), | |
4560 | * which we hold. | |
4561 | */ | |
fba94807 | 4562 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4563 | if (!list_empty(&event->list)) { |
4564 | list_del_init(&event->list); | |
4565 | /* | |
4566 | * We are in atomic context, but cgroup_event_remove() | |
4567 | * may sleep, so we have to call it in workqueue. | |
4568 | */ | |
4569 | schedule_work(&event->remove); | |
4570 | } | |
fba94807 | 4571 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4572 | } |
4573 | ||
4574 | return 0; | |
4575 | } | |
4576 | ||
3bc942f3 | 4577 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4578 | wait_queue_head_t *wqh, poll_table *pt) |
4579 | { | |
3bc942f3 TH |
4580 | struct mem_cgroup_event *event = |
4581 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4582 | |
4583 | event->wqh = wqh; | |
4584 | add_wait_queue(wqh, &event->wait); | |
4585 | } | |
4586 | ||
4587 | /* | |
3bc942f3 TH |
4588 | * DO NOT USE IN NEW FILES. |
4589 | * | |
79bd9814 TH |
4590 | * Parse input and register new cgroup event handler. |
4591 | * | |
4592 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4593 | * Interpretation of args is defined by control file implementation. | |
4594 | */ | |
451af504 TH |
4595 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4596 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4597 | { |
451af504 | 4598 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4599 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4600 | struct mem_cgroup_event *event; |
79bd9814 TH |
4601 | struct cgroup_subsys_state *cfile_css; |
4602 | unsigned int efd, cfd; | |
4603 | struct fd efile; | |
4604 | struct fd cfile; | |
fba94807 | 4605 | const char *name; |
79bd9814 TH |
4606 | char *endp; |
4607 | int ret; | |
4608 | ||
451af504 TH |
4609 | buf = strstrip(buf); |
4610 | ||
4611 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4612 | if (*endp != ' ') |
4613 | return -EINVAL; | |
451af504 | 4614 | buf = endp + 1; |
79bd9814 | 4615 | |
451af504 | 4616 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4617 | if ((*endp != ' ') && (*endp != '\0')) |
4618 | return -EINVAL; | |
451af504 | 4619 | buf = endp + 1; |
79bd9814 TH |
4620 | |
4621 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4622 | if (!event) | |
4623 | return -ENOMEM; | |
4624 | ||
59b6f873 | 4625 | event->memcg = memcg; |
79bd9814 | 4626 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4627 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4628 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4629 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4630 | |
4631 | efile = fdget(efd); | |
4632 | if (!efile.file) { | |
4633 | ret = -EBADF; | |
4634 | goto out_kfree; | |
4635 | } | |
4636 | ||
4637 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4638 | if (IS_ERR(event->eventfd)) { | |
4639 | ret = PTR_ERR(event->eventfd); | |
4640 | goto out_put_efile; | |
4641 | } | |
4642 | ||
4643 | cfile = fdget(cfd); | |
4644 | if (!cfile.file) { | |
4645 | ret = -EBADF; | |
4646 | goto out_put_eventfd; | |
4647 | } | |
4648 | ||
4649 | /* the process need read permission on control file */ | |
4650 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4651 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4652 | if (ret < 0) | |
4653 | goto out_put_cfile; | |
4654 | ||
fba94807 TH |
4655 | /* |
4656 | * Determine the event callbacks and set them in @event. This used | |
4657 | * to be done via struct cftype but cgroup core no longer knows | |
4658 | * about these events. The following is crude but the whole thing | |
4659 | * is for compatibility anyway. | |
3bc942f3 TH |
4660 | * |
4661 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
4662 | */ |
4663 | name = cfile.file->f_dentry->d_name.name; | |
4664 | ||
4665 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4666 | event->register_event = mem_cgroup_usage_register_event; | |
4667 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4668 | } else if (!strcmp(name, "memory.oom_control")) { | |
4669 | event->register_event = mem_cgroup_oom_register_event; | |
4670 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4671 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4672 | event->register_event = vmpressure_register_event; | |
4673 | event->unregister_event = vmpressure_unregister_event; | |
4674 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4675 | event->register_event = memsw_cgroup_usage_register_event; |
4676 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4677 | } else { |
4678 | ret = -EINVAL; | |
4679 | goto out_put_cfile; | |
4680 | } | |
4681 | ||
79bd9814 | 4682 | /* |
b5557c4c TH |
4683 | * Verify @cfile should belong to @css. Also, remaining events are |
4684 | * automatically removed on cgroup destruction but the removal is | |
4685 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4686 | */ |
ec903c0c TH |
4687 | cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent, |
4688 | &memory_cgrp_subsys); | |
79bd9814 | 4689 | ret = -EINVAL; |
5a17f543 | 4690 | if (IS_ERR(cfile_css)) |
79bd9814 | 4691 | goto out_put_cfile; |
5a17f543 TH |
4692 | if (cfile_css != css) { |
4693 | css_put(cfile_css); | |
79bd9814 | 4694 | goto out_put_cfile; |
5a17f543 | 4695 | } |
79bd9814 | 4696 | |
451af504 | 4697 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4698 | if (ret) |
4699 | goto out_put_css; | |
4700 | ||
4701 | efile.file->f_op->poll(efile.file, &event->pt); | |
4702 | ||
fba94807 TH |
4703 | spin_lock(&memcg->event_list_lock); |
4704 | list_add(&event->list, &memcg->event_list); | |
4705 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4706 | |
4707 | fdput(cfile); | |
4708 | fdput(efile); | |
4709 | ||
451af504 | 4710 | return nbytes; |
79bd9814 TH |
4711 | |
4712 | out_put_css: | |
b5557c4c | 4713 | css_put(css); |
79bd9814 TH |
4714 | out_put_cfile: |
4715 | fdput(cfile); | |
4716 | out_put_eventfd: | |
4717 | eventfd_ctx_put(event->eventfd); | |
4718 | out_put_efile: | |
4719 | fdput(efile); | |
4720 | out_kfree: | |
4721 | kfree(event); | |
4722 | ||
4723 | return ret; | |
4724 | } | |
4725 | ||
8cdea7c0 BS |
4726 | static struct cftype mem_cgroup_files[] = { |
4727 | { | |
0eea1030 | 4728 | .name = "usage_in_bytes", |
8c7c6e34 | 4729 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4730 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4731 | }, |
c84872e1 PE |
4732 | { |
4733 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4734 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4735 | .write = mem_cgroup_reset, |
791badbd | 4736 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4737 | }, |
8cdea7c0 | 4738 | { |
0eea1030 | 4739 | .name = "limit_in_bytes", |
8c7c6e34 | 4740 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4741 | .write = mem_cgroup_write, |
791badbd | 4742 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4743 | }, |
296c81d8 BS |
4744 | { |
4745 | .name = "soft_limit_in_bytes", | |
4746 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4747 | .write = mem_cgroup_write, |
791badbd | 4748 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4749 | }, |
8cdea7c0 BS |
4750 | { |
4751 | .name = "failcnt", | |
8c7c6e34 | 4752 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4753 | .write = mem_cgroup_reset, |
791badbd | 4754 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4755 | }, |
d2ceb9b7 KH |
4756 | { |
4757 | .name = "stat", | |
2da8ca82 | 4758 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4759 | }, |
c1e862c1 KH |
4760 | { |
4761 | .name = "force_empty", | |
6770c64e | 4762 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4763 | }, |
18f59ea7 BS |
4764 | { |
4765 | .name = "use_hierarchy", | |
4766 | .write_u64 = mem_cgroup_hierarchy_write, | |
4767 | .read_u64 = mem_cgroup_hierarchy_read, | |
4768 | }, | |
79bd9814 | 4769 | { |
3bc942f3 | 4770 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4771 | .write = memcg_write_event_control, |
79bd9814 TH |
4772 | .flags = CFTYPE_NO_PREFIX, |
4773 | .mode = S_IWUGO, | |
4774 | }, | |
a7885eb8 KM |
4775 | { |
4776 | .name = "swappiness", | |
4777 | .read_u64 = mem_cgroup_swappiness_read, | |
4778 | .write_u64 = mem_cgroup_swappiness_write, | |
4779 | }, | |
7dc74be0 DN |
4780 | { |
4781 | .name = "move_charge_at_immigrate", | |
4782 | .read_u64 = mem_cgroup_move_charge_read, | |
4783 | .write_u64 = mem_cgroup_move_charge_write, | |
4784 | }, | |
9490ff27 KH |
4785 | { |
4786 | .name = "oom_control", | |
2da8ca82 | 4787 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4788 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4789 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4790 | }, | |
70ddf637 AV |
4791 | { |
4792 | .name = "pressure_level", | |
70ddf637 | 4793 | }, |
406eb0c9 YH |
4794 | #ifdef CONFIG_NUMA |
4795 | { | |
4796 | .name = "numa_stat", | |
2da8ca82 | 4797 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4798 | }, |
4799 | #endif | |
510fc4e1 GC |
4800 | #ifdef CONFIG_MEMCG_KMEM |
4801 | { | |
4802 | .name = "kmem.limit_in_bytes", | |
4803 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4804 | .write = mem_cgroup_write, |
791badbd | 4805 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4806 | }, |
4807 | { | |
4808 | .name = "kmem.usage_in_bytes", | |
4809 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4810 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4811 | }, |
4812 | { | |
4813 | .name = "kmem.failcnt", | |
4814 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4815 | .write = mem_cgroup_reset, |
791badbd | 4816 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4817 | }, |
4818 | { | |
4819 | .name = "kmem.max_usage_in_bytes", | |
4820 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4821 | .write = mem_cgroup_reset, |
791badbd | 4822 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4823 | }, |
749c5415 GC |
4824 | #ifdef CONFIG_SLABINFO |
4825 | { | |
4826 | .name = "kmem.slabinfo", | |
2da8ca82 | 4827 | .seq_show = mem_cgroup_slabinfo_read, |
749c5415 GC |
4828 | }, |
4829 | #endif | |
8c7c6e34 | 4830 | #endif |
6bc10349 | 4831 | { }, /* terminate */ |
af36f906 | 4832 | }; |
8c7c6e34 | 4833 | |
2d11085e MH |
4834 | #ifdef CONFIG_MEMCG_SWAP |
4835 | static struct cftype memsw_cgroup_files[] = { | |
4836 | { | |
4837 | .name = "memsw.usage_in_bytes", | |
4838 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 4839 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4840 | }, |
4841 | { | |
4842 | .name = "memsw.max_usage_in_bytes", | |
4843 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 4844 | .write = mem_cgroup_reset, |
791badbd | 4845 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4846 | }, |
4847 | { | |
4848 | .name = "memsw.limit_in_bytes", | |
4849 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 4850 | .write = mem_cgroup_write, |
791badbd | 4851 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4852 | }, |
4853 | { | |
4854 | .name = "memsw.failcnt", | |
4855 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 4856 | .write = mem_cgroup_reset, |
791badbd | 4857 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4858 | }, |
4859 | { }, /* terminate */ | |
4860 | }; | |
4861 | #endif | |
c0ff4b85 | 4862 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4863 | { |
4864 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4865 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4866 | int zone, tmp = node; |
1ecaab2b KH |
4867 | /* |
4868 | * This routine is called against possible nodes. | |
4869 | * But it's BUG to call kmalloc() against offline node. | |
4870 | * | |
4871 | * TODO: this routine can waste much memory for nodes which will | |
4872 | * never be onlined. It's better to use memory hotplug callback | |
4873 | * function. | |
4874 | */ | |
41e3355d KH |
4875 | if (!node_state(node, N_NORMAL_MEMORY)) |
4876 | tmp = -1; | |
17295c88 | 4877 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4878 | if (!pn) |
4879 | return 1; | |
1ecaab2b | 4880 | |
1ecaab2b KH |
4881 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4882 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4883 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4884 | mz->usage_in_excess = 0; |
4885 | mz->on_tree = false; | |
d79154bb | 4886 | mz->memcg = memcg; |
1ecaab2b | 4887 | } |
54f72fe0 | 4888 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4889 | return 0; |
4890 | } | |
4891 | ||
c0ff4b85 | 4892 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4893 | { |
54f72fe0 | 4894 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4895 | } |
4896 | ||
33327948 KH |
4897 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4898 | { | |
d79154bb | 4899 | struct mem_cgroup *memcg; |
8ff69e2c | 4900 | size_t size; |
33327948 | 4901 | |
8ff69e2c VD |
4902 | size = sizeof(struct mem_cgroup); |
4903 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4904 | |
8ff69e2c | 4905 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4906 | if (!memcg) |
e7bbcdf3 DC |
4907 | return NULL; |
4908 | ||
d79154bb HD |
4909 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4910 | if (!memcg->stat) | |
d2e61b8d | 4911 | goto out_free; |
d79154bb HD |
4912 | spin_lock_init(&memcg->pcp_counter_lock); |
4913 | return memcg; | |
d2e61b8d DC |
4914 | |
4915 | out_free: | |
8ff69e2c | 4916 | kfree(memcg); |
d2e61b8d | 4917 | return NULL; |
33327948 KH |
4918 | } |
4919 | ||
59927fb9 | 4920 | /* |
c8b2a36f GC |
4921 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4922 | * (scanning all at force_empty is too costly...) | |
4923 | * | |
4924 | * Instead of clearing all references at force_empty, we remember | |
4925 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4926 | * it goes down to 0. | |
4927 | * | |
4928 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4929 | */ |
c8b2a36f GC |
4930 | |
4931 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4932 | { |
c8b2a36f | 4933 | int node; |
59927fb9 | 4934 | |
bb4cc1a8 | 4935 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4936 | |
4937 | for_each_node(node) | |
4938 | free_mem_cgroup_per_zone_info(memcg, node); | |
4939 | ||
4940 | free_percpu(memcg->stat); | |
4941 | ||
3f134619 GC |
4942 | /* |
4943 | * We need to make sure that (at least for now), the jump label | |
4944 | * destruction code runs outside of the cgroup lock. This is because | |
4945 | * get_online_cpus(), which is called from the static_branch update, | |
4946 | * can't be called inside the cgroup_lock. cpusets are the ones | |
4947 | * enforcing this dependency, so if they ever change, we might as well. | |
4948 | * | |
4949 | * schedule_work() will guarantee this happens. Be careful if you need | |
4950 | * to move this code around, and make sure it is outside | |
4951 | * the cgroup_lock. | |
4952 | */ | |
a8964b9b | 4953 | disarm_static_keys(memcg); |
8ff69e2c | 4954 | kfree(memcg); |
59927fb9 | 4955 | } |
3afe36b1 | 4956 | |
7bcc1bb1 DN |
4957 | /* |
4958 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4959 | */ | |
e1aab161 | 4960 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4961 | { |
3e32cb2e | 4962 | if (!memcg->memory.parent) |
7bcc1bb1 | 4963 | return NULL; |
3e32cb2e | 4964 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4965 | } |
e1aab161 | 4966 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4967 | |
bb4cc1a8 AM |
4968 | static void __init mem_cgroup_soft_limit_tree_init(void) |
4969 | { | |
4970 | struct mem_cgroup_tree_per_node *rtpn; | |
4971 | struct mem_cgroup_tree_per_zone *rtpz; | |
4972 | int tmp, node, zone; | |
4973 | ||
4974 | for_each_node(node) { | |
4975 | tmp = node; | |
4976 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4977 | tmp = -1; | |
4978 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4979 | BUG_ON(!rtpn); | |
4980 | ||
4981 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4982 | ||
4983 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4984 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4985 | rtpz->rb_root = RB_ROOT; | |
4986 | spin_lock_init(&rtpz->lock); | |
4987 | } | |
4988 | } | |
4989 | } | |
4990 | ||
0eb253e2 | 4991 | static struct cgroup_subsys_state * __ref |
eb95419b | 4992 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4993 | { |
d142e3e6 | 4994 | struct mem_cgroup *memcg; |
04046e1a | 4995 | long error = -ENOMEM; |
6d12e2d8 | 4996 | int node; |
8cdea7c0 | 4997 | |
c0ff4b85 R |
4998 | memcg = mem_cgroup_alloc(); |
4999 | if (!memcg) | |
04046e1a | 5000 | return ERR_PTR(error); |
78fb7466 | 5001 | |
3ed28fa1 | 5002 | for_each_node(node) |
c0ff4b85 | 5003 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5004 | goto free_out; |
f64c3f54 | 5005 | |
c077719b | 5006 | /* root ? */ |
eb95419b | 5007 | if (parent_css == NULL) { |
a41c58a6 | 5008 | root_mem_cgroup = memcg; |
3e32cb2e JW |
5009 | page_counter_init(&memcg->memory, NULL); |
5010 | page_counter_init(&memcg->memsw, NULL); | |
5011 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 5012 | } |
28dbc4b6 | 5013 | |
d142e3e6 GC |
5014 | memcg->last_scanned_node = MAX_NUMNODES; |
5015 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
5016 | memcg->move_charge_at_immigrate = 0; |
5017 | mutex_init(&memcg->thresholds_lock); | |
5018 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5019 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5020 | INIT_LIST_HEAD(&memcg->event_list); |
5021 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
5022 | |
5023 | return &memcg->css; | |
5024 | ||
5025 | free_out: | |
5026 | __mem_cgroup_free(memcg); | |
5027 | return ERR_PTR(error); | |
5028 | } | |
5029 | ||
5030 | static int | |
eb95419b | 5031 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 5032 | { |
eb95419b | 5033 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 5034 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 5035 | int ret; |
d142e3e6 | 5036 | |
15a4c835 | 5037 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
5038 | return -ENOSPC; |
5039 | ||
63876986 | 5040 | if (!parent) |
d142e3e6 GC |
5041 | return 0; |
5042 | ||
0999821b | 5043 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
5044 | |
5045 | memcg->use_hierarchy = parent->use_hierarchy; | |
5046 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5047 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5048 | ||
5049 | if (parent->use_hierarchy) { | |
3e32cb2e JW |
5050 | page_counter_init(&memcg->memory, &parent->memory); |
5051 | page_counter_init(&memcg->memsw, &parent->memsw); | |
5052 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 5053 | |
7bcc1bb1 | 5054 | /* |
8d76a979 LZ |
5055 | * No need to take a reference to the parent because cgroup |
5056 | * core guarantees its existence. | |
7bcc1bb1 | 5057 | */ |
18f59ea7 | 5058 | } else { |
3e32cb2e JW |
5059 | page_counter_init(&memcg->memory, NULL); |
5060 | page_counter_init(&memcg->memsw, NULL); | |
5061 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
5062 | /* |
5063 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5064 | * much sense so let cgroup subsystem know about this | |
5065 | * unfortunate state in our controller. | |
5066 | */ | |
d142e3e6 | 5067 | if (parent != root_mem_cgroup) |
073219e9 | 5068 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5069 | } |
0999821b | 5070 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 5071 | |
2f7dd7a4 JW |
5072 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
5073 | if (ret) | |
5074 | return ret; | |
5075 | ||
5076 | /* | |
5077 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
5078 | * orders reading memcg->initialized against its callers | |
5079 | * reading the memcg members. | |
5080 | */ | |
5081 | smp_store_release(&memcg->initialized, 1); | |
5082 | ||
5083 | return 0; | |
8cdea7c0 BS |
5084 | } |
5085 | ||
eb95419b | 5086 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5087 | { |
eb95419b | 5088 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5089 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5090 | |
5091 | /* | |
5092 | * Unregister events and notify userspace. | |
5093 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5094 | * directory to avoid race between userspace and kernelspace. | |
5095 | */ | |
fba94807 TH |
5096 | spin_lock(&memcg->event_list_lock); |
5097 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5098 | list_del_init(&event->list); |
5099 | schedule_work(&event->remove); | |
5100 | } | |
fba94807 | 5101 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5102 | |
776ed0f0 | 5103 | memcg_unregister_all_caches(memcg); |
33cb876e | 5104 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
5105 | } |
5106 | ||
eb95419b | 5107 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5108 | { |
eb95419b | 5109 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 5110 | |
10d5ebf4 | 5111 | memcg_destroy_kmem(memcg); |
465939a1 | 5112 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
5113 | } |
5114 | ||
1ced953b TH |
5115 | /** |
5116 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5117 | * @css: the target css | |
5118 | * | |
5119 | * Reset the states of the mem_cgroup associated with @css. This is | |
5120 | * invoked when the userland requests disabling on the default hierarchy | |
5121 | * but the memcg is pinned through dependency. The memcg should stop | |
5122 | * applying policies and should revert to the vanilla state as it may be | |
5123 | * made visible again. | |
5124 | * | |
5125 | * The current implementation only resets the essential configurations. | |
5126 | * This needs to be expanded to cover all the visible parts. | |
5127 | */ | |
5128 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5129 | { | |
5130 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5131 | ||
3e32cb2e JW |
5132 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
5133 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
5134 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
5135 | memcg->soft_limit = 0; | |
1ced953b TH |
5136 | } |
5137 | ||
02491447 | 5138 | #ifdef CONFIG_MMU |
7dc74be0 | 5139 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5140 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5141 | { |
05b84301 | 5142 | int ret; |
9476db97 JW |
5143 | |
5144 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 5145 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 5146 | if (!ret) { |
854ffa8d | 5147 | mc.precharge += count; |
854ffa8d DN |
5148 | return ret; |
5149 | } | |
692e7c45 | 5150 | if (ret == -EINTR) { |
00501b53 | 5151 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
5152 | return ret; |
5153 | } | |
9476db97 JW |
5154 | |
5155 | /* Try charges one by one with reclaim */ | |
854ffa8d | 5156 | while (count--) { |
00501b53 | 5157 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
5158 | /* |
5159 | * In case of failure, any residual charges against | |
5160 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
5161 | * later on. However, cancel any charges that are |
5162 | * bypassed to root right away or they'll be lost. | |
9476db97 | 5163 | */ |
692e7c45 | 5164 | if (ret == -EINTR) |
00501b53 | 5165 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 5166 | if (ret) |
38c5d72f | 5167 | return ret; |
854ffa8d | 5168 | mc.precharge++; |
9476db97 | 5169 | cond_resched(); |
854ffa8d | 5170 | } |
9476db97 | 5171 | return 0; |
4ffef5fe DN |
5172 | } |
5173 | ||
5174 | /** | |
8d32ff84 | 5175 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5176 | * @vma: the vma the pte to be checked belongs |
5177 | * @addr: the address corresponding to the pte to be checked | |
5178 | * @ptent: the pte to be checked | |
02491447 | 5179 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5180 | * |
5181 | * Returns | |
5182 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5183 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5184 | * move charge. if @target is not NULL, the page is stored in target->page | |
5185 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5186 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5187 | * target for charge migration. if @target is not NULL, the entry is stored | |
5188 | * in target->ent. | |
4ffef5fe DN |
5189 | * |
5190 | * Called with pte lock held. | |
5191 | */ | |
4ffef5fe DN |
5192 | union mc_target { |
5193 | struct page *page; | |
02491447 | 5194 | swp_entry_t ent; |
4ffef5fe DN |
5195 | }; |
5196 | ||
4ffef5fe | 5197 | enum mc_target_type { |
8d32ff84 | 5198 | MC_TARGET_NONE = 0, |
4ffef5fe | 5199 | MC_TARGET_PAGE, |
02491447 | 5200 | MC_TARGET_SWAP, |
4ffef5fe DN |
5201 | }; |
5202 | ||
90254a65 DN |
5203 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5204 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5205 | { |
90254a65 | 5206 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5207 | |
90254a65 DN |
5208 | if (!page || !page_mapped(page)) |
5209 | return NULL; | |
5210 | if (PageAnon(page)) { | |
5211 | /* we don't move shared anon */ | |
4b91355e | 5212 | if (!move_anon()) |
90254a65 | 5213 | return NULL; |
87946a72 DN |
5214 | } else if (!move_file()) |
5215 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5216 | return NULL; |
5217 | if (!get_page_unless_zero(page)) | |
5218 | return NULL; | |
5219 | ||
5220 | return page; | |
5221 | } | |
5222 | ||
4b91355e | 5223 | #ifdef CONFIG_SWAP |
90254a65 DN |
5224 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5225 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5226 | { | |
90254a65 DN |
5227 | struct page *page = NULL; |
5228 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5229 | ||
5230 | if (!move_anon() || non_swap_entry(ent)) | |
5231 | return NULL; | |
4b91355e KH |
5232 | /* |
5233 | * Because lookup_swap_cache() updates some statistics counter, | |
5234 | * we call find_get_page() with swapper_space directly. | |
5235 | */ | |
33806f06 | 5236 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
5237 | if (do_swap_account) |
5238 | entry->val = ent.val; | |
5239 | ||
5240 | return page; | |
5241 | } | |
4b91355e KH |
5242 | #else |
5243 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5244 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5245 | { | |
5246 | return NULL; | |
5247 | } | |
5248 | #endif | |
90254a65 | 5249 | |
87946a72 DN |
5250 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5251 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5252 | { | |
5253 | struct page *page = NULL; | |
87946a72 DN |
5254 | struct address_space *mapping; |
5255 | pgoff_t pgoff; | |
5256 | ||
5257 | if (!vma->vm_file) /* anonymous vma */ | |
5258 | return NULL; | |
5259 | if (!move_file()) | |
5260 | return NULL; | |
5261 | ||
87946a72 DN |
5262 | mapping = vma->vm_file->f_mapping; |
5263 | if (pte_none(ptent)) | |
5264 | pgoff = linear_page_index(vma, addr); | |
5265 | else /* pte_file(ptent) is true */ | |
5266 | pgoff = pte_to_pgoff(ptent); | |
5267 | ||
5268 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5269 | #ifdef CONFIG_SWAP |
5270 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5271 | if (shmem_mapping(mapping)) { |
5272 | page = find_get_entry(mapping, pgoff); | |
5273 | if (radix_tree_exceptional_entry(page)) { | |
5274 | swp_entry_t swp = radix_to_swp_entry(page); | |
5275 | if (do_swap_account) | |
5276 | *entry = swp; | |
5277 | page = find_get_page(swap_address_space(swp), swp.val); | |
5278 | } | |
5279 | } else | |
5280 | page = find_get_page(mapping, pgoff); | |
5281 | #else | |
5282 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5283 | #endif |
87946a72 DN |
5284 | return page; |
5285 | } | |
5286 | ||
8d32ff84 | 5287 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5288 | unsigned long addr, pte_t ptent, union mc_target *target) |
5289 | { | |
5290 | struct page *page = NULL; | |
5291 | struct page_cgroup *pc; | |
8d32ff84 | 5292 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5293 | swp_entry_t ent = { .val = 0 }; |
5294 | ||
5295 | if (pte_present(ptent)) | |
5296 | page = mc_handle_present_pte(vma, addr, ptent); | |
5297 | else if (is_swap_pte(ptent)) | |
5298 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5299 | else if (pte_none(ptent) || pte_file(ptent)) |
5300 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5301 | |
5302 | if (!page && !ent.val) | |
8d32ff84 | 5303 | return ret; |
02491447 DN |
5304 | if (page) { |
5305 | pc = lookup_page_cgroup(page); | |
5306 | /* | |
0a31bc97 JW |
5307 | * Do only loose check w/o serialization. |
5308 | * mem_cgroup_move_account() checks the pc is valid or | |
5309 | * not under LRU exclusion. | |
02491447 DN |
5310 | */ |
5311 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5312 | ret = MC_TARGET_PAGE; | |
5313 | if (target) | |
5314 | target->page = page; | |
5315 | } | |
5316 | if (!ret || !target) | |
5317 | put_page(page); | |
5318 | } | |
90254a65 DN |
5319 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5320 | if (ent.val && !ret && | |
34c00c31 | 5321 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5322 | ret = MC_TARGET_SWAP; |
5323 | if (target) | |
5324 | target->ent = ent; | |
4ffef5fe | 5325 | } |
4ffef5fe DN |
5326 | return ret; |
5327 | } | |
5328 | ||
12724850 NH |
5329 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5330 | /* | |
5331 | * We don't consider swapping or file mapped pages because THP does not | |
5332 | * support them for now. | |
5333 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5334 | */ | |
5335 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5336 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5337 | { | |
5338 | struct page *page = NULL; | |
5339 | struct page_cgroup *pc; | |
5340 | enum mc_target_type ret = MC_TARGET_NONE; | |
5341 | ||
5342 | page = pmd_page(pmd); | |
309381fe | 5343 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
5344 | if (!move_anon()) |
5345 | return ret; | |
5346 | pc = lookup_page_cgroup(page); | |
5347 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5348 | ret = MC_TARGET_PAGE; | |
5349 | if (target) { | |
5350 | get_page(page); | |
5351 | target->page = page; | |
5352 | } | |
5353 | } | |
5354 | return ret; | |
5355 | } | |
5356 | #else | |
5357 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5358 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5359 | { | |
5360 | return MC_TARGET_NONE; | |
5361 | } | |
5362 | #endif | |
5363 | ||
4ffef5fe DN |
5364 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5365 | unsigned long addr, unsigned long end, | |
5366 | struct mm_walk *walk) | |
5367 | { | |
5368 | struct vm_area_struct *vma = walk->private; | |
5369 | pte_t *pte; | |
5370 | spinlock_t *ptl; | |
5371 | ||
bf929152 | 5372 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
5373 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5374 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5375 | spin_unlock(ptl); |
1a5a9906 | 5376 | return 0; |
12724850 | 5377 | } |
03319327 | 5378 | |
45f83cef AA |
5379 | if (pmd_trans_unstable(pmd)) |
5380 | return 0; | |
4ffef5fe DN |
5381 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5382 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5383 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5384 | mc.precharge++; /* increment precharge temporarily */ |
5385 | pte_unmap_unlock(pte - 1, ptl); | |
5386 | cond_resched(); | |
5387 | ||
7dc74be0 DN |
5388 | return 0; |
5389 | } | |
5390 | ||
4ffef5fe DN |
5391 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5392 | { | |
5393 | unsigned long precharge; | |
5394 | struct vm_area_struct *vma; | |
5395 | ||
dfe076b0 | 5396 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5397 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5398 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5399 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5400 | .mm = mm, | |
5401 | .private = vma, | |
5402 | }; | |
5403 | if (is_vm_hugetlb_page(vma)) | |
5404 | continue; | |
4ffef5fe DN |
5405 | walk_page_range(vma->vm_start, vma->vm_end, |
5406 | &mem_cgroup_count_precharge_walk); | |
5407 | } | |
dfe076b0 | 5408 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5409 | |
5410 | precharge = mc.precharge; | |
5411 | mc.precharge = 0; | |
5412 | ||
5413 | return precharge; | |
5414 | } | |
5415 | ||
4ffef5fe DN |
5416 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5417 | { | |
dfe076b0 DN |
5418 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5419 | ||
5420 | VM_BUG_ON(mc.moving_task); | |
5421 | mc.moving_task = current; | |
5422 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5423 | } |
5424 | ||
dfe076b0 DN |
5425 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5426 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5427 | { |
2bd9bb20 KH |
5428 | struct mem_cgroup *from = mc.from; |
5429 | struct mem_cgroup *to = mc.to; | |
5430 | ||
4ffef5fe | 5431 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5432 | if (mc.precharge) { |
00501b53 | 5433 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5434 | mc.precharge = 0; |
5435 | } | |
5436 | /* | |
5437 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5438 | * we must uncharge here. | |
5439 | */ | |
5440 | if (mc.moved_charge) { | |
00501b53 | 5441 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5442 | mc.moved_charge = 0; |
4ffef5fe | 5443 | } |
483c30b5 DN |
5444 | /* we must fixup refcnts and charges */ |
5445 | if (mc.moved_swap) { | |
483c30b5 | 5446 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5447 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5448 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5449 | |
05b84301 | 5450 | /* |
3e32cb2e JW |
5451 | * we charged both to->memory and to->memsw, so we |
5452 | * should uncharge to->memory. | |
05b84301 | 5453 | */ |
ce00a967 | 5454 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5455 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5456 | ||
e8ea14cc | 5457 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 5458 | |
4050377b | 5459 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
5460 | mc.moved_swap = 0; |
5461 | } | |
dfe076b0 DN |
5462 | memcg_oom_recover(from); |
5463 | memcg_oom_recover(to); | |
5464 | wake_up_all(&mc.waitq); | |
5465 | } | |
5466 | ||
5467 | static void mem_cgroup_clear_mc(void) | |
5468 | { | |
5469 | struct mem_cgroup *from = mc.from; | |
5470 | ||
5471 | /* | |
5472 | * we must clear moving_task before waking up waiters at the end of | |
5473 | * task migration. | |
5474 | */ | |
5475 | mc.moving_task = NULL; | |
5476 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5477 | spin_lock(&mc.lock); |
4ffef5fe DN |
5478 | mc.from = NULL; |
5479 | mc.to = NULL; | |
2bd9bb20 | 5480 | spin_unlock(&mc.lock); |
32047e2a | 5481 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5482 | } |
5483 | ||
eb95419b | 5484 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5485 | struct cgroup_taskset *tset) |
7dc74be0 | 5486 | { |
2f7ee569 | 5487 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5488 | int ret = 0; |
eb95419b | 5489 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 5490 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 5491 | |
ee5e8472 GC |
5492 | /* |
5493 | * We are now commited to this value whatever it is. Changes in this | |
5494 | * tunable will only affect upcoming migrations, not the current one. | |
5495 | * So we need to save it, and keep it going. | |
5496 | */ | |
5497 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
5498 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
5499 | struct mm_struct *mm; |
5500 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5501 | ||
c0ff4b85 | 5502 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5503 | |
5504 | mm = get_task_mm(p); | |
5505 | if (!mm) | |
5506 | return 0; | |
7dc74be0 | 5507 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5508 | if (mm->owner == p) { |
5509 | VM_BUG_ON(mc.from); | |
5510 | VM_BUG_ON(mc.to); | |
5511 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5512 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5513 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5514 | mem_cgroup_start_move(from); |
2bd9bb20 | 5515 | spin_lock(&mc.lock); |
4ffef5fe | 5516 | mc.from = from; |
c0ff4b85 | 5517 | mc.to = memcg; |
ee5e8472 | 5518 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 5519 | spin_unlock(&mc.lock); |
dfe076b0 | 5520 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5521 | |
5522 | ret = mem_cgroup_precharge_mc(mm); | |
5523 | if (ret) | |
5524 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5525 | } |
5526 | mmput(mm); | |
7dc74be0 DN |
5527 | } |
5528 | return ret; | |
5529 | } | |
5530 | ||
eb95419b | 5531 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5532 | struct cgroup_taskset *tset) |
7dc74be0 | 5533 | { |
4ffef5fe | 5534 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5535 | } |
5536 | ||
4ffef5fe DN |
5537 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5538 | unsigned long addr, unsigned long end, | |
5539 | struct mm_walk *walk) | |
7dc74be0 | 5540 | { |
4ffef5fe DN |
5541 | int ret = 0; |
5542 | struct vm_area_struct *vma = walk->private; | |
5543 | pte_t *pte; | |
5544 | spinlock_t *ptl; | |
12724850 NH |
5545 | enum mc_target_type target_type; |
5546 | union mc_target target; | |
5547 | struct page *page; | |
5548 | struct page_cgroup *pc; | |
4ffef5fe | 5549 | |
12724850 NH |
5550 | /* |
5551 | * We don't take compound_lock() here but no race with splitting thp | |
5552 | * happens because: | |
5553 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5554 | * under splitting, which means there's no concurrent thp split, | |
5555 | * - if another thread runs into split_huge_page() just after we | |
5556 | * entered this if-block, the thread must wait for page table lock | |
5557 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5558 | * part of thp split is not executed yet. | |
5559 | */ | |
bf929152 | 5560 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5561 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5562 | spin_unlock(ptl); |
12724850 NH |
5563 | return 0; |
5564 | } | |
5565 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5566 | if (target_type == MC_TARGET_PAGE) { | |
5567 | page = target.page; | |
5568 | if (!isolate_lru_page(page)) { | |
5569 | pc = lookup_page_cgroup(page); | |
5570 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5571 | pc, mc.from, mc.to)) { |
12724850 NH |
5572 | mc.precharge -= HPAGE_PMD_NR; |
5573 | mc.moved_charge += HPAGE_PMD_NR; | |
5574 | } | |
5575 | putback_lru_page(page); | |
5576 | } | |
5577 | put_page(page); | |
5578 | } | |
bf929152 | 5579 | spin_unlock(ptl); |
1a5a9906 | 5580 | return 0; |
12724850 NH |
5581 | } |
5582 | ||
45f83cef AA |
5583 | if (pmd_trans_unstable(pmd)) |
5584 | return 0; | |
4ffef5fe DN |
5585 | retry: |
5586 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5587 | for (; addr != end; addr += PAGE_SIZE) { | |
5588 | pte_t ptent = *(pte++); | |
02491447 | 5589 | swp_entry_t ent; |
4ffef5fe DN |
5590 | |
5591 | if (!mc.precharge) | |
5592 | break; | |
5593 | ||
8d32ff84 | 5594 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5595 | case MC_TARGET_PAGE: |
5596 | page = target.page; | |
5597 | if (isolate_lru_page(page)) | |
5598 | goto put; | |
5599 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5600 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5601 | mc.from, mc.to)) { |
4ffef5fe | 5602 | mc.precharge--; |
854ffa8d DN |
5603 | /* we uncharge from mc.from later. */ |
5604 | mc.moved_charge++; | |
4ffef5fe DN |
5605 | } |
5606 | putback_lru_page(page); | |
8d32ff84 | 5607 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5608 | put_page(page); |
5609 | break; | |
02491447 DN |
5610 | case MC_TARGET_SWAP: |
5611 | ent = target.ent; | |
e91cbb42 | 5612 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5613 | mc.precharge--; |
483c30b5 DN |
5614 | /* we fixup refcnts and charges later. */ |
5615 | mc.moved_swap++; | |
5616 | } | |
02491447 | 5617 | break; |
4ffef5fe DN |
5618 | default: |
5619 | break; | |
5620 | } | |
5621 | } | |
5622 | pte_unmap_unlock(pte - 1, ptl); | |
5623 | cond_resched(); | |
5624 | ||
5625 | if (addr != end) { | |
5626 | /* | |
5627 | * We have consumed all precharges we got in can_attach(). | |
5628 | * We try charge one by one, but don't do any additional | |
5629 | * charges to mc.to if we have failed in charge once in attach() | |
5630 | * phase. | |
5631 | */ | |
854ffa8d | 5632 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5633 | if (!ret) |
5634 | goto retry; | |
5635 | } | |
5636 | ||
5637 | return ret; | |
5638 | } | |
5639 | ||
5640 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5641 | { | |
5642 | struct vm_area_struct *vma; | |
5643 | ||
5644 | lru_add_drain_all(); | |
dfe076b0 DN |
5645 | retry: |
5646 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5647 | /* | |
5648 | * Someone who are holding the mmap_sem might be waiting in | |
5649 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5650 | * and retry. Because we cancel precharges, we might not be able | |
5651 | * to move enough charges, but moving charge is a best-effort | |
5652 | * feature anyway, so it wouldn't be a big problem. | |
5653 | */ | |
5654 | __mem_cgroup_clear_mc(); | |
5655 | cond_resched(); | |
5656 | goto retry; | |
5657 | } | |
4ffef5fe DN |
5658 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5659 | int ret; | |
5660 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5661 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5662 | .mm = mm, | |
5663 | .private = vma, | |
5664 | }; | |
5665 | if (is_vm_hugetlb_page(vma)) | |
5666 | continue; | |
4ffef5fe DN |
5667 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5668 | &mem_cgroup_move_charge_walk); | |
5669 | if (ret) | |
5670 | /* | |
5671 | * means we have consumed all precharges and failed in | |
5672 | * doing additional charge. Just abandon here. | |
5673 | */ | |
5674 | break; | |
5675 | } | |
dfe076b0 | 5676 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5677 | } |
5678 | ||
eb95419b | 5679 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5680 | struct cgroup_taskset *tset) |
67e465a7 | 5681 | { |
2f7ee569 | 5682 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5683 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5684 | |
dfe076b0 | 5685 | if (mm) { |
a433658c KM |
5686 | if (mc.to) |
5687 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5688 | mmput(mm); |
5689 | } | |
a433658c KM |
5690 | if (mc.to) |
5691 | mem_cgroup_clear_mc(); | |
67e465a7 | 5692 | } |
5cfb80a7 | 5693 | #else /* !CONFIG_MMU */ |
eb95419b | 5694 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5695 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5696 | { |
5697 | return 0; | |
5698 | } | |
eb95419b | 5699 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5700 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5701 | { |
5702 | } | |
eb95419b | 5703 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5704 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5705 | { |
5706 | } | |
5707 | #endif | |
67e465a7 | 5708 | |
f00baae7 TH |
5709 | /* |
5710 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5711 | * to verify whether we're attached to the default hierarchy on each mount |
5712 | * attempt. | |
f00baae7 | 5713 | */ |
eb95419b | 5714 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5715 | { |
5716 | /* | |
aa6ec29b | 5717 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5718 | * guarantees that @root doesn't have any children, so turning it |
5719 | * on for the root memcg is enough. | |
5720 | */ | |
aa6ec29b | 5721 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 5722 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
5723 | } |
5724 | ||
073219e9 | 5725 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5726 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5727 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5728 | .css_offline = mem_cgroup_css_offline, |
5729 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5730 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5731 | .can_attach = mem_cgroup_can_attach, |
5732 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5733 | .attach = mem_cgroup_move_task, |
f00baae7 | 5734 | .bind = mem_cgroup_bind, |
5577964e | 5735 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 5736 | .early_init = 0, |
8cdea7c0 | 5737 | }; |
c077719b | 5738 | |
c255a458 | 5739 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5740 | static int __init enable_swap_account(char *s) |
5741 | { | |
a2c8990a | 5742 | if (!strcmp(s, "1")) |
a42c390c | 5743 | really_do_swap_account = 1; |
a2c8990a | 5744 | else if (!strcmp(s, "0")) |
a42c390c MH |
5745 | really_do_swap_account = 0; |
5746 | return 1; | |
5747 | } | |
a2c8990a | 5748 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5749 | |
2d11085e MH |
5750 | static void __init memsw_file_init(void) |
5751 | { | |
2cf669a5 TH |
5752 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
5753 | memsw_cgroup_files)); | |
6acc8b02 MH |
5754 | } |
5755 | ||
5756 | static void __init enable_swap_cgroup(void) | |
5757 | { | |
5758 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5759 | do_swap_account = 1; | |
5760 | memsw_file_init(); | |
5761 | } | |
2d11085e | 5762 | } |
6acc8b02 | 5763 | |
2d11085e | 5764 | #else |
6acc8b02 | 5765 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
5766 | { |
5767 | } | |
c077719b | 5768 | #endif |
2d11085e | 5769 | |
0a31bc97 JW |
5770 | #ifdef CONFIG_MEMCG_SWAP |
5771 | /** | |
5772 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5773 | * @page: page whose memsw charge to transfer | |
5774 | * @entry: swap entry to move the charge to | |
5775 | * | |
5776 | * Transfer the memsw charge of @page to @entry. | |
5777 | */ | |
5778 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5779 | { | |
5780 | struct page_cgroup *pc; | |
5781 | unsigned short oldid; | |
5782 | ||
5783 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5784 | VM_BUG_ON_PAGE(page_count(page), page); | |
5785 | ||
5786 | if (!do_swap_account) | |
5787 | return; | |
5788 | ||
5789 | pc = lookup_page_cgroup(page); | |
5790 | ||
5791 | /* Readahead page, never charged */ | |
5792 | if (!PageCgroupUsed(pc)) | |
5793 | return; | |
5794 | ||
5795 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEMSW), page); | |
5796 | ||
5797 | oldid = swap_cgroup_record(entry, mem_cgroup_id(pc->mem_cgroup)); | |
5798 | VM_BUG_ON_PAGE(oldid, page); | |
5799 | ||
5800 | pc->flags &= ~PCG_MEMSW; | |
5801 | css_get(&pc->mem_cgroup->css); | |
5802 | mem_cgroup_swap_statistics(pc->mem_cgroup, true); | |
5803 | } | |
5804 | ||
5805 | /** | |
5806 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5807 | * @entry: swap entry to uncharge | |
5808 | * | |
5809 | * Drop the memsw charge associated with @entry. | |
5810 | */ | |
5811 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5812 | { | |
5813 | struct mem_cgroup *memcg; | |
5814 | unsigned short id; | |
5815 | ||
5816 | if (!do_swap_account) | |
5817 | return; | |
5818 | ||
5819 | id = swap_cgroup_record(entry, 0); | |
5820 | rcu_read_lock(); | |
5821 | memcg = mem_cgroup_lookup(id); | |
5822 | if (memcg) { | |
ce00a967 | 5823 | if (!mem_cgroup_is_root(memcg)) |
3e32cb2e | 5824 | page_counter_uncharge(&memcg->memsw, 1); |
0a31bc97 JW |
5825 | mem_cgroup_swap_statistics(memcg, false); |
5826 | css_put(&memcg->css); | |
5827 | } | |
5828 | rcu_read_unlock(); | |
5829 | } | |
5830 | #endif | |
5831 | ||
00501b53 JW |
5832 | /** |
5833 | * mem_cgroup_try_charge - try charging a page | |
5834 | * @page: page to charge | |
5835 | * @mm: mm context of the victim | |
5836 | * @gfp_mask: reclaim mode | |
5837 | * @memcgp: charged memcg return | |
5838 | * | |
5839 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5840 | * pages according to @gfp_mask if necessary. | |
5841 | * | |
5842 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5843 | * Otherwise, an error code is returned. | |
5844 | * | |
5845 | * After page->mapping has been set up, the caller must finalize the | |
5846 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5847 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5848 | */ | |
5849 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5850 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5851 | { | |
5852 | struct mem_cgroup *memcg = NULL; | |
5853 | unsigned int nr_pages = 1; | |
5854 | int ret = 0; | |
5855 | ||
5856 | if (mem_cgroup_disabled()) | |
5857 | goto out; | |
5858 | ||
5859 | if (PageSwapCache(page)) { | |
5860 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
5861 | /* | |
5862 | * Every swap fault against a single page tries to charge the | |
5863 | * page, bail as early as possible. shmem_unuse() encounters | |
5864 | * already charged pages, too. The USED bit is protected by | |
5865 | * the page lock, which serializes swap cache removal, which | |
5866 | * in turn serializes uncharging. | |
5867 | */ | |
5868 | if (PageCgroupUsed(pc)) | |
5869 | goto out; | |
5870 | } | |
5871 | ||
5872 | if (PageTransHuge(page)) { | |
5873 | nr_pages <<= compound_order(page); | |
5874 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5875 | } | |
5876 | ||
5877 | if (do_swap_account && PageSwapCache(page)) | |
5878 | memcg = try_get_mem_cgroup_from_page(page); | |
5879 | if (!memcg) | |
5880 | memcg = get_mem_cgroup_from_mm(mm); | |
5881 | ||
5882 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5883 | ||
5884 | css_put(&memcg->css); | |
5885 | ||
5886 | if (ret == -EINTR) { | |
5887 | memcg = root_mem_cgroup; | |
5888 | ret = 0; | |
5889 | } | |
5890 | out: | |
5891 | *memcgp = memcg; | |
5892 | return ret; | |
5893 | } | |
5894 | ||
5895 | /** | |
5896 | * mem_cgroup_commit_charge - commit a page charge | |
5897 | * @page: page to charge | |
5898 | * @memcg: memcg to charge the page to | |
5899 | * @lrucare: page might be on LRU already | |
5900 | * | |
5901 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5902 | * after page->mapping has been set up. This must happen atomically | |
5903 | * as part of the page instantiation, i.e. under the page table lock | |
5904 | * for anonymous pages, under the page lock for page and swap cache. | |
5905 | * | |
5906 | * In addition, the page must not be on the LRU during the commit, to | |
5907 | * prevent racing with task migration. If it might be, use @lrucare. | |
5908 | * | |
5909 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5910 | */ | |
5911 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5912 | bool lrucare) | |
5913 | { | |
5914 | unsigned int nr_pages = 1; | |
5915 | ||
5916 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5917 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5918 | ||
5919 | if (mem_cgroup_disabled()) | |
5920 | return; | |
5921 | /* | |
5922 | * Swap faults will attempt to charge the same page multiple | |
5923 | * times. But reuse_swap_page() might have removed the page | |
5924 | * from swapcache already, so we can't check PageSwapCache(). | |
5925 | */ | |
5926 | if (!memcg) | |
5927 | return; | |
5928 | ||
6abb5a86 JW |
5929 | commit_charge(page, memcg, lrucare); |
5930 | ||
00501b53 JW |
5931 | if (PageTransHuge(page)) { |
5932 | nr_pages <<= compound_order(page); | |
5933 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5934 | } | |
5935 | ||
6abb5a86 JW |
5936 | local_irq_disable(); |
5937 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5938 | memcg_check_events(memcg, page); | |
5939 | local_irq_enable(); | |
00501b53 JW |
5940 | |
5941 | if (do_swap_account && PageSwapCache(page)) { | |
5942 | swp_entry_t entry = { .val = page_private(page) }; | |
5943 | /* | |
5944 | * The swap entry might not get freed for a long time, | |
5945 | * let's not wait for it. The page already received a | |
5946 | * memory+swap charge, drop the swap entry duplicate. | |
5947 | */ | |
5948 | mem_cgroup_uncharge_swap(entry); | |
5949 | } | |
5950 | } | |
5951 | ||
5952 | /** | |
5953 | * mem_cgroup_cancel_charge - cancel a page charge | |
5954 | * @page: page to charge | |
5955 | * @memcg: memcg to charge the page to | |
5956 | * | |
5957 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5958 | */ | |
5959 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5960 | { | |
5961 | unsigned int nr_pages = 1; | |
5962 | ||
5963 | if (mem_cgroup_disabled()) | |
5964 | return; | |
5965 | /* | |
5966 | * Swap faults will attempt to charge the same page multiple | |
5967 | * times. But reuse_swap_page() might have removed the page | |
5968 | * from swapcache already, so we can't check PageSwapCache(). | |
5969 | */ | |
5970 | if (!memcg) | |
5971 | return; | |
5972 | ||
5973 | if (PageTransHuge(page)) { | |
5974 | nr_pages <<= compound_order(page); | |
5975 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5976 | } | |
5977 | ||
5978 | cancel_charge(memcg, nr_pages); | |
5979 | } | |
5980 | ||
747db954 JW |
5981 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
5982 | unsigned long nr_mem, unsigned long nr_memsw, | |
5983 | unsigned long nr_anon, unsigned long nr_file, | |
5984 | unsigned long nr_huge, struct page *dummy_page) | |
5985 | { | |
5986 | unsigned long flags; | |
5987 | ||
ce00a967 JW |
5988 | if (!mem_cgroup_is_root(memcg)) { |
5989 | if (nr_mem) | |
3e32cb2e | 5990 | page_counter_uncharge(&memcg->memory, nr_mem); |
ce00a967 | 5991 | if (nr_memsw) |
3e32cb2e | 5992 | page_counter_uncharge(&memcg->memsw, nr_memsw); |
ce00a967 JW |
5993 | memcg_oom_recover(memcg); |
5994 | } | |
747db954 JW |
5995 | |
5996 | local_irq_save(flags); | |
5997 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5998 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5999 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
6000 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
6001 | __this_cpu_add(memcg->stat->nr_page_events, nr_anon + nr_file); | |
6002 | memcg_check_events(memcg, dummy_page); | |
6003 | local_irq_restore(flags); | |
e8ea14cc JW |
6004 | |
6005 | if (!mem_cgroup_is_root(memcg)) | |
6006 | css_put_many(&memcg->css, max(nr_mem, nr_memsw)); | |
747db954 JW |
6007 | } |
6008 | ||
6009 | static void uncharge_list(struct list_head *page_list) | |
6010 | { | |
6011 | struct mem_cgroup *memcg = NULL; | |
6012 | unsigned long nr_memsw = 0; | |
6013 | unsigned long nr_anon = 0; | |
6014 | unsigned long nr_file = 0; | |
6015 | unsigned long nr_huge = 0; | |
6016 | unsigned long pgpgout = 0; | |
6017 | unsigned long nr_mem = 0; | |
6018 | struct list_head *next; | |
6019 | struct page *page; | |
6020 | ||
6021 | next = page_list->next; | |
6022 | do { | |
6023 | unsigned int nr_pages = 1; | |
6024 | struct page_cgroup *pc; | |
6025 | ||
6026 | page = list_entry(next, struct page, lru); | |
6027 | next = page->lru.next; | |
6028 | ||
6029 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6030 | VM_BUG_ON_PAGE(page_count(page), page); | |
6031 | ||
6032 | pc = lookup_page_cgroup(page); | |
6033 | if (!PageCgroupUsed(pc)) | |
6034 | continue; | |
6035 | ||
6036 | /* | |
6037 | * Nobody should be changing or seriously looking at | |
6038 | * pc->mem_cgroup and pc->flags at this point, we have | |
6039 | * fully exclusive access to the page. | |
6040 | */ | |
6041 | ||
6042 | if (memcg != pc->mem_cgroup) { | |
6043 | if (memcg) { | |
6044 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6045 | nr_anon, nr_file, nr_huge, page); | |
6046 | pgpgout = nr_mem = nr_memsw = 0; | |
6047 | nr_anon = nr_file = nr_huge = 0; | |
6048 | } | |
6049 | memcg = pc->mem_cgroup; | |
6050 | } | |
6051 | ||
6052 | if (PageTransHuge(page)) { | |
6053 | nr_pages <<= compound_order(page); | |
6054 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
6055 | nr_huge += nr_pages; | |
6056 | } | |
6057 | ||
6058 | if (PageAnon(page)) | |
6059 | nr_anon += nr_pages; | |
6060 | else | |
6061 | nr_file += nr_pages; | |
6062 | ||
6063 | if (pc->flags & PCG_MEM) | |
6064 | nr_mem += nr_pages; | |
6065 | if (pc->flags & PCG_MEMSW) | |
6066 | nr_memsw += nr_pages; | |
6067 | pc->flags = 0; | |
6068 | ||
6069 | pgpgout++; | |
6070 | } while (next != page_list); | |
6071 | ||
6072 | if (memcg) | |
6073 | uncharge_batch(memcg, pgpgout, nr_mem, nr_memsw, | |
6074 | nr_anon, nr_file, nr_huge, page); | |
6075 | } | |
6076 | ||
0a31bc97 JW |
6077 | /** |
6078 | * mem_cgroup_uncharge - uncharge a page | |
6079 | * @page: page to uncharge | |
6080 | * | |
6081 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6082 | * mem_cgroup_commit_charge(). | |
6083 | */ | |
6084 | void mem_cgroup_uncharge(struct page *page) | |
6085 | { | |
0a31bc97 | 6086 | struct page_cgroup *pc; |
0a31bc97 JW |
6087 | |
6088 | if (mem_cgroup_disabled()) | |
6089 | return; | |
6090 | ||
747db954 | 6091 | /* Don't touch page->lru of any random page, pre-check: */ |
0a31bc97 | 6092 | pc = lookup_page_cgroup(page); |
0a31bc97 JW |
6093 | if (!PageCgroupUsed(pc)) |
6094 | return; | |
6095 | ||
747db954 JW |
6096 | INIT_LIST_HEAD(&page->lru); |
6097 | uncharge_list(&page->lru); | |
6098 | } | |
0a31bc97 | 6099 | |
747db954 JW |
6100 | /** |
6101 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6102 | * @page_list: list of pages to uncharge | |
6103 | * | |
6104 | * Uncharge a list of pages previously charged with | |
6105 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6106 | */ | |
6107 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6108 | { | |
6109 | if (mem_cgroup_disabled()) | |
6110 | return; | |
0a31bc97 | 6111 | |
747db954 JW |
6112 | if (!list_empty(page_list)) |
6113 | uncharge_list(page_list); | |
0a31bc97 JW |
6114 | } |
6115 | ||
6116 | /** | |
6117 | * mem_cgroup_migrate - migrate a charge to another page | |
6118 | * @oldpage: currently charged page | |
6119 | * @newpage: page to transfer the charge to | |
6120 | * @lrucare: both pages might be on the LRU already | |
6121 | * | |
6122 | * Migrate the charge from @oldpage to @newpage. | |
6123 | * | |
6124 | * Both pages must be locked, @newpage->mapping must be set up. | |
6125 | */ | |
6126 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
6127 | bool lrucare) | |
6128 | { | |
0a31bc97 JW |
6129 | struct page_cgroup *pc; |
6130 | int isolated; | |
6131 | ||
6132 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6133 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
6134 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
6135 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
6136 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
6137 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6138 | newpage); | |
0a31bc97 JW |
6139 | |
6140 | if (mem_cgroup_disabled()) | |
6141 | return; | |
6142 | ||
6143 | /* Page cache replacement: new page already charged? */ | |
6144 | pc = lookup_page_cgroup(newpage); | |
6145 | if (PageCgroupUsed(pc)) | |
6146 | return; | |
6147 | ||
7d5e3245 JW |
6148 | /* |
6149 | * Swapcache readahead pages can get migrated before being | |
6150 | * charged, and migration from compaction can happen to an | |
6151 | * uncharged page when the PFN walker finds a page that | |
6152 | * reclaim just put back on the LRU but has not released yet. | |
6153 | */ | |
0a31bc97 JW |
6154 | pc = lookup_page_cgroup(oldpage); |
6155 | if (!PageCgroupUsed(pc)) | |
6156 | return; | |
6157 | ||
6158 | VM_BUG_ON_PAGE(!(pc->flags & PCG_MEM), oldpage); | |
6159 | VM_BUG_ON_PAGE(do_swap_account && !(pc->flags & PCG_MEMSW), oldpage); | |
6160 | ||
0a31bc97 JW |
6161 | if (lrucare) |
6162 | lock_page_lru(oldpage, &isolated); | |
6163 | ||
6164 | pc->flags = 0; | |
6165 | ||
6166 | if (lrucare) | |
6167 | unlock_page_lru(oldpage, isolated); | |
6168 | ||
6abb5a86 | 6169 | commit_charge(newpage, pc->mem_cgroup, lrucare); |
0a31bc97 JW |
6170 | } |
6171 | ||
2d11085e | 6172 | /* |
1081312f MH |
6173 | * subsys_initcall() for memory controller. |
6174 | * | |
6175 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6176 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6177 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6178 | * be initialized from here. | |
2d11085e MH |
6179 | */ |
6180 | static int __init mem_cgroup_init(void) | |
6181 | { | |
6182 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6183 | enable_swap_cgroup(); |
bb4cc1a8 | 6184 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6185 | memcg_stock_init(); |
2d11085e MH |
6186 | return 0; |
6187 | } | |
6188 | subsys_initcall(mem_cgroup_init); |