]>
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 | 1287 | /* |
dfe0e773 | 1288 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1289 | * possibly migrated - before they are charged. |
7512102c | 1290 | */ |
29833315 JW |
1291 | if (!memcg) |
1292 | memcg = root_mem_cgroup; | |
7512102c | 1293 | |
e231875b | 1294 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1295 | lruvec = &mz->lruvec; |
1296 | out: | |
1297 | /* | |
1298 | * Since a node can be onlined after the mem_cgroup was created, | |
1299 | * we have to be prepared to initialize lruvec->zone here; | |
1300 | * and if offlined then reonlined, we need to reinitialize it. | |
1301 | */ | |
1302 | if (unlikely(lruvec->zone != zone)) | |
1303 | lruvec->zone = zone; | |
1304 | return lruvec; | |
08e552c6 | 1305 | } |
b69408e8 | 1306 | |
925b7673 | 1307 | /** |
fa9add64 HD |
1308 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1309 | * @lruvec: mem_cgroup per zone lru vector | |
1310 | * @lru: index of lru list the page is sitting on | |
1311 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1312 | * |
fa9add64 HD |
1313 | * This function must be called when a page is added to or removed from an |
1314 | * lru list. | |
3f58a829 | 1315 | */ |
fa9add64 HD |
1316 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1317 | int nr_pages) | |
3f58a829 MK |
1318 | { |
1319 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1320 | unsigned long *lru_size; |
3f58a829 MK |
1321 | |
1322 | if (mem_cgroup_disabled()) | |
1323 | return; | |
1324 | ||
fa9add64 HD |
1325 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1326 | lru_size = mz->lru_size + lru; | |
1327 | *lru_size += nr_pages; | |
1328 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1329 | } |
544122e5 | 1330 | |
3e92041d | 1331 | /* |
c0ff4b85 | 1332 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1333 | * hierarchy subtree |
1334 | */ | |
c3ac9a8a JW |
1335 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1336 | struct mem_cgroup *memcg) | |
3e92041d | 1337 | { |
91c63734 JW |
1338 | if (root_memcg == memcg) |
1339 | return true; | |
3a981f48 | 1340 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1341 | return false; |
b47f77b5 | 1342 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1343 | } |
1344 | ||
1345 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1346 | struct mem_cgroup *memcg) | |
1347 | { | |
1348 | bool ret; | |
1349 | ||
91c63734 | 1350 | rcu_read_lock(); |
c3ac9a8a | 1351 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1352 | rcu_read_unlock(); |
1353 | return ret; | |
3e92041d MH |
1354 | } |
1355 | ||
ffbdccf5 DR |
1356 | bool task_in_mem_cgroup(struct task_struct *task, |
1357 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1358 | { |
0b7f569e | 1359 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1360 | struct task_struct *p; |
ffbdccf5 | 1361 | bool ret; |
4c4a2214 | 1362 | |
158e0a2d | 1363 | p = find_lock_task_mm(task); |
de077d22 | 1364 | if (p) { |
df381975 | 1365 | curr = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1366 | task_unlock(p); |
1367 | } else { | |
1368 | /* | |
1369 | * All threads may have already detached their mm's, but the oom | |
1370 | * killer still needs to detect if they have already been oom | |
1371 | * killed to prevent needlessly killing additional tasks. | |
1372 | */ | |
ffbdccf5 | 1373 | rcu_read_lock(); |
de077d22 DR |
1374 | curr = mem_cgroup_from_task(task); |
1375 | if (curr) | |
1376 | css_get(&curr->css); | |
ffbdccf5 | 1377 | rcu_read_unlock(); |
de077d22 | 1378 | } |
d31f56db | 1379 | /* |
c0ff4b85 | 1380 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1381 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1382 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1383 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1384 | */ |
c0ff4b85 | 1385 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1386 | css_put(&curr->css); |
4c4a2214 DR |
1387 | return ret; |
1388 | } | |
1389 | ||
c56d5c7d | 1390 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1391 | { |
9b272977 | 1392 | unsigned long inactive_ratio; |
14797e23 | 1393 | unsigned long inactive; |
9b272977 | 1394 | unsigned long active; |
c772be93 | 1395 | unsigned long gb; |
14797e23 | 1396 | |
4d7dcca2 HD |
1397 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1398 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1399 | |
c772be93 KM |
1400 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1401 | if (gb) | |
1402 | inactive_ratio = int_sqrt(10 * gb); | |
1403 | else | |
1404 | inactive_ratio = 1; | |
1405 | ||
9b272977 | 1406 | return inactive * inactive_ratio < active; |
14797e23 KM |
1407 | } |
1408 | ||
3e32cb2e | 1409 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1410 | container_of(counter, struct mem_cgroup, member) |
1411 | ||
19942822 | 1412 | /** |
9d11ea9f | 1413 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1414 | * @memcg: the memory cgroup |
19942822 | 1415 | * |
9d11ea9f | 1416 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1417 | * pages. |
19942822 | 1418 | */ |
c0ff4b85 | 1419 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1420 | { |
3e32cb2e JW |
1421 | unsigned long margin = 0; |
1422 | unsigned long count; | |
1423 | unsigned long limit; | |
9d11ea9f | 1424 | |
3e32cb2e JW |
1425 | count = page_counter_read(&memcg->memory); |
1426 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1427 | if (count < limit) | |
1428 | margin = limit - count; | |
1429 | ||
1430 | if (do_swap_account) { | |
1431 | count = page_counter_read(&memcg->memsw); | |
1432 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1433 | if (count <= limit) | |
1434 | margin = min(margin, limit - count); | |
1435 | } | |
1436 | ||
1437 | return margin; | |
19942822 JW |
1438 | } |
1439 | ||
1f4c025b | 1440 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1441 | { |
a7885eb8 | 1442 | /* root ? */ |
14208b0e | 1443 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1444 | return vm_swappiness; |
1445 | ||
bf1ff263 | 1446 | return memcg->swappiness; |
a7885eb8 KM |
1447 | } |
1448 | ||
32047e2a | 1449 | /* |
bdcbb659 | 1450 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1451 | * |
bdcbb659 QH |
1452 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1453 | * moving cgroups. This is for waiting at high-memory pressure | |
1454 | * caused by "move". | |
32047e2a | 1455 | */ |
c0ff4b85 | 1456 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1457 | { |
2bd9bb20 KH |
1458 | struct mem_cgroup *from; |
1459 | struct mem_cgroup *to; | |
4b534334 | 1460 | bool ret = false; |
2bd9bb20 KH |
1461 | /* |
1462 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1463 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1464 | */ | |
1465 | spin_lock(&mc.lock); | |
1466 | from = mc.from; | |
1467 | to = mc.to; | |
1468 | if (!from) | |
1469 | goto unlock; | |
3e92041d | 1470 | |
c0ff4b85 R |
1471 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1472 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1473 | unlock: |
1474 | spin_unlock(&mc.lock); | |
4b534334 KH |
1475 | return ret; |
1476 | } | |
1477 | ||
c0ff4b85 | 1478 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1479 | { |
1480 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1481 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1482 | DEFINE_WAIT(wait); |
1483 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1484 | /* moving charge context might have finished. */ | |
1485 | if (mc.moving_task) | |
1486 | schedule(); | |
1487 | finish_wait(&mc.waitq, &wait); | |
1488 | return true; | |
1489 | } | |
1490 | } | |
1491 | return false; | |
1492 | } | |
1493 | ||
58cf188e | 1494 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1495 | /** |
58cf188e | 1496 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1497 | * @memcg: The memory cgroup that went over limit |
1498 | * @p: Task that is going to be killed | |
1499 | * | |
1500 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1501 | * enabled | |
1502 | */ | |
1503 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1504 | { | |
e61734c5 | 1505 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1506 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1507 | struct mem_cgroup *iter; |
1508 | unsigned int i; | |
e222432b | 1509 | |
58cf188e | 1510 | if (!p) |
e222432b BS |
1511 | return; |
1512 | ||
08088cb9 | 1513 | mutex_lock(&oom_info_lock); |
e222432b BS |
1514 | rcu_read_lock(); |
1515 | ||
e61734c5 TH |
1516 | pr_info("Task in "); |
1517 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1518 | pr_info(" killed as a result of limit of "); | |
1519 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1520 | pr_info("\n"); | |
e222432b | 1521 | |
e222432b BS |
1522 | rcu_read_unlock(); |
1523 | ||
3e32cb2e JW |
1524 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1525 | K((u64)page_counter_read(&memcg->memory)), | |
1526 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1527 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1528 | K((u64)page_counter_read(&memcg->memsw)), | |
1529 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1530 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1531 | K((u64)page_counter_read(&memcg->kmem)), | |
1532 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1533 | |
1534 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1535 | pr_info("Memory cgroup stats for "); |
1536 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1537 | pr_cont(":"); |
1538 | ||
1539 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1540 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1541 | continue; | |
1542 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1543 | K(mem_cgroup_read_stat(iter, i))); | |
1544 | } | |
1545 | ||
1546 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1547 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1548 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1549 | ||
1550 | pr_cont("\n"); | |
1551 | } | |
08088cb9 | 1552 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1553 | } |
1554 | ||
81d39c20 KH |
1555 | /* |
1556 | * This function returns the number of memcg under hierarchy tree. Returns | |
1557 | * 1(self count) if no children. | |
1558 | */ | |
c0ff4b85 | 1559 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1560 | { |
1561 | int num = 0; | |
7d74b06f KH |
1562 | struct mem_cgroup *iter; |
1563 | ||
c0ff4b85 | 1564 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1565 | num++; |
81d39c20 KH |
1566 | return num; |
1567 | } | |
1568 | ||
a63d83f4 DR |
1569 | /* |
1570 | * Return the memory (and swap, if configured) limit for a memcg. | |
1571 | */ | |
3e32cb2e | 1572 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1573 | { |
3e32cb2e | 1574 | unsigned long limit; |
a63d83f4 | 1575 | |
3e32cb2e | 1576 | limit = memcg->memory.limit; |
9a5a8f19 | 1577 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1578 | unsigned long memsw_limit; |
9a5a8f19 | 1579 | |
3e32cb2e JW |
1580 | memsw_limit = memcg->memsw.limit; |
1581 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1582 | } |
9a5a8f19 | 1583 | return limit; |
a63d83f4 DR |
1584 | } |
1585 | ||
19965460 DR |
1586 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1587 | int order) | |
9cbb78bb DR |
1588 | { |
1589 | struct mem_cgroup *iter; | |
1590 | unsigned long chosen_points = 0; | |
1591 | unsigned long totalpages; | |
1592 | unsigned int points = 0; | |
1593 | struct task_struct *chosen = NULL; | |
1594 | ||
876aafbf | 1595 | /* |
465adcf1 DR |
1596 | * If current has a pending SIGKILL or is exiting, then automatically |
1597 | * select it. The goal is to allow it to allocate so that it may | |
1598 | * quickly exit and free its memory. | |
876aafbf | 1599 | */ |
465adcf1 | 1600 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1601 | set_thread_flag(TIF_MEMDIE); |
1602 | return; | |
1603 | } | |
1604 | ||
1605 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1606 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1607 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1608 | struct css_task_iter it; |
9cbb78bb DR |
1609 | struct task_struct *task; |
1610 | ||
72ec7029 TH |
1611 | css_task_iter_start(&iter->css, &it); |
1612 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1613 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1614 | false)) { | |
1615 | case OOM_SCAN_SELECT: | |
1616 | if (chosen) | |
1617 | put_task_struct(chosen); | |
1618 | chosen = task; | |
1619 | chosen_points = ULONG_MAX; | |
1620 | get_task_struct(chosen); | |
1621 | /* fall through */ | |
1622 | case OOM_SCAN_CONTINUE: | |
1623 | continue; | |
1624 | case OOM_SCAN_ABORT: | |
72ec7029 | 1625 | css_task_iter_end(&it); |
9cbb78bb DR |
1626 | mem_cgroup_iter_break(memcg, iter); |
1627 | if (chosen) | |
1628 | put_task_struct(chosen); | |
1629 | return; | |
1630 | case OOM_SCAN_OK: | |
1631 | break; | |
1632 | }; | |
1633 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1634 | if (!points || points < chosen_points) |
1635 | continue; | |
1636 | /* Prefer thread group leaders for display purposes */ | |
1637 | if (points == chosen_points && | |
1638 | thread_group_leader(chosen)) | |
1639 | continue; | |
1640 | ||
1641 | if (chosen) | |
1642 | put_task_struct(chosen); | |
1643 | chosen = task; | |
1644 | chosen_points = points; | |
1645 | get_task_struct(chosen); | |
9cbb78bb | 1646 | } |
72ec7029 | 1647 | css_task_iter_end(&it); |
9cbb78bb DR |
1648 | } |
1649 | ||
1650 | if (!chosen) | |
1651 | return; | |
1652 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1653 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1654 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1655 | } |
1656 | ||
4d0c066d KH |
1657 | /** |
1658 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1659 | * @memcg: the target memcg |
4d0c066d KH |
1660 | * @nid: the node ID to be checked. |
1661 | * @noswap : specify true here if the user wants flle only information. | |
1662 | * | |
1663 | * This function returns whether the specified memcg contains any | |
1664 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1665 | * pages in the node. | |
1666 | */ | |
c0ff4b85 | 1667 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1668 | int nid, bool noswap) |
1669 | { | |
c0ff4b85 | 1670 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1671 | return true; |
1672 | if (noswap || !total_swap_pages) | |
1673 | return false; | |
c0ff4b85 | 1674 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1675 | return true; |
1676 | return false; | |
1677 | ||
1678 | } | |
bb4cc1a8 | 1679 | #if MAX_NUMNODES > 1 |
889976db YH |
1680 | |
1681 | /* | |
1682 | * Always updating the nodemask is not very good - even if we have an empty | |
1683 | * list or the wrong list here, we can start from some node and traverse all | |
1684 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1685 | * | |
1686 | */ | |
c0ff4b85 | 1687 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1688 | { |
1689 | int nid; | |
453a9bf3 KH |
1690 | /* |
1691 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1692 | * pagein/pageout changes since the last update. | |
1693 | */ | |
c0ff4b85 | 1694 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1695 | return; |
c0ff4b85 | 1696 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1697 | return; |
1698 | ||
889976db | 1699 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1700 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1701 | |
31aaea4a | 1702 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1703 | |
c0ff4b85 R |
1704 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1705 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1706 | } |
453a9bf3 | 1707 | |
c0ff4b85 R |
1708 | atomic_set(&memcg->numainfo_events, 0); |
1709 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1710 | } |
1711 | ||
1712 | /* | |
1713 | * Selecting a node where we start reclaim from. Because what we need is just | |
1714 | * reducing usage counter, start from anywhere is O,K. Considering | |
1715 | * memory reclaim from current node, there are pros. and cons. | |
1716 | * | |
1717 | * Freeing memory from current node means freeing memory from a node which | |
1718 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1719 | * hit limits, it will see a contention on a node. But freeing from remote | |
1720 | * node means more costs for memory reclaim because of memory latency. | |
1721 | * | |
1722 | * Now, we use round-robin. Better algorithm is welcomed. | |
1723 | */ | |
c0ff4b85 | 1724 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1725 | { |
1726 | int node; | |
1727 | ||
c0ff4b85 R |
1728 | mem_cgroup_may_update_nodemask(memcg); |
1729 | node = memcg->last_scanned_node; | |
889976db | 1730 | |
c0ff4b85 | 1731 | node = next_node(node, memcg->scan_nodes); |
889976db | 1732 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1733 | node = first_node(memcg->scan_nodes); |
889976db YH |
1734 | /* |
1735 | * We call this when we hit limit, not when pages are added to LRU. | |
1736 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1737 | * memcg is too small and all pages are not on LRU. In that case, | |
1738 | * we use curret node. | |
1739 | */ | |
1740 | if (unlikely(node == MAX_NUMNODES)) | |
1741 | node = numa_node_id(); | |
1742 | ||
c0ff4b85 | 1743 | memcg->last_scanned_node = node; |
889976db YH |
1744 | return node; |
1745 | } | |
889976db | 1746 | #else |
c0ff4b85 | 1747 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1748 | { |
1749 | return 0; | |
1750 | } | |
1751 | #endif | |
1752 | ||
0608f43d AM |
1753 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1754 | struct zone *zone, | |
1755 | gfp_t gfp_mask, | |
1756 | unsigned long *total_scanned) | |
1757 | { | |
1758 | struct mem_cgroup *victim = NULL; | |
1759 | int total = 0; | |
1760 | int loop = 0; | |
1761 | unsigned long excess; | |
1762 | unsigned long nr_scanned; | |
1763 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1764 | .zone = zone, | |
1765 | .priority = 0, | |
1766 | }; | |
1767 | ||
3e32cb2e | 1768 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1769 | |
1770 | while (1) { | |
1771 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1772 | if (!victim) { | |
1773 | loop++; | |
1774 | if (loop >= 2) { | |
1775 | /* | |
1776 | * If we have not been able to reclaim | |
1777 | * anything, it might because there are | |
1778 | * no reclaimable pages under this hierarchy | |
1779 | */ | |
1780 | if (!total) | |
1781 | break; | |
1782 | /* | |
1783 | * We want to do more targeted reclaim. | |
1784 | * excess >> 2 is not to excessive so as to | |
1785 | * reclaim too much, nor too less that we keep | |
1786 | * coming back to reclaim from this cgroup | |
1787 | */ | |
1788 | if (total >= (excess >> 2) || | |
1789 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1790 | break; | |
1791 | } | |
1792 | continue; | |
1793 | } | |
0608f43d AM |
1794 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1795 | zone, &nr_scanned); | |
1796 | *total_scanned += nr_scanned; | |
3e32cb2e | 1797 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1798 | break; |
6d61ef40 | 1799 | } |
0608f43d AM |
1800 | mem_cgroup_iter_break(root_memcg, victim); |
1801 | return total; | |
6d61ef40 BS |
1802 | } |
1803 | ||
0056f4e6 JW |
1804 | #ifdef CONFIG_LOCKDEP |
1805 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1806 | .name = "memcg_oom_lock", | |
1807 | }; | |
1808 | #endif | |
1809 | ||
fb2a6fc5 JW |
1810 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1811 | ||
867578cb KH |
1812 | /* |
1813 | * Check OOM-Killer is already running under our hierarchy. | |
1814 | * If someone is running, return false. | |
1815 | */ | |
fb2a6fc5 | 1816 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1817 | { |
79dfdacc | 1818 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1819 | |
fb2a6fc5 JW |
1820 | spin_lock(&memcg_oom_lock); |
1821 | ||
9f3a0d09 | 1822 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1823 | if (iter->oom_lock) { |
79dfdacc MH |
1824 | /* |
1825 | * this subtree of our hierarchy is already locked | |
1826 | * so we cannot give a lock. | |
1827 | */ | |
79dfdacc | 1828 | failed = iter; |
9f3a0d09 JW |
1829 | mem_cgroup_iter_break(memcg, iter); |
1830 | break; | |
23751be0 JW |
1831 | } else |
1832 | iter->oom_lock = true; | |
7d74b06f | 1833 | } |
867578cb | 1834 | |
fb2a6fc5 JW |
1835 | if (failed) { |
1836 | /* | |
1837 | * OK, we failed to lock the whole subtree so we have | |
1838 | * to clean up what we set up to the failing subtree | |
1839 | */ | |
1840 | for_each_mem_cgroup_tree(iter, memcg) { | |
1841 | if (iter == failed) { | |
1842 | mem_cgroup_iter_break(memcg, iter); | |
1843 | break; | |
1844 | } | |
1845 | iter->oom_lock = false; | |
79dfdacc | 1846 | } |
0056f4e6 JW |
1847 | } else |
1848 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1849 | |
1850 | spin_unlock(&memcg_oom_lock); | |
1851 | ||
1852 | return !failed; | |
a636b327 | 1853 | } |
0b7f569e | 1854 | |
fb2a6fc5 | 1855 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1856 | { |
7d74b06f KH |
1857 | struct mem_cgroup *iter; |
1858 | ||
fb2a6fc5 | 1859 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1860 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1861 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1862 | iter->oom_lock = false; |
fb2a6fc5 | 1863 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1864 | } |
1865 | ||
c0ff4b85 | 1866 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1867 | { |
1868 | struct mem_cgroup *iter; | |
1869 | ||
c0ff4b85 | 1870 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1871 | atomic_inc(&iter->under_oom); |
1872 | } | |
1873 | ||
c0ff4b85 | 1874 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1875 | { |
1876 | struct mem_cgroup *iter; | |
1877 | ||
867578cb KH |
1878 | /* |
1879 | * When a new child is created while the hierarchy is under oom, | |
1880 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1881 | * atomic_add_unless() here. | |
1882 | */ | |
c0ff4b85 | 1883 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1884 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1885 | } |
1886 | ||
867578cb KH |
1887 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1888 | ||
dc98df5a | 1889 | struct oom_wait_info { |
d79154bb | 1890 | struct mem_cgroup *memcg; |
dc98df5a KH |
1891 | wait_queue_t wait; |
1892 | }; | |
1893 | ||
1894 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1895 | unsigned mode, int sync, void *arg) | |
1896 | { | |
d79154bb HD |
1897 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1898 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1899 | struct oom_wait_info *oom_wait_info; |
1900 | ||
1901 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1902 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1903 | |
dc98df5a | 1904 | /* |
d79154bb | 1905 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1906 | * Then we can use css_is_ancestor without taking care of RCU. |
1907 | */ | |
c0ff4b85 R |
1908 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1909 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1910 | return 0; |
dc98df5a KH |
1911 | return autoremove_wake_function(wait, mode, sync, arg); |
1912 | } | |
1913 | ||
c0ff4b85 | 1914 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1915 | { |
3812c8c8 | 1916 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
1917 | /* for filtering, pass "memcg" as argument. */ |
1918 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1919 | } |
1920 | ||
c0ff4b85 | 1921 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1922 | { |
c0ff4b85 R |
1923 | if (memcg && atomic_read(&memcg->under_oom)) |
1924 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1925 | } |
1926 | ||
3812c8c8 | 1927 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1928 | { |
3812c8c8 JW |
1929 | if (!current->memcg_oom.may_oom) |
1930 | return; | |
867578cb | 1931 | /* |
49426420 JW |
1932 | * We are in the middle of the charge context here, so we |
1933 | * don't want to block when potentially sitting on a callstack | |
1934 | * that holds all kinds of filesystem and mm locks. | |
1935 | * | |
1936 | * Also, the caller may handle a failed allocation gracefully | |
1937 | * (like optional page cache readahead) and so an OOM killer | |
1938 | * invocation might not even be necessary. | |
1939 | * | |
1940 | * That's why we don't do anything here except remember the | |
1941 | * OOM context and then deal with it at the end of the page | |
1942 | * fault when the stack is unwound, the locks are released, | |
1943 | * and when we know whether the fault was overall successful. | |
867578cb | 1944 | */ |
49426420 JW |
1945 | css_get(&memcg->css); |
1946 | current->memcg_oom.memcg = memcg; | |
1947 | current->memcg_oom.gfp_mask = mask; | |
1948 | current->memcg_oom.order = order; | |
3812c8c8 JW |
1949 | } |
1950 | ||
1951 | /** | |
1952 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1953 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1954 | * |
49426420 JW |
1955 | * This has to be called at the end of a page fault if the memcg OOM |
1956 | * handler was enabled. | |
3812c8c8 | 1957 | * |
49426420 | 1958 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1959 | * sleep on a waitqueue until the userspace task resolves the |
1960 | * situation. Sleeping directly in the charge context with all kinds | |
1961 | * of locks held is not a good idea, instead we remember an OOM state | |
1962 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1963 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1964 | * |
1965 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1966 | * completed, %false otherwise. |
3812c8c8 | 1967 | */ |
49426420 | 1968 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1969 | { |
49426420 | 1970 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 1971 | struct oom_wait_info owait; |
49426420 | 1972 | bool locked; |
3812c8c8 JW |
1973 | |
1974 | /* OOM is global, do not handle */ | |
3812c8c8 | 1975 | if (!memcg) |
49426420 | 1976 | return false; |
3812c8c8 | 1977 | |
49426420 JW |
1978 | if (!handle) |
1979 | goto cleanup; | |
3812c8c8 JW |
1980 | |
1981 | owait.memcg = memcg; | |
1982 | owait.wait.flags = 0; | |
1983 | owait.wait.func = memcg_oom_wake_function; | |
1984 | owait.wait.private = current; | |
1985 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1986 | |
3812c8c8 | 1987 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1988 | mem_cgroup_mark_under_oom(memcg); |
1989 | ||
1990 | locked = mem_cgroup_oom_trylock(memcg); | |
1991 | ||
1992 | if (locked) | |
1993 | mem_cgroup_oom_notify(memcg); | |
1994 | ||
1995 | if (locked && !memcg->oom_kill_disable) { | |
1996 | mem_cgroup_unmark_under_oom(memcg); | |
1997 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1998 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
1999 | current->memcg_oom.order); | |
2000 | } else { | |
3812c8c8 | 2001 | schedule(); |
49426420 JW |
2002 | mem_cgroup_unmark_under_oom(memcg); |
2003 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2004 | } | |
2005 | ||
2006 | if (locked) { | |
fb2a6fc5 JW |
2007 | mem_cgroup_oom_unlock(memcg); |
2008 | /* | |
2009 | * There is no guarantee that an OOM-lock contender | |
2010 | * sees the wakeups triggered by the OOM kill | |
2011 | * uncharges. Wake any sleepers explicitely. | |
2012 | */ | |
2013 | memcg_oom_recover(memcg); | |
2014 | } | |
49426420 JW |
2015 | cleanup: |
2016 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2017 | css_put(&memcg->css); |
867578cb | 2018 | return true; |
0b7f569e KH |
2019 | } |
2020 | ||
d7365e78 JW |
2021 | /** |
2022 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
2023 | * @page: page that is going to change accounted state | |
2024 | * @locked: &memcg->move_lock slowpath was taken | |
2025 | * @flags: IRQ-state flags for &memcg->move_lock | |
32047e2a | 2026 | * |
d7365e78 JW |
2027 | * This function must mark the beginning of an accounted page state |
2028 | * change to prevent double accounting when the page is concurrently | |
2029 | * being moved to another memcg: | |
32047e2a | 2030 | * |
d7365e78 JW |
2031 | * memcg = mem_cgroup_begin_page_stat(page, &locked, &flags); |
2032 | * if (TestClearPageState(page)) | |
2033 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
2034 | * mem_cgroup_end_page_stat(memcg, locked, flags); | |
32047e2a | 2035 | * |
d7365e78 JW |
2036 | * The RCU lock is held throughout the transaction. The fast path can |
2037 | * get away without acquiring the memcg->move_lock (@locked is false) | |
2038 | * because page moving starts with an RCU grace period. | |
32047e2a | 2039 | * |
d7365e78 JW |
2040 | * The RCU lock also protects the memcg from being freed when the page |
2041 | * state that is going to change is the only thing preventing the page | |
2042 | * from being uncharged. E.g. end-writeback clearing PageWriteback(), | |
2043 | * which allows migration to go ahead and uncharge the page before the | |
2044 | * account transaction might be complete. | |
d69b042f | 2045 | */ |
d7365e78 JW |
2046 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page, |
2047 | bool *locked, | |
2048 | unsigned long *flags) | |
89c06bd5 KH |
2049 | { |
2050 | struct mem_cgroup *memcg; | |
2051 | struct page_cgroup *pc; | |
2052 | ||
d7365e78 JW |
2053 | rcu_read_lock(); |
2054 | ||
2055 | if (mem_cgroup_disabled()) | |
2056 | return NULL; | |
2057 | ||
89c06bd5 KH |
2058 | pc = lookup_page_cgroup(page); |
2059 | again: | |
2060 | memcg = pc->mem_cgroup; | |
29833315 | 2061 | if (unlikely(!memcg)) |
d7365e78 JW |
2062 | return NULL; |
2063 | ||
2064 | *locked = false; | |
bdcbb659 | 2065 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2066 | return memcg; |
89c06bd5 | 2067 | |
354a4783 | 2068 | spin_lock_irqsave(&memcg->move_lock, *flags); |
29833315 | 2069 | if (memcg != pc->mem_cgroup) { |
354a4783 | 2070 | spin_unlock_irqrestore(&memcg->move_lock, *flags); |
89c06bd5 KH |
2071 | goto again; |
2072 | } | |
2073 | *locked = true; | |
d7365e78 JW |
2074 | |
2075 | return memcg; | |
89c06bd5 KH |
2076 | } |
2077 | ||
d7365e78 JW |
2078 | /** |
2079 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2080 | * @memcg: the memcg that was accounted against | |
2081 | * @locked: value received from mem_cgroup_begin_page_stat() | |
2082 | * @flags: value received from mem_cgroup_begin_page_stat() | |
2083 | */ | |
2084 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg, bool locked, | |
2085 | unsigned long flags) | |
89c06bd5 | 2086 | { |
d7365e78 | 2087 | if (memcg && locked) |
354a4783 | 2088 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 | 2089 | |
d7365e78 | 2090 | rcu_read_unlock(); |
89c06bd5 KH |
2091 | } |
2092 | ||
d7365e78 JW |
2093 | /** |
2094 | * mem_cgroup_update_page_stat - update page state statistics | |
2095 | * @memcg: memcg to account against | |
2096 | * @idx: page state item to account | |
2097 | * @val: number of pages (positive or negative) | |
2098 | * | |
2099 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2100 | */ | |
2101 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2102 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2103 | { |
658b72c5 | 2104 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2105 | |
d7365e78 JW |
2106 | if (memcg) |
2107 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2108 | } |
26174efd | 2109 | |
cdec2e42 KH |
2110 | /* |
2111 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2112 | * TODO: maybe necessary to use big numbers in big irons. | |
2113 | */ | |
7ec99d62 | 2114 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2115 | struct memcg_stock_pcp { |
2116 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2117 | unsigned int nr_pages; |
cdec2e42 | 2118 | struct work_struct work; |
26fe6168 | 2119 | unsigned long flags; |
a0db00fc | 2120 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2121 | }; |
2122 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2123 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2124 | |
a0956d54 SS |
2125 | /** |
2126 | * consume_stock: Try to consume stocked charge on this cpu. | |
2127 | * @memcg: memcg to consume from. | |
2128 | * @nr_pages: how many pages to charge. | |
2129 | * | |
2130 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2131 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2132 | * service an allocation will refill the stock. | |
2133 | * | |
2134 | * returns true if successful, false otherwise. | |
cdec2e42 | 2135 | */ |
a0956d54 | 2136 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2137 | { |
2138 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2139 | bool ret = false; |
cdec2e42 | 2140 | |
a0956d54 | 2141 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2142 | return ret; |
a0956d54 | 2143 | |
cdec2e42 | 2144 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2145 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2146 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2147 | ret = true; |
2148 | } | |
cdec2e42 KH |
2149 | put_cpu_var(memcg_stock); |
2150 | return ret; | |
2151 | } | |
2152 | ||
2153 | /* | |
3e32cb2e | 2154 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2155 | */ |
2156 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2157 | { | |
2158 | struct mem_cgroup *old = stock->cached; | |
2159 | ||
11c9ea4e | 2160 | if (stock->nr_pages) { |
3e32cb2e | 2161 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2162 | if (do_swap_account) |
3e32cb2e | 2163 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2164 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2165 | stock->nr_pages = 0; |
cdec2e42 KH |
2166 | } |
2167 | stock->cached = NULL; | |
cdec2e42 KH |
2168 | } |
2169 | ||
2170 | /* | |
2171 | * This must be called under preempt disabled or must be called by | |
2172 | * a thread which is pinned to local cpu. | |
2173 | */ | |
2174 | static void drain_local_stock(struct work_struct *dummy) | |
2175 | { | |
7c8e0181 | 2176 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2177 | drain_stock(stock); |
26fe6168 | 2178 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2179 | } |
2180 | ||
e4777496 MH |
2181 | static void __init memcg_stock_init(void) |
2182 | { | |
2183 | int cpu; | |
2184 | ||
2185 | for_each_possible_cpu(cpu) { | |
2186 | struct memcg_stock_pcp *stock = | |
2187 | &per_cpu(memcg_stock, cpu); | |
2188 | INIT_WORK(&stock->work, drain_local_stock); | |
2189 | } | |
2190 | } | |
2191 | ||
cdec2e42 | 2192 | /* |
3e32cb2e | 2193 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2194 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2195 | */ |
c0ff4b85 | 2196 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2197 | { |
2198 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2199 | ||
c0ff4b85 | 2200 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2201 | drain_stock(stock); |
c0ff4b85 | 2202 | stock->cached = memcg; |
cdec2e42 | 2203 | } |
11c9ea4e | 2204 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2205 | put_cpu_var(memcg_stock); |
2206 | } | |
2207 | ||
2208 | /* | |
c0ff4b85 | 2209 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2210 | * of the hierarchy under it. |
cdec2e42 | 2211 | */ |
6d3d6aa2 | 2212 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2213 | { |
26fe6168 | 2214 | int cpu, curcpu; |
d38144b7 | 2215 | |
6d3d6aa2 JW |
2216 | /* If someone's already draining, avoid adding running more workers. */ |
2217 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2218 | return; | |
cdec2e42 | 2219 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2220 | get_online_cpus(); |
5af12d0e | 2221 | curcpu = get_cpu(); |
cdec2e42 KH |
2222 | for_each_online_cpu(cpu) { |
2223 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2224 | struct mem_cgroup *memcg; |
26fe6168 | 2225 | |
c0ff4b85 R |
2226 | memcg = stock->cached; |
2227 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2228 | continue; |
c0ff4b85 | 2229 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2230 | continue; |
d1a05b69 MH |
2231 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2232 | if (cpu == curcpu) | |
2233 | drain_local_stock(&stock->work); | |
2234 | else | |
2235 | schedule_work_on(cpu, &stock->work); | |
2236 | } | |
cdec2e42 | 2237 | } |
5af12d0e | 2238 | put_cpu(); |
f894ffa8 | 2239 | put_online_cpus(); |
9f50fad6 | 2240 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2241 | } |
2242 | ||
711d3d2c KH |
2243 | /* |
2244 | * This function drains percpu counter value from DEAD cpu and | |
2245 | * move it to local cpu. Note that this function can be preempted. | |
2246 | */ | |
c0ff4b85 | 2247 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2248 | { |
2249 | int i; | |
2250 | ||
c0ff4b85 | 2251 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2252 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2253 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2254 | |
c0ff4b85 R |
2255 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2256 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2257 | } |
e9f8974f | 2258 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2259 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2260 | |
c0ff4b85 R |
2261 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2262 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2263 | } |
c0ff4b85 | 2264 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2265 | } |
2266 | ||
0db0628d | 2267 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2268 | unsigned long action, |
2269 | void *hcpu) | |
2270 | { | |
2271 | int cpu = (unsigned long)hcpu; | |
2272 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2273 | struct mem_cgroup *iter; |
cdec2e42 | 2274 | |
619d094b | 2275 | if (action == CPU_ONLINE) |
1489ebad | 2276 | return NOTIFY_OK; |
1489ebad | 2277 | |
d833049b | 2278 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2279 | return NOTIFY_OK; |
711d3d2c | 2280 | |
9f3a0d09 | 2281 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2282 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2283 | ||
cdec2e42 KH |
2284 | stock = &per_cpu(memcg_stock, cpu); |
2285 | drain_stock(stock); | |
2286 | return NOTIFY_OK; | |
2287 | } | |
2288 | ||
00501b53 JW |
2289 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2290 | unsigned int nr_pages) | |
8a9f3ccd | 2291 | { |
7ec99d62 | 2292 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2293 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2294 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2295 | struct page_counter *counter; |
6539cc05 | 2296 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2297 | bool may_swap = true; |
2298 | bool drained = false; | |
05b84301 | 2299 | int ret = 0; |
a636b327 | 2300 | |
ce00a967 JW |
2301 | if (mem_cgroup_is_root(memcg)) |
2302 | goto done; | |
6539cc05 | 2303 | retry: |
b6b6cc72 MH |
2304 | if (consume_stock(memcg, nr_pages)) |
2305 | goto done; | |
8a9f3ccd | 2306 | |
3fbe7244 | 2307 | if (!do_swap_account || |
3e32cb2e JW |
2308 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2309 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2310 | goto done_restock; |
3fbe7244 | 2311 | if (do_swap_account) |
3e32cb2e JW |
2312 | page_counter_uncharge(&memcg->memsw, batch); |
2313 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2314 | } else { |
3e32cb2e | 2315 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2316 | may_swap = false; |
3fbe7244 | 2317 | } |
7a81b88c | 2318 | |
6539cc05 JW |
2319 | if (batch > nr_pages) { |
2320 | batch = nr_pages; | |
2321 | goto retry; | |
2322 | } | |
6d61ef40 | 2323 | |
06b078fc JW |
2324 | /* |
2325 | * Unlike in global OOM situations, memcg is not in a physical | |
2326 | * memory shortage. Allow dying and OOM-killed tasks to | |
2327 | * bypass the last charges so that they can exit quickly and | |
2328 | * free their memory. | |
2329 | */ | |
2330 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2331 | fatal_signal_pending(current) || | |
2332 | current->flags & PF_EXITING)) | |
2333 | goto bypass; | |
2334 | ||
2335 | if (unlikely(task_in_memcg_oom(current))) | |
2336 | goto nomem; | |
2337 | ||
6539cc05 JW |
2338 | if (!(gfp_mask & __GFP_WAIT)) |
2339 | goto nomem; | |
4b534334 | 2340 | |
b70a2a21 JW |
2341 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2342 | gfp_mask, may_swap); | |
6539cc05 | 2343 | |
61e02c74 | 2344 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2345 | goto retry; |
28c34c29 | 2346 | |
b70a2a21 | 2347 | if (!drained) { |
6d3d6aa2 | 2348 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2349 | drained = true; |
2350 | goto retry; | |
2351 | } | |
2352 | ||
28c34c29 JW |
2353 | if (gfp_mask & __GFP_NORETRY) |
2354 | goto nomem; | |
6539cc05 JW |
2355 | /* |
2356 | * Even though the limit is exceeded at this point, reclaim | |
2357 | * may have been able to free some pages. Retry the charge | |
2358 | * before killing the task. | |
2359 | * | |
2360 | * Only for regular pages, though: huge pages are rather | |
2361 | * unlikely to succeed so close to the limit, and we fall back | |
2362 | * to regular pages anyway in case of failure. | |
2363 | */ | |
61e02c74 | 2364 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2365 | goto retry; |
2366 | /* | |
2367 | * At task move, charge accounts can be doubly counted. So, it's | |
2368 | * better to wait until the end of task_move if something is going on. | |
2369 | */ | |
2370 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2371 | goto retry; | |
2372 | ||
9b130619 JW |
2373 | if (nr_retries--) |
2374 | goto retry; | |
2375 | ||
06b078fc JW |
2376 | if (gfp_mask & __GFP_NOFAIL) |
2377 | goto bypass; | |
2378 | ||
6539cc05 JW |
2379 | if (fatal_signal_pending(current)) |
2380 | goto bypass; | |
2381 | ||
61e02c74 | 2382 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2383 | nomem: |
6d1fdc48 | 2384 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2385 | return -ENOMEM; |
867578cb | 2386 | bypass: |
ce00a967 | 2387 | return -EINTR; |
6539cc05 JW |
2388 | |
2389 | done_restock: | |
e8ea14cc | 2390 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2391 | if (batch > nr_pages) |
2392 | refill_stock(memcg, batch - nr_pages); | |
2393 | done: | |
05b84301 | 2394 | return ret; |
7a81b88c | 2395 | } |
8a9f3ccd | 2396 | |
00501b53 | 2397 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2398 | { |
ce00a967 JW |
2399 | if (mem_cgroup_is_root(memcg)) |
2400 | return; | |
2401 | ||
3e32cb2e | 2402 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2403 | if (do_swap_account) |
3e32cb2e | 2404 | page_counter_uncharge(&memcg->memsw, nr_pages); |
e8ea14cc JW |
2405 | |
2406 | css_put_many(&memcg->css, nr_pages); | |
d01dd17f KH |
2407 | } |
2408 | ||
a3b2d692 KH |
2409 | /* |
2410 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2411 | * rcu_read_lock(). The caller is responsible for calling |
2412 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2413 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2414 | */ |
2415 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2416 | { | |
a3b2d692 KH |
2417 | /* ID 0 is unused ID */ |
2418 | if (!id) | |
2419 | return NULL; | |
34c00c31 | 2420 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2421 | } |
2422 | ||
0a31bc97 JW |
2423 | /* |
2424 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2425 | * @page: the page | |
2426 | * | |
2427 | * Look up, get a css reference, and return the memcg that owns @page. | |
2428 | * | |
2429 | * The page must be locked to prevent racing with swap-in and page | |
2430 | * cache charges. If coming from an unlocked page table, the caller | |
2431 | * must ensure the page is on the LRU or this can race with charging. | |
2432 | */ | |
e42d9d5d | 2433 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2434 | { |
29833315 | 2435 | struct mem_cgroup *memcg; |
3c776e64 | 2436 | struct page_cgroup *pc; |
a3b2d692 | 2437 | unsigned short id; |
b5a84319 KH |
2438 | swp_entry_t ent; |
2439 | ||
309381fe | 2440 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2441 | |
3c776e64 | 2442 | pc = lookup_page_cgroup(page); |
29833315 JW |
2443 | memcg = pc->mem_cgroup; |
2444 | ||
2445 | if (memcg) { | |
2446 | if (!css_tryget_online(&memcg->css)) | |
c0ff4b85 | 2447 | memcg = NULL; |
e42d9d5d | 2448 | } else if (PageSwapCache(page)) { |
3c776e64 | 2449 | ent.val = page_private(page); |
9fb4b7cc | 2450 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2451 | rcu_read_lock(); |
c0ff4b85 | 2452 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2453 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2454 | memcg = NULL; |
a3b2d692 | 2455 | rcu_read_unlock(); |
3c776e64 | 2456 | } |
c0ff4b85 | 2457 | return memcg; |
b5a84319 KH |
2458 | } |
2459 | ||
0a31bc97 JW |
2460 | static void lock_page_lru(struct page *page, int *isolated) |
2461 | { | |
2462 | struct zone *zone = page_zone(page); | |
2463 | ||
2464 | spin_lock_irq(&zone->lru_lock); | |
2465 | if (PageLRU(page)) { | |
2466 | struct lruvec *lruvec; | |
2467 | ||
2468 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2469 | ClearPageLRU(page); | |
2470 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2471 | *isolated = 1; | |
2472 | } else | |
2473 | *isolated = 0; | |
2474 | } | |
2475 | ||
2476 | static void unlock_page_lru(struct page *page, int isolated) | |
2477 | { | |
2478 | struct zone *zone = page_zone(page); | |
2479 | ||
2480 | if (isolated) { | |
2481 | struct lruvec *lruvec; | |
2482 | ||
2483 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2484 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2485 | SetPageLRU(page); | |
2486 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2487 | } | |
2488 | spin_unlock_irq(&zone->lru_lock); | |
2489 | } | |
2490 | ||
00501b53 | 2491 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2492 | bool lrucare) |
7a81b88c | 2493 | { |
ce587e65 | 2494 | struct page_cgroup *pc = lookup_page_cgroup(page); |
0a31bc97 | 2495 | int isolated; |
9ce70c02 | 2496 | |
29833315 | 2497 | VM_BUG_ON_PAGE(pc->mem_cgroup, page); |
ca3e0214 KH |
2498 | /* |
2499 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2500 | * accessed by any other context at this point. | |
2501 | */ | |
9ce70c02 HD |
2502 | |
2503 | /* | |
2504 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2505 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2506 | */ | |
0a31bc97 JW |
2507 | if (lrucare) |
2508 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2509 | |
0a31bc97 JW |
2510 | /* |
2511 | * Nobody should be changing or seriously looking at | |
29833315 | 2512 | * pc->mem_cgroup at this point: |
0a31bc97 JW |
2513 | * |
2514 | * - the page is uncharged | |
2515 | * | |
2516 | * - the page is off-LRU | |
2517 | * | |
2518 | * - an anonymous fault has exclusive page access, except for | |
2519 | * a locked page table | |
2520 | * | |
2521 | * - a page cache insertion, a swapin fault, or a migration | |
2522 | * have the page locked | |
2523 | */ | |
c0ff4b85 | 2524 | pc->mem_cgroup = memcg; |
9ce70c02 | 2525 | |
0a31bc97 JW |
2526 | if (lrucare) |
2527 | unlock_page_lru(page, isolated); | |
7a81b88c | 2528 | } |
66e1707b | 2529 | |
7ae1e1d0 | 2530 | #ifdef CONFIG_MEMCG_KMEM |
bd673145 VD |
2531 | /* |
2532 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or | |
2533 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. | |
2534 | */ | |
2535 | static DEFINE_MUTEX(memcg_slab_mutex); | |
2536 | ||
1f458cbf GC |
2537 | /* |
2538 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2539 | * in the memcg_cache_params struct. | |
2540 | */ | |
2541 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2542 | { | |
2543 | struct kmem_cache *cachep; | |
2544 | ||
2545 | VM_BUG_ON(p->is_root_cache); | |
2546 | cachep = p->root_cache; | |
7a67d7ab | 2547 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2548 | } |
2549 | ||
749c5415 | 2550 | #ifdef CONFIG_SLABINFO |
2da8ca82 | 2551 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
749c5415 | 2552 | { |
2da8ca82 | 2553 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
749c5415 GC |
2554 | struct memcg_cache_params *params; |
2555 | ||
cf2b8fbf | 2556 | if (!memcg_kmem_is_active(memcg)) |
749c5415 GC |
2557 | return -EIO; |
2558 | ||
2559 | print_slabinfo_header(m); | |
2560 | ||
bd673145 | 2561 | mutex_lock(&memcg_slab_mutex); |
749c5415 GC |
2562 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) |
2563 | cache_show(memcg_params_to_cache(params), m); | |
bd673145 | 2564 | mutex_unlock(&memcg_slab_mutex); |
749c5415 GC |
2565 | |
2566 | return 0; | |
2567 | } | |
2568 | #endif | |
2569 | ||
3e32cb2e JW |
2570 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2571 | unsigned long nr_pages) | |
7ae1e1d0 | 2572 | { |
3e32cb2e | 2573 | struct page_counter *counter; |
7ae1e1d0 | 2574 | int ret = 0; |
7ae1e1d0 | 2575 | |
3e32cb2e JW |
2576 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2577 | if (ret < 0) | |
7ae1e1d0 GC |
2578 | return ret; |
2579 | ||
3e32cb2e | 2580 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2581 | if (ret == -EINTR) { |
2582 | /* | |
00501b53 JW |
2583 | * try_charge() chose to bypass to root due to OOM kill or |
2584 | * fatal signal. Since our only options are to either fail | |
2585 | * the allocation or charge it to this cgroup, do it as a | |
2586 | * temporary condition. But we can't fail. From a kmem/slab | |
2587 | * perspective, the cache has already been selected, by | |
2588 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2589 | * our minds. |
2590 | * | |
2591 | * This condition will only trigger if the task entered | |
00501b53 JW |
2592 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2593 | * during try_charge() above. Tasks that were already dying | |
2594 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2595 | * directed to the root cgroup in memcontrol.h |
2596 | */ | |
3e32cb2e | 2597 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2598 | if (do_swap_account) |
3e32cb2e | 2599 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2600 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2601 | ret = 0; |
2602 | } else if (ret) | |
3e32cb2e | 2603 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2604 | |
2605 | return ret; | |
2606 | } | |
2607 | ||
3e32cb2e JW |
2608 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, |
2609 | unsigned long nr_pages) | |
7ae1e1d0 | 2610 | { |
3e32cb2e | 2611 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2612 | if (do_swap_account) |
3e32cb2e | 2613 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2614 | |
64f21993 | 2615 | page_counter_uncharge(&memcg->kmem, nr_pages); |
e8ea14cc JW |
2616 | |
2617 | css_put_many(&memcg->css, nr_pages); | |
7ae1e1d0 GC |
2618 | } |
2619 | ||
2633d7a0 GC |
2620 | /* |
2621 | * helper for acessing a memcg's index. It will be used as an index in the | |
2622 | * child cache array in kmem_cache, and also to derive its name. This function | |
2623 | * will return -1 when this is not a kmem-limited memcg. | |
2624 | */ | |
2625 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2626 | { | |
2627 | return memcg ? memcg->kmemcg_id : -1; | |
2628 | } | |
2629 | ||
f3bb3043 | 2630 | static int memcg_alloc_cache_id(void) |
55007d84 | 2631 | { |
f3bb3043 VD |
2632 | int id, size; |
2633 | int err; | |
2634 | ||
2635 | id = ida_simple_get(&kmem_limited_groups, | |
2636 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2637 | if (id < 0) | |
2638 | return id; | |
55007d84 | 2639 | |
f3bb3043 VD |
2640 | if (id < memcg_limited_groups_array_size) |
2641 | return id; | |
2642 | ||
2643 | /* | |
2644 | * There's no space for the new id in memcg_caches arrays, | |
2645 | * so we have to grow them. | |
2646 | */ | |
2647 | ||
2648 | size = 2 * (id + 1); | |
55007d84 GC |
2649 | if (size < MEMCG_CACHES_MIN_SIZE) |
2650 | size = MEMCG_CACHES_MIN_SIZE; | |
2651 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2652 | size = MEMCG_CACHES_MAX_SIZE; | |
2653 | ||
f3bb3043 VD |
2654 | mutex_lock(&memcg_slab_mutex); |
2655 | err = memcg_update_all_caches(size); | |
2656 | mutex_unlock(&memcg_slab_mutex); | |
2657 | ||
2658 | if (err) { | |
2659 | ida_simple_remove(&kmem_limited_groups, id); | |
2660 | return err; | |
2661 | } | |
2662 | return id; | |
2663 | } | |
2664 | ||
2665 | static void memcg_free_cache_id(int id) | |
2666 | { | |
2667 | ida_simple_remove(&kmem_limited_groups, id); | |
55007d84 GC |
2668 | } |
2669 | ||
2670 | /* | |
2671 | * We should update the current array size iff all caches updates succeed. This | |
2672 | * can only be done from the slab side. The slab mutex needs to be held when | |
2673 | * calling this. | |
2674 | */ | |
2675 | void memcg_update_array_size(int num) | |
2676 | { | |
f3bb3043 | 2677 | memcg_limited_groups_array_size = num; |
55007d84 GC |
2678 | } |
2679 | ||
776ed0f0 VD |
2680 | static void memcg_register_cache(struct mem_cgroup *memcg, |
2681 | struct kmem_cache *root_cache) | |
2633d7a0 | 2682 | { |
93f39eea VD |
2683 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by |
2684 | memcg_slab_mutex */ | |
bd673145 | 2685 | struct kmem_cache *cachep; |
d7f25f8a GC |
2686 | int id; |
2687 | ||
bd673145 VD |
2688 | lockdep_assert_held(&memcg_slab_mutex); |
2689 | ||
2690 | id = memcg_cache_id(memcg); | |
2691 | ||
2692 | /* | |
2693 | * Since per-memcg caches are created asynchronously on first | |
2694 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
2695 | * create the same cache, but only one of them may succeed. | |
2696 | */ | |
2697 | if (cache_from_memcg_idx(root_cache, id)) | |
1aa13254 VD |
2698 | return; |
2699 | ||
073ee1c6 | 2700 | cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1); |
776ed0f0 | 2701 | cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf); |
2edefe11 | 2702 | /* |
bd673145 VD |
2703 | * If we could not create a memcg cache, do not complain, because |
2704 | * that's not critical at all as we can always proceed with the root | |
2705 | * cache. | |
2edefe11 | 2706 | */ |
bd673145 VD |
2707 | if (!cachep) |
2708 | return; | |
2edefe11 | 2709 | |
33a690c4 | 2710 | css_get(&memcg->css); |
bd673145 | 2711 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
1aa13254 | 2712 | |
d7f25f8a | 2713 | /* |
959c8963 VD |
2714 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
2715 | * barrier here to ensure nobody will see the kmem_cache partially | |
2716 | * initialized. | |
d7f25f8a | 2717 | */ |
959c8963 VD |
2718 | smp_wmb(); |
2719 | ||
bd673145 VD |
2720 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
2721 | root_cache->memcg_params->memcg_caches[id] = cachep; | |
1aa13254 | 2722 | } |
d7f25f8a | 2723 | |
776ed0f0 | 2724 | static void memcg_unregister_cache(struct kmem_cache *cachep) |
1aa13254 | 2725 | { |
bd673145 | 2726 | struct kmem_cache *root_cache; |
1aa13254 VD |
2727 | struct mem_cgroup *memcg; |
2728 | int id; | |
2729 | ||
bd673145 | 2730 | lockdep_assert_held(&memcg_slab_mutex); |
d7f25f8a | 2731 | |
bd673145 | 2732 | BUG_ON(is_root_cache(cachep)); |
2edefe11 | 2733 | |
bd673145 VD |
2734 | root_cache = cachep->memcg_params->root_cache; |
2735 | memcg = cachep->memcg_params->memcg; | |
96403da2 | 2736 | id = memcg_cache_id(memcg); |
d7f25f8a | 2737 | |
bd673145 VD |
2738 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
2739 | root_cache->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a | 2740 | |
bd673145 VD |
2741 | list_del(&cachep->memcg_params->list); |
2742 | ||
2743 | kmem_cache_destroy(cachep); | |
33a690c4 VD |
2744 | |
2745 | /* drop the reference taken in memcg_register_cache */ | |
2746 | css_put(&memcg->css); | |
2633d7a0 GC |
2747 | } |
2748 | ||
0e9d92f2 GC |
2749 | /* |
2750 | * During the creation a new cache, we need to disable our accounting mechanism | |
2751 | * altogether. This is true even if we are not creating, but rather just | |
2752 | * enqueing new caches to be created. | |
2753 | * | |
2754 | * This is because that process will trigger allocations; some visible, like | |
2755 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
2756 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
2757 | * objects during debug. | |
2758 | * | |
2759 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
2760 | * to it. This may not be a bounded recursion: since the first cache creation | |
2761 | * failed to complete (waiting on the allocation), we'll just try to create the | |
2762 | * cache again, failing at the same point. | |
2763 | * | |
2764 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
2765 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
2766 | * inside the following two functions. | |
2767 | */ | |
2768 | static inline void memcg_stop_kmem_account(void) | |
2769 | { | |
2770 | VM_BUG_ON(!current->mm); | |
2771 | current->memcg_kmem_skip_account++; | |
2772 | } | |
2773 | ||
2774 | static inline void memcg_resume_kmem_account(void) | |
2775 | { | |
2776 | VM_BUG_ON(!current->mm); | |
2777 | current->memcg_kmem_skip_account--; | |
2778 | } | |
2779 | ||
776ed0f0 | 2780 | int __memcg_cleanup_cache_params(struct kmem_cache *s) |
7cf27982 GC |
2781 | { |
2782 | struct kmem_cache *c; | |
b8529907 | 2783 | int i, failed = 0; |
7cf27982 | 2784 | |
bd673145 | 2785 | mutex_lock(&memcg_slab_mutex); |
7a67d7ab QH |
2786 | for_each_memcg_cache_index(i) { |
2787 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
2788 | if (!c) |
2789 | continue; | |
2790 | ||
776ed0f0 | 2791 | memcg_unregister_cache(c); |
b8529907 VD |
2792 | |
2793 | if (cache_from_memcg_idx(s, i)) | |
2794 | failed++; | |
7cf27982 | 2795 | } |
bd673145 | 2796 | mutex_unlock(&memcg_slab_mutex); |
b8529907 | 2797 | return failed; |
7cf27982 GC |
2798 | } |
2799 | ||
776ed0f0 | 2800 | static void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
2801 | { |
2802 | struct kmem_cache *cachep; | |
bd673145 | 2803 | struct memcg_cache_params *params, *tmp; |
1f458cbf GC |
2804 | |
2805 | if (!memcg_kmem_is_active(memcg)) | |
2806 | return; | |
2807 | ||
bd673145 VD |
2808 | mutex_lock(&memcg_slab_mutex); |
2809 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { | |
1f458cbf | 2810 | cachep = memcg_params_to_cache(params); |
bd673145 VD |
2811 | kmem_cache_shrink(cachep); |
2812 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
776ed0f0 | 2813 | memcg_unregister_cache(cachep); |
1f458cbf | 2814 | } |
bd673145 | 2815 | mutex_unlock(&memcg_slab_mutex); |
1f458cbf GC |
2816 | } |
2817 | ||
776ed0f0 | 2818 | struct memcg_register_cache_work { |
5722d094 VD |
2819 | struct mem_cgroup *memcg; |
2820 | struct kmem_cache *cachep; | |
2821 | struct work_struct work; | |
2822 | }; | |
2823 | ||
776ed0f0 | 2824 | static void memcg_register_cache_func(struct work_struct *w) |
d7f25f8a | 2825 | { |
776ed0f0 VD |
2826 | struct memcg_register_cache_work *cw = |
2827 | container_of(w, struct memcg_register_cache_work, work); | |
5722d094 VD |
2828 | struct mem_cgroup *memcg = cw->memcg; |
2829 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2830 | |
bd673145 | 2831 | mutex_lock(&memcg_slab_mutex); |
776ed0f0 | 2832 | memcg_register_cache(memcg, cachep); |
bd673145 VD |
2833 | mutex_unlock(&memcg_slab_mutex); |
2834 | ||
5722d094 | 2835 | css_put(&memcg->css); |
d7f25f8a GC |
2836 | kfree(cw); |
2837 | } | |
2838 | ||
2839 | /* | |
2840 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2841 | */ |
776ed0f0 VD |
2842 | static void __memcg_schedule_register_cache(struct mem_cgroup *memcg, |
2843 | struct kmem_cache *cachep) | |
d7f25f8a | 2844 | { |
776ed0f0 | 2845 | struct memcg_register_cache_work *cw; |
d7f25f8a | 2846 | |
776ed0f0 | 2847 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
ca0dde97 LZ |
2848 | if (cw == NULL) { |
2849 | css_put(&memcg->css); | |
d7f25f8a GC |
2850 | return; |
2851 | } | |
2852 | ||
2853 | cw->memcg = memcg; | |
2854 | cw->cachep = cachep; | |
2855 | ||
776ed0f0 | 2856 | INIT_WORK(&cw->work, memcg_register_cache_func); |
d7f25f8a GC |
2857 | schedule_work(&cw->work); |
2858 | } | |
2859 | ||
776ed0f0 VD |
2860 | static void memcg_schedule_register_cache(struct mem_cgroup *memcg, |
2861 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2862 | { |
2863 | /* | |
2864 | * We need to stop accounting when we kmalloc, because if the | |
2865 | * corresponding kmalloc cache is not yet created, the first allocation | |
776ed0f0 | 2866 | * in __memcg_schedule_register_cache will recurse. |
0e9d92f2 GC |
2867 | * |
2868 | * However, it is better to enclose the whole function. Depending on | |
2869 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2870 | * trigger an allocation. This too, will make us recurse. Because at | |
2871 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2872 | * the safest choice is to do it like this, wrapping the whole function. | |
2873 | */ | |
2874 | memcg_stop_kmem_account(); | |
776ed0f0 | 2875 | __memcg_schedule_register_cache(memcg, cachep); |
0e9d92f2 GC |
2876 | memcg_resume_kmem_account(); |
2877 | } | |
c67a8a68 VD |
2878 | |
2879 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) | |
2880 | { | |
3e32cb2e | 2881 | unsigned int nr_pages = 1 << order; |
c67a8a68 VD |
2882 | int res; |
2883 | ||
3e32cb2e | 2884 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, nr_pages); |
c67a8a68 | 2885 | if (!res) |
3e32cb2e | 2886 | atomic_add(nr_pages, &cachep->memcg_params->nr_pages); |
c67a8a68 VD |
2887 | return res; |
2888 | } | |
2889 | ||
2890 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) | |
2891 | { | |
3e32cb2e JW |
2892 | unsigned int nr_pages = 1 << order; |
2893 | ||
2894 | memcg_uncharge_kmem(cachep->memcg_params->memcg, nr_pages); | |
2895 | atomic_sub(nr_pages, &cachep->memcg_params->nr_pages); | |
c67a8a68 VD |
2896 | } |
2897 | ||
d7f25f8a GC |
2898 | /* |
2899 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2900 | * We try to use the current memcg's version of the cache. | |
2901 | * | |
2902 | * If the cache does not exist yet, if we are the first user of it, | |
2903 | * we either create it immediately, if possible, or create it asynchronously | |
2904 | * in a workqueue. | |
2905 | * In the latter case, we will let the current allocation go through with | |
2906 | * the original cache. | |
2907 | * | |
2908 | * Can't be called in interrupt context or from kernel threads. | |
2909 | * This function needs to be called with rcu_read_lock() held. | |
2910 | */ | |
2911 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
2912 | gfp_t gfp) | |
2913 | { | |
2914 | struct mem_cgroup *memcg; | |
959c8963 | 2915 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
2916 | |
2917 | VM_BUG_ON(!cachep->memcg_params); | |
2918 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
2919 | ||
0e9d92f2 GC |
2920 | if (!current->mm || current->memcg_kmem_skip_account) |
2921 | return cachep; | |
2922 | ||
d7f25f8a GC |
2923 | rcu_read_lock(); |
2924 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a | 2925 | |
cf2b8fbf | 2926 | if (!memcg_kmem_is_active(memcg)) |
ca0dde97 | 2927 | goto out; |
d7f25f8a | 2928 | |
959c8963 VD |
2929 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
2930 | if (likely(memcg_cachep)) { | |
2931 | cachep = memcg_cachep; | |
ca0dde97 | 2932 | goto out; |
d7f25f8a GC |
2933 | } |
2934 | ||
ca0dde97 | 2935 | /* The corresponding put will be done in the workqueue. */ |
ec903c0c | 2936 | if (!css_tryget_online(&memcg->css)) |
ca0dde97 LZ |
2937 | goto out; |
2938 | rcu_read_unlock(); | |
2939 | ||
2940 | /* | |
2941 | * If we are in a safe context (can wait, and not in interrupt | |
2942 | * context), we could be be predictable and return right away. | |
2943 | * This would guarantee that the allocation being performed | |
2944 | * already belongs in the new cache. | |
2945 | * | |
2946 | * However, there are some clashes that can arrive from locking. | |
2947 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2948 | * memcg_create_kmem_cache, this means no further allocation |
2949 | * could happen with the slab_mutex held. So it's better to | |
2950 | * defer everything. | |
ca0dde97 | 2951 | */ |
776ed0f0 | 2952 | memcg_schedule_register_cache(memcg, cachep); |
ca0dde97 LZ |
2953 | return cachep; |
2954 | out: | |
2955 | rcu_read_unlock(); | |
2956 | return cachep; | |
d7f25f8a | 2957 | } |
d7f25f8a | 2958 | |
7ae1e1d0 GC |
2959 | /* |
2960 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2961 | * possible for the given order. But the page is not allocated yet, so we'll | |
2962 | * need a further commit step to do the final arrangements. | |
2963 | * | |
2964 | * It is possible for the task to switch cgroups in this mean time, so at | |
2965 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2966 | * the handle argument to return to the caller which cgroup we should commit | |
2967 | * against. We could also return the memcg directly and avoid the pointer | |
2968 | * passing, but a boolean return value gives better semantics considering | |
2969 | * the compiled-out case as well. | |
2970 | * | |
2971 | * Returning true means the allocation is possible. | |
2972 | */ | |
2973 | bool | |
2974 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2975 | { | |
2976 | struct mem_cgroup *memcg; | |
2977 | int ret; | |
2978 | ||
2979 | *_memcg = NULL; | |
6d42c232 GC |
2980 | |
2981 | /* | |
2982 | * Disabling accounting is only relevant for some specific memcg | |
2983 | * internal allocations. Therefore we would initially not have such | |
52383431 VD |
2984 | * check here, since direct calls to the page allocator that are |
2985 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen | |
2986 | * outside memcg core. We are mostly concerned with cache allocations, | |
2987 | * and by having this test at memcg_kmem_get_cache, we are already able | |
2988 | * to relay the allocation to the root cache and bypass the memcg cache | |
2989 | * altogether. | |
6d42c232 GC |
2990 | * |
2991 | * There is one exception, though: the SLUB allocator does not create | |
2992 | * large order caches, but rather service large kmallocs directly from | |
2993 | * the page allocator. Therefore, the following sequence when backed by | |
2994 | * the SLUB allocator: | |
2995 | * | |
f894ffa8 AM |
2996 | * memcg_stop_kmem_account(); |
2997 | * kmalloc(<large_number>) | |
2998 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
2999 | * |
3000 | * would effectively ignore the fact that we should skip accounting, | |
3001 | * since it will drive us directly to this function without passing | |
3002 | * through the cache selector memcg_kmem_get_cache. Such large | |
3003 | * allocations are extremely rare but can happen, for instance, for the | |
3004 | * cache arrays. We bring this test here. | |
3005 | */ | |
3006 | if (!current->mm || current->memcg_kmem_skip_account) | |
3007 | return true; | |
3008 | ||
df381975 | 3009 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 3010 | |
cf2b8fbf | 3011 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
3012 | css_put(&memcg->css); |
3013 | return true; | |
3014 | } | |
3015 | ||
3e32cb2e | 3016 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
3017 | if (!ret) |
3018 | *_memcg = memcg; | |
7ae1e1d0 GC |
3019 | |
3020 | css_put(&memcg->css); | |
3021 | return (ret == 0); | |
3022 | } | |
3023 | ||
3024 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3025 | int order) | |
3026 | { | |
3027 | struct page_cgroup *pc; | |
3028 | ||
3029 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3030 | ||
3031 | /* The page allocation failed. Revert */ | |
3032 | if (!page) { | |
3e32cb2e | 3033 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
3034 | return; |
3035 | } | |
7ae1e1d0 | 3036 | pc = lookup_page_cgroup(page); |
7ae1e1d0 | 3037 | pc->mem_cgroup = memcg; |
7ae1e1d0 GC |
3038 | } |
3039 | ||
3040 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3041 | { | |
29833315 JW |
3042 | struct page_cgroup *pc = lookup_page_cgroup(page); |
3043 | struct mem_cgroup *memcg = pc->mem_cgroup; | |
7ae1e1d0 | 3044 | |
7ae1e1d0 GC |
3045 | if (!memcg) |
3046 | return; | |
3047 | ||
309381fe | 3048 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 3049 | |
3e32cb2e | 3050 | memcg_uncharge_kmem(memcg, 1 << order); |
29833315 | 3051 | pc->mem_cgroup = NULL; |
7ae1e1d0 | 3052 | } |
1f458cbf | 3053 | #else |
776ed0f0 | 3054 | static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3055 | { |
3056 | } | |
7ae1e1d0 GC |
3057 | #endif /* CONFIG_MEMCG_KMEM */ |
3058 | ||
ca3e0214 KH |
3059 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3060 | ||
ca3e0214 KH |
3061 | /* |
3062 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3063 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3064 | * charge/uncharge will be never happen and move_account() is done under | |
3065 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3066 | */ |
e94c8a9c | 3067 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3068 | { |
29833315 | 3069 | struct page_cgroup *pc = lookup_page_cgroup(head); |
e94c8a9c | 3070 | int i; |
ca3e0214 | 3071 | |
3d37c4a9 KH |
3072 | if (mem_cgroup_disabled()) |
3073 | return; | |
b070e65c | 3074 | |
29833315 JW |
3075 | for (i = 1; i < HPAGE_PMD_NR; i++) |
3076 | pc[i].mem_cgroup = pc[0].mem_cgroup; | |
b9982f8d | 3077 | |
29833315 | 3078 | __this_cpu_sub(pc[0].mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 3079 | HPAGE_PMD_NR); |
ca3e0214 | 3080 | } |
12d27107 | 3081 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3082 | |
f817ed48 | 3083 | /** |
de3638d9 | 3084 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3085 | * @page: the page |
7ec99d62 | 3086 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3087 | * @pc: page_cgroup of the page. |
3088 | * @from: mem_cgroup which the page is moved from. | |
3089 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3090 | * | |
3091 | * The caller must confirm following. | |
08e552c6 | 3092 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3093 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3094 | * |
2f3479b1 KH |
3095 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3096 | * from old cgroup. | |
f817ed48 | 3097 | */ |
7ec99d62 JW |
3098 | static int mem_cgroup_move_account(struct page *page, |
3099 | unsigned int nr_pages, | |
3100 | struct page_cgroup *pc, | |
3101 | struct mem_cgroup *from, | |
2f3479b1 | 3102 | struct mem_cgroup *to) |
f817ed48 | 3103 | { |
de3638d9 JW |
3104 | unsigned long flags; |
3105 | int ret; | |
987eba66 | 3106 | |
f817ed48 | 3107 | VM_BUG_ON(from == to); |
309381fe | 3108 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3109 | /* |
3110 | * The page is isolated from LRU. So, collapse function | |
3111 | * will not handle this page. But page splitting can happen. | |
3112 | * Do this check under compound_page_lock(). The caller should | |
3113 | * hold it. | |
3114 | */ | |
3115 | ret = -EBUSY; | |
7ec99d62 | 3116 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3117 | goto out; |
3118 | ||
0a31bc97 JW |
3119 | /* |
3120 | * Prevent mem_cgroup_migrate() from looking at pc->mem_cgroup | |
3121 | * of its source page while we change it: page migration takes | |
3122 | * both pages off the LRU, but page cache replacement doesn't. | |
3123 | */ | |
3124 | if (!trylock_page(page)) | |
3125 | goto out; | |
de3638d9 JW |
3126 | |
3127 | ret = -EINVAL; | |
29833315 | 3128 | if (pc->mem_cgroup != from) |
0a31bc97 | 3129 | goto out_unlock; |
de3638d9 | 3130 | |
354a4783 | 3131 | spin_lock_irqsave(&from->move_lock, flags); |
f817ed48 | 3132 | |
0a31bc97 | 3133 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
3134 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
3135 | nr_pages); | |
3136 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3137 | nr_pages); | |
3138 | } | |
3ea67d06 | 3139 | |
59d1d256 JW |
3140 | if (PageWriteback(page)) { |
3141 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3142 | nr_pages); | |
3143 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3144 | nr_pages); | |
3145 | } | |
3ea67d06 | 3146 | |
0a31bc97 JW |
3147 | /* |
3148 | * It is safe to change pc->mem_cgroup here because the page | |
3149 | * is referenced, charged, and isolated - we can't race with | |
3150 | * uncharging, charging, migration, or LRU putback. | |
3151 | */ | |
d69b042f | 3152 | |
854ffa8d | 3153 | /* caller should have done css_get */ |
08e552c6 | 3154 | pc->mem_cgroup = to; |
354a4783 JW |
3155 | spin_unlock_irqrestore(&from->move_lock, flags); |
3156 | ||
de3638d9 | 3157 | ret = 0; |
0a31bc97 JW |
3158 | |
3159 | local_irq_disable(); | |
3160 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 3161 | memcg_check_events(to, page); |
0a31bc97 | 3162 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 3163 | memcg_check_events(from, page); |
0a31bc97 JW |
3164 | local_irq_enable(); |
3165 | out_unlock: | |
3166 | unlock_page(page); | |
de3638d9 | 3167 | out: |
f817ed48 KH |
3168 | return ret; |
3169 | } | |
3170 | ||
c255a458 | 3171 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
3172 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
3173 | bool charge) | |
d13d1443 | 3174 | { |
0a31bc97 JW |
3175 | int val = (charge) ? 1 : -1; |
3176 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 3177 | } |
02491447 DN |
3178 | |
3179 | /** | |
3180 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3181 | * @entry: swap entry to be moved | |
3182 | * @from: mem_cgroup which the entry is moved from | |
3183 | * @to: mem_cgroup which the entry is moved to | |
3184 | * | |
3185 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3186 | * as the mem_cgroup's id of @from. | |
3187 | * | |
3188 | * Returns 0 on success, -EINVAL on failure. | |
3189 | * | |
3e32cb2e | 3190 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3191 | * both res and memsw, and called css_get(). |
3192 | */ | |
3193 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3194 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3195 | { |
3196 | unsigned short old_id, new_id; | |
3197 | ||
34c00c31 LZ |
3198 | old_id = mem_cgroup_id(from); |
3199 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3200 | |
3201 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3202 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3203 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3204 | /* |
483c30b5 | 3205 | * This function is only called from task migration context now. |
3e32cb2e | 3206 | * It postpones page_counter and refcount handling till the end |
483c30b5 | 3207 | * of task migration(mem_cgroup_clear_mc()) for performance |
4050377b LZ |
3208 | * improvement. But we cannot postpone css_get(to) because if |
3209 | * the process that has been moved to @to does swap-in, the | |
3210 | * refcount of @to might be decreased to 0. | |
3211 | * | |
3212 | * We are in attach() phase, so the cgroup is guaranteed to be | |
3213 | * alive, so we can just call css_get(). | |
02491447 | 3214 | */ |
4050377b | 3215 | css_get(&to->css); |
02491447 DN |
3216 | return 0; |
3217 | } | |
3218 | return -EINVAL; | |
3219 | } | |
3220 | #else | |
3221 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3222 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3223 | { |
3224 | return -EINVAL; | |
3225 | } | |
8c7c6e34 | 3226 | #endif |
d13d1443 | 3227 | |
f212ad7c DN |
3228 | #ifdef CONFIG_DEBUG_VM |
3229 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3230 | { | |
3231 | struct page_cgroup *pc; | |
3232 | ||
3233 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3234 | /* |
3235 | * Can be NULL while feeding pages into the page allocator for | |
3236 | * the first time, i.e. during boot or memory hotplug; | |
3237 | * or when mem_cgroup_disabled(). | |
3238 | */ | |
29833315 | 3239 | if (likely(pc) && pc->mem_cgroup) |
f212ad7c DN |
3240 | return pc; |
3241 | return NULL; | |
3242 | } | |
3243 | ||
3244 | bool mem_cgroup_bad_page_check(struct page *page) | |
3245 | { | |
3246 | if (mem_cgroup_disabled()) | |
3247 | return false; | |
3248 | ||
3249 | return lookup_page_cgroup_used(page) != NULL; | |
3250 | } | |
3251 | ||
3252 | void mem_cgroup_print_bad_page(struct page *page) | |
3253 | { | |
3254 | struct page_cgroup *pc; | |
3255 | ||
3256 | pc = lookup_page_cgroup_used(page); | |
29833315 JW |
3257 | if (pc) |
3258 | pr_alert("pc:%p pc->mem_cgroup:%p\n", pc, pc->mem_cgroup); | |
f212ad7c DN |
3259 | } |
3260 | #endif | |
3261 | ||
3e32cb2e JW |
3262 | static DEFINE_MUTEX(memcg_limit_mutex); |
3263 | ||
d38d2a75 | 3264 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3265 | unsigned long limit) |
628f4235 | 3266 | { |
3e32cb2e JW |
3267 | unsigned long curusage; |
3268 | unsigned long oldusage; | |
3269 | bool enlarge = false; | |
81d39c20 | 3270 | int retry_count; |
3e32cb2e | 3271 | int ret; |
81d39c20 KH |
3272 | |
3273 | /* | |
3274 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3275 | * is depends on callers. We set our retry-count to be function | |
3276 | * of # of children which we should visit in this loop. | |
3277 | */ | |
3e32cb2e JW |
3278 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3279 | mem_cgroup_count_children(memcg); | |
81d39c20 | 3280 | |
3e32cb2e | 3281 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 3282 | |
3e32cb2e | 3283 | do { |
628f4235 KH |
3284 | if (signal_pending(current)) { |
3285 | ret = -EINTR; | |
3286 | break; | |
3287 | } | |
3e32cb2e JW |
3288 | |
3289 | mutex_lock(&memcg_limit_mutex); | |
3290 | if (limit > memcg->memsw.limit) { | |
3291 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3292 | ret = -EINVAL; |
628f4235 KH |
3293 | break; |
3294 | } | |
3e32cb2e JW |
3295 | if (limit > memcg->memory.limit) |
3296 | enlarge = true; | |
3297 | ret = page_counter_limit(&memcg->memory, limit); | |
3298 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3299 | |
3300 | if (!ret) | |
3301 | break; | |
3302 | ||
b70a2a21 JW |
3303 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
3304 | ||
3e32cb2e | 3305 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 3306 | /* Usage is reduced ? */ |
f894ffa8 | 3307 | if (curusage >= oldusage) |
81d39c20 KH |
3308 | retry_count--; |
3309 | else | |
3310 | oldusage = curusage; | |
3e32cb2e JW |
3311 | } while (retry_count); |
3312 | ||
3c11ecf4 KH |
3313 | if (!ret && enlarge) |
3314 | memcg_oom_recover(memcg); | |
14797e23 | 3315 | |
8c7c6e34 KH |
3316 | return ret; |
3317 | } | |
3318 | ||
338c8431 | 3319 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3320 | unsigned long limit) |
8c7c6e34 | 3321 | { |
3e32cb2e JW |
3322 | unsigned long curusage; |
3323 | unsigned long oldusage; | |
3324 | bool enlarge = false; | |
81d39c20 | 3325 | int retry_count; |
3e32cb2e | 3326 | int ret; |
8c7c6e34 | 3327 | |
81d39c20 | 3328 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
3329 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3330 | mem_cgroup_count_children(memcg); | |
3331 | ||
3332 | oldusage = page_counter_read(&memcg->memsw); | |
3333 | ||
3334 | do { | |
8c7c6e34 KH |
3335 | if (signal_pending(current)) { |
3336 | ret = -EINTR; | |
3337 | break; | |
3338 | } | |
3e32cb2e JW |
3339 | |
3340 | mutex_lock(&memcg_limit_mutex); | |
3341 | if (limit < memcg->memory.limit) { | |
3342 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3343 | ret = -EINVAL; |
8c7c6e34 KH |
3344 | break; |
3345 | } | |
3e32cb2e JW |
3346 | if (limit > memcg->memsw.limit) |
3347 | enlarge = true; | |
3348 | ret = page_counter_limit(&memcg->memsw, limit); | |
3349 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3350 | |
3351 | if (!ret) | |
3352 | break; | |
3353 | ||
b70a2a21 JW |
3354 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3355 | ||
3e32cb2e | 3356 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3357 | /* Usage is reduced ? */ |
8c7c6e34 | 3358 | if (curusage >= oldusage) |
628f4235 | 3359 | retry_count--; |
81d39c20 KH |
3360 | else |
3361 | oldusage = curusage; | |
3e32cb2e JW |
3362 | } while (retry_count); |
3363 | ||
3c11ecf4 KH |
3364 | if (!ret && enlarge) |
3365 | memcg_oom_recover(memcg); | |
3e32cb2e | 3366 | |
628f4235 KH |
3367 | return ret; |
3368 | } | |
3369 | ||
0608f43d AM |
3370 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3371 | gfp_t gfp_mask, | |
3372 | unsigned long *total_scanned) | |
3373 | { | |
3374 | unsigned long nr_reclaimed = 0; | |
3375 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3376 | unsigned long reclaimed; | |
3377 | int loop = 0; | |
3378 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3379 | unsigned long excess; |
0608f43d AM |
3380 | unsigned long nr_scanned; |
3381 | ||
3382 | if (order > 0) | |
3383 | return 0; | |
3384 | ||
3385 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3386 | /* | |
3387 | * This loop can run a while, specially if mem_cgroup's continuously | |
3388 | * keep exceeding their soft limit and putting the system under | |
3389 | * pressure | |
3390 | */ | |
3391 | do { | |
3392 | if (next_mz) | |
3393 | mz = next_mz; | |
3394 | else | |
3395 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3396 | if (!mz) | |
3397 | break; | |
3398 | ||
3399 | nr_scanned = 0; | |
3400 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3401 | gfp_mask, &nr_scanned); | |
3402 | nr_reclaimed += reclaimed; | |
3403 | *total_scanned += nr_scanned; | |
0a31bc97 | 3404 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3405 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3406 | |
3407 | /* | |
3408 | * If we failed to reclaim anything from this memory cgroup | |
3409 | * it is time to move on to the next cgroup | |
3410 | */ | |
3411 | next_mz = NULL; | |
bc2f2e7f VD |
3412 | if (!reclaimed) |
3413 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3414 | ||
3e32cb2e | 3415 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3416 | /* |
3417 | * One school of thought says that we should not add | |
3418 | * back the node to the tree if reclaim returns 0. | |
3419 | * But our reclaim could return 0, simply because due | |
3420 | * to priority we are exposing a smaller subset of | |
3421 | * memory to reclaim from. Consider this as a longer | |
3422 | * term TODO. | |
3423 | */ | |
3424 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3425 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3426 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3427 | css_put(&mz->memcg->css); |
3428 | loop++; | |
3429 | /* | |
3430 | * Could not reclaim anything and there are no more | |
3431 | * mem cgroups to try or we seem to be looping without | |
3432 | * reclaiming anything. | |
3433 | */ | |
3434 | if (!nr_reclaimed && | |
3435 | (next_mz == NULL || | |
3436 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3437 | break; | |
3438 | } while (!nr_reclaimed); | |
3439 | if (next_mz) | |
3440 | css_put(&next_mz->memcg->css); | |
3441 | return nr_reclaimed; | |
3442 | } | |
3443 | ||
ea280e7b TH |
3444 | /* |
3445 | * Test whether @memcg has children, dead or alive. Note that this | |
3446 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3447 | * returns %true if there are child csses according to the cgroup | |
3448 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3449 | */ | |
b5f99b53 GC |
3450 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3451 | { | |
ea280e7b TH |
3452 | bool ret; |
3453 | ||
696ac172 | 3454 | /* |
ea280e7b TH |
3455 | * The lock does not prevent addition or deletion of children, but |
3456 | * it prevents a new child from being initialized based on this | |
3457 | * parent in css_online(), so it's enough to decide whether | |
3458 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3459 | */ |
ea280e7b TH |
3460 | lockdep_assert_held(&memcg_create_mutex); |
3461 | ||
3462 | rcu_read_lock(); | |
3463 | ret = css_next_child(NULL, &memcg->css); | |
3464 | rcu_read_unlock(); | |
3465 | return ret; | |
b5f99b53 GC |
3466 | } |
3467 | ||
c26251f9 MH |
3468 | /* |
3469 | * Reclaims as many pages from the given memcg as possible and moves | |
3470 | * the rest to the parent. | |
3471 | * | |
3472 | * Caller is responsible for holding css reference for memcg. | |
3473 | */ | |
3474 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3475 | { | |
3476 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3477 | |
c1e862c1 KH |
3478 | /* we call try-to-free pages for make this cgroup empty */ |
3479 | lru_add_drain_all(); | |
f817ed48 | 3480 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3481 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3482 | int progress; |
c1e862c1 | 3483 | |
c26251f9 MH |
3484 | if (signal_pending(current)) |
3485 | return -EINTR; | |
3486 | ||
b70a2a21 JW |
3487 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3488 | GFP_KERNEL, true); | |
c1e862c1 | 3489 | if (!progress) { |
f817ed48 | 3490 | nr_retries--; |
c1e862c1 | 3491 | /* maybe some writeback is necessary */ |
8aa7e847 | 3492 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3493 | } |
f817ed48 KH |
3494 | |
3495 | } | |
ab5196c2 MH |
3496 | |
3497 | return 0; | |
cc847582 KH |
3498 | } |
3499 | ||
6770c64e TH |
3500 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3501 | char *buf, size_t nbytes, | |
3502 | loff_t off) | |
c1e862c1 | 3503 | { |
6770c64e | 3504 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3505 | |
d8423011 MH |
3506 | if (mem_cgroup_is_root(memcg)) |
3507 | return -EINVAL; | |
6770c64e | 3508 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3509 | } |
3510 | ||
182446d0 TH |
3511 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3512 | struct cftype *cft) | |
18f59ea7 | 3513 | { |
182446d0 | 3514 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3515 | } |
3516 | ||
182446d0 TH |
3517 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3518 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3519 | { |
3520 | int retval = 0; | |
182446d0 | 3521 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3522 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3523 | |
0999821b | 3524 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3525 | |
3526 | if (memcg->use_hierarchy == val) | |
3527 | goto out; | |
3528 | ||
18f59ea7 | 3529 | /* |
af901ca1 | 3530 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3531 | * in the child subtrees. If it is unset, then the change can |
3532 | * occur, provided the current cgroup has no children. | |
3533 | * | |
3534 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3535 | * set if there are no children. | |
3536 | */ | |
c0ff4b85 | 3537 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3538 | (val == 1 || val == 0)) { |
ea280e7b | 3539 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3540 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3541 | else |
3542 | retval = -EBUSY; | |
3543 | } else | |
3544 | retval = -EINVAL; | |
567fb435 GC |
3545 | |
3546 | out: | |
0999821b | 3547 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3548 | |
3549 | return retval; | |
3550 | } | |
3551 | ||
3e32cb2e JW |
3552 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3553 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3554 | { |
3555 | struct mem_cgroup *iter; | |
3556 | long val = 0; | |
3557 | ||
3558 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3559 | for_each_mem_cgroup_tree(iter, memcg) | |
3560 | val += mem_cgroup_read_stat(iter, idx); | |
3561 | ||
3562 | if (val < 0) /* race ? */ | |
3563 | val = 0; | |
3564 | return val; | |
3565 | } | |
3566 | ||
3567 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3568 | { | |
3569 | u64 val; | |
3570 | ||
3e32cb2e JW |
3571 | if (mem_cgroup_is_root(memcg)) { |
3572 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3573 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3574 | if (swap) | |
3575 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3576 | } else { | |
ce00a967 | 3577 | if (!swap) |
3e32cb2e | 3578 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3579 | else |
3e32cb2e | 3580 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3581 | } |
ce00a967 JW |
3582 | return val << PAGE_SHIFT; |
3583 | } | |
3584 | ||
3e32cb2e JW |
3585 | enum { |
3586 | RES_USAGE, | |
3587 | RES_LIMIT, | |
3588 | RES_MAX_USAGE, | |
3589 | RES_FAILCNT, | |
3590 | RES_SOFT_LIMIT, | |
3591 | }; | |
ce00a967 | 3592 | |
791badbd | 3593 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3594 | struct cftype *cft) |
8cdea7c0 | 3595 | { |
182446d0 | 3596 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3597 | struct page_counter *counter; |
af36f906 | 3598 | |
3e32cb2e | 3599 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3600 | case _MEM: |
3e32cb2e JW |
3601 | counter = &memcg->memory; |
3602 | break; | |
8c7c6e34 | 3603 | case _MEMSWAP: |
3e32cb2e JW |
3604 | counter = &memcg->memsw; |
3605 | break; | |
510fc4e1 | 3606 | case _KMEM: |
3e32cb2e | 3607 | counter = &memcg->kmem; |
510fc4e1 | 3608 | break; |
8c7c6e34 KH |
3609 | default: |
3610 | BUG(); | |
8c7c6e34 | 3611 | } |
3e32cb2e JW |
3612 | |
3613 | switch (MEMFILE_ATTR(cft->private)) { | |
3614 | case RES_USAGE: | |
3615 | if (counter == &memcg->memory) | |
3616 | return mem_cgroup_usage(memcg, false); | |
3617 | if (counter == &memcg->memsw) | |
3618 | return mem_cgroup_usage(memcg, true); | |
3619 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3620 | case RES_LIMIT: | |
3621 | return (u64)counter->limit * PAGE_SIZE; | |
3622 | case RES_MAX_USAGE: | |
3623 | return (u64)counter->watermark * PAGE_SIZE; | |
3624 | case RES_FAILCNT: | |
3625 | return counter->failcnt; | |
3626 | case RES_SOFT_LIMIT: | |
3627 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3628 | default: | |
3629 | BUG(); | |
3630 | } | |
8cdea7c0 | 3631 | } |
510fc4e1 | 3632 | |
510fc4e1 | 3633 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
3634 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
3635 | unsigned long nr_pages) | |
d6441637 VD |
3636 | { |
3637 | int err = 0; | |
3638 | int memcg_id; | |
3639 | ||
3640 | if (memcg_kmem_is_active(memcg)) | |
3641 | return 0; | |
3642 | ||
3643 | /* | |
3644 | * We are going to allocate memory for data shared by all memory | |
3645 | * cgroups so let's stop accounting here. | |
3646 | */ | |
3647 | memcg_stop_kmem_account(); | |
3648 | ||
510fc4e1 GC |
3649 | /* |
3650 | * For simplicity, we won't allow this to be disabled. It also can't | |
3651 | * be changed if the cgroup has children already, or if tasks had | |
3652 | * already joined. | |
3653 | * | |
3654 | * If tasks join before we set the limit, a person looking at | |
3655 | * kmem.usage_in_bytes will have no way to determine when it took | |
3656 | * place, which makes the value quite meaningless. | |
3657 | * | |
3658 | * After it first became limited, changes in the value of the limit are | |
3659 | * of course permitted. | |
510fc4e1 | 3660 | */ |
0999821b | 3661 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
3662 | if (cgroup_has_tasks(memcg->css.cgroup) || |
3663 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
3664 | err = -EBUSY; |
3665 | mutex_unlock(&memcg_create_mutex); | |
3666 | if (err) | |
3667 | goto out; | |
510fc4e1 | 3668 | |
f3bb3043 | 3669 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
3670 | if (memcg_id < 0) { |
3671 | err = memcg_id; | |
3672 | goto out; | |
3673 | } | |
3674 | ||
d6441637 VD |
3675 | memcg->kmemcg_id = memcg_id; |
3676 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
d6441637 VD |
3677 | |
3678 | /* | |
3679 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
3680 | * active bit set yet, so this should succeed. | |
3681 | */ | |
3e32cb2e | 3682 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
3683 | VM_BUG_ON(err); |
3684 | ||
3685 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
3686 | /* | |
3687 | * Setting the active bit after enabling static branching will | |
3688 | * guarantee no one starts accounting before all call sites are | |
3689 | * patched. | |
3690 | */ | |
3691 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 3692 | out: |
d6441637 VD |
3693 | memcg_resume_kmem_account(); |
3694 | return err; | |
d6441637 VD |
3695 | } |
3696 | ||
d6441637 | 3697 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3698 | unsigned long limit) |
d6441637 VD |
3699 | { |
3700 | int ret; | |
3701 | ||
3e32cb2e | 3702 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 3703 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 3704 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 3705 | else |
3e32cb2e JW |
3706 | ret = page_counter_limit(&memcg->kmem, limit); |
3707 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
3708 | return ret; |
3709 | } | |
3710 | ||
55007d84 | 3711 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 3712 | { |
55007d84 | 3713 | int ret = 0; |
510fc4e1 | 3714 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 3715 | |
d6441637 VD |
3716 | if (!parent) |
3717 | return 0; | |
55007d84 | 3718 | |
8c0145b6 | 3719 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 3720 | /* |
d6441637 VD |
3721 | * If the parent cgroup is not kmem-active now, it cannot be activated |
3722 | * after this point, because it has at least one child already. | |
55007d84 | 3723 | */ |
d6441637 | 3724 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3725 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3726 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3727 | return ret; |
510fc4e1 | 3728 | } |
d6441637 VD |
3729 | #else |
3730 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3731 | unsigned long limit) |
d6441637 VD |
3732 | { |
3733 | return -EINVAL; | |
3734 | } | |
6d043990 | 3735 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3736 | |
628f4235 KH |
3737 | /* |
3738 | * The user of this function is... | |
3739 | * RES_LIMIT. | |
3740 | */ | |
451af504 TH |
3741 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3742 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3743 | { |
451af504 | 3744 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3745 | unsigned long nr_pages; |
628f4235 KH |
3746 | int ret; |
3747 | ||
451af504 | 3748 | buf = strstrip(buf); |
3e32cb2e JW |
3749 | ret = page_counter_memparse(buf, &nr_pages); |
3750 | if (ret) | |
3751 | return ret; | |
af36f906 | 3752 | |
3e32cb2e | 3753 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3754 | case RES_LIMIT: |
4b3bde4c BS |
3755 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3756 | ret = -EINVAL; | |
3757 | break; | |
3758 | } | |
3e32cb2e JW |
3759 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3760 | case _MEM: | |
3761 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3762 | break; |
3e32cb2e JW |
3763 | case _MEMSWAP: |
3764 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3765 | break; |
3e32cb2e JW |
3766 | case _KMEM: |
3767 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3768 | break; | |
3769 | } | |
296c81d8 | 3770 | break; |
3e32cb2e JW |
3771 | case RES_SOFT_LIMIT: |
3772 | memcg->soft_limit = nr_pages; | |
3773 | ret = 0; | |
628f4235 KH |
3774 | break; |
3775 | } | |
451af504 | 3776 | return ret ?: nbytes; |
8cdea7c0 BS |
3777 | } |
3778 | ||
6770c64e TH |
3779 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3780 | size_t nbytes, loff_t off) | |
c84872e1 | 3781 | { |
6770c64e | 3782 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3783 | struct page_counter *counter; |
c84872e1 | 3784 | |
3e32cb2e JW |
3785 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3786 | case _MEM: | |
3787 | counter = &memcg->memory; | |
3788 | break; | |
3789 | case _MEMSWAP: | |
3790 | counter = &memcg->memsw; | |
3791 | break; | |
3792 | case _KMEM: | |
3793 | counter = &memcg->kmem; | |
3794 | break; | |
3795 | default: | |
3796 | BUG(); | |
3797 | } | |
af36f906 | 3798 | |
3e32cb2e | 3799 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3800 | case RES_MAX_USAGE: |
3e32cb2e | 3801 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3802 | break; |
3803 | case RES_FAILCNT: | |
3e32cb2e | 3804 | counter->failcnt = 0; |
29f2a4da | 3805 | break; |
3e32cb2e JW |
3806 | default: |
3807 | BUG(); | |
29f2a4da | 3808 | } |
f64c3f54 | 3809 | |
6770c64e | 3810 | return nbytes; |
c84872e1 PE |
3811 | } |
3812 | ||
182446d0 | 3813 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3814 | struct cftype *cft) |
3815 | { | |
182446d0 | 3816 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3817 | } |
3818 | ||
02491447 | 3819 | #ifdef CONFIG_MMU |
182446d0 | 3820 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3821 | struct cftype *cft, u64 val) |
3822 | { | |
182446d0 | 3823 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
3824 | |
3825 | if (val >= (1 << NR_MOVE_TYPE)) | |
3826 | return -EINVAL; | |
ee5e8472 | 3827 | |
7dc74be0 | 3828 | /* |
ee5e8472 GC |
3829 | * No kind of locking is needed in here, because ->can_attach() will |
3830 | * check this value once in the beginning of the process, and then carry | |
3831 | * on with stale data. This means that changes to this value will only | |
3832 | * affect task migrations starting after the change. | |
7dc74be0 | 3833 | */ |
c0ff4b85 | 3834 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3835 | return 0; |
3836 | } | |
02491447 | 3837 | #else |
182446d0 | 3838 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3839 | struct cftype *cft, u64 val) |
3840 | { | |
3841 | return -ENOSYS; | |
3842 | } | |
3843 | #endif | |
7dc74be0 | 3844 | |
406eb0c9 | 3845 | #ifdef CONFIG_NUMA |
2da8ca82 | 3846 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3847 | { |
25485de6 GT |
3848 | struct numa_stat { |
3849 | const char *name; | |
3850 | unsigned int lru_mask; | |
3851 | }; | |
3852 | ||
3853 | static const struct numa_stat stats[] = { | |
3854 | { "total", LRU_ALL }, | |
3855 | { "file", LRU_ALL_FILE }, | |
3856 | { "anon", LRU_ALL_ANON }, | |
3857 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3858 | }; | |
3859 | const struct numa_stat *stat; | |
406eb0c9 | 3860 | int nid; |
25485de6 | 3861 | unsigned long nr; |
2da8ca82 | 3862 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3863 | |
25485de6 GT |
3864 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3865 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3866 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3867 | for_each_node_state(nid, N_MEMORY) { | |
3868 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3869 | stat->lru_mask); | |
3870 | seq_printf(m, " N%d=%lu", nid, nr); | |
3871 | } | |
3872 | seq_putc(m, '\n'); | |
406eb0c9 | 3873 | } |
406eb0c9 | 3874 | |
071aee13 YH |
3875 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3876 | struct mem_cgroup *iter; | |
3877 | ||
3878 | nr = 0; | |
3879 | for_each_mem_cgroup_tree(iter, memcg) | |
3880 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3881 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3882 | for_each_node_state(nid, N_MEMORY) { | |
3883 | nr = 0; | |
3884 | for_each_mem_cgroup_tree(iter, memcg) | |
3885 | nr += mem_cgroup_node_nr_lru_pages( | |
3886 | iter, nid, stat->lru_mask); | |
3887 | seq_printf(m, " N%d=%lu", nid, nr); | |
3888 | } | |
3889 | seq_putc(m, '\n'); | |
406eb0c9 | 3890 | } |
406eb0c9 | 3891 | |
406eb0c9 YH |
3892 | return 0; |
3893 | } | |
3894 | #endif /* CONFIG_NUMA */ | |
3895 | ||
af7c4b0e JW |
3896 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
3897 | { | |
3898 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
3899 | } | |
3900 | ||
2da8ca82 | 3901 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3902 | { |
2da8ca82 | 3903 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3904 | unsigned long memory, memsw; |
af7c4b0e JW |
3905 | struct mem_cgroup *mi; |
3906 | unsigned int i; | |
406eb0c9 | 3907 | |
af7c4b0e | 3908 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3909 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3910 | continue; |
af7c4b0e JW |
3911 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
3912 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 3913 | } |
7b854121 | 3914 | |
af7c4b0e JW |
3915 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3916 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3917 | mem_cgroup_read_events(memcg, i)); | |
3918 | ||
3919 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3920 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3921 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3922 | ||
14067bb3 | 3923 | /* Hierarchical information */ |
3e32cb2e JW |
3924 | memory = memsw = PAGE_COUNTER_MAX; |
3925 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3926 | memory = min(memory, mi->memory.limit); | |
3927 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3928 | } |
3e32cb2e JW |
3929 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3930 | (u64)memory * PAGE_SIZE); | |
3931 | if (do_swap_account) | |
3932 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3933 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3934 | |
af7c4b0e JW |
3935 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
3936 | long long val = 0; | |
3937 | ||
bff6bb83 | 3938 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3939 | continue; |
af7c4b0e JW |
3940 | for_each_mem_cgroup_tree(mi, memcg) |
3941 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
3942 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
3943 | } | |
3944 | ||
3945 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3946 | unsigned long long val = 0; | |
3947 | ||
3948 | for_each_mem_cgroup_tree(mi, memcg) | |
3949 | val += mem_cgroup_read_events(mi, i); | |
3950 | seq_printf(m, "total_%s %llu\n", | |
3951 | mem_cgroup_events_names[i], val); | |
3952 | } | |
3953 | ||
3954 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3955 | unsigned long long val = 0; | |
3956 | ||
3957 | for_each_mem_cgroup_tree(mi, memcg) | |
3958 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3959 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3960 | } |
14067bb3 | 3961 | |
7f016ee8 | 3962 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3963 | { |
3964 | int nid, zid; | |
3965 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3966 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3967 | unsigned long recent_rotated[2] = {0, 0}; |
3968 | unsigned long recent_scanned[2] = {0, 0}; | |
3969 | ||
3970 | for_each_online_node(nid) | |
3971 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3972 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3973 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3974 | |
89abfab1 HD |
3975 | recent_rotated[0] += rstat->recent_rotated[0]; |
3976 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3977 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3978 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3979 | } |
78ccf5b5 JW |
3980 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3981 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3982 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3983 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3984 | } |
3985 | #endif | |
3986 | ||
d2ceb9b7 KH |
3987 | return 0; |
3988 | } | |
3989 | ||
182446d0 TH |
3990 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3991 | struct cftype *cft) | |
a7885eb8 | 3992 | { |
182446d0 | 3993 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3994 | |
1f4c025b | 3995 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3996 | } |
3997 | ||
182446d0 TH |
3998 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3999 | struct cftype *cft, u64 val) | |
a7885eb8 | 4000 | { |
182446d0 | 4001 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4002 | |
3dae7fec | 4003 | if (val > 100) |
a7885eb8 KM |
4004 | return -EINVAL; |
4005 | ||
14208b0e | 4006 | if (css->parent) |
3dae7fec JW |
4007 | memcg->swappiness = val; |
4008 | else | |
4009 | vm_swappiness = val; | |
068b38c1 | 4010 | |
a7885eb8 KM |
4011 | return 0; |
4012 | } | |
4013 | ||
2e72b634 KS |
4014 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4015 | { | |
4016 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4017 | unsigned long usage; |
2e72b634 KS |
4018 | int i; |
4019 | ||
4020 | rcu_read_lock(); | |
4021 | if (!swap) | |
2c488db2 | 4022 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4023 | else |
2c488db2 | 4024 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4025 | |
4026 | if (!t) | |
4027 | goto unlock; | |
4028 | ||
ce00a967 | 4029 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4030 | |
4031 | /* | |
748dad36 | 4032 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4033 | * If it's not true, a threshold was crossed after last |
4034 | * call of __mem_cgroup_threshold(). | |
4035 | */ | |
5407a562 | 4036 | i = t->current_threshold; |
2e72b634 KS |
4037 | |
4038 | /* | |
4039 | * Iterate backward over array of thresholds starting from | |
4040 | * current_threshold and check if a threshold is crossed. | |
4041 | * If none of thresholds below usage is crossed, we read | |
4042 | * only one element of the array here. | |
4043 | */ | |
4044 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4045 | eventfd_signal(t->entries[i].eventfd, 1); | |
4046 | ||
4047 | /* i = current_threshold + 1 */ | |
4048 | i++; | |
4049 | ||
4050 | /* | |
4051 | * Iterate forward over array of thresholds starting from | |
4052 | * current_threshold+1 and check if a threshold is crossed. | |
4053 | * If none of thresholds above usage is crossed, we read | |
4054 | * only one element of the array here. | |
4055 | */ | |
4056 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4057 | eventfd_signal(t->entries[i].eventfd, 1); | |
4058 | ||
4059 | /* Update current_threshold */ | |
5407a562 | 4060 | t->current_threshold = i - 1; |
2e72b634 KS |
4061 | unlock: |
4062 | rcu_read_unlock(); | |
4063 | } | |
4064 | ||
4065 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4066 | { | |
ad4ca5f4 KS |
4067 | while (memcg) { |
4068 | __mem_cgroup_threshold(memcg, false); | |
4069 | if (do_swap_account) | |
4070 | __mem_cgroup_threshold(memcg, true); | |
4071 | ||
4072 | memcg = parent_mem_cgroup(memcg); | |
4073 | } | |
2e72b634 KS |
4074 | } |
4075 | ||
4076 | static int compare_thresholds(const void *a, const void *b) | |
4077 | { | |
4078 | const struct mem_cgroup_threshold *_a = a; | |
4079 | const struct mem_cgroup_threshold *_b = b; | |
4080 | ||
2bff24a3 GT |
4081 | if (_a->threshold > _b->threshold) |
4082 | return 1; | |
4083 | ||
4084 | if (_a->threshold < _b->threshold) | |
4085 | return -1; | |
4086 | ||
4087 | return 0; | |
2e72b634 KS |
4088 | } |
4089 | ||
c0ff4b85 | 4090 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4091 | { |
4092 | struct mem_cgroup_eventfd_list *ev; | |
4093 | ||
2bcf2e92 MH |
4094 | spin_lock(&memcg_oom_lock); |
4095 | ||
c0ff4b85 | 4096 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4097 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4098 | |
4099 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4100 | return 0; |
4101 | } | |
4102 | ||
c0ff4b85 | 4103 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4104 | { |
7d74b06f KH |
4105 | struct mem_cgroup *iter; |
4106 | ||
c0ff4b85 | 4107 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4108 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4109 | } |
4110 | ||
59b6f873 | 4111 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4112 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4113 | { |
2c488db2 KS |
4114 | struct mem_cgroup_thresholds *thresholds; |
4115 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4116 | unsigned long threshold; |
4117 | unsigned long usage; | |
2c488db2 | 4118 | int i, size, ret; |
2e72b634 | 4119 | |
3e32cb2e | 4120 | ret = page_counter_memparse(args, &threshold); |
2e72b634 KS |
4121 | if (ret) |
4122 | return ret; | |
4123 | ||
4124 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4125 | |
05b84301 | 4126 | if (type == _MEM) { |
2c488db2 | 4127 | thresholds = &memcg->thresholds; |
ce00a967 | 4128 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4129 | } else if (type == _MEMSWAP) { |
2c488db2 | 4130 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4131 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4132 | } else |
2e72b634 KS |
4133 | BUG(); |
4134 | ||
2e72b634 | 4135 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4136 | if (thresholds->primary) |
2e72b634 KS |
4137 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4138 | ||
2c488db2 | 4139 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4140 | |
4141 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4142 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4143 | GFP_KERNEL); |
2c488db2 | 4144 | if (!new) { |
2e72b634 KS |
4145 | ret = -ENOMEM; |
4146 | goto unlock; | |
4147 | } | |
2c488db2 | 4148 | new->size = size; |
2e72b634 KS |
4149 | |
4150 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4151 | if (thresholds->primary) { |
4152 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4153 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4154 | } |
4155 | ||
2e72b634 | 4156 | /* Add new threshold */ |
2c488db2 KS |
4157 | new->entries[size - 1].eventfd = eventfd; |
4158 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4159 | |
4160 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4161 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4162 | compare_thresholds, NULL); |
4163 | ||
4164 | /* Find current threshold */ | |
2c488db2 | 4165 | new->current_threshold = -1; |
2e72b634 | 4166 | for (i = 0; i < size; i++) { |
748dad36 | 4167 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4168 | /* |
2c488db2 KS |
4169 | * new->current_threshold will not be used until |
4170 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4171 | * it here. |
4172 | */ | |
2c488db2 | 4173 | ++new->current_threshold; |
748dad36 SZ |
4174 | } else |
4175 | break; | |
2e72b634 KS |
4176 | } |
4177 | ||
2c488db2 KS |
4178 | /* Free old spare buffer and save old primary buffer as spare */ |
4179 | kfree(thresholds->spare); | |
4180 | thresholds->spare = thresholds->primary; | |
4181 | ||
4182 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4183 | |
907860ed | 4184 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4185 | synchronize_rcu(); |
4186 | ||
2e72b634 KS |
4187 | unlock: |
4188 | mutex_unlock(&memcg->thresholds_lock); | |
4189 | ||
4190 | return ret; | |
4191 | } | |
4192 | ||
59b6f873 | 4193 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4194 | struct eventfd_ctx *eventfd, const char *args) |
4195 | { | |
59b6f873 | 4196 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4197 | } |
4198 | ||
59b6f873 | 4199 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4200 | struct eventfd_ctx *eventfd, const char *args) |
4201 | { | |
59b6f873 | 4202 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4203 | } |
4204 | ||
59b6f873 | 4205 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4206 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4207 | { |
2c488db2 KS |
4208 | struct mem_cgroup_thresholds *thresholds; |
4209 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4210 | unsigned long usage; |
2c488db2 | 4211 | int i, j, size; |
2e72b634 KS |
4212 | |
4213 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4214 | |
4215 | if (type == _MEM) { | |
2c488db2 | 4216 | thresholds = &memcg->thresholds; |
ce00a967 | 4217 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4218 | } else if (type == _MEMSWAP) { |
2c488db2 | 4219 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4220 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4221 | } else |
2e72b634 KS |
4222 | BUG(); |
4223 | ||
371528ca AV |
4224 | if (!thresholds->primary) |
4225 | goto unlock; | |
4226 | ||
2e72b634 KS |
4227 | /* Check if a threshold crossed before removing */ |
4228 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4229 | ||
4230 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4231 | size = 0; |
4232 | for (i = 0; i < thresholds->primary->size; i++) { | |
4233 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4234 | size++; |
4235 | } | |
4236 | ||
2c488db2 | 4237 | new = thresholds->spare; |
907860ed | 4238 | |
2e72b634 KS |
4239 | /* Set thresholds array to NULL if we don't have thresholds */ |
4240 | if (!size) { | |
2c488db2 KS |
4241 | kfree(new); |
4242 | new = NULL; | |
907860ed | 4243 | goto swap_buffers; |
2e72b634 KS |
4244 | } |
4245 | ||
2c488db2 | 4246 | new->size = size; |
2e72b634 KS |
4247 | |
4248 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4249 | new->current_threshold = -1; |
4250 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4251 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4252 | continue; |
4253 | ||
2c488db2 | 4254 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4255 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4256 | /* |
2c488db2 | 4257 | * new->current_threshold will not be used |
2e72b634 KS |
4258 | * until rcu_assign_pointer(), so it's safe to increment |
4259 | * it here. | |
4260 | */ | |
2c488db2 | 4261 | ++new->current_threshold; |
2e72b634 KS |
4262 | } |
4263 | j++; | |
4264 | } | |
4265 | ||
907860ed | 4266 | swap_buffers: |
2c488db2 KS |
4267 | /* Swap primary and spare array */ |
4268 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4269 | /* If all events are unregistered, free the spare array */ |
4270 | if (!new) { | |
4271 | kfree(thresholds->spare); | |
4272 | thresholds->spare = NULL; | |
4273 | } | |
4274 | ||
2c488db2 | 4275 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4276 | |
907860ed | 4277 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4278 | synchronize_rcu(); |
371528ca | 4279 | unlock: |
2e72b634 | 4280 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4281 | } |
c1e862c1 | 4282 | |
59b6f873 | 4283 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4284 | struct eventfd_ctx *eventfd) |
4285 | { | |
59b6f873 | 4286 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4287 | } |
4288 | ||
59b6f873 | 4289 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4290 | struct eventfd_ctx *eventfd) |
4291 | { | |
59b6f873 | 4292 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4293 | } |
4294 | ||
59b6f873 | 4295 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4296 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4297 | { |
9490ff27 | 4298 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4299 | |
9490ff27 KH |
4300 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4301 | if (!event) | |
4302 | return -ENOMEM; | |
4303 | ||
1af8efe9 | 4304 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4305 | |
4306 | event->eventfd = eventfd; | |
4307 | list_add(&event->list, &memcg->oom_notify); | |
4308 | ||
4309 | /* already in OOM ? */ | |
79dfdacc | 4310 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4311 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4312 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4313 | |
4314 | return 0; | |
4315 | } | |
4316 | ||
59b6f873 | 4317 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4318 | struct eventfd_ctx *eventfd) |
9490ff27 | 4319 | { |
9490ff27 | 4320 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4321 | |
1af8efe9 | 4322 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4323 | |
c0ff4b85 | 4324 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4325 | if (ev->eventfd == eventfd) { |
4326 | list_del(&ev->list); | |
4327 | kfree(ev); | |
4328 | } | |
4329 | } | |
4330 | ||
1af8efe9 | 4331 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4332 | } |
4333 | ||
2da8ca82 | 4334 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4335 | { |
2da8ca82 | 4336 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4337 | |
791badbd TH |
4338 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4339 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4340 | return 0; |
4341 | } | |
4342 | ||
182446d0 | 4343 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4344 | struct cftype *cft, u64 val) |
4345 | { | |
182446d0 | 4346 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4347 | |
4348 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4349 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4350 | return -EINVAL; |
4351 | ||
c0ff4b85 | 4352 | memcg->oom_kill_disable = val; |
4d845ebf | 4353 | if (!val) |
c0ff4b85 | 4354 | memcg_oom_recover(memcg); |
3dae7fec | 4355 | |
3c11ecf4 KH |
4356 | return 0; |
4357 | } | |
4358 | ||
c255a458 | 4359 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4360 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4361 | { |
55007d84 GC |
4362 | int ret; |
4363 | ||
2633d7a0 | 4364 | memcg->kmemcg_id = -1; |
55007d84 GC |
4365 | ret = memcg_propagate_kmem(memcg); |
4366 | if (ret) | |
4367 | return ret; | |
2633d7a0 | 4368 | |
1d62e436 | 4369 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4370 | } |
e5671dfa | 4371 | |
10d5ebf4 | 4372 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4373 | { |
1d62e436 | 4374 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 4375 | } |
e5671dfa | 4376 | #else |
cbe128e3 | 4377 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4378 | { |
4379 | return 0; | |
4380 | } | |
d1a4c0b3 | 4381 | |
10d5ebf4 LZ |
4382 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4383 | { | |
4384 | } | |
e5671dfa GC |
4385 | #endif |
4386 | ||
3bc942f3 TH |
4387 | /* |
4388 | * DO NOT USE IN NEW FILES. | |
4389 | * | |
4390 | * "cgroup.event_control" implementation. | |
4391 | * | |
4392 | * This is way over-engineered. It tries to support fully configurable | |
4393 | * events for each user. Such level of flexibility is completely | |
4394 | * unnecessary especially in the light of the planned unified hierarchy. | |
4395 | * | |
4396 | * Please deprecate this and replace with something simpler if at all | |
4397 | * possible. | |
4398 | */ | |
4399 | ||
79bd9814 TH |
4400 | /* |
4401 | * Unregister event and free resources. | |
4402 | * | |
4403 | * Gets called from workqueue. | |
4404 | */ | |
3bc942f3 | 4405 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4406 | { |
3bc942f3 TH |
4407 | struct mem_cgroup_event *event = |
4408 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4409 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4410 | |
4411 | remove_wait_queue(event->wqh, &event->wait); | |
4412 | ||
59b6f873 | 4413 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4414 | |
4415 | /* Notify userspace the event is going away. */ | |
4416 | eventfd_signal(event->eventfd, 1); | |
4417 | ||
4418 | eventfd_ctx_put(event->eventfd); | |
4419 | kfree(event); | |
59b6f873 | 4420 | css_put(&memcg->css); |
79bd9814 TH |
4421 | } |
4422 | ||
4423 | /* | |
4424 | * Gets called on POLLHUP on eventfd when user closes it. | |
4425 | * | |
4426 | * Called with wqh->lock held and interrupts disabled. | |
4427 | */ | |
3bc942f3 TH |
4428 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4429 | int sync, void *key) | |
79bd9814 | 4430 | { |
3bc942f3 TH |
4431 | struct mem_cgroup_event *event = |
4432 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4433 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4434 | unsigned long flags = (unsigned long)key; |
4435 | ||
4436 | if (flags & POLLHUP) { | |
4437 | /* | |
4438 | * If the event has been detached at cgroup removal, we | |
4439 | * can simply return knowing the other side will cleanup | |
4440 | * for us. | |
4441 | * | |
4442 | * We can't race against event freeing since the other | |
4443 | * side will require wqh->lock via remove_wait_queue(), | |
4444 | * which we hold. | |
4445 | */ | |
fba94807 | 4446 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4447 | if (!list_empty(&event->list)) { |
4448 | list_del_init(&event->list); | |
4449 | /* | |
4450 | * We are in atomic context, but cgroup_event_remove() | |
4451 | * may sleep, so we have to call it in workqueue. | |
4452 | */ | |
4453 | schedule_work(&event->remove); | |
4454 | } | |
fba94807 | 4455 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4456 | } |
4457 | ||
4458 | return 0; | |
4459 | } | |
4460 | ||
3bc942f3 | 4461 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4462 | wait_queue_head_t *wqh, poll_table *pt) |
4463 | { | |
3bc942f3 TH |
4464 | struct mem_cgroup_event *event = |
4465 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4466 | |
4467 | event->wqh = wqh; | |
4468 | add_wait_queue(wqh, &event->wait); | |
4469 | } | |
4470 | ||
4471 | /* | |
3bc942f3 TH |
4472 | * DO NOT USE IN NEW FILES. |
4473 | * | |
79bd9814 TH |
4474 | * Parse input and register new cgroup event handler. |
4475 | * | |
4476 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4477 | * Interpretation of args is defined by control file implementation. | |
4478 | */ | |
451af504 TH |
4479 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4480 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4481 | { |
451af504 | 4482 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4483 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4484 | struct mem_cgroup_event *event; |
79bd9814 TH |
4485 | struct cgroup_subsys_state *cfile_css; |
4486 | unsigned int efd, cfd; | |
4487 | struct fd efile; | |
4488 | struct fd cfile; | |
fba94807 | 4489 | const char *name; |
79bd9814 TH |
4490 | char *endp; |
4491 | int ret; | |
4492 | ||
451af504 TH |
4493 | buf = strstrip(buf); |
4494 | ||
4495 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4496 | if (*endp != ' ') |
4497 | return -EINVAL; | |
451af504 | 4498 | buf = endp + 1; |
79bd9814 | 4499 | |
451af504 | 4500 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4501 | if ((*endp != ' ') && (*endp != '\0')) |
4502 | return -EINVAL; | |
451af504 | 4503 | buf = endp + 1; |
79bd9814 TH |
4504 | |
4505 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4506 | if (!event) | |
4507 | return -ENOMEM; | |
4508 | ||
59b6f873 | 4509 | event->memcg = memcg; |
79bd9814 | 4510 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4511 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4512 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4513 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4514 | |
4515 | efile = fdget(efd); | |
4516 | if (!efile.file) { | |
4517 | ret = -EBADF; | |
4518 | goto out_kfree; | |
4519 | } | |
4520 | ||
4521 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4522 | if (IS_ERR(event->eventfd)) { | |
4523 | ret = PTR_ERR(event->eventfd); | |
4524 | goto out_put_efile; | |
4525 | } | |
4526 | ||
4527 | cfile = fdget(cfd); | |
4528 | if (!cfile.file) { | |
4529 | ret = -EBADF; | |
4530 | goto out_put_eventfd; | |
4531 | } | |
4532 | ||
4533 | /* the process need read permission on control file */ | |
4534 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4535 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4536 | if (ret < 0) | |
4537 | goto out_put_cfile; | |
4538 | ||
fba94807 TH |
4539 | /* |
4540 | * Determine the event callbacks and set them in @event. This used | |
4541 | * to be done via struct cftype but cgroup core no longer knows | |
4542 | * about these events. The following is crude but the whole thing | |
4543 | * is for compatibility anyway. | |
3bc942f3 TH |
4544 | * |
4545 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
4546 | */ |
4547 | name = cfile.file->f_dentry->d_name.name; | |
4548 | ||
4549 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4550 | event->register_event = mem_cgroup_usage_register_event; | |
4551 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4552 | } else if (!strcmp(name, "memory.oom_control")) { | |
4553 | event->register_event = mem_cgroup_oom_register_event; | |
4554 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4555 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4556 | event->register_event = vmpressure_register_event; | |
4557 | event->unregister_event = vmpressure_unregister_event; | |
4558 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4559 | event->register_event = memsw_cgroup_usage_register_event; |
4560 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4561 | } else { |
4562 | ret = -EINVAL; | |
4563 | goto out_put_cfile; | |
4564 | } | |
4565 | ||
79bd9814 | 4566 | /* |
b5557c4c TH |
4567 | * Verify @cfile should belong to @css. Also, remaining events are |
4568 | * automatically removed on cgroup destruction but the removal is | |
4569 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4570 | */ |
ec903c0c TH |
4571 | cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent, |
4572 | &memory_cgrp_subsys); | |
79bd9814 | 4573 | ret = -EINVAL; |
5a17f543 | 4574 | if (IS_ERR(cfile_css)) |
79bd9814 | 4575 | goto out_put_cfile; |
5a17f543 TH |
4576 | if (cfile_css != css) { |
4577 | css_put(cfile_css); | |
79bd9814 | 4578 | goto out_put_cfile; |
5a17f543 | 4579 | } |
79bd9814 | 4580 | |
451af504 | 4581 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4582 | if (ret) |
4583 | goto out_put_css; | |
4584 | ||
4585 | efile.file->f_op->poll(efile.file, &event->pt); | |
4586 | ||
fba94807 TH |
4587 | spin_lock(&memcg->event_list_lock); |
4588 | list_add(&event->list, &memcg->event_list); | |
4589 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4590 | |
4591 | fdput(cfile); | |
4592 | fdput(efile); | |
4593 | ||
451af504 | 4594 | return nbytes; |
79bd9814 TH |
4595 | |
4596 | out_put_css: | |
b5557c4c | 4597 | css_put(css); |
79bd9814 TH |
4598 | out_put_cfile: |
4599 | fdput(cfile); | |
4600 | out_put_eventfd: | |
4601 | eventfd_ctx_put(event->eventfd); | |
4602 | out_put_efile: | |
4603 | fdput(efile); | |
4604 | out_kfree: | |
4605 | kfree(event); | |
4606 | ||
4607 | return ret; | |
4608 | } | |
4609 | ||
8cdea7c0 BS |
4610 | static struct cftype mem_cgroup_files[] = { |
4611 | { | |
0eea1030 | 4612 | .name = "usage_in_bytes", |
8c7c6e34 | 4613 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4614 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4615 | }, |
c84872e1 PE |
4616 | { |
4617 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4618 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4619 | .write = mem_cgroup_reset, |
791badbd | 4620 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4621 | }, |
8cdea7c0 | 4622 | { |
0eea1030 | 4623 | .name = "limit_in_bytes", |
8c7c6e34 | 4624 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4625 | .write = mem_cgroup_write, |
791badbd | 4626 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4627 | }, |
296c81d8 BS |
4628 | { |
4629 | .name = "soft_limit_in_bytes", | |
4630 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4631 | .write = mem_cgroup_write, |
791badbd | 4632 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4633 | }, |
8cdea7c0 BS |
4634 | { |
4635 | .name = "failcnt", | |
8c7c6e34 | 4636 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4637 | .write = mem_cgroup_reset, |
791badbd | 4638 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4639 | }, |
d2ceb9b7 KH |
4640 | { |
4641 | .name = "stat", | |
2da8ca82 | 4642 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4643 | }, |
c1e862c1 KH |
4644 | { |
4645 | .name = "force_empty", | |
6770c64e | 4646 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4647 | }, |
18f59ea7 BS |
4648 | { |
4649 | .name = "use_hierarchy", | |
4650 | .write_u64 = mem_cgroup_hierarchy_write, | |
4651 | .read_u64 = mem_cgroup_hierarchy_read, | |
4652 | }, | |
79bd9814 | 4653 | { |
3bc942f3 | 4654 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4655 | .write = memcg_write_event_control, |
79bd9814 TH |
4656 | .flags = CFTYPE_NO_PREFIX, |
4657 | .mode = S_IWUGO, | |
4658 | }, | |
a7885eb8 KM |
4659 | { |
4660 | .name = "swappiness", | |
4661 | .read_u64 = mem_cgroup_swappiness_read, | |
4662 | .write_u64 = mem_cgroup_swappiness_write, | |
4663 | }, | |
7dc74be0 DN |
4664 | { |
4665 | .name = "move_charge_at_immigrate", | |
4666 | .read_u64 = mem_cgroup_move_charge_read, | |
4667 | .write_u64 = mem_cgroup_move_charge_write, | |
4668 | }, | |
9490ff27 KH |
4669 | { |
4670 | .name = "oom_control", | |
2da8ca82 | 4671 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4672 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4673 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4674 | }, | |
70ddf637 AV |
4675 | { |
4676 | .name = "pressure_level", | |
70ddf637 | 4677 | }, |
406eb0c9 YH |
4678 | #ifdef CONFIG_NUMA |
4679 | { | |
4680 | .name = "numa_stat", | |
2da8ca82 | 4681 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4682 | }, |
4683 | #endif | |
510fc4e1 GC |
4684 | #ifdef CONFIG_MEMCG_KMEM |
4685 | { | |
4686 | .name = "kmem.limit_in_bytes", | |
4687 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4688 | .write = mem_cgroup_write, |
791badbd | 4689 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4690 | }, |
4691 | { | |
4692 | .name = "kmem.usage_in_bytes", | |
4693 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4694 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4695 | }, |
4696 | { | |
4697 | .name = "kmem.failcnt", | |
4698 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4699 | .write = mem_cgroup_reset, |
791badbd | 4700 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4701 | }, |
4702 | { | |
4703 | .name = "kmem.max_usage_in_bytes", | |
4704 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4705 | .write = mem_cgroup_reset, |
791badbd | 4706 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4707 | }, |
749c5415 GC |
4708 | #ifdef CONFIG_SLABINFO |
4709 | { | |
4710 | .name = "kmem.slabinfo", | |
2da8ca82 | 4711 | .seq_show = mem_cgroup_slabinfo_read, |
749c5415 GC |
4712 | }, |
4713 | #endif | |
8c7c6e34 | 4714 | #endif |
6bc10349 | 4715 | { }, /* terminate */ |
af36f906 | 4716 | }; |
8c7c6e34 | 4717 | |
2d11085e MH |
4718 | #ifdef CONFIG_MEMCG_SWAP |
4719 | static struct cftype memsw_cgroup_files[] = { | |
4720 | { | |
4721 | .name = "memsw.usage_in_bytes", | |
4722 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 4723 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4724 | }, |
4725 | { | |
4726 | .name = "memsw.max_usage_in_bytes", | |
4727 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 4728 | .write = mem_cgroup_reset, |
791badbd | 4729 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4730 | }, |
4731 | { | |
4732 | .name = "memsw.limit_in_bytes", | |
4733 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 4734 | .write = mem_cgroup_write, |
791badbd | 4735 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4736 | }, |
4737 | { | |
4738 | .name = "memsw.failcnt", | |
4739 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 4740 | .write = mem_cgroup_reset, |
791badbd | 4741 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4742 | }, |
4743 | { }, /* terminate */ | |
4744 | }; | |
4745 | #endif | |
c0ff4b85 | 4746 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4747 | { |
4748 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4749 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4750 | int zone, tmp = node; |
1ecaab2b KH |
4751 | /* |
4752 | * This routine is called against possible nodes. | |
4753 | * But it's BUG to call kmalloc() against offline node. | |
4754 | * | |
4755 | * TODO: this routine can waste much memory for nodes which will | |
4756 | * never be onlined. It's better to use memory hotplug callback | |
4757 | * function. | |
4758 | */ | |
41e3355d KH |
4759 | if (!node_state(node, N_NORMAL_MEMORY)) |
4760 | tmp = -1; | |
17295c88 | 4761 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4762 | if (!pn) |
4763 | return 1; | |
1ecaab2b | 4764 | |
1ecaab2b KH |
4765 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4766 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4767 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4768 | mz->usage_in_excess = 0; |
4769 | mz->on_tree = false; | |
d79154bb | 4770 | mz->memcg = memcg; |
1ecaab2b | 4771 | } |
54f72fe0 | 4772 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4773 | return 0; |
4774 | } | |
4775 | ||
c0ff4b85 | 4776 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4777 | { |
54f72fe0 | 4778 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4779 | } |
4780 | ||
33327948 KH |
4781 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4782 | { | |
d79154bb | 4783 | struct mem_cgroup *memcg; |
8ff69e2c | 4784 | size_t size; |
33327948 | 4785 | |
8ff69e2c VD |
4786 | size = sizeof(struct mem_cgroup); |
4787 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4788 | |
8ff69e2c | 4789 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4790 | if (!memcg) |
e7bbcdf3 DC |
4791 | return NULL; |
4792 | ||
d79154bb HD |
4793 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4794 | if (!memcg->stat) | |
d2e61b8d | 4795 | goto out_free; |
d79154bb HD |
4796 | spin_lock_init(&memcg->pcp_counter_lock); |
4797 | return memcg; | |
d2e61b8d DC |
4798 | |
4799 | out_free: | |
8ff69e2c | 4800 | kfree(memcg); |
d2e61b8d | 4801 | return NULL; |
33327948 KH |
4802 | } |
4803 | ||
59927fb9 | 4804 | /* |
c8b2a36f GC |
4805 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4806 | * (scanning all at force_empty is too costly...) | |
4807 | * | |
4808 | * Instead of clearing all references at force_empty, we remember | |
4809 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4810 | * it goes down to 0. | |
4811 | * | |
4812 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4813 | */ |
c8b2a36f GC |
4814 | |
4815 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4816 | { |
c8b2a36f | 4817 | int node; |
59927fb9 | 4818 | |
bb4cc1a8 | 4819 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4820 | |
4821 | for_each_node(node) | |
4822 | free_mem_cgroup_per_zone_info(memcg, node); | |
4823 | ||
4824 | free_percpu(memcg->stat); | |
4825 | ||
3f134619 GC |
4826 | /* |
4827 | * We need to make sure that (at least for now), the jump label | |
4828 | * destruction code runs outside of the cgroup lock. This is because | |
4829 | * get_online_cpus(), which is called from the static_branch update, | |
4830 | * can't be called inside the cgroup_lock. cpusets are the ones | |
4831 | * enforcing this dependency, so if they ever change, we might as well. | |
4832 | * | |
4833 | * schedule_work() will guarantee this happens. Be careful if you need | |
4834 | * to move this code around, and make sure it is outside | |
4835 | * the cgroup_lock. | |
4836 | */ | |
a8964b9b | 4837 | disarm_static_keys(memcg); |
8ff69e2c | 4838 | kfree(memcg); |
59927fb9 | 4839 | } |
3afe36b1 | 4840 | |
7bcc1bb1 DN |
4841 | /* |
4842 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4843 | */ | |
e1aab161 | 4844 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4845 | { |
3e32cb2e | 4846 | if (!memcg->memory.parent) |
7bcc1bb1 | 4847 | return NULL; |
3e32cb2e | 4848 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4849 | } |
e1aab161 | 4850 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4851 | |
bb4cc1a8 AM |
4852 | static void __init mem_cgroup_soft_limit_tree_init(void) |
4853 | { | |
4854 | struct mem_cgroup_tree_per_node *rtpn; | |
4855 | struct mem_cgroup_tree_per_zone *rtpz; | |
4856 | int tmp, node, zone; | |
4857 | ||
4858 | for_each_node(node) { | |
4859 | tmp = node; | |
4860 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4861 | tmp = -1; | |
4862 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4863 | BUG_ON(!rtpn); | |
4864 | ||
4865 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4866 | ||
4867 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4868 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4869 | rtpz->rb_root = RB_ROOT; | |
4870 | spin_lock_init(&rtpz->lock); | |
4871 | } | |
4872 | } | |
4873 | } | |
4874 | ||
0eb253e2 | 4875 | static struct cgroup_subsys_state * __ref |
eb95419b | 4876 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4877 | { |
d142e3e6 | 4878 | struct mem_cgroup *memcg; |
04046e1a | 4879 | long error = -ENOMEM; |
6d12e2d8 | 4880 | int node; |
8cdea7c0 | 4881 | |
c0ff4b85 R |
4882 | memcg = mem_cgroup_alloc(); |
4883 | if (!memcg) | |
04046e1a | 4884 | return ERR_PTR(error); |
78fb7466 | 4885 | |
3ed28fa1 | 4886 | for_each_node(node) |
c0ff4b85 | 4887 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4888 | goto free_out; |
f64c3f54 | 4889 | |
c077719b | 4890 | /* root ? */ |
eb95419b | 4891 | if (parent_css == NULL) { |
a41c58a6 | 4892 | root_mem_cgroup = memcg; |
3e32cb2e JW |
4893 | page_counter_init(&memcg->memory, NULL); |
4894 | page_counter_init(&memcg->memsw, NULL); | |
4895 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4896 | } |
28dbc4b6 | 4897 | |
d142e3e6 GC |
4898 | memcg->last_scanned_node = MAX_NUMNODES; |
4899 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4900 | memcg->move_charge_at_immigrate = 0; |
4901 | mutex_init(&memcg->thresholds_lock); | |
4902 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4903 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4904 | INIT_LIST_HEAD(&memcg->event_list); |
4905 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
4906 | |
4907 | return &memcg->css; | |
4908 | ||
4909 | free_out: | |
4910 | __mem_cgroup_free(memcg); | |
4911 | return ERR_PTR(error); | |
4912 | } | |
4913 | ||
4914 | static int | |
eb95419b | 4915 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4916 | { |
eb95419b | 4917 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4918 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4919 | int ret; |
d142e3e6 | 4920 | |
15a4c835 | 4921 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4922 | return -ENOSPC; |
4923 | ||
63876986 | 4924 | if (!parent) |
d142e3e6 GC |
4925 | return 0; |
4926 | ||
0999821b | 4927 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4928 | |
4929 | memcg->use_hierarchy = parent->use_hierarchy; | |
4930 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4931 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4932 | ||
4933 | if (parent->use_hierarchy) { | |
3e32cb2e JW |
4934 | page_counter_init(&memcg->memory, &parent->memory); |
4935 | page_counter_init(&memcg->memsw, &parent->memsw); | |
4936 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4937 | |
7bcc1bb1 | 4938 | /* |
8d76a979 LZ |
4939 | * No need to take a reference to the parent because cgroup |
4940 | * core guarantees its existence. | |
7bcc1bb1 | 4941 | */ |
18f59ea7 | 4942 | } else { |
3e32cb2e JW |
4943 | page_counter_init(&memcg->memory, NULL); |
4944 | page_counter_init(&memcg->memsw, NULL); | |
4945 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4946 | /* |
4947 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4948 | * much sense so let cgroup subsystem know about this | |
4949 | * unfortunate state in our controller. | |
4950 | */ | |
d142e3e6 | 4951 | if (parent != root_mem_cgroup) |
073219e9 | 4952 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4953 | } |
0999821b | 4954 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4955 | |
2f7dd7a4 JW |
4956 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4957 | if (ret) | |
4958 | return ret; | |
4959 | ||
4960 | /* | |
4961 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4962 | * orders reading memcg->initialized against its callers | |
4963 | * reading the memcg members. | |
4964 | */ | |
4965 | smp_store_release(&memcg->initialized, 1); | |
4966 | ||
4967 | return 0; | |
8cdea7c0 BS |
4968 | } |
4969 | ||
eb95419b | 4970 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4971 | { |
eb95419b | 4972 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4973 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4974 | |
4975 | /* | |
4976 | * Unregister events and notify userspace. | |
4977 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4978 | * directory to avoid race between userspace and kernelspace. | |
4979 | */ | |
fba94807 TH |
4980 | spin_lock(&memcg->event_list_lock); |
4981 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4982 | list_del_init(&event->list); |
4983 | schedule_work(&event->remove); | |
4984 | } | |
fba94807 | 4985 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4986 | |
776ed0f0 | 4987 | memcg_unregister_all_caches(memcg); |
33cb876e | 4988 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
4989 | } |
4990 | ||
eb95419b | 4991 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4992 | { |
eb95419b | 4993 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4994 | |
10d5ebf4 | 4995 | memcg_destroy_kmem(memcg); |
465939a1 | 4996 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4997 | } |
4998 | ||
1ced953b TH |
4999 | /** |
5000 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5001 | * @css: the target css | |
5002 | * | |
5003 | * Reset the states of the mem_cgroup associated with @css. This is | |
5004 | * invoked when the userland requests disabling on the default hierarchy | |
5005 | * but the memcg is pinned through dependency. The memcg should stop | |
5006 | * applying policies and should revert to the vanilla state as it may be | |
5007 | * made visible again. | |
5008 | * | |
5009 | * The current implementation only resets the essential configurations. | |
5010 | * This needs to be expanded to cover all the visible parts. | |
5011 | */ | |
5012 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5013 | { | |
5014 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5015 | ||
3e32cb2e JW |
5016 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
5017 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
5018 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
5019 | memcg->soft_limit = 0; | |
1ced953b TH |
5020 | } |
5021 | ||
02491447 | 5022 | #ifdef CONFIG_MMU |
7dc74be0 | 5023 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5024 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5025 | { |
05b84301 | 5026 | int ret; |
9476db97 JW |
5027 | |
5028 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 5029 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 5030 | if (!ret) { |
854ffa8d | 5031 | mc.precharge += count; |
854ffa8d DN |
5032 | return ret; |
5033 | } | |
692e7c45 | 5034 | if (ret == -EINTR) { |
00501b53 | 5035 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
5036 | return ret; |
5037 | } | |
9476db97 JW |
5038 | |
5039 | /* Try charges one by one with reclaim */ | |
854ffa8d | 5040 | while (count--) { |
00501b53 | 5041 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
5042 | /* |
5043 | * In case of failure, any residual charges against | |
5044 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
5045 | * later on. However, cancel any charges that are |
5046 | * bypassed to root right away or they'll be lost. | |
9476db97 | 5047 | */ |
692e7c45 | 5048 | if (ret == -EINTR) |
00501b53 | 5049 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 5050 | if (ret) |
38c5d72f | 5051 | return ret; |
854ffa8d | 5052 | mc.precharge++; |
9476db97 | 5053 | cond_resched(); |
854ffa8d | 5054 | } |
9476db97 | 5055 | return 0; |
4ffef5fe DN |
5056 | } |
5057 | ||
5058 | /** | |
8d32ff84 | 5059 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5060 | * @vma: the vma the pte to be checked belongs |
5061 | * @addr: the address corresponding to the pte to be checked | |
5062 | * @ptent: the pte to be checked | |
02491447 | 5063 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5064 | * |
5065 | * Returns | |
5066 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5067 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5068 | * move charge. if @target is not NULL, the page is stored in target->page | |
5069 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5070 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5071 | * target for charge migration. if @target is not NULL, the entry is stored | |
5072 | * in target->ent. | |
4ffef5fe DN |
5073 | * |
5074 | * Called with pte lock held. | |
5075 | */ | |
4ffef5fe DN |
5076 | union mc_target { |
5077 | struct page *page; | |
02491447 | 5078 | swp_entry_t ent; |
4ffef5fe DN |
5079 | }; |
5080 | ||
4ffef5fe | 5081 | enum mc_target_type { |
8d32ff84 | 5082 | MC_TARGET_NONE = 0, |
4ffef5fe | 5083 | MC_TARGET_PAGE, |
02491447 | 5084 | MC_TARGET_SWAP, |
4ffef5fe DN |
5085 | }; |
5086 | ||
90254a65 DN |
5087 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5088 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5089 | { |
90254a65 | 5090 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5091 | |
90254a65 DN |
5092 | if (!page || !page_mapped(page)) |
5093 | return NULL; | |
5094 | if (PageAnon(page)) { | |
5095 | /* we don't move shared anon */ | |
4b91355e | 5096 | if (!move_anon()) |
90254a65 | 5097 | return NULL; |
87946a72 DN |
5098 | } else if (!move_file()) |
5099 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5100 | return NULL; |
5101 | if (!get_page_unless_zero(page)) | |
5102 | return NULL; | |
5103 | ||
5104 | return page; | |
5105 | } | |
5106 | ||
4b91355e | 5107 | #ifdef CONFIG_SWAP |
90254a65 DN |
5108 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5109 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5110 | { | |
90254a65 DN |
5111 | struct page *page = NULL; |
5112 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5113 | ||
5114 | if (!move_anon() || non_swap_entry(ent)) | |
5115 | return NULL; | |
4b91355e KH |
5116 | /* |
5117 | * Because lookup_swap_cache() updates some statistics counter, | |
5118 | * we call find_get_page() with swapper_space directly. | |
5119 | */ | |
33806f06 | 5120 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
5121 | if (do_swap_account) |
5122 | entry->val = ent.val; | |
5123 | ||
5124 | return page; | |
5125 | } | |
4b91355e KH |
5126 | #else |
5127 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5128 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5129 | { | |
5130 | return NULL; | |
5131 | } | |
5132 | #endif | |
90254a65 | 5133 | |
87946a72 DN |
5134 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5135 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5136 | { | |
5137 | struct page *page = NULL; | |
87946a72 DN |
5138 | struct address_space *mapping; |
5139 | pgoff_t pgoff; | |
5140 | ||
5141 | if (!vma->vm_file) /* anonymous vma */ | |
5142 | return NULL; | |
5143 | if (!move_file()) | |
5144 | return NULL; | |
5145 | ||
87946a72 DN |
5146 | mapping = vma->vm_file->f_mapping; |
5147 | if (pte_none(ptent)) | |
5148 | pgoff = linear_page_index(vma, addr); | |
5149 | else /* pte_file(ptent) is true */ | |
5150 | pgoff = pte_to_pgoff(ptent); | |
5151 | ||
5152 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5153 | #ifdef CONFIG_SWAP |
5154 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5155 | if (shmem_mapping(mapping)) { |
5156 | page = find_get_entry(mapping, pgoff); | |
5157 | if (radix_tree_exceptional_entry(page)) { | |
5158 | swp_entry_t swp = radix_to_swp_entry(page); | |
5159 | if (do_swap_account) | |
5160 | *entry = swp; | |
5161 | page = find_get_page(swap_address_space(swp), swp.val); | |
5162 | } | |
5163 | } else | |
5164 | page = find_get_page(mapping, pgoff); | |
5165 | #else | |
5166 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5167 | #endif |
87946a72 DN |
5168 | return page; |
5169 | } | |
5170 | ||
8d32ff84 | 5171 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5172 | unsigned long addr, pte_t ptent, union mc_target *target) |
5173 | { | |
5174 | struct page *page = NULL; | |
5175 | struct page_cgroup *pc; | |
8d32ff84 | 5176 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5177 | swp_entry_t ent = { .val = 0 }; |
5178 | ||
5179 | if (pte_present(ptent)) | |
5180 | page = mc_handle_present_pte(vma, addr, ptent); | |
5181 | else if (is_swap_pte(ptent)) | |
5182 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5183 | else if (pte_none(ptent) || pte_file(ptent)) |
5184 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5185 | |
5186 | if (!page && !ent.val) | |
8d32ff84 | 5187 | return ret; |
02491447 DN |
5188 | if (page) { |
5189 | pc = lookup_page_cgroup(page); | |
5190 | /* | |
0a31bc97 JW |
5191 | * Do only loose check w/o serialization. |
5192 | * mem_cgroup_move_account() checks the pc is valid or | |
5193 | * not under LRU exclusion. | |
02491447 | 5194 | */ |
29833315 | 5195 | if (pc->mem_cgroup == mc.from) { |
02491447 DN |
5196 | ret = MC_TARGET_PAGE; |
5197 | if (target) | |
5198 | target->page = page; | |
5199 | } | |
5200 | if (!ret || !target) | |
5201 | put_page(page); | |
5202 | } | |
90254a65 DN |
5203 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5204 | if (ent.val && !ret && | |
34c00c31 | 5205 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5206 | ret = MC_TARGET_SWAP; |
5207 | if (target) | |
5208 | target->ent = ent; | |
4ffef5fe | 5209 | } |
4ffef5fe DN |
5210 | return ret; |
5211 | } | |
5212 | ||
12724850 NH |
5213 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5214 | /* | |
5215 | * We don't consider swapping or file mapped pages because THP does not | |
5216 | * support them for now. | |
5217 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5218 | */ | |
5219 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5220 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5221 | { | |
5222 | struct page *page = NULL; | |
5223 | struct page_cgroup *pc; | |
5224 | enum mc_target_type ret = MC_TARGET_NONE; | |
5225 | ||
5226 | page = pmd_page(pmd); | |
309381fe | 5227 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
5228 | if (!move_anon()) |
5229 | return ret; | |
5230 | pc = lookup_page_cgroup(page); | |
29833315 | 5231 | if (pc->mem_cgroup == mc.from) { |
12724850 NH |
5232 | ret = MC_TARGET_PAGE; |
5233 | if (target) { | |
5234 | get_page(page); | |
5235 | target->page = page; | |
5236 | } | |
5237 | } | |
5238 | return ret; | |
5239 | } | |
5240 | #else | |
5241 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5242 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5243 | { | |
5244 | return MC_TARGET_NONE; | |
5245 | } | |
5246 | #endif | |
5247 | ||
4ffef5fe DN |
5248 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5249 | unsigned long addr, unsigned long end, | |
5250 | struct mm_walk *walk) | |
5251 | { | |
5252 | struct vm_area_struct *vma = walk->private; | |
5253 | pte_t *pte; | |
5254 | spinlock_t *ptl; | |
5255 | ||
bf929152 | 5256 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
5257 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5258 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5259 | spin_unlock(ptl); |
1a5a9906 | 5260 | return 0; |
12724850 | 5261 | } |
03319327 | 5262 | |
45f83cef AA |
5263 | if (pmd_trans_unstable(pmd)) |
5264 | return 0; | |
4ffef5fe DN |
5265 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5266 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5267 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5268 | mc.precharge++; /* increment precharge temporarily */ |
5269 | pte_unmap_unlock(pte - 1, ptl); | |
5270 | cond_resched(); | |
5271 | ||
7dc74be0 DN |
5272 | return 0; |
5273 | } | |
5274 | ||
4ffef5fe DN |
5275 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5276 | { | |
5277 | unsigned long precharge; | |
5278 | struct vm_area_struct *vma; | |
5279 | ||
dfe076b0 | 5280 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5281 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5282 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5283 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5284 | .mm = mm, | |
5285 | .private = vma, | |
5286 | }; | |
5287 | if (is_vm_hugetlb_page(vma)) | |
5288 | continue; | |
4ffef5fe DN |
5289 | walk_page_range(vma->vm_start, vma->vm_end, |
5290 | &mem_cgroup_count_precharge_walk); | |
5291 | } | |
dfe076b0 | 5292 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5293 | |
5294 | precharge = mc.precharge; | |
5295 | mc.precharge = 0; | |
5296 | ||
5297 | return precharge; | |
5298 | } | |
5299 | ||
4ffef5fe DN |
5300 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5301 | { | |
dfe076b0 DN |
5302 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5303 | ||
5304 | VM_BUG_ON(mc.moving_task); | |
5305 | mc.moving_task = current; | |
5306 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5307 | } |
5308 | ||
dfe076b0 DN |
5309 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5310 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5311 | { |
2bd9bb20 KH |
5312 | struct mem_cgroup *from = mc.from; |
5313 | struct mem_cgroup *to = mc.to; | |
5314 | ||
4ffef5fe | 5315 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5316 | if (mc.precharge) { |
00501b53 | 5317 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5318 | mc.precharge = 0; |
5319 | } | |
5320 | /* | |
5321 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5322 | * we must uncharge here. | |
5323 | */ | |
5324 | if (mc.moved_charge) { | |
00501b53 | 5325 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5326 | mc.moved_charge = 0; |
4ffef5fe | 5327 | } |
483c30b5 DN |
5328 | /* we must fixup refcnts and charges */ |
5329 | if (mc.moved_swap) { | |
483c30b5 | 5330 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5331 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5332 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5333 | |
05b84301 | 5334 | /* |
3e32cb2e JW |
5335 | * we charged both to->memory and to->memsw, so we |
5336 | * should uncharge to->memory. | |
05b84301 | 5337 | */ |
ce00a967 | 5338 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5339 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5340 | ||
e8ea14cc | 5341 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 5342 | |
4050377b | 5343 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
5344 | mc.moved_swap = 0; |
5345 | } | |
dfe076b0 DN |
5346 | memcg_oom_recover(from); |
5347 | memcg_oom_recover(to); | |
5348 | wake_up_all(&mc.waitq); | |
5349 | } | |
5350 | ||
5351 | static void mem_cgroup_clear_mc(void) | |
5352 | { | |
5353 | struct mem_cgroup *from = mc.from; | |
5354 | ||
5355 | /* | |
5356 | * we must clear moving_task before waking up waiters at the end of | |
5357 | * task migration. | |
5358 | */ | |
5359 | mc.moving_task = NULL; | |
5360 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5361 | spin_lock(&mc.lock); |
4ffef5fe DN |
5362 | mc.from = NULL; |
5363 | mc.to = NULL; | |
2bd9bb20 | 5364 | spin_unlock(&mc.lock); |
247b1447 JW |
5365 | |
5366 | atomic_dec(&from->moving_account); | |
4ffef5fe DN |
5367 | } |
5368 | ||
eb95419b | 5369 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5370 | struct cgroup_taskset *tset) |
7dc74be0 | 5371 | { |
2f7ee569 | 5372 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5373 | int ret = 0; |
eb95419b | 5374 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 5375 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 5376 | |
ee5e8472 GC |
5377 | /* |
5378 | * We are now commited to this value whatever it is. Changes in this | |
5379 | * tunable will only affect upcoming migrations, not the current one. | |
5380 | * So we need to save it, and keep it going. | |
5381 | */ | |
5382 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
5383 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
5384 | struct mm_struct *mm; |
5385 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5386 | ||
c0ff4b85 | 5387 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5388 | |
5389 | mm = get_task_mm(p); | |
5390 | if (!mm) | |
5391 | return 0; | |
7dc74be0 | 5392 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5393 | if (mm->owner == p) { |
5394 | VM_BUG_ON(mc.from); | |
5395 | VM_BUG_ON(mc.to); | |
5396 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5397 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5398 | VM_BUG_ON(mc.moved_swap); |
247b1447 JW |
5399 | |
5400 | /* | |
5401 | * Signal mem_cgroup_begin_page_stat() to take | |
5402 | * the memcg's move_lock while we're moving | |
5403 | * its pages to another memcg. Then wait for | |
5404 | * already started RCU-only updates to finish. | |
5405 | */ | |
5406 | atomic_inc(&from->moving_account); | |
5407 | synchronize_rcu(); | |
5408 | ||
2bd9bb20 | 5409 | spin_lock(&mc.lock); |
4ffef5fe | 5410 | mc.from = from; |
c0ff4b85 | 5411 | mc.to = memcg; |
ee5e8472 | 5412 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 5413 | spin_unlock(&mc.lock); |
dfe076b0 | 5414 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5415 | |
5416 | ret = mem_cgroup_precharge_mc(mm); | |
5417 | if (ret) | |
5418 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5419 | } |
5420 | mmput(mm); | |
7dc74be0 DN |
5421 | } |
5422 | return ret; | |
5423 | } | |
5424 | ||
eb95419b | 5425 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5426 | struct cgroup_taskset *tset) |
7dc74be0 | 5427 | { |
4e2f245d JW |
5428 | if (mc.to) |
5429 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5430 | } |
5431 | ||
4ffef5fe DN |
5432 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5433 | unsigned long addr, unsigned long end, | |
5434 | struct mm_walk *walk) | |
7dc74be0 | 5435 | { |
4ffef5fe DN |
5436 | int ret = 0; |
5437 | struct vm_area_struct *vma = walk->private; | |
5438 | pte_t *pte; | |
5439 | spinlock_t *ptl; | |
12724850 NH |
5440 | enum mc_target_type target_type; |
5441 | union mc_target target; | |
5442 | struct page *page; | |
5443 | struct page_cgroup *pc; | |
4ffef5fe | 5444 | |
12724850 NH |
5445 | /* |
5446 | * We don't take compound_lock() here but no race with splitting thp | |
5447 | * happens because: | |
5448 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5449 | * under splitting, which means there's no concurrent thp split, | |
5450 | * - if another thread runs into split_huge_page() just after we | |
5451 | * entered this if-block, the thread must wait for page table lock | |
5452 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5453 | * part of thp split is not executed yet. | |
5454 | */ | |
bf929152 | 5455 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5456 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5457 | spin_unlock(ptl); |
12724850 NH |
5458 | return 0; |
5459 | } | |
5460 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5461 | if (target_type == MC_TARGET_PAGE) { | |
5462 | page = target.page; | |
5463 | if (!isolate_lru_page(page)) { | |
5464 | pc = lookup_page_cgroup(page); | |
5465 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5466 | pc, mc.from, mc.to)) { |
12724850 NH |
5467 | mc.precharge -= HPAGE_PMD_NR; |
5468 | mc.moved_charge += HPAGE_PMD_NR; | |
5469 | } | |
5470 | putback_lru_page(page); | |
5471 | } | |
5472 | put_page(page); | |
5473 | } | |
bf929152 | 5474 | spin_unlock(ptl); |
1a5a9906 | 5475 | return 0; |
12724850 NH |
5476 | } |
5477 | ||
45f83cef AA |
5478 | if (pmd_trans_unstable(pmd)) |
5479 | return 0; | |
4ffef5fe DN |
5480 | retry: |
5481 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5482 | for (; addr != end; addr += PAGE_SIZE) { | |
5483 | pte_t ptent = *(pte++); | |
02491447 | 5484 | swp_entry_t ent; |
4ffef5fe DN |
5485 | |
5486 | if (!mc.precharge) | |
5487 | break; | |
5488 | ||
8d32ff84 | 5489 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5490 | case MC_TARGET_PAGE: |
5491 | page = target.page; | |
5492 | if (isolate_lru_page(page)) | |
5493 | goto put; | |
5494 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5495 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5496 | mc.from, mc.to)) { |
4ffef5fe | 5497 | mc.precharge--; |
854ffa8d DN |
5498 | /* we uncharge from mc.from later. */ |
5499 | mc.moved_charge++; | |
4ffef5fe DN |
5500 | } |
5501 | putback_lru_page(page); | |
8d32ff84 | 5502 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5503 | put_page(page); |
5504 | break; | |
02491447 DN |
5505 | case MC_TARGET_SWAP: |
5506 | ent = target.ent; | |
e91cbb42 | 5507 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5508 | mc.precharge--; |
483c30b5 DN |
5509 | /* we fixup refcnts and charges later. */ |
5510 | mc.moved_swap++; | |
5511 | } | |
02491447 | 5512 | break; |
4ffef5fe DN |
5513 | default: |
5514 | break; | |
5515 | } | |
5516 | } | |
5517 | pte_unmap_unlock(pte - 1, ptl); | |
5518 | cond_resched(); | |
5519 | ||
5520 | if (addr != end) { | |
5521 | /* | |
5522 | * We have consumed all precharges we got in can_attach(). | |
5523 | * We try charge one by one, but don't do any additional | |
5524 | * charges to mc.to if we have failed in charge once in attach() | |
5525 | * phase. | |
5526 | */ | |
854ffa8d | 5527 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5528 | if (!ret) |
5529 | goto retry; | |
5530 | } | |
5531 | ||
5532 | return ret; | |
5533 | } | |
5534 | ||
5535 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5536 | { | |
5537 | struct vm_area_struct *vma; | |
5538 | ||
5539 | lru_add_drain_all(); | |
dfe076b0 DN |
5540 | retry: |
5541 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5542 | /* | |
5543 | * Someone who are holding the mmap_sem might be waiting in | |
5544 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5545 | * and retry. Because we cancel precharges, we might not be able | |
5546 | * to move enough charges, but moving charge is a best-effort | |
5547 | * feature anyway, so it wouldn't be a big problem. | |
5548 | */ | |
5549 | __mem_cgroup_clear_mc(); | |
5550 | cond_resched(); | |
5551 | goto retry; | |
5552 | } | |
4ffef5fe DN |
5553 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5554 | int ret; | |
5555 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5556 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5557 | .mm = mm, | |
5558 | .private = vma, | |
5559 | }; | |
5560 | if (is_vm_hugetlb_page(vma)) | |
5561 | continue; | |
4ffef5fe DN |
5562 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5563 | &mem_cgroup_move_charge_walk); | |
5564 | if (ret) | |
5565 | /* | |
5566 | * means we have consumed all precharges and failed in | |
5567 | * doing additional charge. Just abandon here. | |
5568 | */ | |
5569 | break; | |
5570 | } | |
dfe076b0 | 5571 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5572 | } |
5573 | ||
eb95419b | 5574 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5575 | struct cgroup_taskset *tset) |
67e465a7 | 5576 | { |
2f7ee569 | 5577 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5578 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5579 | |
dfe076b0 | 5580 | if (mm) { |
a433658c KM |
5581 | if (mc.to) |
5582 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5583 | mmput(mm); |
5584 | } | |
a433658c KM |
5585 | if (mc.to) |
5586 | mem_cgroup_clear_mc(); | |
67e465a7 | 5587 | } |
5cfb80a7 | 5588 | #else /* !CONFIG_MMU */ |
eb95419b | 5589 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5590 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5591 | { |
5592 | return 0; | |
5593 | } | |
eb95419b | 5594 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5595 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5596 | { |
5597 | } | |
eb95419b | 5598 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5599 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5600 | { |
5601 | } | |
5602 | #endif | |
67e465a7 | 5603 | |
f00baae7 TH |
5604 | /* |
5605 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5606 | * to verify whether we're attached to the default hierarchy on each mount |
5607 | * attempt. | |
f00baae7 | 5608 | */ |
eb95419b | 5609 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5610 | { |
5611 | /* | |
aa6ec29b | 5612 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5613 | * guarantees that @root doesn't have any children, so turning it |
5614 | * on for the root memcg is enough. | |
5615 | */ | |
aa6ec29b | 5616 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 5617 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
5618 | } |
5619 | ||
073219e9 | 5620 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5621 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5622 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5623 | .css_offline = mem_cgroup_css_offline, |
5624 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5625 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5626 | .can_attach = mem_cgroup_can_attach, |
5627 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5628 | .attach = mem_cgroup_move_task, |
f00baae7 | 5629 | .bind = mem_cgroup_bind, |
5577964e | 5630 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 5631 | .early_init = 0, |
8cdea7c0 | 5632 | }; |
c077719b | 5633 | |
c255a458 | 5634 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5635 | static int __init enable_swap_account(char *s) |
5636 | { | |
a2c8990a | 5637 | if (!strcmp(s, "1")) |
a42c390c | 5638 | really_do_swap_account = 1; |
a2c8990a | 5639 | else if (!strcmp(s, "0")) |
a42c390c MH |
5640 | really_do_swap_account = 0; |
5641 | return 1; | |
5642 | } | |
a2c8990a | 5643 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5644 | |
2d11085e MH |
5645 | static void __init memsw_file_init(void) |
5646 | { | |
2cf669a5 TH |
5647 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
5648 | memsw_cgroup_files)); | |
6acc8b02 MH |
5649 | } |
5650 | ||
5651 | static void __init enable_swap_cgroup(void) | |
5652 | { | |
5653 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5654 | do_swap_account = 1; | |
5655 | memsw_file_init(); | |
5656 | } | |
2d11085e | 5657 | } |
6acc8b02 | 5658 | |
2d11085e | 5659 | #else |
6acc8b02 | 5660 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
5661 | { |
5662 | } | |
c077719b | 5663 | #endif |
2d11085e | 5664 | |
0a31bc97 JW |
5665 | #ifdef CONFIG_MEMCG_SWAP |
5666 | /** | |
5667 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5668 | * @page: page whose memsw charge to transfer | |
5669 | * @entry: swap entry to move the charge to | |
5670 | * | |
5671 | * Transfer the memsw charge of @page to @entry. | |
5672 | */ | |
5673 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5674 | { | |
7bdd143c | 5675 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5676 | struct page_cgroup *pc; |
5677 | unsigned short oldid; | |
5678 | ||
5679 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5680 | VM_BUG_ON_PAGE(page_count(page), page); | |
5681 | ||
5682 | if (!do_swap_account) | |
5683 | return; | |
5684 | ||
5685 | pc = lookup_page_cgroup(page); | |
29833315 | 5686 | memcg = pc->mem_cgroup; |
0a31bc97 JW |
5687 | |
5688 | /* Readahead page, never charged */ | |
29833315 | 5689 | if (!memcg) |
0a31bc97 JW |
5690 | return; |
5691 | ||
7bdd143c | 5692 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
0a31bc97 | 5693 | VM_BUG_ON_PAGE(oldid, page); |
7bdd143c JW |
5694 | mem_cgroup_swap_statistics(memcg, true); |
5695 | ||
29833315 | 5696 | pc->mem_cgroup = NULL; |
7bdd143c JW |
5697 | |
5698 | if (!mem_cgroup_is_root(memcg)) | |
5699 | page_counter_uncharge(&memcg->memory, 1); | |
5700 | ||
5701 | /* XXX: caller holds IRQ-safe mapping->tree_lock */ | |
5702 | VM_BUG_ON(!irqs_disabled()); | |
0a31bc97 | 5703 | |
7bdd143c JW |
5704 | mem_cgroup_charge_statistics(memcg, page, -1); |
5705 | memcg_check_events(memcg, page); | |
0a31bc97 JW |
5706 | } |
5707 | ||
5708 | /** | |
5709 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5710 | * @entry: swap entry to uncharge | |
5711 | * | |
5712 | * Drop the memsw charge associated with @entry. | |
5713 | */ | |
5714 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5715 | { | |
5716 | struct mem_cgroup *memcg; | |
5717 | unsigned short id; | |
5718 | ||
5719 | if (!do_swap_account) | |
5720 | return; | |
5721 | ||
5722 | id = swap_cgroup_record(entry, 0); | |
5723 | rcu_read_lock(); | |
5724 | memcg = mem_cgroup_lookup(id); | |
5725 | if (memcg) { | |
ce00a967 | 5726 | if (!mem_cgroup_is_root(memcg)) |
3e32cb2e | 5727 | page_counter_uncharge(&memcg->memsw, 1); |
0a31bc97 JW |
5728 | mem_cgroup_swap_statistics(memcg, false); |
5729 | css_put(&memcg->css); | |
5730 | } | |
5731 | rcu_read_unlock(); | |
5732 | } | |
5733 | #endif | |
5734 | ||
00501b53 JW |
5735 | /** |
5736 | * mem_cgroup_try_charge - try charging a page | |
5737 | * @page: page to charge | |
5738 | * @mm: mm context of the victim | |
5739 | * @gfp_mask: reclaim mode | |
5740 | * @memcgp: charged memcg return | |
5741 | * | |
5742 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5743 | * pages according to @gfp_mask if necessary. | |
5744 | * | |
5745 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5746 | * Otherwise, an error code is returned. | |
5747 | * | |
5748 | * After page->mapping has been set up, the caller must finalize the | |
5749 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5750 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5751 | */ | |
5752 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5753 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5754 | { | |
5755 | struct mem_cgroup *memcg = NULL; | |
5756 | unsigned int nr_pages = 1; | |
5757 | int ret = 0; | |
5758 | ||
5759 | if (mem_cgroup_disabled()) | |
5760 | goto out; | |
5761 | ||
5762 | if (PageSwapCache(page)) { | |
5763 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
5764 | /* | |
5765 | * Every swap fault against a single page tries to charge the | |
5766 | * page, bail as early as possible. shmem_unuse() encounters | |
5767 | * already charged pages, too. The USED bit is protected by | |
5768 | * the page lock, which serializes swap cache removal, which | |
5769 | * in turn serializes uncharging. | |
5770 | */ | |
29833315 | 5771 | if (pc->mem_cgroup) |
00501b53 JW |
5772 | goto out; |
5773 | } | |
5774 | ||
5775 | if (PageTransHuge(page)) { | |
5776 | nr_pages <<= compound_order(page); | |
5777 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5778 | } | |
5779 | ||
5780 | if (do_swap_account && PageSwapCache(page)) | |
5781 | memcg = try_get_mem_cgroup_from_page(page); | |
5782 | if (!memcg) | |
5783 | memcg = get_mem_cgroup_from_mm(mm); | |
5784 | ||
5785 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5786 | ||
5787 | css_put(&memcg->css); | |
5788 | ||
5789 | if (ret == -EINTR) { | |
5790 | memcg = root_mem_cgroup; | |
5791 | ret = 0; | |
5792 | } | |
5793 | out: | |
5794 | *memcgp = memcg; | |
5795 | return ret; | |
5796 | } | |
5797 | ||
5798 | /** | |
5799 | * mem_cgroup_commit_charge - commit a page charge | |
5800 | * @page: page to charge | |
5801 | * @memcg: memcg to charge the page to | |
5802 | * @lrucare: page might be on LRU already | |
5803 | * | |
5804 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5805 | * after page->mapping has been set up. This must happen atomically | |
5806 | * as part of the page instantiation, i.e. under the page table lock | |
5807 | * for anonymous pages, under the page lock for page and swap cache. | |
5808 | * | |
5809 | * In addition, the page must not be on the LRU during the commit, to | |
5810 | * prevent racing with task migration. If it might be, use @lrucare. | |
5811 | * | |
5812 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5813 | */ | |
5814 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5815 | bool lrucare) | |
5816 | { | |
5817 | unsigned int nr_pages = 1; | |
5818 | ||
5819 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5820 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5821 | ||
5822 | if (mem_cgroup_disabled()) | |
5823 | return; | |
5824 | /* | |
5825 | * Swap faults will attempt to charge the same page multiple | |
5826 | * times. But reuse_swap_page() might have removed the page | |
5827 | * from swapcache already, so we can't check PageSwapCache(). | |
5828 | */ | |
5829 | if (!memcg) | |
5830 | return; | |
5831 | ||
6abb5a86 JW |
5832 | commit_charge(page, memcg, lrucare); |
5833 | ||
00501b53 JW |
5834 | if (PageTransHuge(page)) { |
5835 | nr_pages <<= compound_order(page); | |
5836 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5837 | } | |
5838 | ||
6abb5a86 JW |
5839 | local_irq_disable(); |
5840 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5841 | memcg_check_events(memcg, page); | |
5842 | local_irq_enable(); | |
00501b53 JW |
5843 | |
5844 | if (do_swap_account && PageSwapCache(page)) { | |
5845 | swp_entry_t entry = { .val = page_private(page) }; | |
5846 | /* | |
5847 | * The swap entry might not get freed for a long time, | |
5848 | * let's not wait for it. The page already received a | |
5849 | * memory+swap charge, drop the swap entry duplicate. | |
5850 | */ | |
5851 | mem_cgroup_uncharge_swap(entry); | |
5852 | } | |
5853 | } | |
5854 | ||
5855 | /** | |
5856 | * mem_cgroup_cancel_charge - cancel a page charge | |
5857 | * @page: page to charge | |
5858 | * @memcg: memcg to charge the page to | |
5859 | * | |
5860 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5861 | */ | |
5862 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5863 | { | |
5864 | unsigned int nr_pages = 1; | |
5865 | ||
5866 | if (mem_cgroup_disabled()) | |
5867 | return; | |
5868 | /* | |
5869 | * Swap faults will attempt to charge the same page multiple | |
5870 | * times. But reuse_swap_page() might have removed the page | |
5871 | * from swapcache already, so we can't check PageSwapCache(). | |
5872 | */ | |
5873 | if (!memcg) | |
5874 | return; | |
5875 | ||
5876 | if (PageTransHuge(page)) { | |
5877 | nr_pages <<= compound_order(page); | |
5878 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5879 | } | |
5880 | ||
5881 | cancel_charge(memcg, nr_pages); | |
5882 | } | |
5883 | ||
747db954 | 5884 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5885 | unsigned long nr_anon, unsigned long nr_file, |
5886 | unsigned long nr_huge, struct page *dummy_page) | |
5887 | { | |
18eca2e6 | 5888 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5889 | unsigned long flags; |
5890 | ||
ce00a967 | 5891 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5892 | page_counter_uncharge(&memcg->memory, nr_pages); |
5893 | if (do_swap_account) | |
5894 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5895 | memcg_oom_recover(memcg); |
5896 | } | |
747db954 JW |
5897 | |
5898 | local_irq_save(flags); | |
5899 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5900 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5901 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5902 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5903 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5904 | memcg_check_events(memcg, dummy_page); |
5905 | local_irq_restore(flags); | |
e8ea14cc JW |
5906 | |
5907 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5908 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5909 | } |
5910 | ||
5911 | static void uncharge_list(struct list_head *page_list) | |
5912 | { | |
5913 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5914 | unsigned long nr_anon = 0; |
5915 | unsigned long nr_file = 0; | |
5916 | unsigned long nr_huge = 0; | |
5917 | unsigned long pgpgout = 0; | |
747db954 JW |
5918 | struct list_head *next; |
5919 | struct page *page; | |
5920 | ||
5921 | next = page_list->next; | |
5922 | do { | |
5923 | unsigned int nr_pages = 1; | |
5924 | struct page_cgroup *pc; | |
5925 | ||
5926 | page = list_entry(next, struct page, lru); | |
5927 | next = page->lru.next; | |
5928 | ||
5929 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5930 | VM_BUG_ON_PAGE(page_count(page), page); | |
5931 | ||
5932 | pc = lookup_page_cgroup(page); | |
29833315 | 5933 | if (!pc->mem_cgroup) |
747db954 JW |
5934 | continue; |
5935 | ||
5936 | /* | |
5937 | * Nobody should be changing or seriously looking at | |
29833315 JW |
5938 | * pc->mem_cgroup at this point, we have fully |
5939 | * exclusive access to the page. | |
747db954 JW |
5940 | */ |
5941 | ||
5942 | if (memcg != pc->mem_cgroup) { | |
5943 | if (memcg) { | |
18eca2e6 JW |
5944 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5945 | nr_huge, page); | |
5946 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 JW |
5947 | } |
5948 | memcg = pc->mem_cgroup; | |
5949 | } | |
5950 | ||
5951 | if (PageTransHuge(page)) { | |
5952 | nr_pages <<= compound_order(page); | |
5953 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5954 | nr_huge += nr_pages; | |
5955 | } | |
5956 | ||
5957 | if (PageAnon(page)) | |
5958 | nr_anon += nr_pages; | |
5959 | else | |
5960 | nr_file += nr_pages; | |
5961 | ||
29833315 | 5962 | pc->mem_cgroup = NULL; |
747db954 JW |
5963 | |
5964 | pgpgout++; | |
5965 | } while (next != page_list); | |
5966 | ||
5967 | if (memcg) | |
18eca2e6 JW |
5968 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5969 | nr_huge, page); | |
747db954 JW |
5970 | } |
5971 | ||
0a31bc97 JW |
5972 | /** |
5973 | * mem_cgroup_uncharge - uncharge a page | |
5974 | * @page: page to uncharge | |
5975 | * | |
5976 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5977 | * mem_cgroup_commit_charge(). | |
5978 | */ | |
5979 | void mem_cgroup_uncharge(struct page *page) | |
5980 | { | |
0a31bc97 | 5981 | struct page_cgroup *pc; |
0a31bc97 JW |
5982 | |
5983 | if (mem_cgroup_disabled()) | |
5984 | return; | |
5985 | ||
747db954 | 5986 | /* Don't touch page->lru of any random page, pre-check: */ |
0a31bc97 | 5987 | pc = lookup_page_cgroup(page); |
29833315 | 5988 | if (!pc->mem_cgroup) |
0a31bc97 JW |
5989 | return; |
5990 | ||
747db954 JW |
5991 | INIT_LIST_HEAD(&page->lru); |
5992 | uncharge_list(&page->lru); | |
5993 | } | |
0a31bc97 | 5994 | |
747db954 JW |
5995 | /** |
5996 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5997 | * @page_list: list of pages to uncharge | |
5998 | * | |
5999 | * Uncharge a list of pages previously charged with | |
6000 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6001 | */ | |
6002 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6003 | { | |
6004 | if (mem_cgroup_disabled()) | |
6005 | return; | |
0a31bc97 | 6006 | |
747db954 JW |
6007 | if (!list_empty(page_list)) |
6008 | uncharge_list(page_list); | |
0a31bc97 JW |
6009 | } |
6010 | ||
6011 | /** | |
6012 | * mem_cgroup_migrate - migrate a charge to another page | |
6013 | * @oldpage: currently charged page | |
6014 | * @newpage: page to transfer the charge to | |
6015 | * @lrucare: both pages might be on the LRU already | |
6016 | * | |
6017 | * Migrate the charge from @oldpage to @newpage. | |
6018 | * | |
6019 | * Both pages must be locked, @newpage->mapping must be set up. | |
6020 | */ | |
6021 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
6022 | bool lrucare) | |
6023 | { | |
29833315 | 6024 | struct mem_cgroup *memcg; |
0a31bc97 JW |
6025 | struct page_cgroup *pc; |
6026 | int isolated; | |
6027 | ||
6028 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6029 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
6030 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
6031 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
6032 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
6033 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6034 | newpage); | |
0a31bc97 JW |
6035 | |
6036 | if (mem_cgroup_disabled()) | |
6037 | return; | |
6038 | ||
6039 | /* Page cache replacement: new page already charged? */ | |
6040 | pc = lookup_page_cgroup(newpage); | |
29833315 | 6041 | if (pc->mem_cgroup) |
0a31bc97 JW |
6042 | return; |
6043 | ||
7d5e3245 JW |
6044 | /* |
6045 | * Swapcache readahead pages can get migrated before being | |
6046 | * charged, and migration from compaction can happen to an | |
6047 | * uncharged page when the PFN walker finds a page that | |
6048 | * reclaim just put back on the LRU but has not released yet. | |
6049 | */ | |
0a31bc97 | 6050 | pc = lookup_page_cgroup(oldpage); |
29833315 JW |
6051 | memcg = pc->mem_cgroup; |
6052 | if (!memcg) | |
0a31bc97 JW |
6053 | return; |
6054 | ||
0a31bc97 JW |
6055 | if (lrucare) |
6056 | lock_page_lru(oldpage, &isolated); | |
6057 | ||
29833315 | 6058 | pc->mem_cgroup = NULL; |
0a31bc97 JW |
6059 | |
6060 | if (lrucare) | |
6061 | unlock_page_lru(oldpage, isolated); | |
6062 | ||
29833315 | 6063 | commit_charge(newpage, memcg, lrucare); |
0a31bc97 JW |
6064 | } |
6065 | ||
2d11085e | 6066 | /* |
1081312f MH |
6067 | * subsys_initcall() for memory controller. |
6068 | * | |
6069 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6070 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6071 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6072 | * be initialized from here. | |
2d11085e MH |
6073 | */ |
6074 | static int __init mem_cgroup_init(void) | |
6075 | { | |
6076 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6077 | enable_swap_cgroup(); |
bb4cc1a8 | 6078 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6079 | memcg_stock_init(); |
2d11085e MH |
6080 | return 0; |
6081 | } | |
6082 | subsys_initcall(mem_cgroup_init); |