]>
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 | |
2314b42d | 1331 | bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root) |
3e92041d | 1332 | { |
2314b42d | 1333 | if (root == memcg) |
91c63734 | 1334 | return true; |
2314b42d | 1335 | if (!root->use_hierarchy) |
91c63734 | 1336 | return false; |
2314b42d | 1337 | return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); |
3e92041d MH |
1338 | } |
1339 | ||
2314b42d | 1340 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
4c4a2214 | 1341 | { |
2314b42d | 1342 | struct mem_cgroup *task_memcg; |
158e0a2d | 1343 | struct task_struct *p; |
ffbdccf5 | 1344 | bool ret; |
4c4a2214 | 1345 | |
158e0a2d | 1346 | p = find_lock_task_mm(task); |
de077d22 | 1347 | if (p) { |
2314b42d | 1348 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1349 | task_unlock(p); |
1350 | } else { | |
1351 | /* | |
1352 | * All threads may have already detached their mm's, but the oom | |
1353 | * killer still needs to detect if they have already been oom | |
1354 | * killed to prevent needlessly killing additional tasks. | |
1355 | */ | |
ffbdccf5 | 1356 | rcu_read_lock(); |
2314b42d JW |
1357 | task_memcg = mem_cgroup_from_task(task); |
1358 | css_get(&task_memcg->css); | |
ffbdccf5 | 1359 | rcu_read_unlock(); |
de077d22 | 1360 | } |
2314b42d JW |
1361 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1362 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1363 | return ret; |
1364 | } | |
1365 | ||
c56d5c7d | 1366 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1367 | { |
9b272977 | 1368 | unsigned long inactive_ratio; |
14797e23 | 1369 | unsigned long inactive; |
9b272977 | 1370 | unsigned long active; |
c772be93 | 1371 | unsigned long gb; |
14797e23 | 1372 | |
4d7dcca2 HD |
1373 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1374 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1375 | |
c772be93 KM |
1376 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1377 | if (gb) | |
1378 | inactive_ratio = int_sqrt(10 * gb); | |
1379 | else | |
1380 | inactive_ratio = 1; | |
1381 | ||
9b272977 | 1382 | return inactive * inactive_ratio < active; |
14797e23 KM |
1383 | } |
1384 | ||
3e32cb2e | 1385 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1386 | container_of(counter, struct mem_cgroup, member) |
1387 | ||
19942822 | 1388 | /** |
9d11ea9f | 1389 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1390 | * @memcg: the memory cgroup |
19942822 | 1391 | * |
9d11ea9f | 1392 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1393 | * pages. |
19942822 | 1394 | */ |
c0ff4b85 | 1395 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1396 | { |
3e32cb2e JW |
1397 | unsigned long margin = 0; |
1398 | unsigned long count; | |
1399 | unsigned long limit; | |
9d11ea9f | 1400 | |
3e32cb2e JW |
1401 | count = page_counter_read(&memcg->memory); |
1402 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1403 | if (count < limit) | |
1404 | margin = limit - count; | |
1405 | ||
1406 | if (do_swap_account) { | |
1407 | count = page_counter_read(&memcg->memsw); | |
1408 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1409 | if (count <= limit) | |
1410 | margin = min(margin, limit - count); | |
1411 | } | |
1412 | ||
1413 | return margin; | |
19942822 JW |
1414 | } |
1415 | ||
1f4c025b | 1416 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1417 | { |
a7885eb8 | 1418 | /* root ? */ |
14208b0e | 1419 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1420 | return vm_swappiness; |
1421 | ||
bf1ff263 | 1422 | return memcg->swappiness; |
a7885eb8 KM |
1423 | } |
1424 | ||
32047e2a | 1425 | /* |
bdcbb659 | 1426 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1427 | * |
bdcbb659 QH |
1428 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1429 | * moving cgroups. This is for waiting at high-memory pressure | |
1430 | * caused by "move". | |
32047e2a | 1431 | */ |
c0ff4b85 | 1432 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1433 | { |
2bd9bb20 KH |
1434 | struct mem_cgroup *from; |
1435 | struct mem_cgroup *to; | |
4b534334 | 1436 | bool ret = false; |
2bd9bb20 KH |
1437 | /* |
1438 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1439 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1440 | */ | |
1441 | spin_lock(&mc.lock); | |
1442 | from = mc.from; | |
1443 | to = mc.to; | |
1444 | if (!from) | |
1445 | goto unlock; | |
3e92041d | 1446 | |
2314b42d JW |
1447 | ret = mem_cgroup_is_descendant(from, memcg) || |
1448 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1449 | unlock: |
1450 | spin_unlock(&mc.lock); | |
4b534334 KH |
1451 | return ret; |
1452 | } | |
1453 | ||
c0ff4b85 | 1454 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1455 | { |
1456 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1457 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1458 | DEFINE_WAIT(wait); |
1459 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1460 | /* moving charge context might have finished. */ | |
1461 | if (mc.moving_task) | |
1462 | schedule(); | |
1463 | finish_wait(&mc.waitq, &wait); | |
1464 | return true; | |
1465 | } | |
1466 | } | |
1467 | return false; | |
1468 | } | |
1469 | ||
58cf188e | 1470 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1471 | /** |
58cf188e | 1472 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1473 | * @memcg: The memory cgroup that went over limit |
1474 | * @p: Task that is going to be killed | |
1475 | * | |
1476 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1477 | * enabled | |
1478 | */ | |
1479 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1480 | { | |
e61734c5 | 1481 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1482 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1483 | struct mem_cgroup *iter; |
1484 | unsigned int i; | |
e222432b | 1485 | |
58cf188e | 1486 | if (!p) |
e222432b BS |
1487 | return; |
1488 | ||
08088cb9 | 1489 | mutex_lock(&oom_info_lock); |
e222432b BS |
1490 | rcu_read_lock(); |
1491 | ||
e61734c5 TH |
1492 | pr_info("Task in "); |
1493 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1494 | pr_info(" killed as a result of limit of "); | |
1495 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1496 | pr_info("\n"); | |
e222432b | 1497 | |
e222432b BS |
1498 | rcu_read_unlock(); |
1499 | ||
3e32cb2e JW |
1500 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1501 | K((u64)page_counter_read(&memcg->memory)), | |
1502 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1503 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1504 | K((u64)page_counter_read(&memcg->memsw)), | |
1505 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1506 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1507 | K((u64)page_counter_read(&memcg->kmem)), | |
1508 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1509 | |
1510 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1511 | pr_info("Memory cgroup stats for "); |
1512 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1513 | pr_cont(":"); |
1514 | ||
1515 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1516 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1517 | continue; | |
1518 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1519 | K(mem_cgroup_read_stat(iter, i))); | |
1520 | } | |
1521 | ||
1522 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1523 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1524 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1525 | ||
1526 | pr_cont("\n"); | |
1527 | } | |
08088cb9 | 1528 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1529 | } |
1530 | ||
81d39c20 KH |
1531 | /* |
1532 | * This function returns the number of memcg under hierarchy tree. Returns | |
1533 | * 1(self count) if no children. | |
1534 | */ | |
c0ff4b85 | 1535 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1536 | { |
1537 | int num = 0; | |
7d74b06f KH |
1538 | struct mem_cgroup *iter; |
1539 | ||
c0ff4b85 | 1540 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1541 | num++; |
81d39c20 KH |
1542 | return num; |
1543 | } | |
1544 | ||
a63d83f4 DR |
1545 | /* |
1546 | * Return the memory (and swap, if configured) limit for a memcg. | |
1547 | */ | |
3e32cb2e | 1548 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1549 | { |
3e32cb2e | 1550 | unsigned long limit; |
a63d83f4 | 1551 | |
3e32cb2e | 1552 | limit = memcg->memory.limit; |
9a5a8f19 | 1553 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1554 | unsigned long memsw_limit; |
9a5a8f19 | 1555 | |
3e32cb2e JW |
1556 | memsw_limit = memcg->memsw.limit; |
1557 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1558 | } |
9a5a8f19 | 1559 | return limit; |
a63d83f4 DR |
1560 | } |
1561 | ||
19965460 DR |
1562 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1563 | int order) | |
9cbb78bb DR |
1564 | { |
1565 | struct mem_cgroup *iter; | |
1566 | unsigned long chosen_points = 0; | |
1567 | unsigned long totalpages; | |
1568 | unsigned int points = 0; | |
1569 | struct task_struct *chosen = NULL; | |
1570 | ||
876aafbf | 1571 | /* |
465adcf1 DR |
1572 | * If current has a pending SIGKILL or is exiting, then automatically |
1573 | * select it. The goal is to allow it to allocate so that it may | |
1574 | * quickly exit and free its memory. | |
876aafbf | 1575 | */ |
465adcf1 | 1576 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1577 | set_thread_flag(TIF_MEMDIE); |
1578 | return; | |
1579 | } | |
1580 | ||
1581 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1582 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1583 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1584 | struct css_task_iter it; |
9cbb78bb DR |
1585 | struct task_struct *task; |
1586 | ||
72ec7029 TH |
1587 | css_task_iter_start(&iter->css, &it); |
1588 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1589 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1590 | false)) { | |
1591 | case OOM_SCAN_SELECT: | |
1592 | if (chosen) | |
1593 | put_task_struct(chosen); | |
1594 | chosen = task; | |
1595 | chosen_points = ULONG_MAX; | |
1596 | get_task_struct(chosen); | |
1597 | /* fall through */ | |
1598 | case OOM_SCAN_CONTINUE: | |
1599 | continue; | |
1600 | case OOM_SCAN_ABORT: | |
72ec7029 | 1601 | css_task_iter_end(&it); |
9cbb78bb DR |
1602 | mem_cgroup_iter_break(memcg, iter); |
1603 | if (chosen) | |
1604 | put_task_struct(chosen); | |
1605 | return; | |
1606 | case OOM_SCAN_OK: | |
1607 | break; | |
1608 | }; | |
1609 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1610 | if (!points || points < chosen_points) |
1611 | continue; | |
1612 | /* Prefer thread group leaders for display purposes */ | |
1613 | if (points == chosen_points && | |
1614 | thread_group_leader(chosen)) | |
1615 | continue; | |
1616 | ||
1617 | if (chosen) | |
1618 | put_task_struct(chosen); | |
1619 | chosen = task; | |
1620 | chosen_points = points; | |
1621 | get_task_struct(chosen); | |
9cbb78bb | 1622 | } |
72ec7029 | 1623 | css_task_iter_end(&it); |
9cbb78bb DR |
1624 | } |
1625 | ||
1626 | if (!chosen) | |
1627 | return; | |
1628 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1629 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1630 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1631 | } |
1632 | ||
4d0c066d KH |
1633 | /** |
1634 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1635 | * @memcg: the target memcg |
4d0c066d KH |
1636 | * @nid: the node ID to be checked. |
1637 | * @noswap : specify true here if the user wants flle only information. | |
1638 | * | |
1639 | * This function returns whether the specified memcg contains any | |
1640 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1641 | * pages in the node. | |
1642 | */ | |
c0ff4b85 | 1643 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1644 | int nid, bool noswap) |
1645 | { | |
c0ff4b85 | 1646 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1647 | return true; |
1648 | if (noswap || !total_swap_pages) | |
1649 | return false; | |
c0ff4b85 | 1650 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1651 | return true; |
1652 | return false; | |
1653 | ||
1654 | } | |
bb4cc1a8 | 1655 | #if MAX_NUMNODES > 1 |
889976db YH |
1656 | |
1657 | /* | |
1658 | * Always updating the nodemask is not very good - even if we have an empty | |
1659 | * list or the wrong list here, we can start from some node and traverse all | |
1660 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1661 | * | |
1662 | */ | |
c0ff4b85 | 1663 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1664 | { |
1665 | int nid; | |
453a9bf3 KH |
1666 | /* |
1667 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1668 | * pagein/pageout changes since the last update. | |
1669 | */ | |
c0ff4b85 | 1670 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1671 | return; |
c0ff4b85 | 1672 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1673 | return; |
1674 | ||
889976db | 1675 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1676 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1677 | |
31aaea4a | 1678 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1679 | |
c0ff4b85 R |
1680 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1681 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1682 | } |
453a9bf3 | 1683 | |
c0ff4b85 R |
1684 | atomic_set(&memcg->numainfo_events, 0); |
1685 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1686 | } |
1687 | ||
1688 | /* | |
1689 | * Selecting a node where we start reclaim from. Because what we need is just | |
1690 | * reducing usage counter, start from anywhere is O,K. Considering | |
1691 | * memory reclaim from current node, there are pros. and cons. | |
1692 | * | |
1693 | * Freeing memory from current node means freeing memory from a node which | |
1694 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1695 | * hit limits, it will see a contention on a node. But freeing from remote | |
1696 | * node means more costs for memory reclaim because of memory latency. | |
1697 | * | |
1698 | * Now, we use round-robin. Better algorithm is welcomed. | |
1699 | */ | |
c0ff4b85 | 1700 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1701 | { |
1702 | int node; | |
1703 | ||
c0ff4b85 R |
1704 | mem_cgroup_may_update_nodemask(memcg); |
1705 | node = memcg->last_scanned_node; | |
889976db | 1706 | |
c0ff4b85 | 1707 | node = next_node(node, memcg->scan_nodes); |
889976db | 1708 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1709 | node = first_node(memcg->scan_nodes); |
889976db YH |
1710 | /* |
1711 | * We call this when we hit limit, not when pages are added to LRU. | |
1712 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1713 | * memcg is too small and all pages are not on LRU. In that case, | |
1714 | * we use curret node. | |
1715 | */ | |
1716 | if (unlikely(node == MAX_NUMNODES)) | |
1717 | node = numa_node_id(); | |
1718 | ||
c0ff4b85 | 1719 | memcg->last_scanned_node = node; |
889976db YH |
1720 | return node; |
1721 | } | |
889976db | 1722 | #else |
c0ff4b85 | 1723 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1724 | { |
1725 | return 0; | |
1726 | } | |
1727 | #endif | |
1728 | ||
0608f43d AM |
1729 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1730 | struct zone *zone, | |
1731 | gfp_t gfp_mask, | |
1732 | unsigned long *total_scanned) | |
1733 | { | |
1734 | struct mem_cgroup *victim = NULL; | |
1735 | int total = 0; | |
1736 | int loop = 0; | |
1737 | unsigned long excess; | |
1738 | unsigned long nr_scanned; | |
1739 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1740 | .zone = zone, | |
1741 | .priority = 0, | |
1742 | }; | |
1743 | ||
3e32cb2e | 1744 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1745 | |
1746 | while (1) { | |
1747 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1748 | if (!victim) { | |
1749 | loop++; | |
1750 | if (loop >= 2) { | |
1751 | /* | |
1752 | * If we have not been able to reclaim | |
1753 | * anything, it might because there are | |
1754 | * no reclaimable pages under this hierarchy | |
1755 | */ | |
1756 | if (!total) | |
1757 | break; | |
1758 | /* | |
1759 | * We want to do more targeted reclaim. | |
1760 | * excess >> 2 is not to excessive so as to | |
1761 | * reclaim too much, nor too less that we keep | |
1762 | * coming back to reclaim from this cgroup | |
1763 | */ | |
1764 | if (total >= (excess >> 2) || | |
1765 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1766 | break; | |
1767 | } | |
1768 | continue; | |
1769 | } | |
0608f43d AM |
1770 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1771 | zone, &nr_scanned); | |
1772 | *total_scanned += nr_scanned; | |
3e32cb2e | 1773 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1774 | break; |
6d61ef40 | 1775 | } |
0608f43d AM |
1776 | mem_cgroup_iter_break(root_memcg, victim); |
1777 | return total; | |
6d61ef40 BS |
1778 | } |
1779 | ||
0056f4e6 JW |
1780 | #ifdef CONFIG_LOCKDEP |
1781 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1782 | .name = "memcg_oom_lock", | |
1783 | }; | |
1784 | #endif | |
1785 | ||
fb2a6fc5 JW |
1786 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1787 | ||
867578cb KH |
1788 | /* |
1789 | * Check OOM-Killer is already running under our hierarchy. | |
1790 | * If someone is running, return false. | |
1791 | */ | |
fb2a6fc5 | 1792 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1793 | { |
79dfdacc | 1794 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1795 | |
fb2a6fc5 JW |
1796 | spin_lock(&memcg_oom_lock); |
1797 | ||
9f3a0d09 | 1798 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1799 | if (iter->oom_lock) { |
79dfdacc MH |
1800 | /* |
1801 | * this subtree of our hierarchy is already locked | |
1802 | * so we cannot give a lock. | |
1803 | */ | |
79dfdacc | 1804 | failed = iter; |
9f3a0d09 JW |
1805 | mem_cgroup_iter_break(memcg, iter); |
1806 | break; | |
23751be0 JW |
1807 | } else |
1808 | iter->oom_lock = true; | |
7d74b06f | 1809 | } |
867578cb | 1810 | |
fb2a6fc5 JW |
1811 | if (failed) { |
1812 | /* | |
1813 | * OK, we failed to lock the whole subtree so we have | |
1814 | * to clean up what we set up to the failing subtree | |
1815 | */ | |
1816 | for_each_mem_cgroup_tree(iter, memcg) { | |
1817 | if (iter == failed) { | |
1818 | mem_cgroup_iter_break(memcg, iter); | |
1819 | break; | |
1820 | } | |
1821 | iter->oom_lock = false; | |
79dfdacc | 1822 | } |
0056f4e6 JW |
1823 | } else |
1824 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1825 | |
1826 | spin_unlock(&memcg_oom_lock); | |
1827 | ||
1828 | return !failed; | |
a636b327 | 1829 | } |
0b7f569e | 1830 | |
fb2a6fc5 | 1831 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1832 | { |
7d74b06f KH |
1833 | struct mem_cgroup *iter; |
1834 | ||
fb2a6fc5 | 1835 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1836 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1837 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1838 | iter->oom_lock = false; |
fb2a6fc5 | 1839 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1840 | } |
1841 | ||
c0ff4b85 | 1842 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1843 | { |
1844 | struct mem_cgroup *iter; | |
1845 | ||
c0ff4b85 | 1846 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1847 | atomic_inc(&iter->under_oom); |
1848 | } | |
1849 | ||
c0ff4b85 | 1850 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1851 | { |
1852 | struct mem_cgroup *iter; | |
1853 | ||
867578cb KH |
1854 | /* |
1855 | * When a new child is created while the hierarchy is under oom, | |
1856 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1857 | * atomic_add_unless() here. | |
1858 | */ | |
c0ff4b85 | 1859 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1860 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1861 | } |
1862 | ||
867578cb KH |
1863 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1864 | ||
dc98df5a | 1865 | struct oom_wait_info { |
d79154bb | 1866 | struct mem_cgroup *memcg; |
dc98df5a KH |
1867 | wait_queue_t wait; |
1868 | }; | |
1869 | ||
1870 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1871 | unsigned mode, int sync, void *arg) | |
1872 | { | |
d79154bb HD |
1873 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1874 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1875 | struct oom_wait_info *oom_wait_info; |
1876 | ||
1877 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1878 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1879 | |
2314b42d JW |
1880 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1881 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1882 | return 0; |
dc98df5a KH |
1883 | return autoremove_wake_function(wait, mode, sync, arg); |
1884 | } | |
1885 | ||
c0ff4b85 | 1886 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1887 | { |
3812c8c8 | 1888 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
1889 | /* for filtering, pass "memcg" as argument. */ |
1890 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1891 | } |
1892 | ||
c0ff4b85 | 1893 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1894 | { |
c0ff4b85 R |
1895 | if (memcg && atomic_read(&memcg->under_oom)) |
1896 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1897 | } |
1898 | ||
3812c8c8 | 1899 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1900 | { |
3812c8c8 JW |
1901 | if (!current->memcg_oom.may_oom) |
1902 | return; | |
867578cb | 1903 | /* |
49426420 JW |
1904 | * We are in the middle of the charge context here, so we |
1905 | * don't want to block when potentially sitting on a callstack | |
1906 | * that holds all kinds of filesystem and mm locks. | |
1907 | * | |
1908 | * Also, the caller may handle a failed allocation gracefully | |
1909 | * (like optional page cache readahead) and so an OOM killer | |
1910 | * invocation might not even be necessary. | |
1911 | * | |
1912 | * That's why we don't do anything here except remember the | |
1913 | * OOM context and then deal with it at the end of the page | |
1914 | * fault when the stack is unwound, the locks are released, | |
1915 | * and when we know whether the fault was overall successful. | |
867578cb | 1916 | */ |
49426420 JW |
1917 | css_get(&memcg->css); |
1918 | current->memcg_oom.memcg = memcg; | |
1919 | current->memcg_oom.gfp_mask = mask; | |
1920 | current->memcg_oom.order = order; | |
3812c8c8 JW |
1921 | } |
1922 | ||
1923 | /** | |
1924 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1925 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1926 | * |
49426420 JW |
1927 | * This has to be called at the end of a page fault if the memcg OOM |
1928 | * handler was enabled. | |
3812c8c8 | 1929 | * |
49426420 | 1930 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1931 | * sleep on a waitqueue until the userspace task resolves the |
1932 | * situation. Sleeping directly in the charge context with all kinds | |
1933 | * of locks held is not a good idea, instead we remember an OOM state | |
1934 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1935 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1936 | * |
1937 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1938 | * completed, %false otherwise. |
3812c8c8 | 1939 | */ |
49426420 | 1940 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1941 | { |
49426420 | 1942 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 1943 | struct oom_wait_info owait; |
49426420 | 1944 | bool locked; |
3812c8c8 JW |
1945 | |
1946 | /* OOM is global, do not handle */ | |
3812c8c8 | 1947 | if (!memcg) |
49426420 | 1948 | return false; |
3812c8c8 | 1949 | |
49426420 JW |
1950 | if (!handle) |
1951 | goto cleanup; | |
3812c8c8 JW |
1952 | |
1953 | owait.memcg = memcg; | |
1954 | owait.wait.flags = 0; | |
1955 | owait.wait.func = memcg_oom_wake_function; | |
1956 | owait.wait.private = current; | |
1957 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1958 | |
3812c8c8 | 1959 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1960 | mem_cgroup_mark_under_oom(memcg); |
1961 | ||
1962 | locked = mem_cgroup_oom_trylock(memcg); | |
1963 | ||
1964 | if (locked) | |
1965 | mem_cgroup_oom_notify(memcg); | |
1966 | ||
1967 | if (locked && !memcg->oom_kill_disable) { | |
1968 | mem_cgroup_unmark_under_oom(memcg); | |
1969 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1970 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
1971 | current->memcg_oom.order); | |
1972 | } else { | |
3812c8c8 | 1973 | schedule(); |
49426420 JW |
1974 | mem_cgroup_unmark_under_oom(memcg); |
1975 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1976 | } | |
1977 | ||
1978 | if (locked) { | |
fb2a6fc5 JW |
1979 | mem_cgroup_oom_unlock(memcg); |
1980 | /* | |
1981 | * There is no guarantee that an OOM-lock contender | |
1982 | * sees the wakeups triggered by the OOM kill | |
1983 | * uncharges. Wake any sleepers explicitely. | |
1984 | */ | |
1985 | memcg_oom_recover(memcg); | |
1986 | } | |
49426420 JW |
1987 | cleanup: |
1988 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 1989 | css_put(&memcg->css); |
867578cb | 1990 | return true; |
0b7f569e KH |
1991 | } |
1992 | ||
d7365e78 JW |
1993 | /** |
1994 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1995 | * @page: page that is going to change accounted state | |
1996 | * @locked: &memcg->move_lock slowpath was taken | |
1997 | * @flags: IRQ-state flags for &memcg->move_lock | |
32047e2a | 1998 | * |
d7365e78 JW |
1999 | * This function must mark the beginning of an accounted page state |
2000 | * change to prevent double accounting when the page is concurrently | |
2001 | * being moved to another memcg: | |
32047e2a | 2002 | * |
d7365e78 JW |
2003 | * memcg = mem_cgroup_begin_page_stat(page, &locked, &flags); |
2004 | * if (TestClearPageState(page)) | |
2005 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
2006 | * mem_cgroup_end_page_stat(memcg, locked, flags); | |
32047e2a | 2007 | * |
d7365e78 JW |
2008 | * The RCU lock is held throughout the transaction. The fast path can |
2009 | * get away without acquiring the memcg->move_lock (@locked is false) | |
2010 | * because page moving starts with an RCU grace period. | |
32047e2a | 2011 | * |
d7365e78 JW |
2012 | * The RCU lock also protects the memcg from being freed when the page |
2013 | * state that is going to change is the only thing preventing the page | |
2014 | * from being uncharged. E.g. end-writeback clearing PageWriteback(), | |
2015 | * which allows migration to go ahead and uncharge the page before the | |
2016 | * account transaction might be complete. | |
d69b042f | 2017 | */ |
d7365e78 JW |
2018 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page, |
2019 | bool *locked, | |
2020 | unsigned long *flags) | |
89c06bd5 KH |
2021 | { |
2022 | struct mem_cgroup *memcg; | |
2023 | struct page_cgroup *pc; | |
2024 | ||
d7365e78 JW |
2025 | rcu_read_lock(); |
2026 | ||
2027 | if (mem_cgroup_disabled()) | |
2028 | return NULL; | |
2029 | ||
89c06bd5 KH |
2030 | pc = lookup_page_cgroup(page); |
2031 | again: | |
2032 | memcg = pc->mem_cgroup; | |
29833315 | 2033 | if (unlikely(!memcg)) |
d7365e78 JW |
2034 | return NULL; |
2035 | ||
2036 | *locked = false; | |
bdcbb659 | 2037 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2038 | return memcg; |
89c06bd5 | 2039 | |
354a4783 | 2040 | spin_lock_irqsave(&memcg->move_lock, *flags); |
29833315 | 2041 | if (memcg != pc->mem_cgroup) { |
354a4783 | 2042 | spin_unlock_irqrestore(&memcg->move_lock, *flags); |
89c06bd5 KH |
2043 | goto again; |
2044 | } | |
2045 | *locked = true; | |
d7365e78 JW |
2046 | |
2047 | return memcg; | |
89c06bd5 KH |
2048 | } |
2049 | ||
d7365e78 JW |
2050 | /** |
2051 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2052 | * @memcg: the memcg that was accounted against | |
2053 | * @locked: value received from mem_cgroup_begin_page_stat() | |
2054 | * @flags: value received from mem_cgroup_begin_page_stat() | |
2055 | */ | |
e4bd6a02 MH |
2056 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg, bool *locked, |
2057 | unsigned long *flags) | |
89c06bd5 | 2058 | { |
e4bd6a02 MH |
2059 | if (memcg && *locked) |
2060 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
89c06bd5 | 2061 | |
d7365e78 | 2062 | rcu_read_unlock(); |
89c06bd5 KH |
2063 | } |
2064 | ||
d7365e78 JW |
2065 | /** |
2066 | * mem_cgroup_update_page_stat - update page state statistics | |
2067 | * @memcg: memcg to account against | |
2068 | * @idx: page state item to account | |
2069 | * @val: number of pages (positive or negative) | |
2070 | * | |
2071 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2072 | */ | |
2073 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2074 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2075 | { |
658b72c5 | 2076 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2077 | |
d7365e78 JW |
2078 | if (memcg) |
2079 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2080 | } |
26174efd | 2081 | |
cdec2e42 KH |
2082 | /* |
2083 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2084 | * TODO: maybe necessary to use big numbers in big irons. | |
2085 | */ | |
7ec99d62 | 2086 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2087 | struct memcg_stock_pcp { |
2088 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2089 | unsigned int nr_pages; |
cdec2e42 | 2090 | struct work_struct work; |
26fe6168 | 2091 | unsigned long flags; |
a0db00fc | 2092 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2093 | }; |
2094 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2095 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2096 | |
a0956d54 SS |
2097 | /** |
2098 | * consume_stock: Try to consume stocked charge on this cpu. | |
2099 | * @memcg: memcg to consume from. | |
2100 | * @nr_pages: how many pages to charge. | |
2101 | * | |
2102 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2103 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2104 | * service an allocation will refill the stock. | |
2105 | * | |
2106 | * returns true if successful, false otherwise. | |
cdec2e42 | 2107 | */ |
a0956d54 | 2108 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2109 | { |
2110 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2111 | bool ret = false; |
cdec2e42 | 2112 | |
a0956d54 | 2113 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2114 | return ret; |
a0956d54 | 2115 | |
cdec2e42 | 2116 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2117 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2118 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2119 | ret = true; |
2120 | } | |
cdec2e42 KH |
2121 | put_cpu_var(memcg_stock); |
2122 | return ret; | |
2123 | } | |
2124 | ||
2125 | /* | |
3e32cb2e | 2126 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2127 | */ |
2128 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2129 | { | |
2130 | struct mem_cgroup *old = stock->cached; | |
2131 | ||
11c9ea4e | 2132 | if (stock->nr_pages) { |
3e32cb2e | 2133 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2134 | if (do_swap_account) |
3e32cb2e | 2135 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2136 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2137 | stock->nr_pages = 0; |
cdec2e42 KH |
2138 | } |
2139 | stock->cached = NULL; | |
cdec2e42 KH |
2140 | } |
2141 | ||
2142 | /* | |
2143 | * This must be called under preempt disabled or must be called by | |
2144 | * a thread which is pinned to local cpu. | |
2145 | */ | |
2146 | static void drain_local_stock(struct work_struct *dummy) | |
2147 | { | |
7c8e0181 | 2148 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2149 | drain_stock(stock); |
26fe6168 | 2150 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2151 | } |
2152 | ||
e4777496 MH |
2153 | static void __init memcg_stock_init(void) |
2154 | { | |
2155 | int cpu; | |
2156 | ||
2157 | for_each_possible_cpu(cpu) { | |
2158 | struct memcg_stock_pcp *stock = | |
2159 | &per_cpu(memcg_stock, cpu); | |
2160 | INIT_WORK(&stock->work, drain_local_stock); | |
2161 | } | |
2162 | } | |
2163 | ||
cdec2e42 | 2164 | /* |
3e32cb2e | 2165 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2166 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2167 | */ |
c0ff4b85 | 2168 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2169 | { |
2170 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2171 | ||
c0ff4b85 | 2172 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2173 | drain_stock(stock); |
c0ff4b85 | 2174 | stock->cached = memcg; |
cdec2e42 | 2175 | } |
11c9ea4e | 2176 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2177 | put_cpu_var(memcg_stock); |
2178 | } | |
2179 | ||
2180 | /* | |
c0ff4b85 | 2181 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2182 | * of the hierarchy under it. |
cdec2e42 | 2183 | */ |
6d3d6aa2 | 2184 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2185 | { |
26fe6168 | 2186 | int cpu, curcpu; |
d38144b7 | 2187 | |
6d3d6aa2 JW |
2188 | /* If someone's already draining, avoid adding running more workers. */ |
2189 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2190 | return; | |
cdec2e42 | 2191 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2192 | get_online_cpus(); |
5af12d0e | 2193 | curcpu = get_cpu(); |
cdec2e42 KH |
2194 | for_each_online_cpu(cpu) { |
2195 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2196 | struct mem_cgroup *memcg; |
26fe6168 | 2197 | |
c0ff4b85 R |
2198 | memcg = stock->cached; |
2199 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2200 | continue; |
2314b42d | 2201 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 2202 | continue; |
d1a05b69 MH |
2203 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2204 | if (cpu == curcpu) | |
2205 | drain_local_stock(&stock->work); | |
2206 | else | |
2207 | schedule_work_on(cpu, &stock->work); | |
2208 | } | |
cdec2e42 | 2209 | } |
5af12d0e | 2210 | put_cpu(); |
f894ffa8 | 2211 | put_online_cpus(); |
9f50fad6 | 2212 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2213 | } |
2214 | ||
711d3d2c KH |
2215 | /* |
2216 | * This function drains percpu counter value from DEAD cpu and | |
2217 | * move it to local cpu. Note that this function can be preempted. | |
2218 | */ | |
c0ff4b85 | 2219 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2220 | { |
2221 | int i; | |
2222 | ||
c0ff4b85 | 2223 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2224 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2225 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2226 | |
c0ff4b85 R |
2227 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2228 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2229 | } |
e9f8974f | 2230 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2231 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2232 | |
c0ff4b85 R |
2233 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2234 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2235 | } |
c0ff4b85 | 2236 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2237 | } |
2238 | ||
0db0628d | 2239 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2240 | unsigned long action, |
2241 | void *hcpu) | |
2242 | { | |
2243 | int cpu = (unsigned long)hcpu; | |
2244 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2245 | struct mem_cgroup *iter; |
cdec2e42 | 2246 | |
619d094b | 2247 | if (action == CPU_ONLINE) |
1489ebad | 2248 | return NOTIFY_OK; |
1489ebad | 2249 | |
d833049b | 2250 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2251 | return NOTIFY_OK; |
711d3d2c | 2252 | |
9f3a0d09 | 2253 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2254 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2255 | ||
cdec2e42 KH |
2256 | stock = &per_cpu(memcg_stock, cpu); |
2257 | drain_stock(stock); | |
2258 | return NOTIFY_OK; | |
2259 | } | |
2260 | ||
00501b53 JW |
2261 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2262 | unsigned int nr_pages) | |
8a9f3ccd | 2263 | { |
7ec99d62 | 2264 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2265 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2266 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2267 | struct page_counter *counter; |
6539cc05 | 2268 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2269 | bool may_swap = true; |
2270 | bool drained = false; | |
05b84301 | 2271 | int ret = 0; |
a636b327 | 2272 | |
ce00a967 JW |
2273 | if (mem_cgroup_is_root(memcg)) |
2274 | goto done; | |
6539cc05 | 2275 | retry: |
b6b6cc72 MH |
2276 | if (consume_stock(memcg, nr_pages)) |
2277 | goto done; | |
8a9f3ccd | 2278 | |
3fbe7244 | 2279 | if (!do_swap_account || |
3e32cb2e JW |
2280 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2281 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2282 | goto done_restock; |
3fbe7244 | 2283 | if (do_swap_account) |
3e32cb2e JW |
2284 | page_counter_uncharge(&memcg->memsw, batch); |
2285 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2286 | } else { |
3e32cb2e | 2287 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2288 | may_swap = false; |
3fbe7244 | 2289 | } |
7a81b88c | 2290 | |
6539cc05 JW |
2291 | if (batch > nr_pages) { |
2292 | batch = nr_pages; | |
2293 | goto retry; | |
2294 | } | |
6d61ef40 | 2295 | |
06b078fc JW |
2296 | /* |
2297 | * Unlike in global OOM situations, memcg is not in a physical | |
2298 | * memory shortage. Allow dying and OOM-killed tasks to | |
2299 | * bypass the last charges so that they can exit quickly and | |
2300 | * free their memory. | |
2301 | */ | |
2302 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2303 | fatal_signal_pending(current) || | |
2304 | current->flags & PF_EXITING)) | |
2305 | goto bypass; | |
2306 | ||
2307 | if (unlikely(task_in_memcg_oom(current))) | |
2308 | goto nomem; | |
2309 | ||
6539cc05 JW |
2310 | if (!(gfp_mask & __GFP_WAIT)) |
2311 | goto nomem; | |
4b534334 | 2312 | |
b70a2a21 JW |
2313 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2314 | gfp_mask, may_swap); | |
6539cc05 | 2315 | |
61e02c74 | 2316 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2317 | goto retry; |
28c34c29 | 2318 | |
b70a2a21 | 2319 | if (!drained) { |
6d3d6aa2 | 2320 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2321 | drained = true; |
2322 | goto retry; | |
2323 | } | |
2324 | ||
28c34c29 JW |
2325 | if (gfp_mask & __GFP_NORETRY) |
2326 | goto nomem; | |
6539cc05 JW |
2327 | /* |
2328 | * Even though the limit is exceeded at this point, reclaim | |
2329 | * may have been able to free some pages. Retry the charge | |
2330 | * before killing the task. | |
2331 | * | |
2332 | * Only for regular pages, though: huge pages are rather | |
2333 | * unlikely to succeed so close to the limit, and we fall back | |
2334 | * to regular pages anyway in case of failure. | |
2335 | */ | |
61e02c74 | 2336 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2337 | goto retry; |
2338 | /* | |
2339 | * At task move, charge accounts can be doubly counted. So, it's | |
2340 | * better to wait until the end of task_move if something is going on. | |
2341 | */ | |
2342 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2343 | goto retry; | |
2344 | ||
9b130619 JW |
2345 | if (nr_retries--) |
2346 | goto retry; | |
2347 | ||
06b078fc JW |
2348 | if (gfp_mask & __GFP_NOFAIL) |
2349 | goto bypass; | |
2350 | ||
6539cc05 JW |
2351 | if (fatal_signal_pending(current)) |
2352 | goto bypass; | |
2353 | ||
61e02c74 | 2354 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2355 | nomem: |
6d1fdc48 | 2356 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2357 | return -ENOMEM; |
867578cb | 2358 | bypass: |
ce00a967 | 2359 | return -EINTR; |
6539cc05 JW |
2360 | |
2361 | done_restock: | |
e8ea14cc | 2362 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2363 | if (batch > nr_pages) |
2364 | refill_stock(memcg, batch - nr_pages); | |
2365 | done: | |
05b84301 | 2366 | return ret; |
7a81b88c | 2367 | } |
8a9f3ccd | 2368 | |
00501b53 | 2369 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2370 | { |
ce00a967 JW |
2371 | if (mem_cgroup_is_root(memcg)) |
2372 | return; | |
2373 | ||
3e32cb2e | 2374 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2375 | if (do_swap_account) |
3e32cb2e | 2376 | page_counter_uncharge(&memcg->memsw, nr_pages); |
e8ea14cc JW |
2377 | |
2378 | css_put_many(&memcg->css, nr_pages); | |
d01dd17f KH |
2379 | } |
2380 | ||
a3b2d692 KH |
2381 | /* |
2382 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2383 | * rcu_read_lock(). The caller is responsible for calling |
2384 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2385 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2386 | */ |
2387 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2388 | { | |
a3b2d692 KH |
2389 | /* ID 0 is unused ID */ |
2390 | if (!id) | |
2391 | return NULL; | |
34c00c31 | 2392 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2393 | } |
2394 | ||
0a31bc97 JW |
2395 | /* |
2396 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2397 | * @page: the page | |
2398 | * | |
2399 | * Look up, get a css reference, and return the memcg that owns @page. | |
2400 | * | |
2401 | * The page must be locked to prevent racing with swap-in and page | |
2402 | * cache charges. If coming from an unlocked page table, the caller | |
2403 | * must ensure the page is on the LRU or this can race with charging. | |
2404 | */ | |
e42d9d5d | 2405 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2406 | { |
29833315 | 2407 | struct mem_cgroup *memcg; |
3c776e64 | 2408 | struct page_cgroup *pc; |
a3b2d692 | 2409 | unsigned short id; |
b5a84319 KH |
2410 | swp_entry_t ent; |
2411 | ||
309381fe | 2412 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2413 | |
3c776e64 | 2414 | pc = lookup_page_cgroup(page); |
29833315 JW |
2415 | memcg = pc->mem_cgroup; |
2416 | ||
2417 | if (memcg) { | |
2418 | if (!css_tryget_online(&memcg->css)) | |
c0ff4b85 | 2419 | memcg = NULL; |
e42d9d5d | 2420 | } else if (PageSwapCache(page)) { |
3c776e64 | 2421 | ent.val = page_private(page); |
9fb4b7cc | 2422 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2423 | rcu_read_lock(); |
c0ff4b85 | 2424 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2425 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2426 | memcg = NULL; |
a3b2d692 | 2427 | rcu_read_unlock(); |
3c776e64 | 2428 | } |
c0ff4b85 | 2429 | return memcg; |
b5a84319 KH |
2430 | } |
2431 | ||
0a31bc97 JW |
2432 | static void lock_page_lru(struct page *page, int *isolated) |
2433 | { | |
2434 | struct zone *zone = page_zone(page); | |
2435 | ||
2436 | spin_lock_irq(&zone->lru_lock); | |
2437 | if (PageLRU(page)) { | |
2438 | struct lruvec *lruvec; | |
2439 | ||
2440 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2441 | ClearPageLRU(page); | |
2442 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2443 | *isolated = 1; | |
2444 | } else | |
2445 | *isolated = 0; | |
2446 | } | |
2447 | ||
2448 | static void unlock_page_lru(struct page *page, int isolated) | |
2449 | { | |
2450 | struct zone *zone = page_zone(page); | |
2451 | ||
2452 | if (isolated) { | |
2453 | struct lruvec *lruvec; | |
2454 | ||
2455 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2456 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2457 | SetPageLRU(page); | |
2458 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2459 | } | |
2460 | spin_unlock_irq(&zone->lru_lock); | |
2461 | } | |
2462 | ||
00501b53 | 2463 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2464 | bool lrucare) |
7a81b88c | 2465 | { |
ce587e65 | 2466 | struct page_cgroup *pc = lookup_page_cgroup(page); |
0a31bc97 | 2467 | int isolated; |
9ce70c02 | 2468 | |
29833315 | 2469 | VM_BUG_ON_PAGE(pc->mem_cgroup, page); |
9ce70c02 HD |
2470 | |
2471 | /* | |
2472 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2473 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2474 | */ | |
0a31bc97 JW |
2475 | if (lrucare) |
2476 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2477 | |
0a31bc97 JW |
2478 | /* |
2479 | * Nobody should be changing or seriously looking at | |
29833315 | 2480 | * pc->mem_cgroup at this point: |
0a31bc97 JW |
2481 | * |
2482 | * - the page is uncharged | |
2483 | * | |
2484 | * - the page is off-LRU | |
2485 | * | |
2486 | * - an anonymous fault has exclusive page access, except for | |
2487 | * a locked page table | |
2488 | * | |
2489 | * - a page cache insertion, a swapin fault, or a migration | |
2490 | * have the page locked | |
2491 | */ | |
c0ff4b85 | 2492 | pc->mem_cgroup = memcg; |
9ce70c02 | 2493 | |
0a31bc97 JW |
2494 | if (lrucare) |
2495 | unlock_page_lru(page, isolated); | |
7a81b88c | 2496 | } |
66e1707b | 2497 | |
7ae1e1d0 | 2498 | #ifdef CONFIG_MEMCG_KMEM |
bd673145 VD |
2499 | /* |
2500 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or | |
2501 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. | |
2502 | */ | |
2503 | static DEFINE_MUTEX(memcg_slab_mutex); | |
2504 | ||
1f458cbf GC |
2505 | /* |
2506 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2507 | * in the memcg_cache_params struct. | |
2508 | */ | |
2509 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2510 | { | |
2511 | struct kmem_cache *cachep; | |
2512 | ||
2513 | VM_BUG_ON(p->is_root_cache); | |
2514 | cachep = p->root_cache; | |
7a67d7ab | 2515 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2516 | } |
2517 | ||
3e32cb2e JW |
2518 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2519 | unsigned long nr_pages) | |
7ae1e1d0 | 2520 | { |
3e32cb2e | 2521 | struct page_counter *counter; |
7ae1e1d0 | 2522 | int ret = 0; |
7ae1e1d0 | 2523 | |
3e32cb2e JW |
2524 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2525 | if (ret < 0) | |
7ae1e1d0 GC |
2526 | return ret; |
2527 | ||
3e32cb2e | 2528 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2529 | if (ret == -EINTR) { |
2530 | /* | |
00501b53 JW |
2531 | * try_charge() chose to bypass to root due to OOM kill or |
2532 | * fatal signal. Since our only options are to either fail | |
2533 | * the allocation or charge it to this cgroup, do it as a | |
2534 | * temporary condition. But we can't fail. From a kmem/slab | |
2535 | * perspective, the cache has already been selected, by | |
2536 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2537 | * our minds. |
2538 | * | |
2539 | * This condition will only trigger if the task entered | |
00501b53 JW |
2540 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2541 | * during try_charge() above. Tasks that were already dying | |
2542 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2543 | * directed to the root cgroup in memcontrol.h |
2544 | */ | |
3e32cb2e | 2545 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2546 | if (do_swap_account) |
3e32cb2e | 2547 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2548 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2549 | ret = 0; |
2550 | } else if (ret) | |
3e32cb2e | 2551 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2552 | |
2553 | return ret; | |
2554 | } | |
2555 | ||
3e32cb2e JW |
2556 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, |
2557 | unsigned long nr_pages) | |
7ae1e1d0 | 2558 | { |
3e32cb2e | 2559 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2560 | if (do_swap_account) |
3e32cb2e | 2561 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2562 | |
64f21993 | 2563 | page_counter_uncharge(&memcg->kmem, nr_pages); |
e8ea14cc JW |
2564 | |
2565 | css_put_many(&memcg->css, nr_pages); | |
7ae1e1d0 GC |
2566 | } |
2567 | ||
2633d7a0 GC |
2568 | /* |
2569 | * helper for acessing a memcg's index. It will be used as an index in the | |
2570 | * child cache array in kmem_cache, and also to derive its name. This function | |
2571 | * will return -1 when this is not a kmem-limited memcg. | |
2572 | */ | |
2573 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2574 | { | |
2575 | return memcg ? memcg->kmemcg_id : -1; | |
2576 | } | |
2577 | ||
f3bb3043 | 2578 | static int memcg_alloc_cache_id(void) |
55007d84 | 2579 | { |
f3bb3043 VD |
2580 | int id, size; |
2581 | int err; | |
2582 | ||
2583 | id = ida_simple_get(&kmem_limited_groups, | |
2584 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2585 | if (id < 0) | |
2586 | return id; | |
55007d84 | 2587 | |
f3bb3043 VD |
2588 | if (id < memcg_limited_groups_array_size) |
2589 | return id; | |
2590 | ||
2591 | /* | |
2592 | * There's no space for the new id in memcg_caches arrays, | |
2593 | * so we have to grow them. | |
2594 | */ | |
2595 | ||
2596 | size = 2 * (id + 1); | |
55007d84 GC |
2597 | if (size < MEMCG_CACHES_MIN_SIZE) |
2598 | size = MEMCG_CACHES_MIN_SIZE; | |
2599 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2600 | size = MEMCG_CACHES_MAX_SIZE; | |
2601 | ||
f3bb3043 VD |
2602 | mutex_lock(&memcg_slab_mutex); |
2603 | err = memcg_update_all_caches(size); | |
2604 | mutex_unlock(&memcg_slab_mutex); | |
2605 | ||
2606 | if (err) { | |
2607 | ida_simple_remove(&kmem_limited_groups, id); | |
2608 | return err; | |
2609 | } | |
2610 | return id; | |
2611 | } | |
2612 | ||
2613 | static void memcg_free_cache_id(int id) | |
2614 | { | |
2615 | ida_simple_remove(&kmem_limited_groups, id); | |
55007d84 GC |
2616 | } |
2617 | ||
2618 | /* | |
2619 | * We should update the current array size iff all caches updates succeed. This | |
2620 | * can only be done from the slab side. The slab mutex needs to be held when | |
2621 | * calling this. | |
2622 | */ | |
2623 | void memcg_update_array_size(int num) | |
2624 | { | |
f3bb3043 | 2625 | memcg_limited_groups_array_size = num; |
55007d84 GC |
2626 | } |
2627 | ||
776ed0f0 VD |
2628 | static void memcg_register_cache(struct mem_cgroup *memcg, |
2629 | struct kmem_cache *root_cache) | |
2633d7a0 | 2630 | { |
93f39eea VD |
2631 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by |
2632 | memcg_slab_mutex */ | |
bd673145 | 2633 | struct kmem_cache *cachep; |
d7f25f8a GC |
2634 | int id; |
2635 | ||
bd673145 VD |
2636 | lockdep_assert_held(&memcg_slab_mutex); |
2637 | ||
2638 | id = memcg_cache_id(memcg); | |
2639 | ||
2640 | /* | |
2641 | * Since per-memcg caches are created asynchronously on first | |
2642 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
2643 | * create the same cache, but only one of them may succeed. | |
2644 | */ | |
2645 | if (cache_from_memcg_idx(root_cache, id)) | |
1aa13254 VD |
2646 | return; |
2647 | ||
073ee1c6 | 2648 | cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1); |
776ed0f0 | 2649 | cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf); |
2edefe11 | 2650 | /* |
bd673145 VD |
2651 | * If we could not create a memcg cache, do not complain, because |
2652 | * that's not critical at all as we can always proceed with the root | |
2653 | * cache. | |
2edefe11 | 2654 | */ |
bd673145 VD |
2655 | if (!cachep) |
2656 | return; | |
2edefe11 | 2657 | |
33a690c4 | 2658 | css_get(&memcg->css); |
bd673145 | 2659 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
1aa13254 | 2660 | |
d7f25f8a | 2661 | /* |
959c8963 VD |
2662 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
2663 | * barrier here to ensure nobody will see the kmem_cache partially | |
2664 | * initialized. | |
d7f25f8a | 2665 | */ |
959c8963 VD |
2666 | smp_wmb(); |
2667 | ||
bd673145 VD |
2668 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
2669 | root_cache->memcg_params->memcg_caches[id] = cachep; | |
1aa13254 | 2670 | } |
d7f25f8a | 2671 | |
776ed0f0 | 2672 | static void memcg_unregister_cache(struct kmem_cache *cachep) |
1aa13254 | 2673 | { |
bd673145 | 2674 | struct kmem_cache *root_cache; |
1aa13254 VD |
2675 | struct mem_cgroup *memcg; |
2676 | int id; | |
2677 | ||
bd673145 | 2678 | lockdep_assert_held(&memcg_slab_mutex); |
d7f25f8a | 2679 | |
bd673145 | 2680 | BUG_ON(is_root_cache(cachep)); |
2edefe11 | 2681 | |
bd673145 VD |
2682 | root_cache = cachep->memcg_params->root_cache; |
2683 | memcg = cachep->memcg_params->memcg; | |
96403da2 | 2684 | id = memcg_cache_id(memcg); |
d7f25f8a | 2685 | |
bd673145 VD |
2686 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
2687 | root_cache->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a | 2688 | |
bd673145 VD |
2689 | list_del(&cachep->memcg_params->list); |
2690 | ||
2691 | kmem_cache_destroy(cachep); | |
33a690c4 VD |
2692 | |
2693 | /* drop the reference taken in memcg_register_cache */ | |
2694 | css_put(&memcg->css); | |
2633d7a0 GC |
2695 | } |
2696 | ||
0e9d92f2 GC |
2697 | /* |
2698 | * During the creation a new cache, we need to disable our accounting mechanism | |
2699 | * altogether. This is true even if we are not creating, but rather just | |
2700 | * enqueing new caches to be created. | |
2701 | * | |
2702 | * This is because that process will trigger allocations; some visible, like | |
2703 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
2704 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
2705 | * objects during debug. | |
2706 | * | |
2707 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
2708 | * to it. This may not be a bounded recursion: since the first cache creation | |
2709 | * failed to complete (waiting on the allocation), we'll just try to create the | |
2710 | * cache again, failing at the same point. | |
2711 | * | |
2712 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
2713 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
2714 | * inside the following two functions. | |
2715 | */ | |
2716 | static inline void memcg_stop_kmem_account(void) | |
2717 | { | |
2718 | VM_BUG_ON(!current->mm); | |
2719 | current->memcg_kmem_skip_account++; | |
2720 | } | |
2721 | ||
2722 | static inline void memcg_resume_kmem_account(void) | |
2723 | { | |
2724 | VM_BUG_ON(!current->mm); | |
2725 | current->memcg_kmem_skip_account--; | |
2726 | } | |
2727 | ||
776ed0f0 | 2728 | int __memcg_cleanup_cache_params(struct kmem_cache *s) |
7cf27982 GC |
2729 | { |
2730 | struct kmem_cache *c; | |
b8529907 | 2731 | int i, failed = 0; |
7cf27982 | 2732 | |
bd673145 | 2733 | mutex_lock(&memcg_slab_mutex); |
7a67d7ab QH |
2734 | for_each_memcg_cache_index(i) { |
2735 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
2736 | if (!c) |
2737 | continue; | |
2738 | ||
776ed0f0 | 2739 | memcg_unregister_cache(c); |
b8529907 VD |
2740 | |
2741 | if (cache_from_memcg_idx(s, i)) | |
2742 | failed++; | |
7cf27982 | 2743 | } |
bd673145 | 2744 | mutex_unlock(&memcg_slab_mutex); |
b8529907 | 2745 | return failed; |
7cf27982 GC |
2746 | } |
2747 | ||
776ed0f0 | 2748 | static void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
2749 | { |
2750 | struct kmem_cache *cachep; | |
bd673145 | 2751 | struct memcg_cache_params *params, *tmp; |
1f458cbf GC |
2752 | |
2753 | if (!memcg_kmem_is_active(memcg)) | |
2754 | return; | |
2755 | ||
bd673145 VD |
2756 | mutex_lock(&memcg_slab_mutex); |
2757 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { | |
1f458cbf | 2758 | cachep = memcg_params_to_cache(params); |
bd673145 VD |
2759 | kmem_cache_shrink(cachep); |
2760 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
776ed0f0 | 2761 | memcg_unregister_cache(cachep); |
1f458cbf | 2762 | } |
bd673145 | 2763 | mutex_unlock(&memcg_slab_mutex); |
1f458cbf GC |
2764 | } |
2765 | ||
776ed0f0 | 2766 | struct memcg_register_cache_work { |
5722d094 VD |
2767 | struct mem_cgroup *memcg; |
2768 | struct kmem_cache *cachep; | |
2769 | struct work_struct work; | |
2770 | }; | |
2771 | ||
776ed0f0 | 2772 | static void memcg_register_cache_func(struct work_struct *w) |
d7f25f8a | 2773 | { |
776ed0f0 VD |
2774 | struct memcg_register_cache_work *cw = |
2775 | container_of(w, struct memcg_register_cache_work, work); | |
5722d094 VD |
2776 | struct mem_cgroup *memcg = cw->memcg; |
2777 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2778 | |
bd673145 | 2779 | mutex_lock(&memcg_slab_mutex); |
776ed0f0 | 2780 | memcg_register_cache(memcg, cachep); |
bd673145 VD |
2781 | mutex_unlock(&memcg_slab_mutex); |
2782 | ||
5722d094 | 2783 | css_put(&memcg->css); |
d7f25f8a GC |
2784 | kfree(cw); |
2785 | } | |
2786 | ||
2787 | /* | |
2788 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2789 | */ |
776ed0f0 VD |
2790 | static void __memcg_schedule_register_cache(struct mem_cgroup *memcg, |
2791 | struct kmem_cache *cachep) | |
d7f25f8a | 2792 | { |
776ed0f0 | 2793 | struct memcg_register_cache_work *cw; |
d7f25f8a | 2794 | |
776ed0f0 | 2795 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
ca0dde97 LZ |
2796 | if (cw == NULL) { |
2797 | css_put(&memcg->css); | |
d7f25f8a GC |
2798 | return; |
2799 | } | |
2800 | ||
2801 | cw->memcg = memcg; | |
2802 | cw->cachep = cachep; | |
2803 | ||
776ed0f0 | 2804 | INIT_WORK(&cw->work, memcg_register_cache_func); |
d7f25f8a GC |
2805 | schedule_work(&cw->work); |
2806 | } | |
2807 | ||
776ed0f0 VD |
2808 | static void memcg_schedule_register_cache(struct mem_cgroup *memcg, |
2809 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2810 | { |
2811 | /* | |
2812 | * We need to stop accounting when we kmalloc, because if the | |
2813 | * corresponding kmalloc cache is not yet created, the first allocation | |
776ed0f0 | 2814 | * in __memcg_schedule_register_cache will recurse. |
0e9d92f2 GC |
2815 | * |
2816 | * However, it is better to enclose the whole function. Depending on | |
2817 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2818 | * trigger an allocation. This too, will make us recurse. Because at | |
2819 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2820 | * the safest choice is to do it like this, wrapping the whole function. | |
2821 | */ | |
2822 | memcg_stop_kmem_account(); | |
776ed0f0 | 2823 | __memcg_schedule_register_cache(memcg, cachep); |
0e9d92f2 GC |
2824 | memcg_resume_kmem_account(); |
2825 | } | |
c67a8a68 VD |
2826 | |
2827 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) | |
2828 | { | |
3e32cb2e | 2829 | unsigned int nr_pages = 1 << order; |
c67a8a68 VD |
2830 | int res; |
2831 | ||
3e32cb2e | 2832 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, nr_pages); |
c67a8a68 | 2833 | if (!res) |
3e32cb2e | 2834 | atomic_add(nr_pages, &cachep->memcg_params->nr_pages); |
c67a8a68 VD |
2835 | return res; |
2836 | } | |
2837 | ||
2838 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) | |
2839 | { | |
3e32cb2e JW |
2840 | unsigned int nr_pages = 1 << order; |
2841 | ||
2842 | memcg_uncharge_kmem(cachep->memcg_params->memcg, nr_pages); | |
2843 | atomic_sub(nr_pages, &cachep->memcg_params->nr_pages); | |
c67a8a68 VD |
2844 | } |
2845 | ||
d7f25f8a GC |
2846 | /* |
2847 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2848 | * We try to use the current memcg's version of the cache. | |
2849 | * | |
2850 | * If the cache does not exist yet, if we are the first user of it, | |
2851 | * we either create it immediately, if possible, or create it asynchronously | |
2852 | * in a workqueue. | |
2853 | * In the latter case, we will let the current allocation go through with | |
2854 | * the original cache. | |
2855 | * | |
2856 | * Can't be called in interrupt context or from kernel threads. | |
2857 | * This function needs to be called with rcu_read_lock() held. | |
2858 | */ | |
2859 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
2860 | gfp_t gfp) | |
2861 | { | |
2862 | struct mem_cgroup *memcg; | |
959c8963 | 2863 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
2864 | |
2865 | VM_BUG_ON(!cachep->memcg_params); | |
2866 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
2867 | ||
0e9d92f2 GC |
2868 | if (!current->mm || current->memcg_kmem_skip_account) |
2869 | return cachep; | |
2870 | ||
d7f25f8a GC |
2871 | rcu_read_lock(); |
2872 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a | 2873 | |
cf2b8fbf | 2874 | if (!memcg_kmem_is_active(memcg)) |
ca0dde97 | 2875 | goto out; |
d7f25f8a | 2876 | |
959c8963 VD |
2877 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
2878 | if (likely(memcg_cachep)) { | |
2879 | cachep = memcg_cachep; | |
ca0dde97 | 2880 | goto out; |
d7f25f8a GC |
2881 | } |
2882 | ||
ca0dde97 | 2883 | /* The corresponding put will be done in the workqueue. */ |
ec903c0c | 2884 | if (!css_tryget_online(&memcg->css)) |
ca0dde97 LZ |
2885 | goto out; |
2886 | rcu_read_unlock(); | |
2887 | ||
2888 | /* | |
2889 | * If we are in a safe context (can wait, and not in interrupt | |
2890 | * context), we could be be predictable and return right away. | |
2891 | * This would guarantee that the allocation being performed | |
2892 | * already belongs in the new cache. | |
2893 | * | |
2894 | * However, there are some clashes that can arrive from locking. | |
2895 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2896 | * memcg_create_kmem_cache, this means no further allocation |
2897 | * could happen with the slab_mutex held. So it's better to | |
2898 | * defer everything. | |
ca0dde97 | 2899 | */ |
776ed0f0 | 2900 | memcg_schedule_register_cache(memcg, cachep); |
ca0dde97 LZ |
2901 | return cachep; |
2902 | out: | |
2903 | rcu_read_unlock(); | |
2904 | return cachep; | |
d7f25f8a | 2905 | } |
d7f25f8a | 2906 | |
7ae1e1d0 GC |
2907 | /* |
2908 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2909 | * possible for the given order. But the page is not allocated yet, so we'll | |
2910 | * need a further commit step to do the final arrangements. | |
2911 | * | |
2912 | * It is possible for the task to switch cgroups in this mean time, so at | |
2913 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2914 | * the handle argument to return to the caller which cgroup we should commit | |
2915 | * against. We could also return the memcg directly and avoid the pointer | |
2916 | * passing, but a boolean return value gives better semantics considering | |
2917 | * the compiled-out case as well. | |
2918 | * | |
2919 | * Returning true means the allocation is possible. | |
2920 | */ | |
2921 | bool | |
2922 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2923 | { | |
2924 | struct mem_cgroup *memcg; | |
2925 | int ret; | |
2926 | ||
2927 | *_memcg = NULL; | |
6d42c232 GC |
2928 | |
2929 | /* | |
2930 | * Disabling accounting is only relevant for some specific memcg | |
2931 | * internal allocations. Therefore we would initially not have such | |
52383431 VD |
2932 | * check here, since direct calls to the page allocator that are |
2933 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen | |
2934 | * outside memcg core. We are mostly concerned with cache allocations, | |
2935 | * and by having this test at memcg_kmem_get_cache, we are already able | |
2936 | * to relay the allocation to the root cache and bypass the memcg cache | |
2937 | * altogether. | |
6d42c232 GC |
2938 | * |
2939 | * There is one exception, though: the SLUB allocator does not create | |
2940 | * large order caches, but rather service large kmallocs directly from | |
2941 | * the page allocator. Therefore, the following sequence when backed by | |
2942 | * the SLUB allocator: | |
2943 | * | |
f894ffa8 AM |
2944 | * memcg_stop_kmem_account(); |
2945 | * kmalloc(<large_number>) | |
2946 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
2947 | * |
2948 | * would effectively ignore the fact that we should skip accounting, | |
2949 | * since it will drive us directly to this function without passing | |
2950 | * through the cache selector memcg_kmem_get_cache. Such large | |
2951 | * allocations are extremely rare but can happen, for instance, for the | |
2952 | * cache arrays. We bring this test here. | |
2953 | */ | |
2954 | if (!current->mm || current->memcg_kmem_skip_account) | |
2955 | return true; | |
2956 | ||
df381975 | 2957 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 2958 | |
cf2b8fbf | 2959 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
2960 | css_put(&memcg->css); |
2961 | return true; | |
2962 | } | |
2963 | ||
3e32cb2e | 2964 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
2965 | if (!ret) |
2966 | *_memcg = memcg; | |
7ae1e1d0 GC |
2967 | |
2968 | css_put(&memcg->css); | |
2969 | return (ret == 0); | |
2970 | } | |
2971 | ||
2972 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
2973 | int order) | |
2974 | { | |
2975 | struct page_cgroup *pc; | |
2976 | ||
2977 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
2978 | ||
2979 | /* The page allocation failed. Revert */ | |
2980 | if (!page) { | |
3e32cb2e | 2981 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
2982 | return; |
2983 | } | |
7ae1e1d0 | 2984 | pc = lookup_page_cgroup(page); |
7ae1e1d0 | 2985 | pc->mem_cgroup = memcg; |
7ae1e1d0 GC |
2986 | } |
2987 | ||
2988 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
2989 | { | |
29833315 JW |
2990 | struct page_cgroup *pc = lookup_page_cgroup(page); |
2991 | struct mem_cgroup *memcg = pc->mem_cgroup; | |
7ae1e1d0 | 2992 | |
7ae1e1d0 GC |
2993 | if (!memcg) |
2994 | return; | |
2995 | ||
309381fe | 2996 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2997 | |
3e32cb2e | 2998 | memcg_uncharge_kmem(memcg, 1 << order); |
29833315 | 2999 | pc->mem_cgroup = NULL; |
7ae1e1d0 | 3000 | } |
1f458cbf | 3001 | #else |
776ed0f0 | 3002 | static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3003 | { |
3004 | } | |
7ae1e1d0 GC |
3005 | #endif /* CONFIG_MEMCG_KMEM */ |
3006 | ||
ca3e0214 KH |
3007 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3008 | ||
ca3e0214 KH |
3009 | /* |
3010 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3011 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3012 | * charge/uncharge will be never happen and move_account() is done under | |
3013 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3014 | */ |
e94c8a9c | 3015 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3016 | { |
29833315 | 3017 | struct page_cgroup *pc = lookup_page_cgroup(head); |
e94c8a9c | 3018 | int i; |
ca3e0214 | 3019 | |
3d37c4a9 KH |
3020 | if (mem_cgroup_disabled()) |
3021 | return; | |
b070e65c | 3022 | |
29833315 JW |
3023 | for (i = 1; i < HPAGE_PMD_NR; i++) |
3024 | pc[i].mem_cgroup = pc[0].mem_cgroup; | |
b9982f8d | 3025 | |
29833315 | 3026 | __this_cpu_sub(pc[0].mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 3027 | HPAGE_PMD_NR); |
ca3e0214 | 3028 | } |
12d27107 | 3029 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3030 | |
f817ed48 | 3031 | /** |
de3638d9 | 3032 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3033 | * @page: the page |
7ec99d62 | 3034 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3035 | * @pc: page_cgroup of the page. |
3036 | * @from: mem_cgroup which the page is moved from. | |
3037 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3038 | * | |
3039 | * The caller must confirm following. | |
08e552c6 | 3040 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3041 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3042 | * |
2f3479b1 KH |
3043 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3044 | * from old cgroup. | |
f817ed48 | 3045 | */ |
7ec99d62 JW |
3046 | static int mem_cgroup_move_account(struct page *page, |
3047 | unsigned int nr_pages, | |
3048 | struct page_cgroup *pc, | |
3049 | struct mem_cgroup *from, | |
2f3479b1 | 3050 | struct mem_cgroup *to) |
f817ed48 | 3051 | { |
de3638d9 JW |
3052 | unsigned long flags; |
3053 | int ret; | |
987eba66 | 3054 | |
f817ed48 | 3055 | VM_BUG_ON(from == to); |
309381fe | 3056 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3057 | /* |
3058 | * The page is isolated from LRU. So, collapse function | |
3059 | * will not handle this page. But page splitting can happen. | |
3060 | * Do this check under compound_page_lock(). The caller should | |
3061 | * hold it. | |
3062 | */ | |
3063 | ret = -EBUSY; | |
7ec99d62 | 3064 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3065 | goto out; |
3066 | ||
0a31bc97 JW |
3067 | /* |
3068 | * Prevent mem_cgroup_migrate() from looking at pc->mem_cgroup | |
3069 | * of its source page while we change it: page migration takes | |
3070 | * both pages off the LRU, but page cache replacement doesn't. | |
3071 | */ | |
3072 | if (!trylock_page(page)) | |
3073 | goto out; | |
de3638d9 JW |
3074 | |
3075 | ret = -EINVAL; | |
29833315 | 3076 | if (pc->mem_cgroup != from) |
0a31bc97 | 3077 | goto out_unlock; |
de3638d9 | 3078 | |
354a4783 | 3079 | spin_lock_irqsave(&from->move_lock, flags); |
f817ed48 | 3080 | |
0a31bc97 | 3081 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
3082 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
3083 | nr_pages); | |
3084 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3085 | nr_pages); | |
3086 | } | |
3ea67d06 | 3087 | |
59d1d256 JW |
3088 | if (PageWriteback(page)) { |
3089 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3090 | nr_pages); | |
3091 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3092 | nr_pages); | |
3093 | } | |
3ea67d06 | 3094 | |
0a31bc97 JW |
3095 | /* |
3096 | * It is safe to change pc->mem_cgroup here because the page | |
3097 | * is referenced, charged, and isolated - we can't race with | |
3098 | * uncharging, charging, migration, or LRU putback. | |
3099 | */ | |
d69b042f | 3100 | |
854ffa8d | 3101 | /* caller should have done css_get */ |
08e552c6 | 3102 | pc->mem_cgroup = to; |
354a4783 JW |
3103 | spin_unlock_irqrestore(&from->move_lock, flags); |
3104 | ||
de3638d9 | 3105 | ret = 0; |
0a31bc97 JW |
3106 | |
3107 | local_irq_disable(); | |
3108 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 3109 | memcg_check_events(to, page); |
0a31bc97 | 3110 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 3111 | memcg_check_events(from, page); |
0a31bc97 JW |
3112 | local_irq_enable(); |
3113 | out_unlock: | |
3114 | unlock_page(page); | |
de3638d9 | 3115 | out: |
f817ed48 KH |
3116 | return ret; |
3117 | } | |
3118 | ||
c255a458 | 3119 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
3120 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
3121 | bool charge) | |
d13d1443 | 3122 | { |
0a31bc97 JW |
3123 | int val = (charge) ? 1 : -1; |
3124 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 3125 | } |
02491447 DN |
3126 | |
3127 | /** | |
3128 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3129 | * @entry: swap entry to be moved | |
3130 | * @from: mem_cgroup which the entry is moved from | |
3131 | * @to: mem_cgroup which the entry is moved to | |
3132 | * | |
3133 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3134 | * as the mem_cgroup's id of @from. | |
3135 | * | |
3136 | * Returns 0 on success, -EINVAL on failure. | |
3137 | * | |
3e32cb2e | 3138 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3139 | * both res and memsw, and called css_get(). |
3140 | */ | |
3141 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3142 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3143 | { |
3144 | unsigned short old_id, new_id; | |
3145 | ||
34c00c31 LZ |
3146 | old_id = mem_cgroup_id(from); |
3147 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3148 | |
3149 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3150 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3151 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3152 | /* |
483c30b5 | 3153 | * This function is only called from task migration context now. |
3e32cb2e | 3154 | * It postpones page_counter and refcount handling till the end |
483c30b5 | 3155 | * of task migration(mem_cgroup_clear_mc()) for performance |
4050377b LZ |
3156 | * improvement. But we cannot postpone css_get(to) because if |
3157 | * the process that has been moved to @to does swap-in, the | |
3158 | * refcount of @to might be decreased to 0. | |
3159 | * | |
3160 | * We are in attach() phase, so the cgroup is guaranteed to be | |
3161 | * alive, so we can just call css_get(). | |
02491447 | 3162 | */ |
4050377b | 3163 | css_get(&to->css); |
02491447 DN |
3164 | return 0; |
3165 | } | |
3166 | return -EINVAL; | |
3167 | } | |
3168 | #else | |
3169 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3170 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3171 | { |
3172 | return -EINVAL; | |
3173 | } | |
8c7c6e34 | 3174 | #endif |
d13d1443 | 3175 | |
f212ad7c DN |
3176 | #ifdef CONFIG_DEBUG_VM |
3177 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3178 | { | |
3179 | struct page_cgroup *pc; | |
3180 | ||
3181 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3182 | /* |
3183 | * Can be NULL while feeding pages into the page allocator for | |
3184 | * the first time, i.e. during boot or memory hotplug; | |
3185 | * or when mem_cgroup_disabled(). | |
3186 | */ | |
29833315 | 3187 | if (likely(pc) && pc->mem_cgroup) |
f212ad7c DN |
3188 | return pc; |
3189 | return NULL; | |
3190 | } | |
3191 | ||
3192 | bool mem_cgroup_bad_page_check(struct page *page) | |
3193 | { | |
3194 | if (mem_cgroup_disabled()) | |
3195 | return false; | |
3196 | ||
3197 | return lookup_page_cgroup_used(page) != NULL; | |
3198 | } | |
3199 | ||
3200 | void mem_cgroup_print_bad_page(struct page *page) | |
3201 | { | |
3202 | struct page_cgroup *pc; | |
3203 | ||
3204 | pc = lookup_page_cgroup_used(page); | |
29833315 JW |
3205 | if (pc) |
3206 | pr_alert("pc:%p pc->mem_cgroup:%p\n", pc, pc->mem_cgroup); | |
f212ad7c DN |
3207 | } |
3208 | #endif | |
3209 | ||
3e32cb2e JW |
3210 | static DEFINE_MUTEX(memcg_limit_mutex); |
3211 | ||
d38d2a75 | 3212 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3213 | unsigned long limit) |
628f4235 | 3214 | { |
3e32cb2e JW |
3215 | unsigned long curusage; |
3216 | unsigned long oldusage; | |
3217 | bool enlarge = false; | |
81d39c20 | 3218 | int retry_count; |
3e32cb2e | 3219 | int ret; |
81d39c20 KH |
3220 | |
3221 | /* | |
3222 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3223 | * is depends on callers. We set our retry-count to be function | |
3224 | * of # of children which we should visit in this loop. | |
3225 | */ | |
3e32cb2e JW |
3226 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3227 | mem_cgroup_count_children(memcg); | |
81d39c20 | 3228 | |
3e32cb2e | 3229 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 3230 | |
3e32cb2e | 3231 | do { |
628f4235 KH |
3232 | if (signal_pending(current)) { |
3233 | ret = -EINTR; | |
3234 | break; | |
3235 | } | |
3e32cb2e JW |
3236 | |
3237 | mutex_lock(&memcg_limit_mutex); | |
3238 | if (limit > memcg->memsw.limit) { | |
3239 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3240 | ret = -EINVAL; |
628f4235 KH |
3241 | break; |
3242 | } | |
3e32cb2e JW |
3243 | if (limit > memcg->memory.limit) |
3244 | enlarge = true; | |
3245 | ret = page_counter_limit(&memcg->memory, limit); | |
3246 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3247 | |
3248 | if (!ret) | |
3249 | break; | |
3250 | ||
b70a2a21 JW |
3251 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
3252 | ||
3e32cb2e | 3253 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 3254 | /* Usage is reduced ? */ |
f894ffa8 | 3255 | if (curusage >= oldusage) |
81d39c20 KH |
3256 | retry_count--; |
3257 | else | |
3258 | oldusage = curusage; | |
3e32cb2e JW |
3259 | } while (retry_count); |
3260 | ||
3c11ecf4 KH |
3261 | if (!ret && enlarge) |
3262 | memcg_oom_recover(memcg); | |
14797e23 | 3263 | |
8c7c6e34 KH |
3264 | return ret; |
3265 | } | |
3266 | ||
338c8431 | 3267 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3268 | unsigned long limit) |
8c7c6e34 | 3269 | { |
3e32cb2e JW |
3270 | unsigned long curusage; |
3271 | unsigned long oldusage; | |
3272 | bool enlarge = false; | |
81d39c20 | 3273 | int retry_count; |
3e32cb2e | 3274 | int ret; |
8c7c6e34 | 3275 | |
81d39c20 | 3276 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
3277 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3278 | mem_cgroup_count_children(memcg); | |
3279 | ||
3280 | oldusage = page_counter_read(&memcg->memsw); | |
3281 | ||
3282 | do { | |
8c7c6e34 KH |
3283 | if (signal_pending(current)) { |
3284 | ret = -EINTR; | |
3285 | break; | |
3286 | } | |
3e32cb2e JW |
3287 | |
3288 | mutex_lock(&memcg_limit_mutex); | |
3289 | if (limit < memcg->memory.limit) { | |
3290 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3291 | ret = -EINVAL; |
8c7c6e34 KH |
3292 | break; |
3293 | } | |
3e32cb2e JW |
3294 | if (limit > memcg->memsw.limit) |
3295 | enlarge = true; | |
3296 | ret = page_counter_limit(&memcg->memsw, limit); | |
3297 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3298 | |
3299 | if (!ret) | |
3300 | break; | |
3301 | ||
b70a2a21 JW |
3302 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3303 | ||
3e32cb2e | 3304 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3305 | /* Usage is reduced ? */ |
8c7c6e34 | 3306 | if (curusage >= oldusage) |
628f4235 | 3307 | retry_count--; |
81d39c20 KH |
3308 | else |
3309 | oldusage = curusage; | |
3e32cb2e JW |
3310 | } while (retry_count); |
3311 | ||
3c11ecf4 KH |
3312 | if (!ret && enlarge) |
3313 | memcg_oom_recover(memcg); | |
3e32cb2e | 3314 | |
628f4235 KH |
3315 | return ret; |
3316 | } | |
3317 | ||
0608f43d AM |
3318 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3319 | gfp_t gfp_mask, | |
3320 | unsigned long *total_scanned) | |
3321 | { | |
3322 | unsigned long nr_reclaimed = 0; | |
3323 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3324 | unsigned long reclaimed; | |
3325 | int loop = 0; | |
3326 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3327 | unsigned long excess; |
0608f43d AM |
3328 | unsigned long nr_scanned; |
3329 | ||
3330 | if (order > 0) | |
3331 | return 0; | |
3332 | ||
3333 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3334 | /* | |
3335 | * This loop can run a while, specially if mem_cgroup's continuously | |
3336 | * keep exceeding their soft limit and putting the system under | |
3337 | * pressure | |
3338 | */ | |
3339 | do { | |
3340 | if (next_mz) | |
3341 | mz = next_mz; | |
3342 | else | |
3343 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3344 | if (!mz) | |
3345 | break; | |
3346 | ||
3347 | nr_scanned = 0; | |
3348 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3349 | gfp_mask, &nr_scanned); | |
3350 | nr_reclaimed += reclaimed; | |
3351 | *total_scanned += nr_scanned; | |
0a31bc97 | 3352 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3353 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3354 | |
3355 | /* | |
3356 | * If we failed to reclaim anything from this memory cgroup | |
3357 | * it is time to move on to the next cgroup | |
3358 | */ | |
3359 | next_mz = NULL; | |
bc2f2e7f VD |
3360 | if (!reclaimed) |
3361 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3362 | ||
3e32cb2e | 3363 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3364 | /* |
3365 | * One school of thought says that we should not add | |
3366 | * back the node to the tree if reclaim returns 0. | |
3367 | * But our reclaim could return 0, simply because due | |
3368 | * to priority we are exposing a smaller subset of | |
3369 | * memory to reclaim from. Consider this as a longer | |
3370 | * term TODO. | |
3371 | */ | |
3372 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3373 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3374 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3375 | css_put(&mz->memcg->css); |
3376 | loop++; | |
3377 | /* | |
3378 | * Could not reclaim anything and there are no more | |
3379 | * mem cgroups to try or we seem to be looping without | |
3380 | * reclaiming anything. | |
3381 | */ | |
3382 | if (!nr_reclaimed && | |
3383 | (next_mz == NULL || | |
3384 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3385 | break; | |
3386 | } while (!nr_reclaimed); | |
3387 | if (next_mz) | |
3388 | css_put(&next_mz->memcg->css); | |
3389 | return nr_reclaimed; | |
3390 | } | |
3391 | ||
ea280e7b TH |
3392 | /* |
3393 | * Test whether @memcg has children, dead or alive. Note that this | |
3394 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3395 | * returns %true if there are child csses according to the cgroup | |
3396 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3397 | */ | |
b5f99b53 GC |
3398 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3399 | { | |
ea280e7b TH |
3400 | bool ret; |
3401 | ||
696ac172 | 3402 | /* |
ea280e7b TH |
3403 | * The lock does not prevent addition or deletion of children, but |
3404 | * it prevents a new child from being initialized based on this | |
3405 | * parent in css_online(), so it's enough to decide whether | |
3406 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3407 | */ |
ea280e7b TH |
3408 | lockdep_assert_held(&memcg_create_mutex); |
3409 | ||
3410 | rcu_read_lock(); | |
3411 | ret = css_next_child(NULL, &memcg->css); | |
3412 | rcu_read_unlock(); | |
3413 | return ret; | |
b5f99b53 GC |
3414 | } |
3415 | ||
c26251f9 MH |
3416 | /* |
3417 | * Reclaims as many pages from the given memcg as possible and moves | |
3418 | * the rest to the parent. | |
3419 | * | |
3420 | * Caller is responsible for holding css reference for memcg. | |
3421 | */ | |
3422 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3423 | { | |
3424 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3425 | |
c1e862c1 KH |
3426 | /* we call try-to-free pages for make this cgroup empty */ |
3427 | lru_add_drain_all(); | |
f817ed48 | 3428 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3429 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3430 | int progress; |
c1e862c1 | 3431 | |
c26251f9 MH |
3432 | if (signal_pending(current)) |
3433 | return -EINTR; | |
3434 | ||
b70a2a21 JW |
3435 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3436 | GFP_KERNEL, true); | |
c1e862c1 | 3437 | if (!progress) { |
f817ed48 | 3438 | nr_retries--; |
c1e862c1 | 3439 | /* maybe some writeback is necessary */ |
8aa7e847 | 3440 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3441 | } |
f817ed48 KH |
3442 | |
3443 | } | |
ab5196c2 MH |
3444 | |
3445 | return 0; | |
cc847582 KH |
3446 | } |
3447 | ||
6770c64e TH |
3448 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3449 | char *buf, size_t nbytes, | |
3450 | loff_t off) | |
c1e862c1 | 3451 | { |
6770c64e | 3452 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3453 | |
d8423011 MH |
3454 | if (mem_cgroup_is_root(memcg)) |
3455 | return -EINVAL; | |
6770c64e | 3456 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3457 | } |
3458 | ||
182446d0 TH |
3459 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3460 | struct cftype *cft) | |
18f59ea7 | 3461 | { |
182446d0 | 3462 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3463 | } |
3464 | ||
182446d0 TH |
3465 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3466 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3467 | { |
3468 | int retval = 0; | |
182446d0 | 3469 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3470 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3471 | |
0999821b | 3472 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3473 | |
3474 | if (memcg->use_hierarchy == val) | |
3475 | goto out; | |
3476 | ||
18f59ea7 | 3477 | /* |
af901ca1 | 3478 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3479 | * in the child subtrees. If it is unset, then the change can |
3480 | * occur, provided the current cgroup has no children. | |
3481 | * | |
3482 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3483 | * set if there are no children. | |
3484 | */ | |
c0ff4b85 | 3485 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3486 | (val == 1 || val == 0)) { |
ea280e7b | 3487 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3488 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3489 | else |
3490 | retval = -EBUSY; | |
3491 | } else | |
3492 | retval = -EINVAL; | |
567fb435 GC |
3493 | |
3494 | out: | |
0999821b | 3495 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3496 | |
3497 | return retval; | |
3498 | } | |
3499 | ||
3e32cb2e JW |
3500 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3501 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3502 | { |
3503 | struct mem_cgroup *iter; | |
3504 | long val = 0; | |
3505 | ||
3506 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3507 | for_each_mem_cgroup_tree(iter, memcg) | |
3508 | val += mem_cgroup_read_stat(iter, idx); | |
3509 | ||
3510 | if (val < 0) /* race ? */ | |
3511 | val = 0; | |
3512 | return val; | |
3513 | } | |
3514 | ||
3515 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3516 | { | |
3517 | u64 val; | |
3518 | ||
3e32cb2e JW |
3519 | if (mem_cgroup_is_root(memcg)) { |
3520 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3521 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3522 | if (swap) | |
3523 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3524 | } else { | |
ce00a967 | 3525 | if (!swap) |
3e32cb2e | 3526 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3527 | else |
3e32cb2e | 3528 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3529 | } |
ce00a967 JW |
3530 | return val << PAGE_SHIFT; |
3531 | } | |
3532 | ||
3e32cb2e JW |
3533 | enum { |
3534 | RES_USAGE, | |
3535 | RES_LIMIT, | |
3536 | RES_MAX_USAGE, | |
3537 | RES_FAILCNT, | |
3538 | RES_SOFT_LIMIT, | |
3539 | }; | |
ce00a967 | 3540 | |
791badbd | 3541 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3542 | struct cftype *cft) |
8cdea7c0 | 3543 | { |
182446d0 | 3544 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3545 | struct page_counter *counter; |
af36f906 | 3546 | |
3e32cb2e | 3547 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3548 | case _MEM: |
3e32cb2e JW |
3549 | counter = &memcg->memory; |
3550 | break; | |
8c7c6e34 | 3551 | case _MEMSWAP: |
3e32cb2e JW |
3552 | counter = &memcg->memsw; |
3553 | break; | |
510fc4e1 | 3554 | case _KMEM: |
3e32cb2e | 3555 | counter = &memcg->kmem; |
510fc4e1 | 3556 | break; |
8c7c6e34 KH |
3557 | default: |
3558 | BUG(); | |
8c7c6e34 | 3559 | } |
3e32cb2e JW |
3560 | |
3561 | switch (MEMFILE_ATTR(cft->private)) { | |
3562 | case RES_USAGE: | |
3563 | if (counter == &memcg->memory) | |
3564 | return mem_cgroup_usage(memcg, false); | |
3565 | if (counter == &memcg->memsw) | |
3566 | return mem_cgroup_usage(memcg, true); | |
3567 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3568 | case RES_LIMIT: | |
3569 | return (u64)counter->limit * PAGE_SIZE; | |
3570 | case RES_MAX_USAGE: | |
3571 | return (u64)counter->watermark * PAGE_SIZE; | |
3572 | case RES_FAILCNT: | |
3573 | return counter->failcnt; | |
3574 | case RES_SOFT_LIMIT: | |
3575 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3576 | default: | |
3577 | BUG(); | |
3578 | } | |
8cdea7c0 | 3579 | } |
510fc4e1 | 3580 | |
510fc4e1 | 3581 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
3582 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
3583 | unsigned long nr_pages) | |
d6441637 VD |
3584 | { |
3585 | int err = 0; | |
3586 | int memcg_id; | |
3587 | ||
3588 | if (memcg_kmem_is_active(memcg)) | |
3589 | return 0; | |
3590 | ||
3591 | /* | |
3592 | * We are going to allocate memory for data shared by all memory | |
3593 | * cgroups so let's stop accounting here. | |
3594 | */ | |
3595 | memcg_stop_kmem_account(); | |
3596 | ||
510fc4e1 GC |
3597 | /* |
3598 | * For simplicity, we won't allow this to be disabled. It also can't | |
3599 | * be changed if the cgroup has children already, or if tasks had | |
3600 | * already joined. | |
3601 | * | |
3602 | * If tasks join before we set the limit, a person looking at | |
3603 | * kmem.usage_in_bytes will have no way to determine when it took | |
3604 | * place, which makes the value quite meaningless. | |
3605 | * | |
3606 | * After it first became limited, changes in the value of the limit are | |
3607 | * of course permitted. | |
510fc4e1 | 3608 | */ |
0999821b | 3609 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
3610 | if (cgroup_has_tasks(memcg->css.cgroup) || |
3611 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
3612 | err = -EBUSY; |
3613 | mutex_unlock(&memcg_create_mutex); | |
3614 | if (err) | |
3615 | goto out; | |
510fc4e1 | 3616 | |
f3bb3043 | 3617 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
3618 | if (memcg_id < 0) { |
3619 | err = memcg_id; | |
3620 | goto out; | |
3621 | } | |
3622 | ||
d6441637 VD |
3623 | memcg->kmemcg_id = memcg_id; |
3624 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
d6441637 VD |
3625 | |
3626 | /* | |
3627 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
3628 | * active bit set yet, so this should succeed. | |
3629 | */ | |
3e32cb2e | 3630 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
3631 | VM_BUG_ON(err); |
3632 | ||
3633 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
3634 | /* | |
3635 | * Setting the active bit after enabling static branching will | |
3636 | * guarantee no one starts accounting before all call sites are | |
3637 | * patched. | |
3638 | */ | |
3639 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 3640 | out: |
d6441637 VD |
3641 | memcg_resume_kmem_account(); |
3642 | return err; | |
d6441637 VD |
3643 | } |
3644 | ||
d6441637 | 3645 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3646 | unsigned long limit) |
d6441637 VD |
3647 | { |
3648 | int ret; | |
3649 | ||
3e32cb2e | 3650 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 3651 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 3652 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 3653 | else |
3e32cb2e JW |
3654 | ret = page_counter_limit(&memcg->kmem, limit); |
3655 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
3656 | return ret; |
3657 | } | |
3658 | ||
55007d84 | 3659 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 3660 | { |
55007d84 | 3661 | int ret = 0; |
510fc4e1 | 3662 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 3663 | |
d6441637 VD |
3664 | if (!parent) |
3665 | return 0; | |
55007d84 | 3666 | |
8c0145b6 | 3667 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 3668 | /* |
d6441637 VD |
3669 | * If the parent cgroup is not kmem-active now, it cannot be activated |
3670 | * after this point, because it has at least one child already. | |
55007d84 | 3671 | */ |
d6441637 | 3672 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3673 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3674 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3675 | return ret; |
510fc4e1 | 3676 | } |
d6441637 VD |
3677 | #else |
3678 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3679 | unsigned long limit) |
d6441637 VD |
3680 | { |
3681 | return -EINVAL; | |
3682 | } | |
6d043990 | 3683 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3684 | |
628f4235 KH |
3685 | /* |
3686 | * The user of this function is... | |
3687 | * RES_LIMIT. | |
3688 | */ | |
451af504 TH |
3689 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3690 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3691 | { |
451af504 | 3692 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3693 | unsigned long nr_pages; |
628f4235 KH |
3694 | int ret; |
3695 | ||
451af504 | 3696 | buf = strstrip(buf); |
3e32cb2e JW |
3697 | ret = page_counter_memparse(buf, &nr_pages); |
3698 | if (ret) | |
3699 | return ret; | |
af36f906 | 3700 | |
3e32cb2e | 3701 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3702 | case RES_LIMIT: |
4b3bde4c BS |
3703 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3704 | ret = -EINVAL; | |
3705 | break; | |
3706 | } | |
3e32cb2e JW |
3707 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3708 | case _MEM: | |
3709 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3710 | break; |
3e32cb2e JW |
3711 | case _MEMSWAP: |
3712 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3713 | break; |
3e32cb2e JW |
3714 | case _KMEM: |
3715 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3716 | break; | |
3717 | } | |
296c81d8 | 3718 | break; |
3e32cb2e JW |
3719 | case RES_SOFT_LIMIT: |
3720 | memcg->soft_limit = nr_pages; | |
3721 | ret = 0; | |
628f4235 KH |
3722 | break; |
3723 | } | |
451af504 | 3724 | return ret ?: nbytes; |
8cdea7c0 BS |
3725 | } |
3726 | ||
6770c64e TH |
3727 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3728 | size_t nbytes, loff_t off) | |
c84872e1 | 3729 | { |
6770c64e | 3730 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3731 | struct page_counter *counter; |
c84872e1 | 3732 | |
3e32cb2e JW |
3733 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3734 | case _MEM: | |
3735 | counter = &memcg->memory; | |
3736 | break; | |
3737 | case _MEMSWAP: | |
3738 | counter = &memcg->memsw; | |
3739 | break; | |
3740 | case _KMEM: | |
3741 | counter = &memcg->kmem; | |
3742 | break; | |
3743 | default: | |
3744 | BUG(); | |
3745 | } | |
af36f906 | 3746 | |
3e32cb2e | 3747 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3748 | case RES_MAX_USAGE: |
3e32cb2e | 3749 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3750 | break; |
3751 | case RES_FAILCNT: | |
3e32cb2e | 3752 | counter->failcnt = 0; |
29f2a4da | 3753 | break; |
3e32cb2e JW |
3754 | default: |
3755 | BUG(); | |
29f2a4da | 3756 | } |
f64c3f54 | 3757 | |
6770c64e | 3758 | return nbytes; |
c84872e1 PE |
3759 | } |
3760 | ||
182446d0 | 3761 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3762 | struct cftype *cft) |
3763 | { | |
182446d0 | 3764 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3765 | } |
3766 | ||
02491447 | 3767 | #ifdef CONFIG_MMU |
182446d0 | 3768 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3769 | struct cftype *cft, u64 val) |
3770 | { | |
182446d0 | 3771 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
3772 | |
3773 | if (val >= (1 << NR_MOVE_TYPE)) | |
3774 | return -EINVAL; | |
ee5e8472 | 3775 | |
7dc74be0 | 3776 | /* |
ee5e8472 GC |
3777 | * No kind of locking is needed in here, because ->can_attach() will |
3778 | * check this value once in the beginning of the process, and then carry | |
3779 | * on with stale data. This means that changes to this value will only | |
3780 | * affect task migrations starting after the change. | |
7dc74be0 | 3781 | */ |
c0ff4b85 | 3782 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3783 | return 0; |
3784 | } | |
02491447 | 3785 | #else |
182446d0 | 3786 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3787 | struct cftype *cft, u64 val) |
3788 | { | |
3789 | return -ENOSYS; | |
3790 | } | |
3791 | #endif | |
7dc74be0 | 3792 | |
406eb0c9 | 3793 | #ifdef CONFIG_NUMA |
2da8ca82 | 3794 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3795 | { |
25485de6 GT |
3796 | struct numa_stat { |
3797 | const char *name; | |
3798 | unsigned int lru_mask; | |
3799 | }; | |
3800 | ||
3801 | static const struct numa_stat stats[] = { | |
3802 | { "total", LRU_ALL }, | |
3803 | { "file", LRU_ALL_FILE }, | |
3804 | { "anon", LRU_ALL_ANON }, | |
3805 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3806 | }; | |
3807 | const struct numa_stat *stat; | |
406eb0c9 | 3808 | int nid; |
25485de6 | 3809 | unsigned long nr; |
2da8ca82 | 3810 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3811 | |
25485de6 GT |
3812 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3813 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3814 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3815 | for_each_node_state(nid, N_MEMORY) { | |
3816 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3817 | stat->lru_mask); | |
3818 | seq_printf(m, " N%d=%lu", nid, nr); | |
3819 | } | |
3820 | seq_putc(m, '\n'); | |
406eb0c9 | 3821 | } |
406eb0c9 | 3822 | |
071aee13 YH |
3823 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3824 | struct mem_cgroup *iter; | |
3825 | ||
3826 | nr = 0; | |
3827 | for_each_mem_cgroup_tree(iter, memcg) | |
3828 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3829 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3830 | for_each_node_state(nid, N_MEMORY) { | |
3831 | nr = 0; | |
3832 | for_each_mem_cgroup_tree(iter, memcg) | |
3833 | nr += mem_cgroup_node_nr_lru_pages( | |
3834 | iter, nid, stat->lru_mask); | |
3835 | seq_printf(m, " N%d=%lu", nid, nr); | |
3836 | } | |
3837 | seq_putc(m, '\n'); | |
406eb0c9 | 3838 | } |
406eb0c9 | 3839 | |
406eb0c9 YH |
3840 | return 0; |
3841 | } | |
3842 | #endif /* CONFIG_NUMA */ | |
3843 | ||
af7c4b0e JW |
3844 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
3845 | { | |
3846 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
3847 | } | |
3848 | ||
2da8ca82 | 3849 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3850 | { |
2da8ca82 | 3851 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3852 | unsigned long memory, memsw; |
af7c4b0e JW |
3853 | struct mem_cgroup *mi; |
3854 | unsigned int i; | |
406eb0c9 | 3855 | |
af7c4b0e | 3856 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3857 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3858 | continue; |
af7c4b0e JW |
3859 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
3860 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 3861 | } |
7b854121 | 3862 | |
af7c4b0e JW |
3863 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3864 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3865 | mem_cgroup_read_events(memcg, i)); | |
3866 | ||
3867 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3868 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3869 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3870 | ||
14067bb3 | 3871 | /* Hierarchical information */ |
3e32cb2e JW |
3872 | memory = memsw = PAGE_COUNTER_MAX; |
3873 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3874 | memory = min(memory, mi->memory.limit); | |
3875 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3876 | } |
3e32cb2e JW |
3877 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3878 | (u64)memory * PAGE_SIZE); | |
3879 | if (do_swap_account) | |
3880 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3881 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3882 | |
af7c4b0e JW |
3883 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
3884 | long long val = 0; | |
3885 | ||
bff6bb83 | 3886 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3887 | continue; |
af7c4b0e JW |
3888 | for_each_mem_cgroup_tree(mi, memcg) |
3889 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
3890 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
3891 | } | |
3892 | ||
3893 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3894 | unsigned long long val = 0; | |
3895 | ||
3896 | for_each_mem_cgroup_tree(mi, memcg) | |
3897 | val += mem_cgroup_read_events(mi, i); | |
3898 | seq_printf(m, "total_%s %llu\n", | |
3899 | mem_cgroup_events_names[i], val); | |
3900 | } | |
3901 | ||
3902 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3903 | unsigned long long val = 0; | |
3904 | ||
3905 | for_each_mem_cgroup_tree(mi, memcg) | |
3906 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3907 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3908 | } |
14067bb3 | 3909 | |
7f016ee8 | 3910 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3911 | { |
3912 | int nid, zid; | |
3913 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3914 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3915 | unsigned long recent_rotated[2] = {0, 0}; |
3916 | unsigned long recent_scanned[2] = {0, 0}; | |
3917 | ||
3918 | for_each_online_node(nid) | |
3919 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3920 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3921 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3922 | |
89abfab1 HD |
3923 | recent_rotated[0] += rstat->recent_rotated[0]; |
3924 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3925 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3926 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3927 | } |
78ccf5b5 JW |
3928 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3929 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3930 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3931 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3932 | } |
3933 | #endif | |
3934 | ||
d2ceb9b7 KH |
3935 | return 0; |
3936 | } | |
3937 | ||
182446d0 TH |
3938 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3939 | struct cftype *cft) | |
a7885eb8 | 3940 | { |
182446d0 | 3941 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3942 | |
1f4c025b | 3943 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3944 | } |
3945 | ||
182446d0 TH |
3946 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3947 | struct cftype *cft, u64 val) | |
a7885eb8 | 3948 | { |
182446d0 | 3949 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3950 | |
3dae7fec | 3951 | if (val > 100) |
a7885eb8 KM |
3952 | return -EINVAL; |
3953 | ||
14208b0e | 3954 | if (css->parent) |
3dae7fec JW |
3955 | memcg->swappiness = val; |
3956 | else | |
3957 | vm_swappiness = val; | |
068b38c1 | 3958 | |
a7885eb8 KM |
3959 | return 0; |
3960 | } | |
3961 | ||
2e72b634 KS |
3962 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3963 | { | |
3964 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3965 | unsigned long usage; |
2e72b634 KS |
3966 | int i; |
3967 | ||
3968 | rcu_read_lock(); | |
3969 | if (!swap) | |
2c488db2 | 3970 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3971 | else |
2c488db2 | 3972 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3973 | |
3974 | if (!t) | |
3975 | goto unlock; | |
3976 | ||
ce00a967 | 3977 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3978 | |
3979 | /* | |
748dad36 | 3980 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3981 | * If it's not true, a threshold was crossed after last |
3982 | * call of __mem_cgroup_threshold(). | |
3983 | */ | |
5407a562 | 3984 | i = t->current_threshold; |
2e72b634 KS |
3985 | |
3986 | /* | |
3987 | * Iterate backward over array of thresholds starting from | |
3988 | * current_threshold and check if a threshold is crossed. | |
3989 | * If none of thresholds below usage is crossed, we read | |
3990 | * only one element of the array here. | |
3991 | */ | |
3992 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3993 | eventfd_signal(t->entries[i].eventfd, 1); | |
3994 | ||
3995 | /* i = current_threshold + 1 */ | |
3996 | i++; | |
3997 | ||
3998 | /* | |
3999 | * Iterate forward over array of thresholds starting from | |
4000 | * current_threshold+1 and check if a threshold is crossed. | |
4001 | * If none of thresholds above usage is crossed, we read | |
4002 | * only one element of the array here. | |
4003 | */ | |
4004 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4005 | eventfd_signal(t->entries[i].eventfd, 1); | |
4006 | ||
4007 | /* Update current_threshold */ | |
5407a562 | 4008 | t->current_threshold = i - 1; |
2e72b634 KS |
4009 | unlock: |
4010 | rcu_read_unlock(); | |
4011 | } | |
4012 | ||
4013 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4014 | { | |
ad4ca5f4 KS |
4015 | while (memcg) { |
4016 | __mem_cgroup_threshold(memcg, false); | |
4017 | if (do_swap_account) | |
4018 | __mem_cgroup_threshold(memcg, true); | |
4019 | ||
4020 | memcg = parent_mem_cgroup(memcg); | |
4021 | } | |
2e72b634 KS |
4022 | } |
4023 | ||
4024 | static int compare_thresholds(const void *a, const void *b) | |
4025 | { | |
4026 | const struct mem_cgroup_threshold *_a = a; | |
4027 | const struct mem_cgroup_threshold *_b = b; | |
4028 | ||
2bff24a3 GT |
4029 | if (_a->threshold > _b->threshold) |
4030 | return 1; | |
4031 | ||
4032 | if (_a->threshold < _b->threshold) | |
4033 | return -1; | |
4034 | ||
4035 | return 0; | |
2e72b634 KS |
4036 | } |
4037 | ||
c0ff4b85 | 4038 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4039 | { |
4040 | struct mem_cgroup_eventfd_list *ev; | |
4041 | ||
2bcf2e92 MH |
4042 | spin_lock(&memcg_oom_lock); |
4043 | ||
c0ff4b85 | 4044 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4045 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4046 | |
4047 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4048 | return 0; |
4049 | } | |
4050 | ||
c0ff4b85 | 4051 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4052 | { |
7d74b06f KH |
4053 | struct mem_cgroup *iter; |
4054 | ||
c0ff4b85 | 4055 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4056 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4057 | } |
4058 | ||
59b6f873 | 4059 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4060 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4061 | { |
2c488db2 KS |
4062 | struct mem_cgroup_thresholds *thresholds; |
4063 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4064 | unsigned long threshold; |
4065 | unsigned long usage; | |
2c488db2 | 4066 | int i, size, ret; |
2e72b634 | 4067 | |
3e32cb2e | 4068 | ret = page_counter_memparse(args, &threshold); |
2e72b634 KS |
4069 | if (ret) |
4070 | return ret; | |
4071 | ||
4072 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4073 | |
05b84301 | 4074 | if (type == _MEM) { |
2c488db2 | 4075 | thresholds = &memcg->thresholds; |
ce00a967 | 4076 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4077 | } else if (type == _MEMSWAP) { |
2c488db2 | 4078 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4079 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4080 | } else |
2e72b634 KS |
4081 | BUG(); |
4082 | ||
2e72b634 | 4083 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4084 | if (thresholds->primary) |
2e72b634 KS |
4085 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4086 | ||
2c488db2 | 4087 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4088 | |
4089 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4090 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4091 | GFP_KERNEL); |
2c488db2 | 4092 | if (!new) { |
2e72b634 KS |
4093 | ret = -ENOMEM; |
4094 | goto unlock; | |
4095 | } | |
2c488db2 | 4096 | new->size = size; |
2e72b634 KS |
4097 | |
4098 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4099 | if (thresholds->primary) { |
4100 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4101 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4102 | } |
4103 | ||
2e72b634 | 4104 | /* Add new threshold */ |
2c488db2 KS |
4105 | new->entries[size - 1].eventfd = eventfd; |
4106 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4107 | |
4108 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4109 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4110 | compare_thresholds, NULL); |
4111 | ||
4112 | /* Find current threshold */ | |
2c488db2 | 4113 | new->current_threshold = -1; |
2e72b634 | 4114 | for (i = 0; i < size; i++) { |
748dad36 | 4115 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4116 | /* |
2c488db2 KS |
4117 | * new->current_threshold will not be used until |
4118 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4119 | * it here. |
4120 | */ | |
2c488db2 | 4121 | ++new->current_threshold; |
748dad36 SZ |
4122 | } else |
4123 | break; | |
2e72b634 KS |
4124 | } |
4125 | ||
2c488db2 KS |
4126 | /* Free old spare buffer and save old primary buffer as spare */ |
4127 | kfree(thresholds->spare); | |
4128 | thresholds->spare = thresholds->primary; | |
4129 | ||
4130 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4131 | |
907860ed | 4132 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4133 | synchronize_rcu(); |
4134 | ||
2e72b634 KS |
4135 | unlock: |
4136 | mutex_unlock(&memcg->thresholds_lock); | |
4137 | ||
4138 | return ret; | |
4139 | } | |
4140 | ||
59b6f873 | 4141 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4142 | struct eventfd_ctx *eventfd, const char *args) |
4143 | { | |
59b6f873 | 4144 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4145 | } |
4146 | ||
59b6f873 | 4147 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4148 | struct eventfd_ctx *eventfd, const char *args) |
4149 | { | |
59b6f873 | 4150 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4151 | } |
4152 | ||
59b6f873 | 4153 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4154 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4155 | { |
2c488db2 KS |
4156 | struct mem_cgroup_thresholds *thresholds; |
4157 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4158 | unsigned long usage; |
2c488db2 | 4159 | int i, j, size; |
2e72b634 KS |
4160 | |
4161 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4162 | |
4163 | if (type == _MEM) { | |
2c488db2 | 4164 | thresholds = &memcg->thresholds; |
ce00a967 | 4165 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4166 | } else if (type == _MEMSWAP) { |
2c488db2 | 4167 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4168 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4169 | } else |
2e72b634 KS |
4170 | BUG(); |
4171 | ||
371528ca AV |
4172 | if (!thresholds->primary) |
4173 | goto unlock; | |
4174 | ||
2e72b634 KS |
4175 | /* Check if a threshold crossed before removing */ |
4176 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4177 | ||
4178 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4179 | size = 0; |
4180 | for (i = 0; i < thresholds->primary->size; i++) { | |
4181 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4182 | size++; |
4183 | } | |
4184 | ||
2c488db2 | 4185 | new = thresholds->spare; |
907860ed | 4186 | |
2e72b634 KS |
4187 | /* Set thresholds array to NULL if we don't have thresholds */ |
4188 | if (!size) { | |
2c488db2 KS |
4189 | kfree(new); |
4190 | new = NULL; | |
907860ed | 4191 | goto swap_buffers; |
2e72b634 KS |
4192 | } |
4193 | ||
2c488db2 | 4194 | new->size = size; |
2e72b634 KS |
4195 | |
4196 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4197 | new->current_threshold = -1; |
4198 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4199 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4200 | continue; |
4201 | ||
2c488db2 | 4202 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4203 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4204 | /* |
2c488db2 | 4205 | * new->current_threshold will not be used |
2e72b634 KS |
4206 | * until rcu_assign_pointer(), so it's safe to increment |
4207 | * it here. | |
4208 | */ | |
2c488db2 | 4209 | ++new->current_threshold; |
2e72b634 KS |
4210 | } |
4211 | j++; | |
4212 | } | |
4213 | ||
907860ed | 4214 | swap_buffers: |
2c488db2 KS |
4215 | /* Swap primary and spare array */ |
4216 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4217 | /* If all events are unregistered, free the spare array */ |
4218 | if (!new) { | |
4219 | kfree(thresholds->spare); | |
4220 | thresholds->spare = NULL; | |
4221 | } | |
4222 | ||
2c488db2 | 4223 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4224 | |
907860ed | 4225 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4226 | synchronize_rcu(); |
371528ca | 4227 | unlock: |
2e72b634 | 4228 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4229 | } |
c1e862c1 | 4230 | |
59b6f873 | 4231 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4232 | struct eventfd_ctx *eventfd) |
4233 | { | |
59b6f873 | 4234 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4235 | } |
4236 | ||
59b6f873 | 4237 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4238 | struct eventfd_ctx *eventfd) |
4239 | { | |
59b6f873 | 4240 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4241 | } |
4242 | ||
59b6f873 | 4243 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4244 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4245 | { |
9490ff27 | 4246 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4247 | |
9490ff27 KH |
4248 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4249 | if (!event) | |
4250 | return -ENOMEM; | |
4251 | ||
1af8efe9 | 4252 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4253 | |
4254 | event->eventfd = eventfd; | |
4255 | list_add(&event->list, &memcg->oom_notify); | |
4256 | ||
4257 | /* already in OOM ? */ | |
79dfdacc | 4258 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4259 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4260 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4261 | |
4262 | return 0; | |
4263 | } | |
4264 | ||
59b6f873 | 4265 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4266 | struct eventfd_ctx *eventfd) |
9490ff27 | 4267 | { |
9490ff27 | 4268 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4269 | |
1af8efe9 | 4270 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4271 | |
c0ff4b85 | 4272 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4273 | if (ev->eventfd == eventfd) { |
4274 | list_del(&ev->list); | |
4275 | kfree(ev); | |
4276 | } | |
4277 | } | |
4278 | ||
1af8efe9 | 4279 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4280 | } |
4281 | ||
2da8ca82 | 4282 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4283 | { |
2da8ca82 | 4284 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4285 | |
791badbd TH |
4286 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4287 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4288 | return 0; |
4289 | } | |
4290 | ||
182446d0 | 4291 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4292 | struct cftype *cft, u64 val) |
4293 | { | |
182446d0 | 4294 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4295 | |
4296 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4297 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4298 | return -EINVAL; |
4299 | ||
c0ff4b85 | 4300 | memcg->oom_kill_disable = val; |
4d845ebf | 4301 | if (!val) |
c0ff4b85 | 4302 | memcg_oom_recover(memcg); |
3dae7fec | 4303 | |
3c11ecf4 KH |
4304 | return 0; |
4305 | } | |
4306 | ||
c255a458 | 4307 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4308 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4309 | { |
55007d84 GC |
4310 | int ret; |
4311 | ||
2633d7a0 | 4312 | memcg->kmemcg_id = -1; |
55007d84 GC |
4313 | ret = memcg_propagate_kmem(memcg); |
4314 | if (ret) | |
4315 | return ret; | |
2633d7a0 | 4316 | |
1d62e436 | 4317 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4318 | } |
e5671dfa | 4319 | |
10d5ebf4 | 4320 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4321 | { |
1d62e436 | 4322 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 4323 | } |
e5671dfa | 4324 | #else |
cbe128e3 | 4325 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4326 | { |
4327 | return 0; | |
4328 | } | |
d1a4c0b3 | 4329 | |
10d5ebf4 LZ |
4330 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4331 | { | |
4332 | } | |
e5671dfa GC |
4333 | #endif |
4334 | ||
3bc942f3 TH |
4335 | /* |
4336 | * DO NOT USE IN NEW FILES. | |
4337 | * | |
4338 | * "cgroup.event_control" implementation. | |
4339 | * | |
4340 | * This is way over-engineered. It tries to support fully configurable | |
4341 | * events for each user. Such level of flexibility is completely | |
4342 | * unnecessary especially in the light of the planned unified hierarchy. | |
4343 | * | |
4344 | * Please deprecate this and replace with something simpler if at all | |
4345 | * possible. | |
4346 | */ | |
4347 | ||
79bd9814 TH |
4348 | /* |
4349 | * Unregister event and free resources. | |
4350 | * | |
4351 | * Gets called from workqueue. | |
4352 | */ | |
3bc942f3 | 4353 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4354 | { |
3bc942f3 TH |
4355 | struct mem_cgroup_event *event = |
4356 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4357 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4358 | |
4359 | remove_wait_queue(event->wqh, &event->wait); | |
4360 | ||
59b6f873 | 4361 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4362 | |
4363 | /* Notify userspace the event is going away. */ | |
4364 | eventfd_signal(event->eventfd, 1); | |
4365 | ||
4366 | eventfd_ctx_put(event->eventfd); | |
4367 | kfree(event); | |
59b6f873 | 4368 | css_put(&memcg->css); |
79bd9814 TH |
4369 | } |
4370 | ||
4371 | /* | |
4372 | * Gets called on POLLHUP on eventfd when user closes it. | |
4373 | * | |
4374 | * Called with wqh->lock held and interrupts disabled. | |
4375 | */ | |
3bc942f3 TH |
4376 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4377 | int sync, void *key) | |
79bd9814 | 4378 | { |
3bc942f3 TH |
4379 | struct mem_cgroup_event *event = |
4380 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4381 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4382 | unsigned long flags = (unsigned long)key; |
4383 | ||
4384 | if (flags & POLLHUP) { | |
4385 | /* | |
4386 | * If the event has been detached at cgroup removal, we | |
4387 | * can simply return knowing the other side will cleanup | |
4388 | * for us. | |
4389 | * | |
4390 | * We can't race against event freeing since the other | |
4391 | * side will require wqh->lock via remove_wait_queue(), | |
4392 | * which we hold. | |
4393 | */ | |
fba94807 | 4394 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4395 | if (!list_empty(&event->list)) { |
4396 | list_del_init(&event->list); | |
4397 | /* | |
4398 | * We are in atomic context, but cgroup_event_remove() | |
4399 | * may sleep, so we have to call it in workqueue. | |
4400 | */ | |
4401 | schedule_work(&event->remove); | |
4402 | } | |
fba94807 | 4403 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4404 | } |
4405 | ||
4406 | return 0; | |
4407 | } | |
4408 | ||
3bc942f3 | 4409 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4410 | wait_queue_head_t *wqh, poll_table *pt) |
4411 | { | |
3bc942f3 TH |
4412 | struct mem_cgroup_event *event = |
4413 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4414 | |
4415 | event->wqh = wqh; | |
4416 | add_wait_queue(wqh, &event->wait); | |
4417 | } | |
4418 | ||
4419 | /* | |
3bc942f3 TH |
4420 | * DO NOT USE IN NEW FILES. |
4421 | * | |
79bd9814 TH |
4422 | * Parse input and register new cgroup event handler. |
4423 | * | |
4424 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4425 | * Interpretation of args is defined by control file implementation. | |
4426 | */ | |
451af504 TH |
4427 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4428 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4429 | { |
451af504 | 4430 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4431 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4432 | struct mem_cgroup_event *event; |
79bd9814 TH |
4433 | struct cgroup_subsys_state *cfile_css; |
4434 | unsigned int efd, cfd; | |
4435 | struct fd efile; | |
4436 | struct fd cfile; | |
fba94807 | 4437 | const char *name; |
79bd9814 TH |
4438 | char *endp; |
4439 | int ret; | |
4440 | ||
451af504 TH |
4441 | buf = strstrip(buf); |
4442 | ||
4443 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4444 | if (*endp != ' ') |
4445 | return -EINVAL; | |
451af504 | 4446 | buf = endp + 1; |
79bd9814 | 4447 | |
451af504 | 4448 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4449 | if ((*endp != ' ') && (*endp != '\0')) |
4450 | return -EINVAL; | |
451af504 | 4451 | buf = endp + 1; |
79bd9814 TH |
4452 | |
4453 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4454 | if (!event) | |
4455 | return -ENOMEM; | |
4456 | ||
59b6f873 | 4457 | event->memcg = memcg; |
79bd9814 | 4458 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4459 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4460 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4461 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4462 | |
4463 | efile = fdget(efd); | |
4464 | if (!efile.file) { | |
4465 | ret = -EBADF; | |
4466 | goto out_kfree; | |
4467 | } | |
4468 | ||
4469 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4470 | if (IS_ERR(event->eventfd)) { | |
4471 | ret = PTR_ERR(event->eventfd); | |
4472 | goto out_put_efile; | |
4473 | } | |
4474 | ||
4475 | cfile = fdget(cfd); | |
4476 | if (!cfile.file) { | |
4477 | ret = -EBADF; | |
4478 | goto out_put_eventfd; | |
4479 | } | |
4480 | ||
4481 | /* the process need read permission on control file */ | |
4482 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4483 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4484 | if (ret < 0) | |
4485 | goto out_put_cfile; | |
4486 | ||
fba94807 TH |
4487 | /* |
4488 | * Determine the event callbacks and set them in @event. This used | |
4489 | * to be done via struct cftype but cgroup core no longer knows | |
4490 | * about these events. The following is crude but the whole thing | |
4491 | * is for compatibility anyway. | |
3bc942f3 TH |
4492 | * |
4493 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
4494 | */ |
4495 | name = cfile.file->f_dentry->d_name.name; | |
4496 | ||
4497 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4498 | event->register_event = mem_cgroup_usage_register_event; | |
4499 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4500 | } else if (!strcmp(name, "memory.oom_control")) { | |
4501 | event->register_event = mem_cgroup_oom_register_event; | |
4502 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4503 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4504 | event->register_event = vmpressure_register_event; | |
4505 | event->unregister_event = vmpressure_unregister_event; | |
4506 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4507 | event->register_event = memsw_cgroup_usage_register_event; |
4508 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4509 | } else { |
4510 | ret = -EINVAL; | |
4511 | goto out_put_cfile; | |
4512 | } | |
4513 | ||
79bd9814 | 4514 | /* |
b5557c4c TH |
4515 | * Verify @cfile should belong to @css. Also, remaining events are |
4516 | * automatically removed on cgroup destruction but the removal is | |
4517 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4518 | */ |
ec903c0c TH |
4519 | cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent, |
4520 | &memory_cgrp_subsys); | |
79bd9814 | 4521 | ret = -EINVAL; |
5a17f543 | 4522 | if (IS_ERR(cfile_css)) |
79bd9814 | 4523 | goto out_put_cfile; |
5a17f543 TH |
4524 | if (cfile_css != css) { |
4525 | css_put(cfile_css); | |
79bd9814 | 4526 | goto out_put_cfile; |
5a17f543 | 4527 | } |
79bd9814 | 4528 | |
451af504 | 4529 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4530 | if (ret) |
4531 | goto out_put_css; | |
4532 | ||
4533 | efile.file->f_op->poll(efile.file, &event->pt); | |
4534 | ||
fba94807 TH |
4535 | spin_lock(&memcg->event_list_lock); |
4536 | list_add(&event->list, &memcg->event_list); | |
4537 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4538 | |
4539 | fdput(cfile); | |
4540 | fdput(efile); | |
4541 | ||
451af504 | 4542 | return nbytes; |
79bd9814 TH |
4543 | |
4544 | out_put_css: | |
b5557c4c | 4545 | css_put(css); |
79bd9814 TH |
4546 | out_put_cfile: |
4547 | fdput(cfile); | |
4548 | out_put_eventfd: | |
4549 | eventfd_ctx_put(event->eventfd); | |
4550 | out_put_efile: | |
4551 | fdput(efile); | |
4552 | out_kfree: | |
4553 | kfree(event); | |
4554 | ||
4555 | return ret; | |
4556 | } | |
4557 | ||
8cdea7c0 BS |
4558 | static struct cftype mem_cgroup_files[] = { |
4559 | { | |
0eea1030 | 4560 | .name = "usage_in_bytes", |
8c7c6e34 | 4561 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4562 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4563 | }, |
c84872e1 PE |
4564 | { |
4565 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4566 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4567 | .write = mem_cgroup_reset, |
791badbd | 4568 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4569 | }, |
8cdea7c0 | 4570 | { |
0eea1030 | 4571 | .name = "limit_in_bytes", |
8c7c6e34 | 4572 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4573 | .write = mem_cgroup_write, |
791badbd | 4574 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4575 | }, |
296c81d8 BS |
4576 | { |
4577 | .name = "soft_limit_in_bytes", | |
4578 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4579 | .write = mem_cgroup_write, |
791badbd | 4580 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4581 | }, |
8cdea7c0 BS |
4582 | { |
4583 | .name = "failcnt", | |
8c7c6e34 | 4584 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4585 | .write = mem_cgroup_reset, |
791badbd | 4586 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4587 | }, |
d2ceb9b7 KH |
4588 | { |
4589 | .name = "stat", | |
2da8ca82 | 4590 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4591 | }, |
c1e862c1 KH |
4592 | { |
4593 | .name = "force_empty", | |
6770c64e | 4594 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4595 | }, |
18f59ea7 BS |
4596 | { |
4597 | .name = "use_hierarchy", | |
4598 | .write_u64 = mem_cgroup_hierarchy_write, | |
4599 | .read_u64 = mem_cgroup_hierarchy_read, | |
4600 | }, | |
79bd9814 | 4601 | { |
3bc942f3 | 4602 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4603 | .write = memcg_write_event_control, |
79bd9814 TH |
4604 | .flags = CFTYPE_NO_PREFIX, |
4605 | .mode = S_IWUGO, | |
4606 | }, | |
a7885eb8 KM |
4607 | { |
4608 | .name = "swappiness", | |
4609 | .read_u64 = mem_cgroup_swappiness_read, | |
4610 | .write_u64 = mem_cgroup_swappiness_write, | |
4611 | }, | |
7dc74be0 DN |
4612 | { |
4613 | .name = "move_charge_at_immigrate", | |
4614 | .read_u64 = mem_cgroup_move_charge_read, | |
4615 | .write_u64 = mem_cgroup_move_charge_write, | |
4616 | }, | |
9490ff27 KH |
4617 | { |
4618 | .name = "oom_control", | |
2da8ca82 | 4619 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4620 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4621 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4622 | }, | |
70ddf637 AV |
4623 | { |
4624 | .name = "pressure_level", | |
70ddf637 | 4625 | }, |
406eb0c9 YH |
4626 | #ifdef CONFIG_NUMA |
4627 | { | |
4628 | .name = "numa_stat", | |
2da8ca82 | 4629 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4630 | }, |
4631 | #endif | |
510fc4e1 GC |
4632 | #ifdef CONFIG_MEMCG_KMEM |
4633 | { | |
4634 | .name = "kmem.limit_in_bytes", | |
4635 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4636 | .write = mem_cgroup_write, |
791badbd | 4637 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4638 | }, |
4639 | { | |
4640 | .name = "kmem.usage_in_bytes", | |
4641 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4642 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4643 | }, |
4644 | { | |
4645 | .name = "kmem.failcnt", | |
4646 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4647 | .write = mem_cgroup_reset, |
791badbd | 4648 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4649 | }, |
4650 | { | |
4651 | .name = "kmem.max_usage_in_bytes", | |
4652 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4653 | .write = mem_cgroup_reset, |
791badbd | 4654 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4655 | }, |
749c5415 GC |
4656 | #ifdef CONFIG_SLABINFO |
4657 | { | |
4658 | .name = "kmem.slabinfo", | |
b047501c VD |
4659 | .seq_start = slab_start, |
4660 | .seq_next = slab_next, | |
4661 | .seq_stop = slab_stop, | |
4662 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4663 | }, |
4664 | #endif | |
8c7c6e34 | 4665 | #endif |
6bc10349 | 4666 | { }, /* terminate */ |
af36f906 | 4667 | }; |
8c7c6e34 | 4668 | |
2d11085e MH |
4669 | #ifdef CONFIG_MEMCG_SWAP |
4670 | static struct cftype memsw_cgroup_files[] = { | |
4671 | { | |
4672 | .name = "memsw.usage_in_bytes", | |
4673 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 4674 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4675 | }, |
4676 | { | |
4677 | .name = "memsw.max_usage_in_bytes", | |
4678 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 4679 | .write = mem_cgroup_reset, |
791badbd | 4680 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4681 | }, |
4682 | { | |
4683 | .name = "memsw.limit_in_bytes", | |
4684 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 4685 | .write = mem_cgroup_write, |
791badbd | 4686 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4687 | }, |
4688 | { | |
4689 | .name = "memsw.failcnt", | |
4690 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 4691 | .write = mem_cgroup_reset, |
791badbd | 4692 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4693 | }, |
4694 | { }, /* terminate */ | |
4695 | }; | |
4696 | #endif | |
c0ff4b85 | 4697 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4698 | { |
4699 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4700 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4701 | int zone, tmp = node; |
1ecaab2b KH |
4702 | /* |
4703 | * This routine is called against possible nodes. | |
4704 | * But it's BUG to call kmalloc() against offline node. | |
4705 | * | |
4706 | * TODO: this routine can waste much memory for nodes which will | |
4707 | * never be onlined. It's better to use memory hotplug callback | |
4708 | * function. | |
4709 | */ | |
41e3355d KH |
4710 | if (!node_state(node, N_NORMAL_MEMORY)) |
4711 | tmp = -1; | |
17295c88 | 4712 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4713 | if (!pn) |
4714 | return 1; | |
1ecaab2b | 4715 | |
1ecaab2b KH |
4716 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4717 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4718 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4719 | mz->usage_in_excess = 0; |
4720 | mz->on_tree = false; | |
d79154bb | 4721 | mz->memcg = memcg; |
1ecaab2b | 4722 | } |
54f72fe0 | 4723 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4724 | return 0; |
4725 | } | |
4726 | ||
c0ff4b85 | 4727 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4728 | { |
54f72fe0 | 4729 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4730 | } |
4731 | ||
33327948 KH |
4732 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4733 | { | |
d79154bb | 4734 | struct mem_cgroup *memcg; |
8ff69e2c | 4735 | size_t size; |
33327948 | 4736 | |
8ff69e2c VD |
4737 | size = sizeof(struct mem_cgroup); |
4738 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4739 | |
8ff69e2c | 4740 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4741 | if (!memcg) |
e7bbcdf3 DC |
4742 | return NULL; |
4743 | ||
d79154bb HD |
4744 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4745 | if (!memcg->stat) | |
d2e61b8d | 4746 | goto out_free; |
d79154bb HD |
4747 | spin_lock_init(&memcg->pcp_counter_lock); |
4748 | return memcg; | |
d2e61b8d DC |
4749 | |
4750 | out_free: | |
8ff69e2c | 4751 | kfree(memcg); |
d2e61b8d | 4752 | return NULL; |
33327948 KH |
4753 | } |
4754 | ||
59927fb9 | 4755 | /* |
c8b2a36f GC |
4756 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4757 | * (scanning all at force_empty is too costly...) | |
4758 | * | |
4759 | * Instead of clearing all references at force_empty, we remember | |
4760 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4761 | * it goes down to 0. | |
4762 | * | |
4763 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4764 | */ |
c8b2a36f GC |
4765 | |
4766 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4767 | { |
c8b2a36f | 4768 | int node; |
59927fb9 | 4769 | |
bb4cc1a8 | 4770 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4771 | |
4772 | for_each_node(node) | |
4773 | free_mem_cgroup_per_zone_info(memcg, node); | |
4774 | ||
4775 | free_percpu(memcg->stat); | |
4776 | ||
3f134619 GC |
4777 | /* |
4778 | * We need to make sure that (at least for now), the jump label | |
4779 | * destruction code runs outside of the cgroup lock. This is because | |
4780 | * get_online_cpus(), which is called from the static_branch update, | |
4781 | * can't be called inside the cgroup_lock. cpusets are the ones | |
4782 | * enforcing this dependency, so if they ever change, we might as well. | |
4783 | * | |
4784 | * schedule_work() will guarantee this happens. Be careful if you need | |
4785 | * to move this code around, and make sure it is outside | |
4786 | * the cgroup_lock. | |
4787 | */ | |
a8964b9b | 4788 | disarm_static_keys(memcg); |
8ff69e2c | 4789 | kfree(memcg); |
59927fb9 | 4790 | } |
3afe36b1 | 4791 | |
7bcc1bb1 DN |
4792 | /* |
4793 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4794 | */ | |
e1aab161 | 4795 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4796 | { |
3e32cb2e | 4797 | if (!memcg->memory.parent) |
7bcc1bb1 | 4798 | return NULL; |
3e32cb2e | 4799 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4800 | } |
e1aab161 | 4801 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4802 | |
bb4cc1a8 AM |
4803 | static void __init mem_cgroup_soft_limit_tree_init(void) |
4804 | { | |
4805 | struct mem_cgroup_tree_per_node *rtpn; | |
4806 | struct mem_cgroup_tree_per_zone *rtpz; | |
4807 | int tmp, node, zone; | |
4808 | ||
4809 | for_each_node(node) { | |
4810 | tmp = node; | |
4811 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4812 | tmp = -1; | |
4813 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4814 | BUG_ON(!rtpn); | |
4815 | ||
4816 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4817 | ||
4818 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4819 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4820 | rtpz->rb_root = RB_ROOT; | |
4821 | spin_lock_init(&rtpz->lock); | |
4822 | } | |
4823 | } | |
4824 | } | |
4825 | ||
0eb253e2 | 4826 | static struct cgroup_subsys_state * __ref |
eb95419b | 4827 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4828 | { |
d142e3e6 | 4829 | struct mem_cgroup *memcg; |
04046e1a | 4830 | long error = -ENOMEM; |
6d12e2d8 | 4831 | int node; |
8cdea7c0 | 4832 | |
c0ff4b85 R |
4833 | memcg = mem_cgroup_alloc(); |
4834 | if (!memcg) | |
04046e1a | 4835 | return ERR_PTR(error); |
78fb7466 | 4836 | |
3ed28fa1 | 4837 | for_each_node(node) |
c0ff4b85 | 4838 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4839 | goto free_out; |
f64c3f54 | 4840 | |
c077719b | 4841 | /* root ? */ |
eb95419b | 4842 | if (parent_css == NULL) { |
a41c58a6 | 4843 | root_mem_cgroup = memcg; |
3e32cb2e JW |
4844 | page_counter_init(&memcg->memory, NULL); |
4845 | page_counter_init(&memcg->memsw, NULL); | |
4846 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4847 | } |
28dbc4b6 | 4848 | |
d142e3e6 GC |
4849 | memcg->last_scanned_node = MAX_NUMNODES; |
4850 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4851 | memcg->move_charge_at_immigrate = 0; |
4852 | mutex_init(&memcg->thresholds_lock); | |
4853 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4854 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4855 | INIT_LIST_HEAD(&memcg->event_list); |
4856 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
4857 | |
4858 | return &memcg->css; | |
4859 | ||
4860 | free_out: | |
4861 | __mem_cgroup_free(memcg); | |
4862 | return ERR_PTR(error); | |
4863 | } | |
4864 | ||
4865 | static int | |
eb95419b | 4866 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4867 | { |
eb95419b | 4868 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4869 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4870 | int ret; |
d142e3e6 | 4871 | |
15a4c835 | 4872 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4873 | return -ENOSPC; |
4874 | ||
63876986 | 4875 | if (!parent) |
d142e3e6 GC |
4876 | return 0; |
4877 | ||
0999821b | 4878 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4879 | |
4880 | memcg->use_hierarchy = parent->use_hierarchy; | |
4881 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4882 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4883 | ||
4884 | if (parent->use_hierarchy) { | |
3e32cb2e JW |
4885 | page_counter_init(&memcg->memory, &parent->memory); |
4886 | page_counter_init(&memcg->memsw, &parent->memsw); | |
4887 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4888 | |
7bcc1bb1 | 4889 | /* |
8d76a979 LZ |
4890 | * No need to take a reference to the parent because cgroup |
4891 | * core guarantees its existence. | |
7bcc1bb1 | 4892 | */ |
18f59ea7 | 4893 | } else { |
3e32cb2e JW |
4894 | page_counter_init(&memcg->memory, NULL); |
4895 | page_counter_init(&memcg->memsw, NULL); | |
4896 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4897 | /* |
4898 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4899 | * much sense so let cgroup subsystem know about this | |
4900 | * unfortunate state in our controller. | |
4901 | */ | |
d142e3e6 | 4902 | if (parent != root_mem_cgroup) |
073219e9 | 4903 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4904 | } |
0999821b | 4905 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4906 | |
2f7dd7a4 JW |
4907 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4908 | if (ret) | |
4909 | return ret; | |
4910 | ||
4911 | /* | |
4912 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4913 | * orders reading memcg->initialized against its callers | |
4914 | * reading the memcg members. | |
4915 | */ | |
4916 | smp_store_release(&memcg->initialized, 1); | |
4917 | ||
4918 | return 0; | |
8cdea7c0 BS |
4919 | } |
4920 | ||
eb95419b | 4921 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4922 | { |
eb95419b | 4923 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4924 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4925 | |
4926 | /* | |
4927 | * Unregister events and notify userspace. | |
4928 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4929 | * directory to avoid race between userspace and kernelspace. | |
4930 | */ | |
fba94807 TH |
4931 | spin_lock(&memcg->event_list_lock); |
4932 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4933 | list_del_init(&event->list); |
4934 | schedule_work(&event->remove); | |
4935 | } | |
fba94807 | 4936 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4937 | |
776ed0f0 | 4938 | memcg_unregister_all_caches(memcg); |
33cb876e | 4939 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
4940 | } |
4941 | ||
eb95419b | 4942 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4943 | { |
eb95419b | 4944 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4945 | |
10d5ebf4 | 4946 | memcg_destroy_kmem(memcg); |
465939a1 | 4947 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4948 | } |
4949 | ||
1ced953b TH |
4950 | /** |
4951 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4952 | * @css: the target css | |
4953 | * | |
4954 | * Reset the states of the mem_cgroup associated with @css. This is | |
4955 | * invoked when the userland requests disabling on the default hierarchy | |
4956 | * but the memcg is pinned through dependency. The memcg should stop | |
4957 | * applying policies and should revert to the vanilla state as it may be | |
4958 | * made visible again. | |
4959 | * | |
4960 | * The current implementation only resets the essential configurations. | |
4961 | * This needs to be expanded to cover all the visible parts. | |
4962 | */ | |
4963 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4964 | { | |
4965 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4966 | ||
3e32cb2e JW |
4967 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4968 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4969 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
4970 | memcg->soft_limit = 0; | |
1ced953b TH |
4971 | } |
4972 | ||
02491447 | 4973 | #ifdef CONFIG_MMU |
7dc74be0 | 4974 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4975 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4976 | { |
05b84301 | 4977 | int ret; |
9476db97 JW |
4978 | |
4979 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 4980 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 4981 | if (!ret) { |
854ffa8d | 4982 | mc.precharge += count; |
854ffa8d DN |
4983 | return ret; |
4984 | } | |
692e7c45 | 4985 | if (ret == -EINTR) { |
00501b53 | 4986 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
4987 | return ret; |
4988 | } | |
9476db97 JW |
4989 | |
4990 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4991 | while (count--) { |
00501b53 | 4992 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
4993 | /* |
4994 | * In case of failure, any residual charges against | |
4995 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
4996 | * later on. However, cancel any charges that are |
4997 | * bypassed to root right away or they'll be lost. | |
9476db97 | 4998 | */ |
692e7c45 | 4999 | if (ret == -EINTR) |
00501b53 | 5000 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 5001 | if (ret) |
38c5d72f | 5002 | return ret; |
854ffa8d | 5003 | mc.precharge++; |
9476db97 | 5004 | cond_resched(); |
854ffa8d | 5005 | } |
9476db97 | 5006 | return 0; |
4ffef5fe DN |
5007 | } |
5008 | ||
5009 | /** | |
8d32ff84 | 5010 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5011 | * @vma: the vma the pte to be checked belongs |
5012 | * @addr: the address corresponding to the pte to be checked | |
5013 | * @ptent: the pte to be checked | |
02491447 | 5014 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5015 | * |
5016 | * Returns | |
5017 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5018 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5019 | * move charge. if @target is not NULL, the page is stored in target->page | |
5020 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5021 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5022 | * target for charge migration. if @target is not NULL, the entry is stored | |
5023 | * in target->ent. | |
4ffef5fe DN |
5024 | * |
5025 | * Called with pte lock held. | |
5026 | */ | |
4ffef5fe DN |
5027 | union mc_target { |
5028 | struct page *page; | |
02491447 | 5029 | swp_entry_t ent; |
4ffef5fe DN |
5030 | }; |
5031 | ||
4ffef5fe | 5032 | enum mc_target_type { |
8d32ff84 | 5033 | MC_TARGET_NONE = 0, |
4ffef5fe | 5034 | MC_TARGET_PAGE, |
02491447 | 5035 | MC_TARGET_SWAP, |
4ffef5fe DN |
5036 | }; |
5037 | ||
90254a65 DN |
5038 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5039 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5040 | { |
90254a65 | 5041 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5042 | |
90254a65 DN |
5043 | if (!page || !page_mapped(page)) |
5044 | return NULL; | |
5045 | if (PageAnon(page)) { | |
5046 | /* we don't move shared anon */ | |
4b91355e | 5047 | if (!move_anon()) |
90254a65 | 5048 | return NULL; |
87946a72 DN |
5049 | } else if (!move_file()) |
5050 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5051 | return NULL; |
5052 | if (!get_page_unless_zero(page)) | |
5053 | return NULL; | |
5054 | ||
5055 | return page; | |
5056 | } | |
5057 | ||
4b91355e | 5058 | #ifdef CONFIG_SWAP |
90254a65 DN |
5059 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5060 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5061 | { | |
90254a65 DN |
5062 | struct page *page = NULL; |
5063 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5064 | ||
5065 | if (!move_anon() || non_swap_entry(ent)) | |
5066 | return NULL; | |
4b91355e KH |
5067 | /* |
5068 | * Because lookup_swap_cache() updates some statistics counter, | |
5069 | * we call find_get_page() with swapper_space directly. | |
5070 | */ | |
33806f06 | 5071 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
5072 | if (do_swap_account) |
5073 | entry->val = ent.val; | |
5074 | ||
5075 | return page; | |
5076 | } | |
4b91355e KH |
5077 | #else |
5078 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5079 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5080 | { | |
5081 | return NULL; | |
5082 | } | |
5083 | #endif | |
90254a65 | 5084 | |
87946a72 DN |
5085 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5086 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5087 | { | |
5088 | struct page *page = NULL; | |
87946a72 DN |
5089 | struct address_space *mapping; |
5090 | pgoff_t pgoff; | |
5091 | ||
5092 | if (!vma->vm_file) /* anonymous vma */ | |
5093 | return NULL; | |
5094 | if (!move_file()) | |
5095 | return NULL; | |
5096 | ||
87946a72 DN |
5097 | mapping = vma->vm_file->f_mapping; |
5098 | if (pte_none(ptent)) | |
5099 | pgoff = linear_page_index(vma, addr); | |
5100 | else /* pte_file(ptent) is true */ | |
5101 | pgoff = pte_to_pgoff(ptent); | |
5102 | ||
5103 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5104 | #ifdef CONFIG_SWAP |
5105 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5106 | if (shmem_mapping(mapping)) { |
5107 | page = find_get_entry(mapping, pgoff); | |
5108 | if (radix_tree_exceptional_entry(page)) { | |
5109 | swp_entry_t swp = radix_to_swp_entry(page); | |
5110 | if (do_swap_account) | |
5111 | *entry = swp; | |
5112 | page = find_get_page(swap_address_space(swp), swp.val); | |
5113 | } | |
5114 | } else | |
5115 | page = find_get_page(mapping, pgoff); | |
5116 | #else | |
5117 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5118 | #endif |
87946a72 DN |
5119 | return page; |
5120 | } | |
5121 | ||
8d32ff84 | 5122 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5123 | unsigned long addr, pte_t ptent, union mc_target *target) |
5124 | { | |
5125 | struct page *page = NULL; | |
5126 | struct page_cgroup *pc; | |
8d32ff84 | 5127 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5128 | swp_entry_t ent = { .val = 0 }; |
5129 | ||
5130 | if (pte_present(ptent)) | |
5131 | page = mc_handle_present_pte(vma, addr, ptent); | |
5132 | else if (is_swap_pte(ptent)) | |
5133 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5134 | else if (pte_none(ptent) || pte_file(ptent)) |
5135 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5136 | |
5137 | if (!page && !ent.val) | |
8d32ff84 | 5138 | return ret; |
02491447 DN |
5139 | if (page) { |
5140 | pc = lookup_page_cgroup(page); | |
5141 | /* | |
0a31bc97 JW |
5142 | * Do only loose check w/o serialization. |
5143 | * mem_cgroup_move_account() checks the pc is valid or | |
5144 | * not under LRU exclusion. | |
02491447 | 5145 | */ |
29833315 | 5146 | if (pc->mem_cgroup == mc.from) { |
02491447 DN |
5147 | ret = MC_TARGET_PAGE; |
5148 | if (target) | |
5149 | target->page = page; | |
5150 | } | |
5151 | if (!ret || !target) | |
5152 | put_page(page); | |
5153 | } | |
90254a65 DN |
5154 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5155 | if (ent.val && !ret && | |
34c00c31 | 5156 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5157 | ret = MC_TARGET_SWAP; |
5158 | if (target) | |
5159 | target->ent = ent; | |
4ffef5fe | 5160 | } |
4ffef5fe DN |
5161 | return ret; |
5162 | } | |
5163 | ||
12724850 NH |
5164 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5165 | /* | |
5166 | * We don't consider swapping or file mapped pages because THP does not | |
5167 | * support them for now. | |
5168 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5169 | */ | |
5170 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5171 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5172 | { | |
5173 | struct page *page = NULL; | |
5174 | struct page_cgroup *pc; | |
5175 | enum mc_target_type ret = MC_TARGET_NONE; | |
5176 | ||
5177 | page = pmd_page(pmd); | |
309381fe | 5178 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
5179 | if (!move_anon()) |
5180 | return ret; | |
5181 | pc = lookup_page_cgroup(page); | |
29833315 | 5182 | if (pc->mem_cgroup == mc.from) { |
12724850 NH |
5183 | ret = MC_TARGET_PAGE; |
5184 | if (target) { | |
5185 | get_page(page); | |
5186 | target->page = page; | |
5187 | } | |
5188 | } | |
5189 | return ret; | |
5190 | } | |
5191 | #else | |
5192 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5193 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5194 | { | |
5195 | return MC_TARGET_NONE; | |
5196 | } | |
5197 | #endif | |
5198 | ||
4ffef5fe DN |
5199 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5200 | unsigned long addr, unsigned long end, | |
5201 | struct mm_walk *walk) | |
5202 | { | |
5203 | struct vm_area_struct *vma = walk->private; | |
5204 | pte_t *pte; | |
5205 | spinlock_t *ptl; | |
5206 | ||
bf929152 | 5207 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
5208 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5209 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5210 | spin_unlock(ptl); |
1a5a9906 | 5211 | return 0; |
12724850 | 5212 | } |
03319327 | 5213 | |
45f83cef AA |
5214 | if (pmd_trans_unstable(pmd)) |
5215 | return 0; | |
4ffef5fe DN |
5216 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5217 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5218 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5219 | mc.precharge++; /* increment precharge temporarily */ |
5220 | pte_unmap_unlock(pte - 1, ptl); | |
5221 | cond_resched(); | |
5222 | ||
7dc74be0 DN |
5223 | return 0; |
5224 | } | |
5225 | ||
4ffef5fe DN |
5226 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5227 | { | |
5228 | unsigned long precharge; | |
5229 | struct vm_area_struct *vma; | |
5230 | ||
dfe076b0 | 5231 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5232 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5233 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5234 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5235 | .mm = mm, | |
5236 | .private = vma, | |
5237 | }; | |
5238 | if (is_vm_hugetlb_page(vma)) | |
5239 | continue; | |
4ffef5fe DN |
5240 | walk_page_range(vma->vm_start, vma->vm_end, |
5241 | &mem_cgroup_count_precharge_walk); | |
5242 | } | |
dfe076b0 | 5243 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5244 | |
5245 | precharge = mc.precharge; | |
5246 | mc.precharge = 0; | |
5247 | ||
5248 | return precharge; | |
5249 | } | |
5250 | ||
4ffef5fe DN |
5251 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5252 | { | |
dfe076b0 DN |
5253 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5254 | ||
5255 | VM_BUG_ON(mc.moving_task); | |
5256 | mc.moving_task = current; | |
5257 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5258 | } |
5259 | ||
dfe076b0 DN |
5260 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5261 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5262 | { |
2bd9bb20 KH |
5263 | struct mem_cgroup *from = mc.from; |
5264 | struct mem_cgroup *to = mc.to; | |
5265 | ||
4ffef5fe | 5266 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5267 | if (mc.precharge) { |
00501b53 | 5268 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5269 | mc.precharge = 0; |
5270 | } | |
5271 | /* | |
5272 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5273 | * we must uncharge here. | |
5274 | */ | |
5275 | if (mc.moved_charge) { | |
00501b53 | 5276 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5277 | mc.moved_charge = 0; |
4ffef5fe | 5278 | } |
483c30b5 DN |
5279 | /* we must fixup refcnts and charges */ |
5280 | if (mc.moved_swap) { | |
483c30b5 | 5281 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5282 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5283 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5284 | |
05b84301 | 5285 | /* |
3e32cb2e JW |
5286 | * we charged both to->memory and to->memsw, so we |
5287 | * should uncharge to->memory. | |
05b84301 | 5288 | */ |
ce00a967 | 5289 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5290 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5291 | ||
e8ea14cc | 5292 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 5293 | |
4050377b | 5294 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
5295 | mc.moved_swap = 0; |
5296 | } | |
dfe076b0 DN |
5297 | memcg_oom_recover(from); |
5298 | memcg_oom_recover(to); | |
5299 | wake_up_all(&mc.waitq); | |
5300 | } | |
5301 | ||
5302 | static void mem_cgroup_clear_mc(void) | |
5303 | { | |
dfe076b0 DN |
5304 | /* |
5305 | * we must clear moving_task before waking up waiters at the end of | |
5306 | * task migration. | |
5307 | */ | |
5308 | mc.moving_task = NULL; | |
5309 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5310 | spin_lock(&mc.lock); |
4ffef5fe DN |
5311 | mc.from = NULL; |
5312 | mc.to = NULL; | |
2bd9bb20 | 5313 | spin_unlock(&mc.lock); |
4ffef5fe DN |
5314 | } |
5315 | ||
eb95419b | 5316 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5317 | struct cgroup_taskset *tset) |
7dc74be0 | 5318 | { |
2f7ee569 | 5319 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5320 | int ret = 0; |
eb95419b | 5321 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 5322 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 5323 | |
ee5e8472 GC |
5324 | /* |
5325 | * We are now commited to this value whatever it is. Changes in this | |
5326 | * tunable will only affect upcoming migrations, not the current one. | |
5327 | * So we need to save it, and keep it going. | |
5328 | */ | |
5329 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
5330 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
5331 | struct mm_struct *mm; |
5332 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5333 | ||
c0ff4b85 | 5334 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5335 | |
5336 | mm = get_task_mm(p); | |
5337 | if (!mm) | |
5338 | return 0; | |
7dc74be0 | 5339 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5340 | if (mm->owner == p) { |
5341 | VM_BUG_ON(mc.from); | |
5342 | VM_BUG_ON(mc.to); | |
5343 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5344 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5345 | VM_BUG_ON(mc.moved_swap); |
247b1447 | 5346 | |
2bd9bb20 | 5347 | spin_lock(&mc.lock); |
4ffef5fe | 5348 | mc.from = from; |
c0ff4b85 | 5349 | mc.to = memcg; |
ee5e8472 | 5350 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 5351 | spin_unlock(&mc.lock); |
dfe076b0 | 5352 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5353 | |
5354 | ret = mem_cgroup_precharge_mc(mm); | |
5355 | if (ret) | |
5356 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5357 | } |
5358 | mmput(mm); | |
7dc74be0 DN |
5359 | } |
5360 | return ret; | |
5361 | } | |
5362 | ||
eb95419b | 5363 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5364 | struct cgroup_taskset *tset) |
7dc74be0 | 5365 | { |
4e2f245d JW |
5366 | if (mc.to) |
5367 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5368 | } |
5369 | ||
4ffef5fe DN |
5370 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5371 | unsigned long addr, unsigned long end, | |
5372 | struct mm_walk *walk) | |
7dc74be0 | 5373 | { |
4ffef5fe DN |
5374 | int ret = 0; |
5375 | struct vm_area_struct *vma = walk->private; | |
5376 | pte_t *pte; | |
5377 | spinlock_t *ptl; | |
12724850 NH |
5378 | enum mc_target_type target_type; |
5379 | union mc_target target; | |
5380 | struct page *page; | |
5381 | struct page_cgroup *pc; | |
4ffef5fe | 5382 | |
12724850 NH |
5383 | /* |
5384 | * We don't take compound_lock() here but no race with splitting thp | |
5385 | * happens because: | |
5386 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5387 | * under splitting, which means there's no concurrent thp split, | |
5388 | * - if another thread runs into split_huge_page() just after we | |
5389 | * entered this if-block, the thread must wait for page table lock | |
5390 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5391 | * part of thp split is not executed yet. | |
5392 | */ | |
bf929152 | 5393 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5394 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5395 | spin_unlock(ptl); |
12724850 NH |
5396 | return 0; |
5397 | } | |
5398 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5399 | if (target_type == MC_TARGET_PAGE) { | |
5400 | page = target.page; | |
5401 | if (!isolate_lru_page(page)) { | |
5402 | pc = lookup_page_cgroup(page); | |
5403 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5404 | pc, mc.from, mc.to)) { |
12724850 NH |
5405 | mc.precharge -= HPAGE_PMD_NR; |
5406 | mc.moved_charge += HPAGE_PMD_NR; | |
5407 | } | |
5408 | putback_lru_page(page); | |
5409 | } | |
5410 | put_page(page); | |
5411 | } | |
bf929152 | 5412 | spin_unlock(ptl); |
1a5a9906 | 5413 | return 0; |
12724850 NH |
5414 | } |
5415 | ||
45f83cef AA |
5416 | if (pmd_trans_unstable(pmd)) |
5417 | return 0; | |
4ffef5fe DN |
5418 | retry: |
5419 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5420 | for (; addr != end; addr += PAGE_SIZE) { | |
5421 | pte_t ptent = *(pte++); | |
02491447 | 5422 | swp_entry_t ent; |
4ffef5fe DN |
5423 | |
5424 | if (!mc.precharge) | |
5425 | break; | |
5426 | ||
8d32ff84 | 5427 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5428 | case MC_TARGET_PAGE: |
5429 | page = target.page; | |
5430 | if (isolate_lru_page(page)) | |
5431 | goto put; | |
5432 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5433 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5434 | mc.from, mc.to)) { |
4ffef5fe | 5435 | mc.precharge--; |
854ffa8d DN |
5436 | /* we uncharge from mc.from later. */ |
5437 | mc.moved_charge++; | |
4ffef5fe DN |
5438 | } |
5439 | putback_lru_page(page); | |
8d32ff84 | 5440 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5441 | put_page(page); |
5442 | break; | |
02491447 DN |
5443 | case MC_TARGET_SWAP: |
5444 | ent = target.ent; | |
e91cbb42 | 5445 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5446 | mc.precharge--; |
483c30b5 DN |
5447 | /* we fixup refcnts and charges later. */ |
5448 | mc.moved_swap++; | |
5449 | } | |
02491447 | 5450 | break; |
4ffef5fe DN |
5451 | default: |
5452 | break; | |
5453 | } | |
5454 | } | |
5455 | pte_unmap_unlock(pte - 1, ptl); | |
5456 | cond_resched(); | |
5457 | ||
5458 | if (addr != end) { | |
5459 | /* | |
5460 | * We have consumed all precharges we got in can_attach(). | |
5461 | * We try charge one by one, but don't do any additional | |
5462 | * charges to mc.to if we have failed in charge once in attach() | |
5463 | * phase. | |
5464 | */ | |
854ffa8d | 5465 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5466 | if (!ret) |
5467 | goto retry; | |
5468 | } | |
5469 | ||
5470 | return ret; | |
5471 | } | |
5472 | ||
5473 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5474 | { | |
5475 | struct vm_area_struct *vma; | |
5476 | ||
5477 | lru_add_drain_all(); | |
312722cb JW |
5478 | /* |
5479 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5480 | * move_lock while we're moving its pages to another memcg. | |
5481 | * Then wait for already started RCU-only updates to finish. | |
5482 | */ | |
5483 | atomic_inc(&mc.from->moving_account); | |
5484 | synchronize_rcu(); | |
dfe076b0 DN |
5485 | retry: |
5486 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5487 | /* | |
5488 | * Someone who are holding the mmap_sem might be waiting in | |
5489 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5490 | * and retry. Because we cancel precharges, we might not be able | |
5491 | * to move enough charges, but moving charge is a best-effort | |
5492 | * feature anyway, so it wouldn't be a big problem. | |
5493 | */ | |
5494 | __mem_cgroup_clear_mc(); | |
5495 | cond_resched(); | |
5496 | goto retry; | |
5497 | } | |
4ffef5fe DN |
5498 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5499 | int ret; | |
5500 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5501 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5502 | .mm = mm, | |
5503 | .private = vma, | |
5504 | }; | |
5505 | if (is_vm_hugetlb_page(vma)) | |
5506 | continue; | |
4ffef5fe DN |
5507 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5508 | &mem_cgroup_move_charge_walk); | |
5509 | if (ret) | |
5510 | /* | |
5511 | * means we have consumed all precharges and failed in | |
5512 | * doing additional charge. Just abandon here. | |
5513 | */ | |
5514 | break; | |
5515 | } | |
dfe076b0 | 5516 | up_read(&mm->mmap_sem); |
312722cb | 5517 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5518 | } |
5519 | ||
eb95419b | 5520 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5521 | struct cgroup_taskset *tset) |
67e465a7 | 5522 | { |
2f7ee569 | 5523 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5524 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5525 | |
dfe076b0 | 5526 | if (mm) { |
a433658c KM |
5527 | if (mc.to) |
5528 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5529 | mmput(mm); |
5530 | } | |
a433658c KM |
5531 | if (mc.to) |
5532 | mem_cgroup_clear_mc(); | |
67e465a7 | 5533 | } |
5cfb80a7 | 5534 | #else /* !CONFIG_MMU */ |
eb95419b | 5535 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5536 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5537 | { |
5538 | return 0; | |
5539 | } | |
eb95419b | 5540 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5541 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5542 | { |
5543 | } | |
eb95419b | 5544 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5545 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5546 | { |
5547 | } | |
5548 | #endif | |
67e465a7 | 5549 | |
f00baae7 TH |
5550 | /* |
5551 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5552 | * to verify whether we're attached to the default hierarchy on each mount |
5553 | * attempt. | |
f00baae7 | 5554 | */ |
eb95419b | 5555 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5556 | { |
5557 | /* | |
aa6ec29b | 5558 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5559 | * guarantees that @root doesn't have any children, so turning it |
5560 | * on for the root memcg is enough. | |
5561 | */ | |
aa6ec29b | 5562 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 5563 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
5564 | } |
5565 | ||
073219e9 | 5566 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5567 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5568 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5569 | .css_offline = mem_cgroup_css_offline, |
5570 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5571 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5572 | .can_attach = mem_cgroup_can_attach, |
5573 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5574 | .attach = mem_cgroup_move_task, |
f00baae7 | 5575 | .bind = mem_cgroup_bind, |
5577964e | 5576 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 5577 | .early_init = 0, |
8cdea7c0 | 5578 | }; |
c077719b | 5579 | |
c255a458 | 5580 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5581 | static int __init enable_swap_account(char *s) |
5582 | { | |
a2c8990a | 5583 | if (!strcmp(s, "1")) |
a42c390c | 5584 | really_do_swap_account = 1; |
a2c8990a | 5585 | else if (!strcmp(s, "0")) |
a42c390c MH |
5586 | really_do_swap_account = 0; |
5587 | return 1; | |
5588 | } | |
a2c8990a | 5589 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5590 | |
2d11085e MH |
5591 | static void __init memsw_file_init(void) |
5592 | { | |
2cf669a5 TH |
5593 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
5594 | memsw_cgroup_files)); | |
6acc8b02 MH |
5595 | } |
5596 | ||
5597 | static void __init enable_swap_cgroup(void) | |
5598 | { | |
5599 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5600 | do_swap_account = 1; | |
5601 | memsw_file_init(); | |
5602 | } | |
2d11085e | 5603 | } |
6acc8b02 | 5604 | |
2d11085e | 5605 | #else |
6acc8b02 | 5606 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
5607 | { |
5608 | } | |
c077719b | 5609 | #endif |
2d11085e | 5610 | |
0a31bc97 JW |
5611 | #ifdef CONFIG_MEMCG_SWAP |
5612 | /** | |
5613 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5614 | * @page: page whose memsw charge to transfer | |
5615 | * @entry: swap entry to move the charge to | |
5616 | * | |
5617 | * Transfer the memsw charge of @page to @entry. | |
5618 | */ | |
5619 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5620 | { | |
7bdd143c | 5621 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5622 | struct page_cgroup *pc; |
5623 | unsigned short oldid; | |
5624 | ||
5625 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5626 | VM_BUG_ON_PAGE(page_count(page), page); | |
5627 | ||
5628 | if (!do_swap_account) | |
5629 | return; | |
5630 | ||
5631 | pc = lookup_page_cgroup(page); | |
29833315 | 5632 | memcg = pc->mem_cgroup; |
0a31bc97 JW |
5633 | |
5634 | /* Readahead page, never charged */ | |
29833315 | 5635 | if (!memcg) |
0a31bc97 JW |
5636 | return; |
5637 | ||
7bdd143c | 5638 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
0a31bc97 | 5639 | VM_BUG_ON_PAGE(oldid, page); |
7bdd143c JW |
5640 | mem_cgroup_swap_statistics(memcg, true); |
5641 | ||
29833315 | 5642 | pc->mem_cgroup = NULL; |
7bdd143c JW |
5643 | |
5644 | if (!mem_cgroup_is_root(memcg)) | |
5645 | page_counter_uncharge(&memcg->memory, 1); | |
5646 | ||
5647 | /* XXX: caller holds IRQ-safe mapping->tree_lock */ | |
5648 | VM_BUG_ON(!irqs_disabled()); | |
0a31bc97 | 5649 | |
7bdd143c JW |
5650 | mem_cgroup_charge_statistics(memcg, page, -1); |
5651 | memcg_check_events(memcg, page); | |
0a31bc97 JW |
5652 | } |
5653 | ||
5654 | /** | |
5655 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5656 | * @entry: swap entry to uncharge | |
5657 | * | |
5658 | * Drop the memsw charge associated with @entry. | |
5659 | */ | |
5660 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5661 | { | |
5662 | struct mem_cgroup *memcg; | |
5663 | unsigned short id; | |
5664 | ||
5665 | if (!do_swap_account) | |
5666 | return; | |
5667 | ||
5668 | id = swap_cgroup_record(entry, 0); | |
5669 | rcu_read_lock(); | |
5670 | memcg = mem_cgroup_lookup(id); | |
5671 | if (memcg) { | |
ce00a967 | 5672 | if (!mem_cgroup_is_root(memcg)) |
3e32cb2e | 5673 | page_counter_uncharge(&memcg->memsw, 1); |
0a31bc97 JW |
5674 | mem_cgroup_swap_statistics(memcg, false); |
5675 | css_put(&memcg->css); | |
5676 | } | |
5677 | rcu_read_unlock(); | |
5678 | } | |
5679 | #endif | |
5680 | ||
00501b53 JW |
5681 | /** |
5682 | * mem_cgroup_try_charge - try charging a page | |
5683 | * @page: page to charge | |
5684 | * @mm: mm context of the victim | |
5685 | * @gfp_mask: reclaim mode | |
5686 | * @memcgp: charged memcg return | |
5687 | * | |
5688 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5689 | * pages according to @gfp_mask if necessary. | |
5690 | * | |
5691 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5692 | * Otherwise, an error code is returned. | |
5693 | * | |
5694 | * After page->mapping has been set up, the caller must finalize the | |
5695 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5696 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5697 | */ | |
5698 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5699 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5700 | { | |
5701 | struct mem_cgroup *memcg = NULL; | |
5702 | unsigned int nr_pages = 1; | |
5703 | int ret = 0; | |
5704 | ||
5705 | if (mem_cgroup_disabled()) | |
5706 | goto out; | |
5707 | ||
5708 | if (PageSwapCache(page)) { | |
5709 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
5710 | /* | |
5711 | * Every swap fault against a single page tries to charge the | |
5712 | * page, bail as early as possible. shmem_unuse() encounters | |
5713 | * already charged pages, too. The USED bit is protected by | |
5714 | * the page lock, which serializes swap cache removal, which | |
5715 | * in turn serializes uncharging. | |
5716 | */ | |
29833315 | 5717 | if (pc->mem_cgroup) |
00501b53 JW |
5718 | goto out; |
5719 | } | |
5720 | ||
5721 | if (PageTransHuge(page)) { | |
5722 | nr_pages <<= compound_order(page); | |
5723 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5724 | } | |
5725 | ||
5726 | if (do_swap_account && PageSwapCache(page)) | |
5727 | memcg = try_get_mem_cgroup_from_page(page); | |
5728 | if (!memcg) | |
5729 | memcg = get_mem_cgroup_from_mm(mm); | |
5730 | ||
5731 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5732 | ||
5733 | css_put(&memcg->css); | |
5734 | ||
5735 | if (ret == -EINTR) { | |
5736 | memcg = root_mem_cgroup; | |
5737 | ret = 0; | |
5738 | } | |
5739 | out: | |
5740 | *memcgp = memcg; | |
5741 | return ret; | |
5742 | } | |
5743 | ||
5744 | /** | |
5745 | * mem_cgroup_commit_charge - commit a page charge | |
5746 | * @page: page to charge | |
5747 | * @memcg: memcg to charge the page to | |
5748 | * @lrucare: page might be on LRU already | |
5749 | * | |
5750 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5751 | * after page->mapping has been set up. This must happen atomically | |
5752 | * as part of the page instantiation, i.e. under the page table lock | |
5753 | * for anonymous pages, under the page lock for page and swap cache. | |
5754 | * | |
5755 | * In addition, the page must not be on the LRU during the commit, to | |
5756 | * prevent racing with task migration. If it might be, use @lrucare. | |
5757 | * | |
5758 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5759 | */ | |
5760 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5761 | bool lrucare) | |
5762 | { | |
5763 | unsigned int nr_pages = 1; | |
5764 | ||
5765 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5766 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5767 | ||
5768 | if (mem_cgroup_disabled()) | |
5769 | return; | |
5770 | /* | |
5771 | * Swap faults will attempt to charge the same page multiple | |
5772 | * times. But reuse_swap_page() might have removed the page | |
5773 | * from swapcache already, so we can't check PageSwapCache(). | |
5774 | */ | |
5775 | if (!memcg) | |
5776 | return; | |
5777 | ||
6abb5a86 JW |
5778 | commit_charge(page, memcg, lrucare); |
5779 | ||
00501b53 JW |
5780 | if (PageTransHuge(page)) { |
5781 | nr_pages <<= compound_order(page); | |
5782 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5783 | } | |
5784 | ||
6abb5a86 JW |
5785 | local_irq_disable(); |
5786 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5787 | memcg_check_events(memcg, page); | |
5788 | local_irq_enable(); | |
00501b53 JW |
5789 | |
5790 | if (do_swap_account && PageSwapCache(page)) { | |
5791 | swp_entry_t entry = { .val = page_private(page) }; | |
5792 | /* | |
5793 | * The swap entry might not get freed for a long time, | |
5794 | * let's not wait for it. The page already received a | |
5795 | * memory+swap charge, drop the swap entry duplicate. | |
5796 | */ | |
5797 | mem_cgroup_uncharge_swap(entry); | |
5798 | } | |
5799 | } | |
5800 | ||
5801 | /** | |
5802 | * mem_cgroup_cancel_charge - cancel a page charge | |
5803 | * @page: page to charge | |
5804 | * @memcg: memcg to charge the page to | |
5805 | * | |
5806 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5807 | */ | |
5808 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5809 | { | |
5810 | unsigned int nr_pages = 1; | |
5811 | ||
5812 | if (mem_cgroup_disabled()) | |
5813 | return; | |
5814 | /* | |
5815 | * Swap faults will attempt to charge the same page multiple | |
5816 | * times. But reuse_swap_page() might have removed the page | |
5817 | * from swapcache already, so we can't check PageSwapCache(). | |
5818 | */ | |
5819 | if (!memcg) | |
5820 | return; | |
5821 | ||
5822 | if (PageTransHuge(page)) { | |
5823 | nr_pages <<= compound_order(page); | |
5824 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5825 | } | |
5826 | ||
5827 | cancel_charge(memcg, nr_pages); | |
5828 | } | |
5829 | ||
747db954 | 5830 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5831 | unsigned long nr_anon, unsigned long nr_file, |
5832 | unsigned long nr_huge, struct page *dummy_page) | |
5833 | { | |
18eca2e6 | 5834 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5835 | unsigned long flags; |
5836 | ||
ce00a967 | 5837 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5838 | page_counter_uncharge(&memcg->memory, nr_pages); |
5839 | if (do_swap_account) | |
5840 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5841 | memcg_oom_recover(memcg); |
5842 | } | |
747db954 JW |
5843 | |
5844 | local_irq_save(flags); | |
5845 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5846 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5847 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5848 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5849 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5850 | memcg_check_events(memcg, dummy_page); |
5851 | local_irq_restore(flags); | |
e8ea14cc JW |
5852 | |
5853 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5854 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5855 | } |
5856 | ||
5857 | static void uncharge_list(struct list_head *page_list) | |
5858 | { | |
5859 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5860 | unsigned long nr_anon = 0; |
5861 | unsigned long nr_file = 0; | |
5862 | unsigned long nr_huge = 0; | |
5863 | unsigned long pgpgout = 0; | |
747db954 JW |
5864 | struct list_head *next; |
5865 | struct page *page; | |
5866 | ||
5867 | next = page_list->next; | |
5868 | do { | |
5869 | unsigned int nr_pages = 1; | |
5870 | struct page_cgroup *pc; | |
5871 | ||
5872 | page = list_entry(next, struct page, lru); | |
5873 | next = page->lru.next; | |
5874 | ||
5875 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5876 | VM_BUG_ON_PAGE(page_count(page), page); | |
5877 | ||
5878 | pc = lookup_page_cgroup(page); | |
29833315 | 5879 | if (!pc->mem_cgroup) |
747db954 JW |
5880 | continue; |
5881 | ||
5882 | /* | |
5883 | * Nobody should be changing or seriously looking at | |
29833315 JW |
5884 | * pc->mem_cgroup at this point, we have fully |
5885 | * exclusive access to the page. | |
747db954 JW |
5886 | */ |
5887 | ||
5888 | if (memcg != pc->mem_cgroup) { | |
5889 | if (memcg) { | |
18eca2e6 JW |
5890 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5891 | nr_huge, page); | |
5892 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 JW |
5893 | } |
5894 | memcg = pc->mem_cgroup; | |
5895 | } | |
5896 | ||
5897 | if (PageTransHuge(page)) { | |
5898 | nr_pages <<= compound_order(page); | |
5899 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5900 | nr_huge += nr_pages; | |
5901 | } | |
5902 | ||
5903 | if (PageAnon(page)) | |
5904 | nr_anon += nr_pages; | |
5905 | else | |
5906 | nr_file += nr_pages; | |
5907 | ||
29833315 | 5908 | pc->mem_cgroup = NULL; |
747db954 JW |
5909 | |
5910 | pgpgout++; | |
5911 | } while (next != page_list); | |
5912 | ||
5913 | if (memcg) | |
18eca2e6 JW |
5914 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5915 | nr_huge, page); | |
747db954 JW |
5916 | } |
5917 | ||
0a31bc97 JW |
5918 | /** |
5919 | * mem_cgroup_uncharge - uncharge a page | |
5920 | * @page: page to uncharge | |
5921 | * | |
5922 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5923 | * mem_cgroup_commit_charge(). | |
5924 | */ | |
5925 | void mem_cgroup_uncharge(struct page *page) | |
5926 | { | |
0a31bc97 | 5927 | struct page_cgroup *pc; |
0a31bc97 JW |
5928 | |
5929 | if (mem_cgroup_disabled()) | |
5930 | return; | |
5931 | ||
747db954 | 5932 | /* Don't touch page->lru of any random page, pre-check: */ |
0a31bc97 | 5933 | pc = lookup_page_cgroup(page); |
29833315 | 5934 | if (!pc->mem_cgroup) |
0a31bc97 JW |
5935 | return; |
5936 | ||
747db954 JW |
5937 | INIT_LIST_HEAD(&page->lru); |
5938 | uncharge_list(&page->lru); | |
5939 | } | |
0a31bc97 | 5940 | |
747db954 JW |
5941 | /** |
5942 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5943 | * @page_list: list of pages to uncharge | |
5944 | * | |
5945 | * Uncharge a list of pages previously charged with | |
5946 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5947 | */ | |
5948 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5949 | { | |
5950 | if (mem_cgroup_disabled()) | |
5951 | return; | |
0a31bc97 | 5952 | |
747db954 JW |
5953 | if (!list_empty(page_list)) |
5954 | uncharge_list(page_list); | |
0a31bc97 JW |
5955 | } |
5956 | ||
5957 | /** | |
5958 | * mem_cgroup_migrate - migrate a charge to another page | |
5959 | * @oldpage: currently charged page | |
5960 | * @newpage: page to transfer the charge to | |
5961 | * @lrucare: both pages might be on the LRU already | |
5962 | * | |
5963 | * Migrate the charge from @oldpage to @newpage. | |
5964 | * | |
5965 | * Both pages must be locked, @newpage->mapping must be set up. | |
5966 | */ | |
5967 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
5968 | bool lrucare) | |
5969 | { | |
29833315 | 5970 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5971 | struct page_cgroup *pc; |
5972 | int isolated; | |
5973 | ||
5974 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5975 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
5976 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
5977 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
5978 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
5979 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5980 | newpage); | |
0a31bc97 JW |
5981 | |
5982 | if (mem_cgroup_disabled()) | |
5983 | return; | |
5984 | ||
5985 | /* Page cache replacement: new page already charged? */ | |
5986 | pc = lookup_page_cgroup(newpage); | |
29833315 | 5987 | if (pc->mem_cgroup) |
0a31bc97 JW |
5988 | return; |
5989 | ||
7d5e3245 JW |
5990 | /* |
5991 | * Swapcache readahead pages can get migrated before being | |
5992 | * charged, and migration from compaction can happen to an | |
5993 | * uncharged page when the PFN walker finds a page that | |
5994 | * reclaim just put back on the LRU but has not released yet. | |
5995 | */ | |
0a31bc97 | 5996 | pc = lookup_page_cgroup(oldpage); |
29833315 JW |
5997 | memcg = pc->mem_cgroup; |
5998 | if (!memcg) | |
0a31bc97 JW |
5999 | return; |
6000 | ||
0a31bc97 JW |
6001 | if (lrucare) |
6002 | lock_page_lru(oldpage, &isolated); | |
6003 | ||
29833315 | 6004 | pc->mem_cgroup = NULL; |
0a31bc97 JW |
6005 | |
6006 | if (lrucare) | |
6007 | unlock_page_lru(oldpage, isolated); | |
6008 | ||
29833315 | 6009 | commit_charge(newpage, memcg, lrucare); |
0a31bc97 JW |
6010 | } |
6011 | ||
2d11085e | 6012 | /* |
1081312f MH |
6013 | * subsys_initcall() for memory controller. |
6014 | * | |
6015 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6016 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6017 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6018 | * be initialized from here. | |
2d11085e MH |
6019 | */ |
6020 | static int __init mem_cgroup_init(void) | |
6021 | { | |
6022 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6023 | enable_swap_cgroup(); |
bb4cc1a8 | 6024 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6025 | memcg_stock_init(); |
2d11085e MH |
6026 | return 0; |
6027 | } | |
6028 | subsys_initcall(mem_cgroup_init); |