]>
Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
8cdea7c0 BS |
17 | * This program is free software; you can redistribute it and/or modify |
18 | * it under the terms of the GNU General Public License as published by | |
19 | * the Free Software Foundation; either version 2 of the License, or | |
20 | * (at your option) any later version. | |
21 | * | |
22 | * This program is distributed in the hope that it will be useful, | |
23 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
24 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
25 | * GNU General Public License for more details. | |
26 | */ | |
27 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
29 | #include <linux/memcontrol.h> |
30 | #include <linux/cgroup.h> | |
78fb7466 | 31 | #include <linux/mm.h> |
4ffef5fe | 32 | #include <linux/hugetlb.h> |
d13d1443 | 33 | #include <linux/pagemap.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
bb4cc1a8 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 | 47 | #include <linux/eventfd.h> |
79bd9814 | 48 | #include <linux/poll.h> |
2e72b634 | 49 | #include <linux/sort.h> |
66e1707b | 50 | #include <linux/fs.h> |
d2ceb9b7 | 51 | #include <linux/seq_file.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
52d4b9ac | 54 | #include <linux/page_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
79bd9814 | 58 | #include <linux/file.h> |
08e552c6 | 59 | #include "internal.h" |
d1a4c0b3 | 60 | #include <net/sock.h> |
4bd2c1ee | 61 | #include <net/ip.h> |
d1a4c0b3 | 62 | #include <net/tcp_memcontrol.h> |
f35c3a8e | 63 | #include "slab.h" |
8cdea7c0 | 64 | |
8697d331 BS |
65 | #include <asm/uaccess.h> |
66 | ||
cc8e970c KM |
67 | #include <trace/events/vmscan.h> |
68 | ||
073219e9 TH |
69 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
70 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 71 | |
a181b0e8 | 72 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 73 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 74 | |
c255a458 | 75 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 76 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 77 | int do_swap_account __read_mostly; |
a42c390c MH |
78 | |
79 | /* for remember boot option*/ | |
c255a458 | 80 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
81 | static int really_do_swap_account __initdata = 1; |
82 | #else | |
ada4ba59 | 83 | static int really_do_swap_account __initdata; |
a42c390c MH |
84 | #endif |
85 | ||
c077719b | 86 | #else |
a0db00fc | 87 | #define do_swap_account 0 |
c077719b KH |
88 | #endif |
89 | ||
90 | ||
af7c4b0e JW |
91 | static const char * const mem_cgroup_stat_names[] = { |
92 | "cache", | |
93 | "rss", | |
b070e65c | 94 | "rss_huge", |
af7c4b0e | 95 | "mapped_file", |
3ea67d06 | 96 | "writeback", |
af7c4b0e JW |
97 | "swap", |
98 | }; | |
99 | ||
e9f8974f JW |
100 | enum mem_cgroup_events_index { |
101 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
102 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
103 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
104 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
105 | MEM_CGROUP_EVENTS_NSTATS, |
106 | }; | |
af7c4b0e JW |
107 | |
108 | static const char * const mem_cgroup_events_names[] = { | |
109 | "pgpgin", | |
110 | "pgpgout", | |
111 | "pgfault", | |
112 | "pgmajfault", | |
113 | }; | |
114 | ||
58cf188e SZ |
115 | static const char * const mem_cgroup_lru_names[] = { |
116 | "inactive_anon", | |
117 | "active_anon", | |
118 | "inactive_file", | |
119 | "active_file", | |
120 | "unevictable", | |
121 | }; | |
122 | ||
7a159cc9 JW |
123 | /* |
124 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
125 | * it will be incremated by the number of pages. This counter is used for | |
126 | * for trigger some periodic events. This is straightforward and better | |
127 | * than using jiffies etc. to handle periodic memcg event. | |
128 | */ | |
129 | enum mem_cgroup_events_target { | |
130 | MEM_CGROUP_TARGET_THRESH, | |
bb4cc1a8 | 131 | MEM_CGROUP_TARGET_SOFTLIMIT, |
453a9bf3 | 132 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
133 | MEM_CGROUP_NTARGETS, |
134 | }; | |
a0db00fc KS |
135 | #define THRESHOLDS_EVENTS_TARGET 128 |
136 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
137 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 138 | |
d52aa412 | 139 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 140 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 141 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 142 | unsigned long nr_page_events; |
7a159cc9 | 143 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
144 | }; |
145 | ||
5ac8fb31 JW |
146 | struct reclaim_iter { |
147 | struct mem_cgroup *position; | |
527a5ec9 JW |
148 | /* scan generation, increased every round-trip */ |
149 | unsigned int generation; | |
150 | }; | |
151 | ||
6d12e2d8 KH |
152 | /* |
153 | * per-zone information in memory controller. | |
154 | */ | |
6d12e2d8 | 155 | struct mem_cgroup_per_zone { |
6290df54 | 156 | struct lruvec lruvec; |
1eb49272 | 157 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 158 | |
5ac8fb31 | 159 | struct reclaim_iter iter[DEF_PRIORITY + 1]; |
527a5ec9 | 160 | |
bb4cc1a8 | 161 | struct rb_node tree_node; /* RB tree node */ |
3e32cb2e | 162 | unsigned long usage_in_excess;/* Set to the value by which */ |
bb4cc1a8 AM |
163 | /* the soft limit is exceeded*/ |
164 | bool on_tree; | |
d79154bb | 165 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 166 | /* use container_of */ |
6d12e2d8 | 167 | }; |
6d12e2d8 KH |
168 | |
169 | struct mem_cgroup_per_node { | |
170 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
171 | }; | |
172 | ||
bb4cc1a8 AM |
173 | /* |
174 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
175 | * their hierarchy representation | |
176 | */ | |
177 | ||
178 | struct mem_cgroup_tree_per_zone { | |
179 | struct rb_root rb_root; | |
180 | spinlock_t lock; | |
181 | }; | |
182 | ||
183 | struct mem_cgroup_tree_per_node { | |
184 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
185 | }; | |
186 | ||
187 | struct mem_cgroup_tree { | |
188 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
189 | }; | |
190 | ||
191 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
192 | ||
2e72b634 KS |
193 | struct mem_cgroup_threshold { |
194 | struct eventfd_ctx *eventfd; | |
3e32cb2e | 195 | unsigned long threshold; |
2e72b634 KS |
196 | }; |
197 | ||
9490ff27 | 198 | /* For threshold */ |
2e72b634 | 199 | struct mem_cgroup_threshold_ary { |
748dad36 | 200 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 201 | int current_threshold; |
2e72b634 KS |
202 | /* Size of entries[] */ |
203 | unsigned int size; | |
204 | /* Array of thresholds */ | |
205 | struct mem_cgroup_threshold entries[0]; | |
206 | }; | |
2c488db2 KS |
207 | |
208 | struct mem_cgroup_thresholds { | |
209 | /* Primary thresholds array */ | |
210 | struct mem_cgroup_threshold_ary *primary; | |
211 | /* | |
212 | * Spare threshold array. | |
213 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
214 | * It must be able to store at least primary->size - 1 entries. | |
215 | */ | |
216 | struct mem_cgroup_threshold_ary *spare; | |
217 | }; | |
218 | ||
9490ff27 KH |
219 | /* for OOM */ |
220 | struct mem_cgroup_eventfd_list { | |
221 | struct list_head list; | |
222 | struct eventfd_ctx *eventfd; | |
223 | }; | |
2e72b634 | 224 | |
79bd9814 TH |
225 | /* |
226 | * cgroup_event represents events which userspace want to receive. | |
227 | */ | |
3bc942f3 | 228 | struct mem_cgroup_event { |
79bd9814 | 229 | /* |
59b6f873 | 230 | * memcg which the event belongs to. |
79bd9814 | 231 | */ |
59b6f873 | 232 | struct mem_cgroup *memcg; |
79bd9814 TH |
233 | /* |
234 | * eventfd to signal userspace about the event. | |
235 | */ | |
236 | struct eventfd_ctx *eventfd; | |
237 | /* | |
238 | * Each of these stored in a list by the cgroup. | |
239 | */ | |
240 | struct list_head list; | |
fba94807 TH |
241 | /* |
242 | * register_event() callback will be used to add new userspace | |
243 | * waiter for changes related to this event. Use eventfd_signal() | |
244 | * on eventfd to send notification to userspace. | |
245 | */ | |
59b6f873 | 246 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 247 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
248 | /* |
249 | * unregister_event() callback will be called when userspace closes | |
250 | * the eventfd or on cgroup removing. This callback must be set, | |
251 | * if you want provide notification functionality. | |
252 | */ | |
59b6f873 | 253 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 254 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
255 | /* |
256 | * All fields below needed to unregister event when | |
257 | * userspace closes eventfd. | |
258 | */ | |
259 | poll_table pt; | |
260 | wait_queue_head_t *wqh; | |
261 | wait_queue_t wait; | |
262 | struct work_struct remove; | |
263 | }; | |
264 | ||
c0ff4b85 R |
265 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
266 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 267 | |
8cdea7c0 BS |
268 | /* |
269 | * The memory controller data structure. The memory controller controls both | |
270 | * page cache and RSS per cgroup. We would eventually like to provide | |
271 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
272 | * to help the administrator determine what knobs to tune. | |
273 | * | |
274 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
275 | * we hit the water mark. May be even add a low water mark, such that |
276 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
277 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
278 | */ |
279 | struct mem_cgroup { | |
280 | struct cgroup_subsys_state css; | |
3e32cb2e JW |
281 | |
282 | /* Accounted resources */ | |
283 | struct page_counter memory; | |
284 | struct page_counter memsw; | |
285 | struct page_counter kmem; | |
286 | ||
287 | unsigned long soft_limit; | |
59927fb9 | 288 | |
70ddf637 AV |
289 | /* vmpressure notifications */ |
290 | struct vmpressure vmpressure; | |
291 | ||
2f7dd7a4 JW |
292 | /* css_online() has been completed */ |
293 | int initialized; | |
294 | ||
18f59ea7 BS |
295 | /* |
296 | * Should the accounting and control be hierarchical, per subtree? | |
297 | */ | |
298 | bool use_hierarchy; | |
510fc4e1 | 299 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
300 | |
301 | bool oom_lock; | |
302 | atomic_t under_oom; | |
3812c8c8 | 303 | atomic_t oom_wakeups; |
79dfdacc | 304 | |
1f4c025b | 305 | int swappiness; |
3c11ecf4 KH |
306 | /* OOM-Killer disable */ |
307 | int oom_kill_disable; | |
a7885eb8 | 308 | |
2e72b634 KS |
309 | /* protect arrays of thresholds */ |
310 | struct mutex thresholds_lock; | |
311 | ||
312 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 313 | struct mem_cgroup_thresholds thresholds; |
907860ed | 314 | |
2e72b634 | 315 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 316 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 317 | |
9490ff27 KH |
318 | /* For oom notifier event fd */ |
319 | struct list_head oom_notify; | |
185efc0f | 320 | |
7dc74be0 DN |
321 | /* |
322 | * Should we move charges of a task when a task is moved into this | |
323 | * mem_cgroup ? And what type of charges should we move ? | |
324 | */ | |
f894ffa8 | 325 | unsigned long move_charge_at_immigrate; |
619d094b KH |
326 | /* |
327 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
328 | */ | |
329 | atomic_t moving_account; | |
312734c0 KH |
330 | /* taken only while moving_account > 0 */ |
331 | spinlock_t move_lock; | |
d52aa412 | 332 | /* |
c62b1a3b | 333 | * percpu counter. |
d52aa412 | 334 | */ |
3a7951b4 | 335 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
336 | /* |
337 | * used when a cpu is offlined or other synchronizations | |
338 | * See mem_cgroup_read_stat(). | |
339 | */ | |
340 | struct mem_cgroup_stat_cpu nocpu_base; | |
341 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 342 | |
4bd2c1ee | 343 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
2e685cad | 344 | struct cg_proto tcp_mem; |
d1a4c0b3 | 345 | #endif |
2633d7a0 | 346 | #if defined(CONFIG_MEMCG_KMEM) |
bd673145 VD |
347 | /* analogous to slab_common's slab_caches list, but per-memcg; |
348 | * protected by memcg_slab_mutex */ | |
2633d7a0 | 349 | struct list_head memcg_slab_caches; |
2633d7a0 GC |
350 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ |
351 | int kmemcg_id; | |
352 | #endif | |
45cf7ebd GC |
353 | |
354 | int last_scanned_node; | |
355 | #if MAX_NUMNODES > 1 | |
356 | nodemask_t scan_nodes; | |
357 | atomic_t numainfo_events; | |
358 | atomic_t numainfo_updating; | |
359 | #endif | |
70ddf637 | 360 | |
fba94807 TH |
361 | /* List of events which userspace want to receive */ |
362 | struct list_head event_list; | |
363 | spinlock_t event_list_lock; | |
364 | ||
54f72fe0 JW |
365 | struct mem_cgroup_per_node *nodeinfo[0]; |
366 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
367 | }; |
368 | ||
510fc4e1 GC |
369 | /* internal only representation about the status of kmem accounting. */ |
370 | enum { | |
6de64beb | 371 | KMEM_ACCOUNTED_ACTIVE, /* accounted by this cgroup itself */ |
510fc4e1 GC |
372 | }; |
373 | ||
510fc4e1 GC |
374 | #ifdef CONFIG_MEMCG_KMEM |
375 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
376 | { | |
377 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
378 | } | |
7de37682 GC |
379 | |
380 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
381 | { | |
382 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
383 | } | |
384 | ||
510fc4e1 GC |
385 | #endif |
386 | ||
7dc74be0 DN |
387 | /* Stuffs for move charges at task migration. */ |
388 | /* | |
ee5e8472 GC |
389 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
390 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
391 | */ |
392 | enum move_type { | |
4ffef5fe | 393 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 394 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
395 | NR_MOVE_TYPE, |
396 | }; | |
397 | ||
4ffef5fe DN |
398 | /* "mc" and its members are protected by cgroup_mutex */ |
399 | static struct move_charge_struct { | |
b1dd693e | 400 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
401 | struct mem_cgroup *from; |
402 | struct mem_cgroup *to; | |
ee5e8472 | 403 | unsigned long immigrate_flags; |
4ffef5fe | 404 | unsigned long precharge; |
854ffa8d | 405 | unsigned long moved_charge; |
483c30b5 | 406 | unsigned long moved_swap; |
8033b97c DN |
407 | struct task_struct *moving_task; /* a task moving charges */ |
408 | wait_queue_head_t waitq; /* a waitq for other context */ | |
409 | } mc = { | |
2bd9bb20 | 410 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
411 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
412 | }; | |
4ffef5fe | 413 | |
90254a65 DN |
414 | static bool move_anon(void) |
415 | { | |
ee5e8472 | 416 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
417 | } |
418 | ||
87946a72 DN |
419 | static bool move_file(void) |
420 | { | |
ee5e8472 | 421 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
422 | } |
423 | ||
4e416953 BS |
424 | /* |
425 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
426 | * limit reclaim to prevent infinite loops, if they ever occur. | |
427 | */ | |
a0db00fc | 428 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 429 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 430 | |
217bc319 KH |
431 | enum charge_type { |
432 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 433 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 434 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 435 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
436 | NR_CHARGE_TYPE, |
437 | }; | |
438 | ||
8c7c6e34 | 439 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
440 | enum res_type { |
441 | _MEM, | |
442 | _MEMSWAP, | |
443 | _OOM_TYPE, | |
510fc4e1 | 444 | _KMEM, |
86ae53e1 GC |
445 | }; |
446 | ||
a0db00fc KS |
447 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
448 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 449 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
450 | /* Used for OOM nofiier */ |
451 | #define OOM_CONTROL (0) | |
8c7c6e34 | 452 | |
0999821b GC |
453 | /* |
454 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
455 | * As a consequence, any change that needs to protect against new child cgroups | |
456 | * appearing has to hold it as well. | |
457 | */ | |
458 | static DEFINE_MUTEX(memcg_create_mutex); | |
459 | ||
b2145145 WL |
460 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
461 | { | |
a7c6d554 | 462 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
463 | } |
464 | ||
70ddf637 AV |
465 | /* Some nice accessors for the vmpressure. */ |
466 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
467 | { | |
468 | if (!memcg) | |
469 | memcg = root_mem_cgroup; | |
470 | return &memcg->vmpressure; | |
471 | } | |
472 | ||
473 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
474 | { | |
475 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
476 | } | |
477 | ||
7ffc0edc MH |
478 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
479 | { | |
480 | return (memcg == root_mem_cgroup); | |
481 | } | |
482 | ||
4219b2da LZ |
483 | /* |
484 | * We restrict the id in the range of [1, 65535], so it can fit into | |
485 | * an unsigned short. | |
486 | */ | |
487 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
488 | ||
34c00c31 LZ |
489 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
490 | { | |
15a4c835 | 491 | return memcg->css.id; |
34c00c31 LZ |
492 | } |
493 | ||
494 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
495 | { | |
496 | struct cgroup_subsys_state *css; | |
497 | ||
7d699ddb | 498 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
499 | return mem_cgroup_from_css(css); |
500 | } | |
501 | ||
e1aab161 | 502 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 503 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 504 | |
e1aab161 GC |
505 | void sock_update_memcg(struct sock *sk) |
506 | { | |
376be5ff | 507 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 508 | struct mem_cgroup *memcg; |
3f134619 | 509 | struct cg_proto *cg_proto; |
e1aab161 GC |
510 | |
511 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
512 | ||
f3f511e1 GC |
513 | /* Socket cloning can throw us here with sk_cgrp already |
514 | * filled. It won't however, necessarily happen from | |
515 | * process context. So the test for root memcg given | |
516 | * the current task's memcg won't help us in this case. | |
517 | * | |
518 | * Respecting the original socket's memcg is a better | |
519 | * decision in this case. | |
520 | */ | |
521 | if (sk->sk_cgrp) { | |
522 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 523 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
524 | return; |
525 | } | |
526 | ||
e1aab161 GC |
527 | rcu_read_lock(); |
528 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 529 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 530 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
531 | memcg_proto_active(cg_proto) && |
532 | css_tryget_online(&memcg->css)) { | |
3f134619 | 533 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
534 | } |
535 | rcu_read_unlock(); | |
536 | } | |
537 | } | |
538 | EXPORT_SYMBOL(sock_update_memcg); | |
539 | ||
540 | void sock_release_memcg(struct sock *sk) | |
541 | { | |
376be5ff | 542 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
543 | struct mem_cgroup *memcg; |
544 | WARN_ON(!sk->sk_cgrp->memcg); | |
545 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 546 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
547 | } |
548 | } | |
d1a4c0b3 GC |
549 | |
550 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
551 | { | |
552 | if (!memcg || mem_cgroup_is_root(memcg)) | |
553 | return NULL; | |
554 | ||
2e685cad | 555 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
556 | } |
557 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 558 | |
3f134619 GC |
559 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
560 | { | |
2e685cad | 561 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
562 | return; |
563 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
564 | } | |
565 | #else | |
566 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
567 | { | |
568 | } | |
569 | #endif | |
570 | ||
a8964b9b | 571 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
572 | /* |
573 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
b8627835 LZ |
574 | * The main reason for not using cgroup id for this: |
575 | * this works better in sparse environments, where we have a lot of memcgs, | |
576 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
577 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
578 | * 200 entry array for that. | |
55007d84 GC |
579 | * |
580 | * The current size of the caches array is stored in | |
581 | * memcg_limited_groups_array_size. It will double each time we have to | |
582 | * increase it. | |
583 | */ | |
584 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
585 | int memcg_limited_groups_array_size; |
586 | ||
55007d84 GC |
587 | /* |
588 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
589 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
590 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
591 | * tunable, but that is strictly not necessary. | |
592 | * | |
b8627835 | 593 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
594 | * this constant directly from cgroup, but it is understandable that this is |
595 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 596 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
597 | * increase ours as well if it increases. |
598 | */ | |
599 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 600 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 601 | |
d7f25f8a GC |
602 | /* |
603 | * A lot of the calls to the cache allocation functions are expected to be | |
604 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
605 | * conditional to this static branch, we'll have to allow modules that does | |
606 | * kmem_cache_alloc and the such to see this symbol as well | |
607 | */ | |
a8964b9b | 608 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 609 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 610 | |
f3bb3043 VD |
611 | static void memcg_free_cache_id(int id); |
612 | ||
a8964b9b GC |
613 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
614 | { | |
55007d84 | 615 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 616 | static_key_slow_dec(&memcg_kmem_enabled_key); |
f3bb3043 | 617 | memcg_free_cache_id(memcg->kmemcg_id); |
55007d84 | 618 | } |
bea207c8 GC |
619 | /* |
620 | * This check can't live in kmem destruction function, | |
621 | * since the charges will outlive the cgroup | |
622 | */ | |
3e32cb2e | 623 | WARN_ON(page_counter_read(&memcg->kmem)); |
a8964b9b GC |
624 | } |
625 | #else | |
626 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
627 | { | |
628 | } | |
629 | #endif /* CONFIG_MEMCG_KMEM */ | |
630 | ||
631 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
632 | { | |
633 | disarm_sock_keys(memcg); | |
634 | disarm_kmem_keys(memcg); | |
635 | } | |
636 | ||
f64c3f54 | 637 | static struct mem_cgroup_per_zone * |
e231875b | 638 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 639 | { |
e231875b JZ |
640 | int nid = zone_to_nid(zone); |
641 | int zid = zone_idx(zone); | |
642 | ||
54f72fe0 | 643 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
644 | } |
645 | ||
c0ff4b85 | 646 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 647 | { |
c0ff4b85 | 648 | return &memcg->css; |
d324236b WF |
649 | } |
650 | ||
f64c3f54 | 651 | static struct mem_cgroup_per_zone * |
e231875b | 652 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 653 | { |
97a6c37b JW |
654 | int nid = page_to_nid(page); |
655 | int zid = page_zonenum(page); | |
f64c3f54 | 656 | |
e231875b | 657 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
658 | } |
659 | ||
bb4cc1a8 AM |
660 | static struct mem_cgroup_tree_per_zone * |
661 | soft_limit_tree_node_zone(int nid, int zid) | |
662 | { | |
663 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
664 | } | |
665 | ||
666 | static struct mem_cgroup_tree_per_zone * | |
667 | soft_limit_tree_from_page(struct page *page) | |
668 | { | |
669 | int nid = page_to_nid(page); | |
670 | int zid = page_zonenum(page); | |
671 | ||
672 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
673 | } | |
674 | ||
cf2c8127 JW |
675 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
676 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 677 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
678 | { |
679 | struct rb_node **p = &mctz->rb_root.rb_node; | |
680 | struct rb_node *parent = NULL; | |
681 | struct mem_cgroup_per_zone *mz_node; | |
682 | ||
683 | if (mz->on_tree) | |
684 | return; | |
685 | ||
686 | mz->usage_in_excess = new_usage_in_excess; | |
687 | if (!mz->usage_in_excess) | |
688 | return; | |
689 | while (*p) { | |
690 | parent = *p; | |
691 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
692 | tree_node); | |
693 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
694 | p = &(*p)->rb_left; | |
695 | /* | |
696 | * We can't avoid mem cgroups that are over their soft | |
697 | * limit by the same amount | |
698 | */ | |
699 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
700 | p = &(*p)->rb_right; | |
701 | } | |
702 | rb_link_node(&mz->tree_node, parent, p); | |
703 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
704 | mz->on_tree = true; | |
705 | } | |
706 | ||
cf2c8127 JW |
707 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
708 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
709 | { |
710 | if (!mz->on_tree) | |
711 | return; | |
712 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
713 | mz->on_tree = false; | |
714 | } | |
715 | ||
cf2c8127 JW |
716 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
717 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 718 | { |
0a31bc97 JW |
719 | unsigned long flags; |
720 | ||
721 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 722 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 723 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
724 | } |
725 | ||
3e32cb2e JW |
726 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
727 | { | |
728 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
729 | unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit); | |
730 | unsigned long excess = 0; | |
731 | ||
732 | if (nr_pages > soft_limit) | |
733 | excess = nr_pages - soft_limit; | |
734 | ||
735 | return excess; | |
736 | } | |
bb4cc1a8 AM |
737 | |
738 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
739 | { | |
3e32cb2e | 740 | unsigned long excess; |
bb4cc1a8 AM |
741 | struct mem_cgroup_per_zone *mz; |
742 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 743 | |
e231875b | 744 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
745 | /* |
746 | * Necessary to update all ancestors when hierarchy is used. | |
747 | * because their event counter is not touched. | |
748 | */ | |
749 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 750 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 751 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
752 | /* |
753 | * We have to update the tree if mz is on RB-tree or | |
754 | * mem is over its softlimit. | |
755 | */ | |
756 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
757 | unsigned long flags; |
758 | ||
759 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
760 | /* if on-tree, remove it */ |
761 | if (mz->on_tree) | |
cf2c8127 | 762 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
763 | /* |
764 | * Insert again. mz->usage_in_excess will be updated. | |
765 | * If excess is 0, no tree ops. | |
766 | */ | |
cf2c8127 | 767 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 768 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
769 | } |
770 | } | |
771 | } | |
772 | ||
773 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
774 | { | |
bb4cc1a8 | 775 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
776 | struct mem_cgroup_per_zone *mz; |
777 | int nid, zid; | |
bb4cc1a8 | 778 | |
e231875b JZ |
779 | for_each_node(nid) { |
780 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
781 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
782 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 783 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
784 | } |
785 | } | |
786 | } | |
787 | ||
788 | static struct mem_cgroup_per_zone * | |
789 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
790 | { | |
791 | struct rb_node *rightmost = NULL; | |
792 | struct mem_cgroup_per_zone *mz; | |
793 | ||
794 | retry: | |
795 | mz = NULL; | |
796 | rightmost = rb_last(&mctz->rb_root); | |
797 | if (!rightmost) | |
798 | goto done; /* Nothing to reclaim from */ | |
799 | ||
800 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
801 | /* | |
802 | * Remove the node now but someone else can add it back, | |
803 | * we will to add it back at the end of reclaim to its correct | |
804 | * position in the tree. | |
805 | */ | |
cf2c8127 | 806 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 807 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 808 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
809 | goto retry; |
810 | done: | |
811 | return mz; | |
812 | } | |
813 | ||
814 | static struct mem_cgroup_per_zone * | |
815 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
816 | { | |
817 | struct mem_cgroup_per_zone *mz; | |
818 | ||
0a31bc97 | 819 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 820 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 821 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
822 | return mz; |
823 | } | |
824 | ||
711d3d2c KH |
825 | /* |
826 | * Implementation Note: reading percpu statistics for memcg. | |
827 | * | |
828 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
829 | * synchronization to implement "quick" read. There are trade-off between | |
830 | * reading cost and precision of value. Then, we may have a chance to implement | |
831 | * a periodic synchronizion of counter in memcg's counter. | |
832 | * | |
833 | * But this _read() function is used for user interface now. The user accounts | |
834 | * memory usage by memory cgroup and he _always_ requires exact value because | |
835 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
836 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
837 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
838 | * | |
839 | * If there are kernel internal actions which can make use of some not-exact | |
840 | * value, and reading all cpu value can be performance bottleneck in some | |
841 | * common workload, threashold and synchonization as vmstat[] should be | |
842 | * implemented. | |
843 | */ | |
c0ff4b85 | 844 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 845 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 846 | { |
7a159cc9 | 847 | long val = 0; |
c62b1a3b | 848 | int cpu; |
c62b1a3b | 849 | |
711d3d2c KH |
850 | get_online_cpus(); |
851 | for_each_online_cpu(cpu) | |
c0ff4b85 | 852 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 853 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
854 | spin_lock(&memcg->pcp_counter_lock); |
855 | val += memcg->nocpu_base.count[idx]; | |
856 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
857 | #endif |
858 | put_online_cpus(); | |
c62b1a3b KH |
859 | return val; |
860 | } | |
861 | ||
c0ff4b85 | 862 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
863 | enum mem_cgroup_events_index idx) |
864 | { | |
865 | unsigned long val = 0; | |
866 | int cpu; | |
867 | ||
9c567512 | 868 | get_online_cpus(); |
e9f8974f | 869 | for_each_online_cpu(cpu) |
c0ff4b85 | 870 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 871 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
872 | spin_lock(&memcg->pcp_counter_lock); |
873 | val += memcg->nocpu_base.events[idx]; | |
874 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 875 | #endif |
9c567512 | 876 | put_online_cpus(); |
e9f8974f JW |
877 | return val; |
878 | } | |
879 | ||
c0ff4b85 | 880 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 881 | struct page *page, |
0a31bc97 | 882 | int nr_pages) |
d52aa412 | 883 | { |
b2402857 KH |
884 | /* |
885 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
886 | * counted as CACHE even if it's on ANON LRU. | |
887 | */ | |
0a31bc97 | 888 | if (PageAnon(page)) |
b2402857 | 889 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 890 | nr_pages); |
d52aa412 | 891 | else |
b2402857 | 892 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 893 | nr_pages); |
55e462b0 | 894 | |
b070e65c DR |
895 | if (PageTransHuge(page)) |
896 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
897 | nr_pages); | |
898 | ||
e401f176 KH |
899 | /* pagein of a big page is an event. So, ignore page size */ |
900 | if (nr_pages > 0) | |
c0ff4b85 | 901 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 902 | else { |
c0ff4b85 | 903 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
904 | nr_pages = -nr_pages; /* for event */ |
905 | } | |
e401f176 | 906 | |
13114716 | 907 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
908 | } |
909 | ||
e231875b | 910 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
911 | { |
912 | struct mem_cgroup_per_zone *mz; | |
913 | ||
914 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
915 | return mz->lru_size[lru]; | |
916 | } | |
917 | ||
e231875b JZ |
918 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
919 | int nid, | |
920 | unsigned int lru_mask) | |
bb2a0de9 | 921 | { |
e231875b | 922 | unsigned long nr = 0; |
889976db YH |
923 | int zid; |
924 | ||
e231875b | 925 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 926 | |
e231875b JZ |
927 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
928 | struct mem_cgroup_per_zone *mz; | |
929 | enum lru_list lru; | |
930 | ||
931 | for_each_lru(lru) { | |
932 | if (!(BIT(lru) & lru_mask)) | |
933 | continue; | |
934 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
935 | nr += mz->lru_size[lru]; | |
936 | } | |
937 | } | |
938 | return nr; | |
889976db | 939 | } |
bb2a0de9 | 940 | |
c0ff4b85 | 941 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 942 | unsigned int lru_mask) |
6d12e2d8 | 943 | { |
e231875b | 944 | unsigned long nr = 0; |
889976db | 945 | int nid; |
6d12e2d8 | 946 | |
31aaea4a | 947 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
948 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
949 | return nr; | |
d52aa412 KH |
950 | } |
951 | ||
f53d7ce3 JW |
952 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
953 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
954 | { |
955 | unsigned long val, next; | |
956 | ||
13114716 | 957 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 958 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 959 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
960 | if ((long)next - (long)val < 0) { |
961 | switch (target) { | |
962 | case MEM_CGROUP_TARGET_THRESH: | |
963 | next = val + THRESHOLDS_EVENTS_TARGET; | |
964 | break; | |
bb4cc1a8 AM |
965 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
966 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
967 | break; | |
f53d7ce3 JW |
968 | case MEM_CGROUP_TARGET_NUMAINFO: |
969 | next = val + NUMAINFO_EVENTS_TARGET; | |
970 | break; | |
971 | default: | |
972 | break; | |
973 | } | |
974 | __this_cpu_write(memcg->stat->targets[target], next); | |
975 | return true; | |
7a159cc9 | 976 | } |
f53d7ce3 | 977 | return false; |
d2265e6f KH |
978 | } |
979 | ||
980 | /* | |
981 | * Check events in order. | |
982 | * | |
983 | */ | |
c0ff4b85 | 984 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
985 | { |
986 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
987 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
988 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 989 | bool do_softlimit; |
82b3f2a7 | 990 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 991 | |
bb4cc1a8 AM |
992 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
993 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
994 | #if MAX_NUMNODES > 1 |
995 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
996 | MEM_CGROUP_TARGET_NUMAINFO); | |
997 | #endif | |
c0ff4b85 | 998 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
999 | if (unlikely(do_softlimit)) |
1000 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 1001 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1002 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1003 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1004 | #endif |
0a31bc97 | 1005 | } |
d2265e6f KH |
1006 | } |
1007 | ||
cf475ad2 | 1008 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1009 | { |
31a78f23 BS |
1010 | /* |
1011 | * mm_update_next_owner() may clear mm->owner to NULL | |
1012 | * if it races with swapoff, page migration, etc. | |
1013 | * So this can be called with p == NULL. | |
1014 | */ | |
1015 | if (unlikely(!p)) | |
1016 | return NULL; | |
1017 | ||
073219e9 | 1018 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1019 | } |
1020 | ||
df381975 | 1021 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1022 | { |
c0ff4b85 | 1023 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 1024 | |
54595fe2 KH |
1025 | rcu_read_lock(); |
1026 | do { | |
6f6acb00 MH |
1027 | /* |
1028 | * Page cache insertions can happen withou an | |
1029 | * actual mm context, e.g. during disk probing | |
1030 | * on boot, loopback IO, acct() writes etc. | |
1031 | */ | |
1032 | if (unlikely(!mm)) | |
df381975 | 1033 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1034 | else { |
1035 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1036 | if (unlikely(!memcg)) | |
1037 | memcg = root_mem_cgroup; | |
1038 | } | |
ec903c0c | 1039 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1040 | rcu_read_unlock(); |
c0ff4b85 | 1041 | return memcg; |
54595fe2 KH |
1042 | } |
1043 | ||
5660048c JW |
1044 | /** |
1045 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1046 | * @root: hierarchy root | |
1047 | * @prev: previously returned memcg, NULL on first invocation | |
1048 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1049 | * | |
1050 | * Returns references to children of the hierarchy below @root, or | |
1051 | * @root itself, or %NULL after a full round-trip. | |
1052 | * | |
1053 | * Caller must pass the return value in @prev on subsequent | |
1054 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1055 | * to cancel a hierarchy walk before the round-trip is complete. | |
1056 | * | |
1057 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1058 | * divide up the memcgs in the hierarchy among all concurrent | |
1059 | * reclaimers operating on the same zone and priority. | |
1060 | */ | |
694fbc0f | 1061 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1062 | struct mem_cgroup *prev, |
694fbc0f | 1063 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1064 | { |
5ac8fb31 JW |
1065 | struct reclaim_iter *uninitialized_var(iter); |
1066 | struct cgroup_subsys_state *css = NULL; | |
9f3a0d09 | 1067 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1068 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1069 | |
694fbc0f AM |
1070 | if (mem_cgroup_disabled()) |
1071 | return NULL; | |
5660048c | 1072 | |
9f3a0d09 JW |
1073 | if (!root) |
1074 | root = root_mem_cgroup; | |
7d74b06f | 1075 | |
9f3a0d09 | 1076 | if (prev && !reclaim) |
5ac8fb31 | 1077 | pos = prev; |
14067bb3 | 1078 | |
9f3a0d09 JW |
1079 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1080 | if (prev) | |
5ac8fb31 | 1081 | goto out; |
694fbc0f | 1082 | return root; |
9f3a0d09 | 1083 | } |
14067bb3 | 1084 | |
542f85f9 | 1085 | rcu_read_lock(); |
5f578161 | 1086 | |
5ac8fb31 JW |
1087 | if (reclaim) { |
1088 | struct mem_cgroup_per_zone *mz; | |
1089 | ||
1090 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
1091 | iter = &mz->iter[reclaim->priority]; | |
1092 | ||
1093 | if (prev && reclaim->generation != iter->generation) | |
1094 | goto out_unlock; | |
1095 | ||
1096 | do { | |
1097 | pos = ACCESS_ONCE(iter->position); | |
1098 | /* | |
1099 | * A racing update may change the position and | |
1100 | * put the last reference, hence css_tryget(), | |
1101 | * or retry to see the updated position. | |
1102 | */ | |
1103 | } while (pos && !css_tryget(&pos->css)); | |
1104 | } | |
1105 | ||
1106 | if (pos) | |
1107 | css = &pos->css; | |
1108 | ||
1109 | for (;;) { | |
1110 | css = css_next_descendant_pre(css, &root->css); | |
1111 | if (!css) { | |
1112 | /* | |
1113 | * Reclaimers share the hierarchy walk, and a | |
1114 | * new one might jump in right at the end of | |
1115 | * the hierarchy - make sure they see at least | |
1116 | * one group and restart from the beginning. | |
1117 | */ | |
1118 | if (!prev) | |
1119 | continue; | |
1120 | break; | |
527a5ec9 | 1121 | } |
7d74b06f | 1122 | |
5ac8fb31 JW |
1123 | /* |
1124 | * Verify the css and acquire a reference. The root | |
1125 | * is provided by the caller, so we know it's alive | |
1126 | * and kicking, and don't take an extra reference. | |
1127 | */ | |
1128 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1129 | |
5ac8fb31 JW |
1130 | if (css == &root->css) |
1131 | break; | |
542f85f9 | 1132 | |
b2052564 | 1133 | if (css_tryget(css)) { |
5ac8fb31 JW |
1134 | /* |
1135 | * Make sure the memcg is initialized: | |
1136 | * mem_cgroup_css_online() orders the the | |
1137 | * initialization against setting the flag. | |
1138 | */ | |
1139 | if (smp_load_acquire(&memcg->initialized)) | |
1140 | break; | |
1141 | ||
1142 | css_put(css); | |
527a5ec9 | 1143 | } |
9f3a0d09 | 1144 | |
5ac8fb31 JW |
1145 | memcg = NULL; |
1146 | } | |
1147 | ||
1148 | if (reclaim) { | |
1149 | if (cmpxchg(&iter->position, pos, memcg) == pos) { | |
1150 | if (memcg) | |
1151 | css_get(&memcg->css); | |
1152 | if (pos) | |
1153 | css_put(&pos->css); | |
1154 | } | |
1155 | ||
1156 | /* | |
1157 | * pairs with css_tryget when dereferencing iter->position | |
1158 | * above. | |
1159 | */ | |
1160 | if (pos) | |
1161 | css_put(&pos->css); | |
1162 | ||
1163 | if (!memcg) | |
1164 | iter->generation++; | |
1165 | else if (!prev) | |
1166 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1167 | } |
5ac8fb31 | 1168 | |
542f85f9 MH |
1169 | out_unlock: |
1170 | rcu_read_unlock(); | |
5ac8fb31 | 1171 | out: |
c40046f3 MH |
1172 | if (prev && prev != root) |
1173 | css_put(&prev->css); | |
1174 | ||
9f3a0d09 | 1175 | return memcg; |
14067bb3 | 1176 | } |
7d74b06f | 1177 | |
5660048c JW |
1178 | /** |
1179 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1180 | * @root: hierarchy root | |
1181 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1182 | */ | |
1183 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1184 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1185 | { |
1186 | if (!root) | |
1187 | root = root_mem_cgroup; | |
1188 | if (prev && prev != root) | |
1189 | css_put(&prev->css); | |
1190 | } | |
7d74b06f | 1191 | |
9f3a0d09 JW |
1192 | /* |
1193 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1194 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1195 | * be used for reference counting. | |
1196 | */ | |
1197 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1198 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1199 | iter != NULL; \ |
527a5ec9 | 1200 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1201 | |
9f3a0d09 | 1202 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1203 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1204 | iter != NULL; \ |
527a5ec9 | 1205 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1206 | |
68ae564b | 1207 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1208 | { |
c0ff4b85 | 1209 | struct mem_cgroup *memcg; |
456f998e | 1210 | |
456f998e | 1211 | rcu_read_lock(); |
c0ff4b85 R |
1212 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1213 | if (unlikely(!memcg)) | |
456f998e YH |
1214 | goto out; |
1215 | ||
1216 | switch (idx) { | |
456f998e | 1217 | case PGFAULT: |
0e574a93 JW |
1218 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1219 | break; | |
1220 | case PGMAJFAULT: | |
1221 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1222 | break; |
1223 | default: | |
1224 | BUG(); | |
1225 | } | |
1226 | out: | |
1227 | rcu_read_unlock(); | |
1228 | } | |
68ae564b | 1229 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1230 | |
925b7673 JW |
1231 | /** |
1232 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1233 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1234 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1235 | * |
1236 | * Returns the lru list vector holding pages for the given @zone and | |
1237 | * @mem. This can be the global zone lruvec, if the memory controller | |
1238 | * is disabled. | |
1239 | */ | |
1240 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1241 | struct mem_cgroup *memcg) | |
1242 | { | |
1243 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1244 | struct lruvec *lruvec; |
925b7673 | 1245 | |
bea8c150 HD |
1246 | if (mem_cgroup_disabled()) { |
1247 | lruvec = &zone->lruvec; | |
1248 | goto out; | |
1249 | } | |
925b7673 | 1250 | |
e231875b | 1251 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1252 | lruvec = &mz->lruvec; |
1253 | out: | |
1254 | /* | |
1255 | * Since a node can be onlined after the mem_cgroup was created, | |
1256 | * we have to be prepared to initialize lruvec->zone here; | |
1257 | * and if offlined then reonlined, we need to reinitialize it. | |
1258 | */ | |
1259 | if (unlikely(lruvec->zone != zone)) | |
1260 | lruvec->zone = zone; | |
1261 | return lruvec; | |
925b7673 JW |
1262 | } |
1263 | ||
925b7673 | 1264 | /** |
dfe0e773 | 1265 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1266 | * @page: the page |
fa9add64 | 1267 | * @zone: zone of the page |
dfe0e773 JW |
1268 | * |
1269 | * This function is only safe when following the LRU page isolation | |
1270 | * and putback protocol: the LRU lock must be held, and the page must | |
1271 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1272 | */ |
fa9add64 | 1273 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1274 | { |
08e552c6 | 1275 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1276 | struct mem_cgroup *memcg; |
1277 | struct page_cgroup *pc; | |
bea8c150 | 1278 | struct lruvec *lruvec; |
6d12e2d8 | 1279 | |
bea8c150 HD |
1280 | if (mem_cgroup_disabled()) { |
1281 | lruvec = &zone->lruvec; | |
1282 | goto out; | |
1283 | } | |
925b7673 | 1284 | |
08e552c6 | 1285 | pc = lookup_page_cgroup(page); |
38c5d72f | 1286 | memcg = pc->mem_cgroup; |
7512102c | 1287 | /* |
dfe0e773 | 1288 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1289 | * possibly migrated - before they are charged. |
7512102c | 1290 | */ |
29833315 JW |
1291 | if (!memcg) |
1292 | memcg = root_mem_cgroup; | |
7512102c | 1293 | |
e231875b | 1294 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1295 | lruvec = &mz->lruvec; |
1296 | out: | |
1297 | /* | |
1298 | * Since a node can be onlined after the mem_cgroup was created, | |
1299 | * we have to be prepared to initialize lruvec->zone here; | |
1300 | * and if offlined then reonlined, we need to reinitialize it. | |
1301 | */ | |
1302 | if (unlikely(lruvec->zone != zone)) | |
1303 | lruvec->zone = zone; | |
1304 | return lruvec; | |
08e552c6 | 1305 | } |
b69408e8 | 1306 | |
925b7673 | 1307 | /** |
fa9add64 HD |
1308 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1309 | * @lruvec: mem_cgroup per zone lru vector | |
1310 | * @lru: index of lru list the page is sitting on | |
1311 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1312 | * |
fa9add64 HD |
1313 | * This function must be called when a page is added to or removed from an |
1314 | * lru list. | |
3f58a829 | 1315 | */ |
fa9add64 HD |
1316 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1317 | int nr_pages) | |
3f58a829 MK |
1318 | { |
1319 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1320 | unsigned long *lru_size; |
3f58a829 MK |
1321 | |
1322 | if (mem_cgroup_disabled()) | |
1323 | return; | |
1324 | ||
fa9add64 HD |
1325 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1326 | lru_size = mz->lru_size + lru; | |
1327 | *lru_size += nr_pages; | |
1328 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1329 | } |
544122e5 | 1330 | |
3e92041d | 1331 | /* |
c0ff4b85 | 1332 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1333 | * hierarchy subtree |
1334 | */ | |
c3ac9a8a JW |
1335 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1336 | struct mem_cgroup *memcg) | |
3e92041d | 1337 | { |
91c63734 JW |
1338 | if (root_memcg == memcg) |
1339 | return true; | |
413918bb | 1340 | if (!root_memcg->use_hierarchy) |
91c63734 | 1341 | return false; |
b47f77b5 | 1342 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
c3ac9a8a JW |
1343 | } |
1344 | ||
1345 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1346 | struct mem_cgroup *memcg) | |
1347 | { | |
1348 | bool ret; | |
1349 | ||
91c63734 | 1350 | rcu_read_lock(); |
c3ac9a8a | 1351 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1352 | rcu_read_unlock(); |
1353 | return ret; | |
3e92041d MH |
1354 | } |
1355 | ||
ffbdccf5 DR |
1356 | bool task_in_mem_cgroup(struct task_struct *task, |
1357 | const struct mem_cgroup *memcg) | |
4c4a2214 | 1358 | { |
c01f46c7 | 1359 | struct mem_cgroup *curr; |
158e0a2d | 1360 | struct task_struct *p; |
ffbdccf5 | 1361 | bool ret; |
4c4a2214 | 1362 | |
158e0a2d | 1363 | p = find_lock_task_mm(task); |
de077d22 | 1364 | if (p) { |
df381975 | 1365 | curr = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1366 | task_unlock(p); |
1367 | } else { | |
1368 | /* | |
1369 | * All threads may have already detached their mm's, but the oom | |
1370 | * killer still needs to detect if they have already been oom | |
1371 | * killed to prevent needlessly killing additional tasks. | |
1372 | */ | |
ffbdccf5 | 1373 | rcu_read_lock(); |
de077d22 | 1374 | curr = mem_cgroup_from_task(task); |
c01f46c7 | 1375 | css_get(&curr->css); |
ffbdccf5 | 1376 | rcu_read_unlock(); |
de077d22 | 1377 | } |
d31f56db | 1378 | /* |
c0ff4b85 | 1379 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1380 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1381 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1382 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1383 | */ |
c0ff4b85 | 1384 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1385 | css_put(&curr->css); |
4c4a2214 DR |
1386 | return ret; |
1387 | } | |
1388 | ||
c56d5c7d | 1389 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1390 | { |
9b272977 | 1391 | unsigned long inactive_ratio; |
14797e23 | 1392 | unsigned long inactive; |
9b272977 | 1393 | unsigned long active; |
c772be93 | 1394 | unsigned long gb; |
14797e23 | 1395 | |
4d7dcca2 HD |
1396 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1397 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1398 | |
c772be93 KM |
1399 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1400 | if (gb) | |
1401 | inactive_ratio = int_sqrt(10 * gb); | |
1402 | else | |
1403 | inactive_ratio = 1; | |
1404 | ||
9b272977 | 1405 | return inactive * inactive_ratio < active; |
14797e23 KM |
1406 | } |
1407 | ||
3e32cb2e | 1408 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1409 | container_of(counter, struct mem_cgroup, member) |
1410 | ||
19942822 | 1411 | /** |
9d11ea9f | 1412 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1413 | * @memcg: the memory cgroup |
19942822 | 1414 | * |
9d11ea9f | 1415 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1416 | * pages. |
19942822 | 1417 | */ |
c0ff4b85 | 1418 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1419 | { |
3e32cb2e JW |
1420 | unsigned long margin = 0; |
1421 | unsigned long count; | |
1422 | unsigned long limit; | |
9d11ea9f | 1423 | |
3e32cb2e JW |
1424 | count = page_counter_read(&memcg->memory); |
1425 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1426 | if (count < limit) | |
1427 | margin = limit - count; | |
1428 | ||
1429 | if (do_swap_account) { | |
1430 | count = page_counter_read(&memcg->memsw); | |
1431 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1432 | if (count <= limit) | |
1433 | margin = min(margin, limit - count); | |
1434 | } | |
1435 | ||
1436 | return margin; | |
19942822 JW |
1437 | } |
1438 | ||
1f4c025b | 1439 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1440 | { |
a7885eb8 | 1441 | /* root ? */ |
14208b0e | 1442 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1443 | return vm_swappiness; |
1444 | ||
bf1ff263 | 1445 | return memcg->swappiness; |
a7885eb8 KM |
1446 | } |
1447 | ||
32047e2a | 1448 | /* |
bdcbb659 | 1449 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1450 | * |
bdcbb659 QH |
1451 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1452 | * moving cgroups. This is for waiting at high-memory pressure | |
1453 | * caused by "move". | |
32047e2a | 1454 | */ |
c0ff4b85 | 1455 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1456 | { |
2bd9bb20 KH |
1457 | struct mem_cgroup *from; |
1458 | struct mem_cgroup *to; | |
4b534334 | 1459 | bool ret = false; |
2bd9bb20 KH |
1460 | /* |
1461 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1462 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1463 | */ | |
1464 | spin_lock(&mc.lock); | |
1465 | from = mc.from; | |
1466 | to = mc.to; | |
1467 | if (!from) | |
1468 | goto unlock; | |
3e92041d | 1469 | |
c0ff4b85 R |
1470 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1471 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1472 | unlock: |
1473 | spin_unlock(&mc.lock); | |
4b534334 KH |
1474 | return ret; |
1475 | } | |
1476 | ||
c0ff4b85 | 1477 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1478 | { |
1479 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1480 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1481 | DEFINE_WAIT(wait); |
1482 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1483 | /* moving charge context might have finished. */ | |
1484 | if (mc.moving_task) | |
1485 | schedule(); | |
1486 | finish_wait(&mc.waitq, &wait); | |
1487 | return true; | |
1488 | } | |
1489 | } | |
1490 | return false; | |
1491 | } | |
1492 | ||
58cf188e | 1493 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1494 | /** |
58cf188e | 1495 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1496 | * @memcg: The memory cgroup that went over limit |
1497 | * @p: Task that is going to be killed | |
1498 | * | |
1499 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1500 | * enabled | |
1501 | */ | |
1502 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1503 | { | |
e61734c5 | 1504 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1505 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1506 | struct mem_cgroup *iter; |
1507 | unsigned int i; | |
e222432b | 1508 | |
58cf188e | 1509 | if (!p) |
e222432b BS |
1510 | return; |
1511 | ||
08088cb9 | 1512 | mutex_lock(&oom_info_lock); |
e222432b BS |
1513 | rcu_read_lock(); |
1514 | ||
e61734c5 TH |
1515 | pr_info("Task in "); |
1516 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1517 | pr_info(" killed as a result of limit of "); | |
1518 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1519 | pr_info("\n"); | |
e222432b | 1520 | |
e222432b BS |
1521 | rcu_read_unlock(); |
1522 | ||
3e32cb2e JW |
1523 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1524 | K((u64)page_counter_read(&memcg->memory)), | |
1525 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1526 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1527 | K((u64)page_counter_read(&memcg->memsw)), | |
1528 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1529 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1530 | K((u64)page_counter_read(&memcg->kmem)), | |
1531 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1532 | |
1533 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1534 | pr_info("Memory cgroup stats for "); |
1535 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1536 | pr_cont(":"); |
1537 | ||
1538 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1539 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1540 | continue; | |
1541 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1542 | K(mem_cgroup_read_stat(iter, i))); | |
1543 | } | |
1544 | ||
1545 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1546 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1547 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1548 | ||
1549 | pr_cont("\n"); | |
1550 | } | |
08088cb9 | 1551 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1552 | } |
1553 | ||
81d39c20 KH |
1554 | /* |
1555 | * This function returns the number of memcg under hierarchy tree. Returns | |
1556 | * 1(self count) if no children. | |
1557 | */ | |
c0ff4b85 | 1558 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1559 | { |
1560 | int num = 0; | |
7d74b06f KH |
1561 | struct mem_cgroup *iter; |
1562 | ||
c0ff4b85 | 1563 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1564 | num++; |
81d39c20 KH |
1565 | return num; |
1566 | } | |
1567 | ||
a63d83f4 DR |
1568 | /* |
1569 | * Return the memory (and swap, if configured) limit for a memcg. | |
1570 | */ | |
3e32cb2e | 1571 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1572 | { |
3e32cb2e | 1573 | unsigned long limit; |
a63d83f4 | 1574 | |
3e32cb2e | 1575 | limit = memcg->memory.limit; |
9a5a8f19 | 1576 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1577 | unsigned long memsw_limit; |
9a5a8f19 | 1578 | |
3e32cb2e JW |
1579 | memsw_limit = memcg->memsw.limit; |
1580 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1581 | } |
9a5a8f19 | 1582 | return limit; |
a63d83f4 DR |
1583 | } |
1584 | ||
19965460 DR |
1585 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1586 | int order) | |
9cbb78bb DR |
1587 | { |
1588 | struct mem_cgroup *iter; | |
1589 | unsigned long chosen_points = 0; | |
1590 | unsigned long totalpages; | |
1591 | unsigned int points = 0; | |
1592 | struct task_struct *chosen = NULL; | |
1593 | ||
876aafbf | 1594 | /* |
465adcf1 DR |
1595 | * If current has a pending SIGKILL or is exiting, then automatically |
1596 | * select it. The goal is to allow it to allocate so that it may | |
1597 | * quickly exit and free its memory. | |
876aafbf | 1598 | */ |
465adcf1 | 1599 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1600 | set_thread_flag(TIF_MEMDIE); |
1601 | return; | |
1602 | } | |
1603 | ||
1604 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1605 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1606 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1607 | struct css_task_iter it; |
9cbb78bb DR |
1608 | struct task_struct *task; |
1609 | ||
72ec7029 TH |
1610 | css_task_iter_start(&iter->css, &it); |
1611 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1612 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1613 | false)) { | |
1614 | case OOM_SCAN_SELECT: | |
1615 | if (chosen) | |
1616 | put_task_struct(chosen); | |
1617 | chosen = task; | |
1618 | chosen_points = ULONG_MAX; | |
1619 | get_task_struct(chosen); | |
1620 | /* fall through */ | |
1621 | case OOM_SCAN_CONTINUE: | |
1622 | continue; | |
1623 | case OOM_SCAN_ABORT: | |
72ec7029 | 1624 | css_task_iter_end(&it); |
9cbb78bb DR |
1625 | mem_cgroup_iter_break(memcg, iter); |
1626 | if (chosen) | |
1627 | put_task_struct(chosen); | |
1628 | return; | |
1629 | case OOM_SCAN_OK: | |
1630 | break; | |
1631 | }; | |
1632 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1633 | if (!points || points < chosen_points) |
1634 | continue; | |
1635 | /* Prefer thread group leaders for display purposes */ | |
1636 | if (points == chosen_points && | |
1637 | thread_group_leader(chosen)) | |
1638 | continue; | |
1639 | ||
1640 | if (chosen) | |
1641 | put_task_struct(chosen); | |
1642 | chosen = task; | |
1643 | chosen_points = points; | |
1644 | get_task_struct(chosen); | |
9cbb78bb | 1645 | } |
72ec7029 | 1646 | css_task_iter_end(&it); |
9cbb78bb DR |
1647 | } |
1648 | ||
1649 | if (!chosen) | |
1650 | return; | |
1651 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1652 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1653 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1654 | } |
1655 | ||
4d0c066d KH |
1656 | /** |
1657 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1658 | * @memcg: the target memcg |
4d0c066d KH |
1659 | * @nid: the node ID to be checked. |
1660 | * @noswap : specify true here if the user wants flle only information. | |
1661 | * | |
1662 | * This function returns whether the specified memcg contains any | |
1663 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1664 | * pages in the node. | |
1665 | */ | |
c0ff4b85 | 1666 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1667 | int nid, bool noswap) |
1668 | { | |
c0ff4b85 | 1669 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1670 | return true; |
1671 | if (noswap || !total_swap_pages) | |
1672 | return false; | |
c0ff4b85 | 1673 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1674 | return true; |
1675 | return false; | |
1676 | ||
1677 | } | |
bb4cc1a8 | 1678 | #if MAX_NUMNODES > 1 |
889976db YH |
1679 | |
1680 | /* | |
1681 | * Always updating the nodemask is not very good - even if we have an empty | |
1682 | * list or the wrong list here, we can start from some node and traverse all | |
1683 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1684 | * | |
1685 | */ | |
c0ff4b85 | 1686 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1687 | { |
1688 | int nid; | |
453a9bf3 KH |
1689 | /* |
1690 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1691 | * pagein/pageout changes since the last update. | |
1692 | */ | |
c0ff4b85 | 1693 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1694 | return; |
c0ff4b85 | 1695 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1696 | return; |
1697 | ||
889976db | 1698 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1699 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1700 | |
31aaea4a | 1701 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1702 | |
c0ff4b85 R |
1703 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1704 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1705 | } |
453a9bf3 | 1706 | |
c0ff4b85 R |
1707 | atomic_set(&memcg->numainfo_events, 0); |
1708 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1709 | } |
1710 | ||
1711 | /* | |
1712 | * Selecting a node where we start reclaim from. Because what we need is just | |
1713 | * reducing usage counter, start from anywhere is O,K. Considering | |
1714 | * memory reclaim from current node, there are pros. and cons. | |
1715 | * | |
1716 | * Freeing memory from current node means freeing memory from a node which | |
1717 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1718 | * hit limits, it will see a contention on a node. But freeing from remote | |
1719 | * node means more costs for memory reclaim because of memory latency. | |
1720 | * | |
1721 | * Now, we use round-robin. Better algorithm is welcomed. | |
1722 | */ | |
c0ff4b85 | 1723 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1724 | { |
1725 | int node; | |
1726 | ||
c0ff4b85 R |
1727 | mem_cgroup_may_update_nodemask(memcg); |
1728 | node = memcg->last_scanned_node; | |
889976db | 1729 | |
c0ff4b85 | 1730 | node = next_node(node, memcg->scan_nodes); |
889976db | 1731 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1732 | node = first_node(memcg->scan_nodes); |
889976db YH |
1733 | /* |
1734 | * We call this when we hit limit, not when pages are added to LRU. | |
1735 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1736 | * memcg is too small and all pages are not on LRU. In that case, | |
1737 | * we use curret node. | |
1738 | */ | |
1739 | if (unlikely(node == MAX_NUMNODES)) | |
1740 | node = numa_node_id(); | |
1741 | ||
c0ff4b85 | 1742 | memcg->last_scanned_node = node; |
889976db YH |
1743 | return node; |
1744 | } | |
889976db | 1745 | #else |
c0ff4b85 | 1746 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1747 | { |
1748 | return 0; | |
1749 | } | |
1750 | #endif | |
1751 | ||
0608f43d AM |
1752 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1753 | struct zone *zone, | |
1754 | gfp_t gfp_mask, | |
1755 | unsigned long *total_scanned) | |
1756 | { | |
1757 | struct mem_cgroup *victim = NULL; | |
1758 | int total = 0; | |
1759 | int loop = 0; | |
1760 | unsigned long excess; | |
1761 | unsigned long nr_scanned; | |
1762 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1763 | .zone = zone, | |
1764 | .priority = 0, | |
1765 | }; | |
1766 | ||
3e32cb2e | 1767 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1768 | |
1769 | while (1) { | |
1770 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1771 | if (!victim) { | |
1772 | loop++; | |
1773 | if (loop >= 2) { | |
1774 | /* | |
1775 | * If we have not been able to reclaim | |
1776 | * anything, it might because there are | |
1777 | * no reclaimable pages under this hierarchy | |
1778 | */ | |
1779 | if (!total) | |
1780 | break; | |
1781 | /* | |
1782 | * We want to do more targeted reclaim. | |
1783 | * excess >> 2 is not to excessive so as to | |
1784 | * reclaim too much, nor too less that we keep | |
1785 | * coming back to reclaim from this cgroup | |
1786 | */ | |
1787 | if (total >= (excess >> 2) || | |
1788 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1789 | break; | |
1790 | } | |
1791 | continue; | |
1792 | } | |
0608f43d AM |
1793 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1794 | zone, &nr_scanned); | |
1795 | *total_scanned += nr_scanned; | |
3e32cb2e | 1796 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1797 | break; |
6d61ef40 | 1798 | } |
0608f43d AM |
1799 | mem_cgroup_iter_break(root_memcg, victim); |
1800 | return total; | |
6d61ef40 BS |
1801 | } |
1802 | ||
0056f4e6 JW |
1803 | #ifdef CONFIG_LOCKDEP |
1804 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1805 | .name = "memcg_oom_lock", | |
1806 | }; | |
1807 | #endif | |
1808 | ||
fb2a6fc5 JW |
1809 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1810 | ||
867578cb KH |
1811 | /* |
1812 | * Check OOM-Killer is already running under our hierarchy. | |
1813 | * If someone is running, return false. | |
1814 | */ | |
fb2a6fc5 | 1815 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1816 | { |
79dfdacc | 1817 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1818 | |
fb2a6fc5 JW |
1819 | spin_lock(&memcg_oom_lock); |
1820 | ||
9f3a0d09 | 1821 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1822 | if (iter->oom_lock) { |
79dfdacc MH |
1823 | /* |
1824 | * this subtree of our hierarchy is already locked | |
1825 | * so we cannot give a lock. | |
1826 | */ | |
79dfdacc | 1827 | failed = iter; |
9f3a0d09 JW |
1828 | mem_cgroup_iter_break(memcg, iter); |
1829 | break; | |
23751be0 JW |
1830 | } else |
1831 | iter->oom_lock = true; | |
7d74b06f | 1832 | } |
867578cb | 1833 | |
fb2a6fc5 JW |
1834 | if (failed) { |
1835 | /* | |
1836 | * OK, we failed to lock the whole subtree so we have | |
1837 | * to clean up what we set up to the failing subtree | |
1838 | */ | |
1839 | for_each_mem_cgroup_tree(iter, memcg) { | |
1840 | if (iter == failed) { | |
1841 | mem_cgroup_iter_break(memcg, iter); | |
1842 | break; | |
1843 | } | |
1844 | iter->oom_lock = false; | |
79dfdacc | 1845 | } |
0056f4e6 JW |
1846 | } else |
1847 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1848 | |
1849 | spin_unlock(&memcg_oom_lock); | |
1850 | ||
1851 | return !failed; | |
a636b327 | 1852 | } |
0b7f569e | 1853 | |
fb2a6fc5 | 1854 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1855 | { |
7d74b06f KH |
1856 | struct mem_cgroup *iter; |
1857 | ||
fb2a6fc5 | 1858 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1859 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1860 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1861 | iter->oom_lock = false; |
fb2a6fc5 | 1862 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1863 | } |
1864 | ||
c0ff4b85 | 1865 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1866 | { |
1867 | struct mem_cgroup *iter; | |
1868 | ||
c0ff4b85 | 1869 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1870 | atomic_inc(&iter->under_oom); |
1871 | } | |
1872 | ||
c0ff4b85 | 1873 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1874 | { |
1875 | struct mem_cgroup *iter; | |
1876 | ||
867578cb KH |
1877 | /* |
1878 | * When a new child is created while the hierarchy is under oom, | |
1879 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1880 | * atomic_add_unless() here. | |
1881 | */ | |
c0ff4b85 | 1882 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1883 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1884 | } |
1885 | ||
867578cb KH |
1886 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1887 | ||
dc98df5a | 1888 | struct oom_wait_info { |
d79154bb | 1889 | struct mem_cgroup *memcg; |
dc98df5a KH |
1890 | wait_queue_t wait; |
1891 | }; | |
1892 | ||
1893 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1894 | unsigned mode, int sync, void *arg) | |
1895 | { | |
d79154bb HD |
1896 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1897 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1898 | struct oom_wait_info *oom_wait_info; |
1899 | ||
1900 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1901 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1902 | |
dc98df5a | 1903 | /* |
d79154bb | 1904 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1905 | * Then we can use css_is_ancestor without taking care of RCU. |
1906 | */ | |
c0ff4b85 R |
1907 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1908 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1909 | return 0; |
dc98df5a KH |
1910 | return autoremove_wake_function(wait, mode, sync, arg); |
1911 | } | |
1912 | ||
c0ff4b85 | 1913 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1914 | { |
3812c8c8 | 1915 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
1916 | /* for filtering, pass "memcg" as argument. */ |
1917 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1918 | } |
1919 | ||
c0ff4b85 | 1920 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1921 | { |
c0ff4b85 R |
1922 | if (memcg && atomic_read(&memcg->under_oom)) |
1923 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1924 | } |
1925 | ||
3812c8c8 | 1926 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1927 | { |
3812c8c8 JW |
1928 | if (!current->memcg_oom.may_oom) |
1929 | return; | |
867578cb | 1930 | /* |
49426420 JW |
1931 | * We are in the middle of the charge context here, so we |
1932 | * don't want to block when potentially sitting on a callstack | |
1933 | * that holds all kinds of filesystem and mm locks. | |
1934 | * | |
1935 | * Also, the caller may handle a failed allocation gracefully | |
1936 | * (like optional page cache readahead) and so an OOM killer | |
1937 | * invocation might not even be necessary. | |
1938 | * | |
1939 | * That's why we don't do anything here except remember the | |
1940 | * OOM context and then deal with it at the end of the page | |
1941 | * fault when the stack is unwound, the locks are released, | |
1942 | * and when we know whether the fault was overall successful. | |
867578cb | 1943 | */ |
49426420 JW |
1944 | css_get(&memcg->css); |
1945 | current->memcg_oom.memcg = memcg; | |
1946 | current->memcg_oom.gfp_mask = mask; | |
1947 | current->memcg_oom.order = order; | |
3812c8c8 JW |
1948 | } |
1949 | ||
1950 | /** | |
1951 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1952 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1953 | * |
49426420 JW |
1954 | * This has to be called at the end of a page fault if the memcg OOM |
1955 | * handler was enabled. | |
3812c8c8 | 1956 | * |
49426420 | 1957 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1958 | * sleep on a waitqueue until the userspace task resolves the |
1959 | * situation. Sleeping directly in the charge context with all kinds | |
1960 | * of locks held is not a good idea, instead we remember an OOM state | |
1961 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1962 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1963 | * |
1964 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1965 | * completed, %false otherwise. |
3812c8c8 | 1966 | */ |
49426420 | 1967 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1968 | { |
49426420 | 1969 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 1970 | struct oom_wait_info owait; |
49426420 | 1971 | bool locked; |
3812c8c8 JW |
1972 | |
1973 | /* OOM is global, do not handle */ | |
3812c8c8 | 1974 | if (!memcg) |
49426420 | 1975 | return false; |
3812c8c8 | 1976 | |
49426420 JW |
1977 | if (!handle) |
1978 | goto cleanup; | |
3812c8c8 JW |
1979 | |
1980 | owait.memcg = memcg; | |
1981 | owait.wait.flags = 0; | |
1982 | owait.wait.func = memcg_oom_wake_function; | |
1983 | owait.wait.private = current; | |
1984 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1985 | |
3812c8c8 | 1986 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1987 | mem_cgroup_mark_under_oom(memcg); |
1988 | ||
1989 | locked = mem_cgroup_oom_trylock(memcg); | |
1990 | ||
1991 | if (locked) | |
1992 | mem_cgroup_oom_notify(memcg); | |
1993 | ||
1994 | if (locked && !memcg->oom_kill_disable) { | |
1995 | mem_cgroup_unmark_under_oom(memcg); | |
1996 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1997 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
1998 | current->memcg_oom.order); | |
1999 | } else { | |
3812c8c8 | 2000 | schedule(); |
49426420 JW |
2001 | mem_cgroup_unmark_under_oom(memcg); |
2002 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2003 | } | |
2004 | ||
2005 | if (locked) { | |
fb2a6fc5 JW |
2006 | mem_cgroup_oom_unlock(memcg); |
2007 | /* | |
2008 | * There is no guarantee that an OOM-lock contender | |
2009 | * sees the wakeups triggered by the OOM kill | |
2010 | * uncharges. Wake any sleepers explicitely. | |
2011 | */ | |
2012 | memcg_oom_recover(memcg); | |
2013 | } | |
49426420 JW |
2014 | cleanup: |
2015 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 2016 | css_put(&memcg->css); |
867578cb | 2017 | return true; |
0b7f569e KH |
2018 | } |
2019 | ||
d7365e78 JW |
2020 | /** |
2021 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
2022 | * @page: page that is going to change accounted state | |
2023 | * @locked: &memcg->move_lock slowpath was taken | |
2024 | * @flags: IRQ-state flags for &memcg->move_lock | |
32047e2a | 2025 | * |
d7365e78 JW |
2026 | * This function must mark the beginning of an accounted page state |
2027 | * change to prevent double accounting when the page is concurrently | |
2028 | * being moved to another memcg: | |
32047e2a | 2029 | * |
d7365e78 JW |
2030 | * memcg = mem_cgroup_begin_page_stat(page, &locked, &flags); |
2031 | * if (TestClearPageState(page)) | |
2032 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
2033 | * mem_cgroup_end_page_stat(memcg, locked, flags); | |
32047e2a | 2034 | * |
d7365e78 JW |
2035 | * The RCU lock is held throughout the transaction. The fast path can |
2036 | * get away without acquiring the memcg->move_lock (@locked is false) | |
2037 | * because page moving starts with an RCU grace period. | |
32047e2a | 2038 | * |
d7365e78 JW |
2039 | * The RCU lock also protects the memcg from being freed when the page |
2040 | * state that is going to change is the only thing preventing the page | |
2041 | * from being uncharged. E.g. end-writeback clearing PageWriteback(), | |
2042 | * which allows migration to go ahead and uncharge the page before the | |
2043 | * account transaction might be complete. | |
d69b042f | 2044 | */ |
d7365e78 JW |
2045 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page, |
2046 | bool *locked, | |
2047 | unsigned long *flags) | |
89c06bd5 KH |
2048 | { |
2049 | struct mem_cgroup *memcg; | |
2050 | struct page_cgroup *pc; | |
2051 | ||
d7365e78 JW |
2052 | rcu_read_lock(); |
2053 | ||
2054 | if (mem_cgroup_disabled()) | |
2055 | return NULL; | |
2056 | ||
89c06bd5 KH |
2057 | pc = lookup_page_cgroup(page); |
2058 | again: | |
2059 | memcg = pc->mem_cgroup; | |
29833315 | 2060 | if (unlikely(!memcg)) |
d7365e78 JW |
2061 | return NULL; |
2062 | ||
2063 | *locked = false; | |
bdcbb659 | 2064 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2065 | return memcg; |
89c06bd5 | 2066 | |
354a4783 | 2067 | spin_lock_irqsave(&memcg->move_lock, *flags); |
29833315 | 2068 | if (memcg != pc->mem_cgroup) { |
354a4783 | 2069 | spin_unlock_irqrestore(&memcg->move_lock, *flags); |
89c06bd5 KH |
2070 | goto again; |
2071 | } | |
2072 | *locked = true; | |
d7365e78 JW |
2073 | |
2074 | return memcg; | |
89c06bd5 KH |
2075 | } |
2076 | ||
d7365e78 JW |
2077 | /** |
2078 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2079 | * @memcg: the memcg that was accounted against | |
2080 | * @locked: value received from mem_cgroup_begin_page_stat() | |
2081 | * @flags: value received from mem_cgroup_begin_page_stat() | |
2082 | */ | |
2083 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg, bool locked, | |
2084 | unsigned long flags) | |
89c06bd5 | 2085 | { |
d7365e78 | 2086 | if (memcg && locked) |
354a4783 | 2087 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 | 2088 | |
d7365e78 | 2089 | rcu_read_unlock(); |
89c06bd5 KH |
2090 | } |
2091 | ||
d7365e78 JW |
2092 | /** |
2093 | * mem_cgroup_update_page_stat - update page state statistics | |
2094 | * @memcg: memcg to account against | |
2095 | * @idx: page state item to account | |
2096 | * @val: number of pages (positive or negative) | |
2097 | * | |
2098 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2099 | */ | |
2100 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2101 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2102 | { |
658b72c5 | 2103 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2104 | |
d7365e78 JW |
2105 | if (memcg) |
2106 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2107 | } |
26174efd | 2108 | |
cdec2e42 KH |
2109 | /* |
2110 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2111 | * TODO: maybe necessary to use big numbers in big irons. | |
2112 | */ | |
7ec99d62 | 2113 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2114 | struct memcg_stock_pcp { |
2115 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2116 | unsigned int nr_pages; |
cdec2e42 | 2117 | struct work_struct work; |
26fe6168 | 2118 | unsigned long flags; |
a0db00fc | 2119 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2120 | }; |
2121 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2122 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2123 | |
a0956d54 SS |
2124 | /** |
2125 | * consume_stock: Try to consume stocked charge on this cpu. | |
2126 | * @memcg: memcg to consume from. | |
2127 | * @nr_pages: how many pages to charge. | |
2128 | * | |
2129 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2130 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2131 | * service an allocation will refill the stock. | |
2132 | * | |
2133 | * returns true if successful, false otherwise. | |
cdec2e42 | 2134 | */ |
a0956d54 | 2135 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2136 | { |
2137 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2138 | bool ret = false; |
cdec2e42 | 2139 | |
a0956d54 | 2140 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2141 | return ret; |
a0956d54 | 2142 | |
cdec2e42 | 2143 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2144 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2145 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2146 | ret = true; |
2147 | } | |
cdec2e42 KH |
2148 | put_cpu_var(memcg_stock); |
2149 | return ret; | |
2150 | } | |
2151 | ||
2152 | /* | |
3e32cb2e | 2153 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2154 | */ |
2155 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2156 | { | |
2157 | struct mem_cgroup *old = stock->cached; | |
2158 | ||
11c9ea4e | 2159 | if (stock->nr_pages) { |
3e32cb2e | 2160 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2161 | if (do_swap_account) |
3e32cb2e | 2162 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2163 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2164 | stock->nr_pages = 0; |
cdec2e42 KH |
2165 | } |
2166 | stock->cached = NULL; | |
cdec2e42 KH |
2167 | } |
2168 | ||
2169 | /* | |
2170 | * This must be called under preempt disabled or must be called by | |
2171 | * a thread which is pinned to local cpu. | |
2172 | */ | |
2173 | static void drain_local_stock(struct work_struct *dummy) | |
2174 | { | |
7c8e0181 | 2175 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2176 | drain_stock(stock); |
26fe6168 | 2177 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2178 | } |
2179 | ||
e4777496 MH |
2180 | static void __init memcg_stock_init(void) |
2181 | { | |
2182 | int cpu; | |
2183 | ||
2184 | for_each_possible_cpu(cpu) { | |
2185 | struct memcg_stock_pcp *stock = | |
2186 | &per_cpu(memcg_stock, cpu); | |
2187 | INIT_WORK(&stock->work, drain_local_stock); | |
2188 | } | |
2189 | } | |
2190 | ||
cdec2e42 | 2191 | /* |
3e32cb2e | 2192 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2193 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2194 | */ |
c0ff4b85 | 2195 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2196 | { |
2197 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2198 | ||
c0ff4b85 | 2199 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2200 | drain_stock(stock); |
c0ff4b85 | 2201 | stock->cached = memcg; |
cdec2e42 | 2202 | } |
11c9ea4e | 2203 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2204 | put_cpu_var(memcg_stock); |
2205 | } | |
2206 | ||
2207 | /* | |
c0ff4b85 | 2208 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2209 | * of the hierarchy under it. |
cdec2e42 | 2210 | */ |
6d3d6aa2 | 2211 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2212 | { |
26fe6168 | 2213 | int cpu, curcpu; |
d38144b7 | 2214 | |
6d3d6aa2 JW |
2215 | /* If someone's already draining, avoid adding running more workers. */ |
2216 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2217 | return; | |
cdec2e42 | 2218 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2219 | get_online_cpus(); |
5af12d0e | 2220 | curcpu = get_cpu(); |
cdec2e42 KH |
2221 | for_each_online_cpu(cpu) { |
2222 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2223 | struct mem_cgroup *memcg; |
26fe6168 | 2224 | |
c0ff4b85 R |
2225 | memcg = stock->cached; |
2226 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2227 | continue; |
c0ff4b85 | 2228 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2229 | continue; |
d1a05b69 MH |
2230 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2231 | if (cpu == curcpu) | |
2232 | drain_local_stock(&stock->work); | |
2233 | else | |
2234 | schedule_work_on(cpu, &stock->work); | |
2235 | } | |
cdec2e42 | 2236 | } |
5af12d0e | 2237 | put_cpu(); |
f894ffa8 | 2238 | put_online_cpus(); |
9f50fad6 | 2239 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2240 | } |
2241 | ||
711d3d2c KH |
2242 | /* |
2243 | * This function drains percpu counter value from DEAD cpu and | |
2244 | * move it to local cpu. Note that this function can be preempted. | |
2245 | */ | |
c0ff4b85 | 2246 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2247 | { |
2248 | int i; | |
2249 | ||
c0ff4b85 | 2250 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2251 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2252 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2253 | |
c0ff4b85 R |
2254 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2255 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2256 | } |
e9f8974f | 2257 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2258 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2259 | |
c0ff4b85 R |
2260 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2261 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2262 | } |
c0ff4b85 | 2263 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2264 | } |
2265 | ||
0db0628d | 2266 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2267 | unsigned long action, |
2268 | void *hcpu) | |
2269 | { | |
2270 | int cpu = (unsigned long)hcpu; | |
2271 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2272 | struct mem_cgroup *iter; |
cdec2e42 | 2273 | |
619d094b | 2274 | if (action == CPU_ONLINE) |
1489ebad | 2275 | return NOTIFY_OK; |
1489ebad | 2276 | |
d833049b | 2277 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2278 | return NOTIFY_OK; |
711d3d2c | 2279 | |
9f3a0d09 | 2280 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2281 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2282 | ||
cdec2e42 KH |
2283 | stock = &per_cpu(memcg_stock, cpu); |
2284 | drain_stock(stock); | |
2285 | return NOTIFY_OK; | |
2286 | } | |
2287 | ||
00501b53 JW |
2288 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2289 | unsigned int nr_pages) | |
8a9f3ccd | 2290 | { |
7ec99d62 | 2291 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2292 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2293 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2294 | struct page_counter *counter; |
6539cc05 | 2295 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2296 | bool may_swap = true; |
2297 | bool drained = false; | |
05b84301 | 2298 | int ret = 0; |
a636b327 | 2299 | |
ce00a967 JW |
2300 | if (mem_cgroup_is_root(memcg)) |
2301 | goto done; | |
6539cc05 | 2302 | retry: |
b6b6cc72 MH |
2303 | if (consume_stock(memcg, nr_pages)) |
2304 | goto done; | |
8a9f3ccd | 2305 | |
3fbe7244 | 2306 | if (!do_swap_account || |
3e32cb2e JW |
2307 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2308 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2309 | goto done_restock; |
3fbe7244 | 2310 | if (do_swap_account) |
3e32cb2e JW |
2311 | page_counter_uncharge(&memcg->memsw, batch); |
2312 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2313 | } else { |
3e32cb2e | 2314 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2315 | may_swap = false; |
3fbe7244 | 2316 | } |
7a81b88c | 2317 | |
6539cc05 JW |
2318 | if (batch > nr_pages) { |
2319 | batch = nr_pages; | |
2320 | goto retry; | |
2321 | } | |
6d61ef40 | 2322 | |
06b078fc JW |
2323 | /* |
2324 | * Unlike in global OOM situations, memcg is not in a physical | |
2325 | * memory shortage. Allow dying and OOM-killed tasks to | |
2326 | * bypass the last charges so that they can exit quickly and | |
2327 | * free their memory. | |
2328 | */ | |
2329 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2330 | fatal_signal_pending(current) || | |
2331 | current->flags & PF_EXITING)) | |
2332 | goto bypass; | |
2333 | ||
2334 | if (unlikely(task_in_memcg_oom(current))) | |
2335 | goto nomem; | |
2336 | ||
6539cc05 JW |
2337 | if (!(gfp_mask & __GFP_WAIT)) |
2338 | goto nomem; | |
4b534334 | 2339 | |
b70a2a21 JW |
2340 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2341 | gfp_mask, may_swap); | |
6539cc05 | 2342 | |
61e02c74 | 2343 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2344 | goto retry; |
28c34c29 | 2345 | |
b70a2a21 | 2346 | if (!drained) { |
6d3d6aa2 | 2347 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2348 | drained = true; |
2349 | goto retry; | |
2350 | } | |
2351 | ||
28c34c29 JW |
2352 | if (gfp_mask & __GFP_NORETRY) |
2353 | goto nomem; | |
6539cc05 JW |
2354 | /* |
2355 | * Even though the limit is exceeded at this point, reclaim | |
2356 | * may have been able to free some pages. Retry the charge | |
2357 | * before killing the task. | |
2358 | * | |
2359 | * Only for regular pages, though: huge pages are rather | |
2360 | * unlikely to succeed so close to the limit, and we fall back | |
2361 | * to regular pages anyway in case of failure. | |
2362 | */ | |
61e02c74 | 2363 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2364 | goto retry; |
2365 | /* | |
2366 | * At task move, charge accounts can be doubly counted. So, it's | |
2367 | * better to wait until the end of task_move if something is going on. | |
2368 | */ | |
2369 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2370 | goto retry; | |
2371 | ||
9b130619 JW |
2372 | if (nr_retries--) |
2373 | goto retry; | |
2374 | ||
06b078fc JW |
2375 | if (gfp_mask & __GFP_NOFAIL) |
2376 | goto bypass; | |
2377 | ||
6539cc05 JW |
2378 | if (fatal_signal_pending(current)) |
2379 | goto bypass; | |
2380 | ||
61e02c74 | 2381 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2382 | nomem: |
6d1fdc48 | 2383 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2384 | return -ENOMEM; |
867578cb | 2385 | bypass: |
ce00a967 | 2386 | return -EINTR; |
6539cc05 JW |
2387 | |
2388 | done_restock: | |
e8ea14cc | 2389 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2390 | if (batch > nr_pages) |
2391 | refill_stock(memcg, batch - nr_pages); | |
2392 | done: | |
05b84301 | 2393 | return ret; |
7a81b88c | 2394 | } |
8a9f3ccd | 2395 | |
00501b53 | 2396 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2397 | { |
ce00a967 JW |
2398 | if (mem_cgroup_is_root(memcg)) |
2399 | return; | |
2400 | ||
3e32cb2e | 2401 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2402 | if (do_swap_account) |
3e32cb2e | 2403 | page_counter_uncharge(&memcg->memsw, nr_pages); |
e8ea14cc JW |
2404 | |
2405 | css_put_many(&memcg->css, nr_pages); | |
d01dd17f KH |
2406 | } |
2407 | ||
a3b2d692 KH |
2408 | /* |
2409 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2410 | * rcu_read_lock(). The caller is responsible for calling |
2411 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2412 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2413 | */ |
2414 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2415 | { | |
a3b2d692 KH |
2416 | /* ID 0 is unused ID */ |
2417 | if (!id) | |
2418 | return NULL; | |
34c00c31 | 2419 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2420 | } |
2421 | ||
0a31bc97 JW |
2422 | /* |
2423 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2424 | * @page: the page | |
2425 | * | |
2426 | * Look up, get a css reference, and return the memcg that owns @page. | |
2427 | * | |
2428 | * The page must be locked to prevent racing with swap-in and page | |
2429 | * cache charges. If coming from an unlocked page table, the caller | |
2430 | * must ensure the page is on the LRU or this can race with charging. | |
2431 | */ | |
e42d9d5d | 2432 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2433 | { |
29833315 | 2434 | struct mem_cgroup *memcg; |
3c776e64 | 2435 | struct page_cgroup *pc; |
a3b2d692 | 2436 | unsigned short id; |
b5a84319 KH |
2437 | swp_entry_t ent; |
2438 | ||
309381fe | 2439 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2440 | |
3c776e64 | 2441 | pc = lookup_page_cgroup(page); |
29833315 JW |
2442 | memcg = pc->mem_cgroup; |
2443 | ||
2444 | if (memcg) { | |
2445 | if (!css_tryget_online(&memcg->css)) | |
c0ff4b85 | 2446 | memcg = NULL; |
e42d9d5d | 2447 | } else if (PageSwapCache(page)) { |
3c776e64 | 2448 | ent.val = page_private(page); |
9fb4b7cc | 2449 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2450 | rcu_read_lock(); |
c0ff4b85 | 2451 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2452 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2453 | memcg = NULL; |
a3b2d692 | 2454 | rcu_read_unlock(); |
3c776e64 | 2455 | } |
c0ff4b85 | 2456 | return memcg; |
b5a84319 KH |
2457 | } |
2458 | ||
0a31bc97 JW |
2459 | static void lock_page_lru(struct page *page, int *isolated) |
2460 | { | |
2461 | struct zone *zone = page_zone(page); | |
2462 | ||
2463 | spin_lock_irq(&zone->lru_lock); | |
2464 | if (PageLRU(page)) { | |
2465 | struct lruvec *lruvec; | |
2466 | ||
2467 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2468 | ClearPageLRU(page); | |
2469 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2470 | *isolated = 1; | |
2471 | } else | |
2472 | *isolated = 0; | |
2473 | } | |
2474 | ||
2475 | static void unlock_page_lru(struct page *page, int isolated) | |
2476 | { | |
2477 | struct zone *zone = page_zone(page); | |
2478 | ||
2479 | if (isolated) { | |
2480 | struct lruvec *lruvec; | |
2481 | ||
2482 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2483 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2484 | SetPageLRU(page); | |
2485 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2486 | } | |
2487 | spin_unlock_irq(&zone->lru_lock); | |
2488 | } | |
2489 | ||
00501b53 | 2490 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2491 | bool lrucare) |
7a81b88c | 2492 | { |
ce587e65 | 2493 | struct page_cgroup *pc = lookup_page_cgroup(page); |
0a31bc97 | 2494 | int isolated; |
9ce70c02 | 2495 | |
29833315 | 2496 | VM_BUG_ON_PAGE(pc->mem_cgroup, page); |
ca3e0214 KH |
2497 | /* |
2498 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2499 | * accessed by any other context at this point. | |
2500 | */ | |
9ce70c02 HD |
2501 | |
2502 | /* | |
2503 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2504 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2505 | */ | |
0a31bc97 JW |
2506 | if (lrucare) |
2507 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2508 | |
0a31bc97 JW |
2509 | /* |
2510 | * Nobody should be changing or seriously looking at | |
29833315 | 2511 | * pc->mem_cgroup at this point: |
0a31bc97 JW |
2512 | * |
2513 | * - the page is uncharged | |
2514 | * | |
2515 | * - the page is off-LRU | |
2516 | * | |
2517 | * - an anonymous fault has exclusive page access, except for | |
2518 | * a locked page table | |
2519 | * | |
2520 | * - a page cache insertion, a swapin fault, or a migration | |
2521 | * have the page locked | |
2522 | */ | |
c0ff4b85 | 2523 | pc->mem_cgroup = memcg; |
9ce70c02 | 2524 | |
0a31bc97 JW |
2525 | if (lrucare) |
2526 | unlock_page_lru(page, isolated); | |
7a81b88c | 2527 | } |
66e1707b | 2528 | |
7ae1e1d0 | 2529 | #ifdef CONFIG_MEMCG_KMEM |
bd673145 VD |
2530 | /* |
2531 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or | |
2532 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. | |
2533 | */ | |
2534 | static DEFINE_MUTEX(memcg_slab_mutex); | |
2535 | ||
1f458cbf GC |
2536 | /* |
2537 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2538 | * in the memcg_cache_params struct. | |
2539 | */ | |
2540 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2541 | { | |
2542 | struct kmem_cache *cachep; | |
2543 | ||
2544 | VM_BUG_ON(p->is_root_cache); | |
2545 | cachep = p->root_cache; | |
7a67d7ab | 2546 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
1f458cbf GC |
2547 | } |
2548 | ||
3e32cb2e JW |
2549 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2550 | unsigned long nr_pages) | |
7ae1e1d0 | 2551 | { |
3e32cb2e | 2552 | struct page_counter *counter; |
7ae1e1d0 | 2553 | int ret = 0; |
7ae1e1d0 | 2554 | |
3e32cb2e JW |
2555 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2556 | if (ret < 0) | |
7ae1e1d0 GC |
2557 | return ret; |
2558 | ||
3e32cb2e | 2559 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2560 | if (ret == -EINTR) { |
2561 | /* | |
00501b53 JW |
2562 | * try_charge() chose to bypass to root due to OOM kill or |
2563 | * fatal signal. Since our only options are to either fail | |
2564 | * the allocation or charge it to this cgroup, do it as a | |
2565 | * temporary condition. But we can't fail. From a kmem/slab | |
2566 | * perspective, the cache has already been selected, by | |
2567 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2568 | * our minds. |
2569 | * | |
2570 | * This condition will only trigger if the task entered | |
00501b53 JW |
2571 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2572 | * during try_charge() above. Tasks that were already dying | |
2573 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2574 | * directed to the root cgroup in memcontrol.h |
2575 | */ | |
3e32cb2e | 2576 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2577 | if (do_swap_account) |
3e32cb2e | 2578 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2579 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2580 | ret = 0; |
2581 | } else if (ret) | |
3e32cb2e | 2582 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2583 | |
2584 | return ret; | |
2585 | } | |
2586 | ||
3e32cb2e JW |
2587 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, |
2588 | unsigned long nr_pages) | |
7ae1e1d0 | 2589 | { |
3e32cb2e | 2590 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2591 | if (do_swap_account) |
3e32cb2e | 2592 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2593 | |
64f21993 | 2594 | page_counter_uncharge(&memcg->kmem, nr_pages); |
e8ea14cc JW |
2595 | |
2596 | css_put_many(&memcg->css, nr_pages); | |
7ae1e1d0 GC |
2597 | } |
2598 | ||
2633d7a0 GC |
2599 | /* |
2600 | * helper for acessing a memcg's index. It will be used as an index in the | |
2601 | * child cache array in kmem_cache, and also to derive its name. This function | |
2602 | * will return -1 when this is not a kmem-limited memcg. | |
2603 | */ | |
2604 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2605 | { | |
2606 | return memcg ? memcg->kmemcg_id : -1; | |
2607 | } | |
2608 | ||
f3bb3043 | 2609 | static int memcg_alloc_cache_id(void) |
55007d84 | 2610 | { |
f3bb3043 VD |
2611 | int id, size; |
2612 | int err; | |
2613 | ||
2614 | id = ida_simple_get(&kmem_limited_groups, | |
2615 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2616 | if (id < 0) | |
2617 | return id; | |
55007d84 | 2618 | |
f3bb3043 VD |
2619 | if (id < memcg_limited_groups_array_size) |
2620 | return id; | |
2621 | ||
2622 | /* | |
2623 | * There's no space for the new id in memcg_caches arrays, | |
2624 | * so we have to grow them. | |
2625 | */ | |
2626 | ||
2627 | size = 2 * (id + 1); | |
55007d84 GC |
2628 | if (size < MEMCG_CACHES_MIN_SIZE) |
2629 | size = MEMCG_CACHES_MIN_SIZE; | |
2630 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2631 | size = MEMCG_CACHES_MAX_SIZE; | |
2632 | ||
f3bb3043 VD |
2633 | mutex_lock(&memcg_slab_mutex); |
2634 | err = memcg_update_all_caches(size); | |
2635 | mutex_unlock(&memcg_slab_mutex); | |
2636 | ||
2637 | if (err) { | |
2638 | ida_simple_remove(&kmem_limited_groups, id); | |
2639 | return err; | |
2640 | } | |
2641 | return id; | |
2642 | } | |
2643 | ||
2644 | static void memcg_free_cache_id(int id) | |
2645 | { | |
2646 | ida_simple_remove(&kmem_limited_groups, id); | |
55007d84 GC |
2647 | } |
2648 | ||
2649 | /* | |
2650 | * We should update the current array size iff all caches updates succeed. This | |
2651 | * can only be done from the slab side. The slab mutex needs to be held when | |
2652 | * calling this. | |
2653 | */ | |
2654 | void memcg_update_array_size(int num) | |
2655 | { | |
f3bb3043 | 2656 | memcg_limited_groups_array_size = num; |
55007d84 GC |
2657 | } |
2658 | ||
776ed0f0 VD |
2659 | static void memcg_register_cache(struct mem_cgroup *memcg, |
2660 | struct kmem_cache *root_cache) | |
2633d7a0 | 2661 | { |
93f39eea VD |
2662 | static char memcg_name_buf[NAME_MAX + 1]; /* protected by |
2663 | memcg_slab_mutex */ | |
bd673145 | 2664 | struct kmem_cache *cachep; |
d7f25f8a GC |
2665 | int id; |
2666 | ||
bd673145 VD |
2667 | lockdep_assert_held(&memcg_slab_mutex); |
2668 | ||
2669 | id = memcg_cache_id(memcg); | |
2670 | ||
2671 | /* | |
2672 | * Since per-memcg caches are created asynchronously on first | |
2673 | * allocation (see memcg_kmem_get_cache()), several threads can try to | |
2674 | * create the same cache, but only one of them may succeed. | |
2675 | */ | |
2676 | if (cache_from_memcg_idx(root_cache, id)) | |
1aa13254 VD |
2677 | return; |
2678 | ||
073ee1c6 | 2679 | cgroup_name(memcg->css.cgroup, memcg_name_buf, NAME_MAX + 1); |
776ed0f0 | 2680 | cachep = memcg_create_kmem_cache(memcg, root_cache, memcg_name_buf); |
2edefe11 | 2681 | /* |
bd673145 VD |
2682 | * If we could not create a memcg cache, do not complain, because |
2683 | * that's not critical at all as we can always proceed with the root | |
2684 | * cache. | |
2edefe11 | 2685 | */ |
bd673145 VD |
2686 | if (!cachep) |
2687 | return; | |
2edefe11 | 2688 | |
33a690c4 | 2689 | css_get(&memcg->css); |
bd673145 | 2690 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
1aa13254 | 2691 | |
d7f25f8a | 2692 | /* |
959c8963 VD |
2693 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
2694 | * barrier here to ensure nobody will see the kmem_cache partially | |
2695 | * initialized. | |
d7f25f8a | 2696 | */ |
959c8963 VD |
2697 | smp_wmb(); |
2698 | ||
bd673145 VD |
2699 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
2700 | root_cache->memcg_params->memcg_caches[id] = cachep; | |
1aa13254 | 2701 | } |
d7f25f8a | 2702 | |
776ed0f0 | 2703 | static void memcg_unregister_cache(struct kmem_cache *cachep) |
1aa13254 | 2704 | { |
bd673145 | 2705 | struct kmem_cache *root_cache; |
1aa13254 VD |
2706 | struct mem_cgroup *memcg; |
2707 | int id; | |
2708 | ||
bd673145 | 2709 | lockdep_assert_held(&memcg_slab_mutex); |
d7f25f8a | 2710 | |
bd673145 | 2711 | BUG_ON(is_root_cache(cachep)); |
2edefe11 | 2712 | |
bd673145 VD |
2713 | root_cache = cachep->memcg_params->root_cache; |
2714 | memcg = cachep->memcg_params->memcg; | |
96403da2 | 2715 | id = memcg_cache_id(memcg); |
d7f25f8a | 2716 | |
bd673145 VD |
2717 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
2718 | root_cache->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a | 2719 | |
bd673145 VD |
2720 | list_del(&cachep->memcg_params->list); |
2721 | ||
2722 | kmem_cache_destroy(cachep); | |
33a690c4 VD |
2723 | |
2724 | /* drop the reference taken in memcg_register_cache */ | |
2725 | css_put(&memcg->css); | |
2633d7a0 GC |
2726 | } |
2727 | ||
0e9d92f2 GC |
2728 | /* |
2729 | * During the creation a new cache, we need to disable our accounting mechanism | |
2730 | * altogether. This is true even if we are not creating, but rather just | |
2731 | * enqueing new caches to be created. | |
2732 | * | |
2733 | * This is because that process will trigger allocations; some visible, like | |
2734 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
2735 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
2736 | * objects during debug. | |
2737 | * | |
2738 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
2739 | * to it. This may not be a bounded recursion: since the first cache creation | |
2740 | * failed to complete (waiting on the allocation), we'll just try to create the | |
2741 | * cache again, failing at the same point. | |
2742 | * | |
2743 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
2744 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
2745 | * inside the following two functions. | |
2746 | */ | |
2747 | static inline void memcg_stop_kmem_account(void) | |
2748 | { | |
2749 | VM_BUG_ON(!current->mm); | |
2750 | current->memcg_kmem_skip_account++; | |
2751 | } | |
2752 | ||
2753 | static inline void memcg_resume_kmem_account(void) | |
2754 | { | |
2755 | VM_BUG_ON(!current->mm); | |
2756 | current->memcg_kmem_skip_account--; | |
2757 | } | |
2758 | ||
776ed0f0 | 2759 | int __memcg_cleanup_cache_params(struct kmem_cache *s) |
7cf27982 GC |
2760 | { |
2761 | struct kmem_cache *c; | |
b8529907 | 2762 | int i, failed = 0; |
7cf27982 | 2763 | |
bd673145 | 2764 | mutex_lock(&memcg_slab_mutex); |
7a67d7ab QH |
2765 | for_each_memcg_cache_index(i) { |
2766 | c = cache_from_memcg_idx(s, i); | |
7cf27982 GC |
2767 | if (!c) |
2768 | continue; | |
2769 | ||
776ed0f0 | 2770 | memcg_unregister_cache(c); |
b8529907 VD |
2771 | |
2772 | if (cache_from_memcg_idx(s, i)) | |
2773 | failed++; | |
7cf27982 | 2774 | } |
bd673145 | 2775 | mutex_unlock(&memcg_slab_mutex); |
b8529907 | 2776 | return failed; |
7cf27982 GC |
2777 | } |
2778 | ||
776ed0f0 | 2779 | static void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
2780 | { |
2781 | struct kmem_cache *cachep; | |
bd673145 | 2782 | struct memcg_cache_params *params, *tmp; |
1f458cbf GC |
2783 | |
2784 | if (!memcg_kmem_is_active(memcg)) | |
2785 | return; | |
2786 | ||
bd673145 VD |
2787 | mutex_lock(&memcg_slab_mutex); |
2788 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { | |
1f458cbf | 2789 | cachep = memcg_params_to_cache(params); |
bd673145 VD |
2790 | kmem_cache_shrink(cachep); |
2791 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
776ed0f0 | 2792 | memcg_unregister_cache(cachep); |
1f458cbf | 2793 | } |
bd673145 | 2794 | mutex_unlock(&memcg_slab_mutex); |
1f458cbf GC |
2795 | } |
2796 | ||
776ed0f0 | 2797 | struct memcg_register_cache_work { |
5722d094 VD |
2798 | struct mem_cgroup *memcg; |
2799 | struct kmem_cache *cachep; | |
2800 | struct work_struct work; | |
2801 | }; | |
2802 | ||
776ed0f0 | 2803 | static void memcg_register_cache_func(struct work_struct *w) |
d7f25f8a | 2804 | { |
776ed0f0 VD |
2805 | struct memcg_register_cache_work *cw = |
2806 | container_of(w, struct memcg_register_cache_work, work); | |
5722d094 VD |
2807 | struct mem_cgroup *memcg = cw->memcg; |
2808 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2809 | |
bd673145 | 2810 | mutex_lock(&memcg_slab_mutex); |
776ed0f0 | 2811 | memcg_register_cache(memcg, cachep); |
bd673145 VD |
2812 | mutex_unlock(&memcg_slab_mutex); |
2813 | ||
5722d094 | 2814 | css_put(&memcg->css); |
d7f25f8a GC |
2815 | kfree(cw); |
2816 | } | |
2817 | ||
2818 | /* | |
2819 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2820 | */ |
776ed0f0 VD |
2821 | static void __memcg_schedule_register_cache(struct mem_cgroup *memcg, |
2822 | struct kmem_cache *cachep) | |
d7f25f8a | 2823 | { |
776ed0f0 | 2824 | struct memcg_register_cache_work *cw; |
d7f25f8a | 2825 | |
776ed0f0 | 2826 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
ca0dde97 LZ |
2827 | if (cw == NULL) { |
2828 | css_put(&memcg->css); | |
d7f25f8a GC |
2829 | return; |
2830 | } | |
2831 | ||
2832 | cw->memcg = memcg; | |
2833 | cw->cachep = cachep; | |
2834 | ||
776ed0f0 | 2835 | INIT_WORK(&cw->work, memcg_register_cache_func); |
d7f25f8a GC |
2836 | schedule_work(&cw->work); |
2837 | } | |
2838 | ||
776ed0f0 VD |
2839 | static void memcg_schedule_register_cache(struct mem_cgroup *memcg, |
2840 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2841 | { |
2842 | /* | |
2843 | * We need to stop accounting when we kmalloc, because if the | |
2844 | * corresponding kmalloc cache is not yet created, the first allocation | |
776ed0f0 | 2845 | * in __memcg_schedule_register_cache will recurse. |
0e9d92f2 GC |
2846 | * |
2847 | * However, it is better to enclose the whole function. Depending on | |
2848 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2849 | * trigger an allocation. This too, will make us recurse. Because at | |
2850 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2851 | * the safest choice is to do it like this, wrapping the whole function. | |
2852 | */ | |
2853 | memcg_stop_kmem_account(); | |
776ed0f0 | 2854 | __memcg_schedule_register_cache(memcg, cachep); |
0e9d92f2 GC |
2855 | memcg_resume_kmem_account(); |
2856 | } | |
c67a8a68 VD |
2857 | |
2858 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) | |
2859 | { | |
3e32cb2e | 2860 | unsigned int nr_pages = 1 << order; |
c67a8a68 VD |
2861 | int res; |
2862 | ||
3e32cb2e | 2863 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, nr_pages); |
c67a8a68 | 2864 | if (!res) |
3e32cb2e | 2865 | atomic_add(nr_pages, &cachep->memcg_params->nr_pages); |
c67a8a68 VD |
2866 | return res; |
2867 | } | |
2868 | ||
2869 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) | |
2870 | { | |
3e32cb2e JW |
2871 | unsigned int nr_pages = 1 << order; |
2872 | ||
2873 | memcg_uncharge_kmem(cachep->memcg_params->memcg, nr_pages); | |
2874 | atomic_sub(nr_pages, &cachep->memcg_params->nr_pages); | |
c67a8a68 VD |
2875 | } |
2876 | ||
d7f25f8a GC |
2877 | /* |
2878 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2879 | * We try to use the current memcg's version of the cache. | |
2880 | * | |
2881 | * If the cache does not exist yet, if we are the first user of it, | |
2882 | * we either create it immediately, if possible, or create it asynchronously | |
2883 | * in a workqueue. | |
2884 | * In the latter case, we will let the current allocation go through with | |
2885 | * the original cache. | |
2886 | * | |
2887 | * Can't be called in interrupt context or from kernel threads. | |
2888 | * This function needs to be called with rcu_read_lock() held. | |
2889 | */ | |
2890 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
2891 | gfp_t gfp) | |
2892 | { | |
2893 | struct mem_cgroup *memcg; | |
959c8963 | 2894 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
2895 | |
2896 | VM_BUG_ON(!cachep->memcg_params); | |
2897 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
2898 | ||
0e9d92f2 GC |
2899 | if (!current->mm || current->memcg_kmem_skip_account) |
2900 | return cachep; | |
2901 | ||
d7f25f8a GC |
2902 | rcu_read_lock(); |
2903 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a | 2904 | |
cf2b8fbf | 2905 | if (!memcg_kmem_is_active(memcg)) |
ca0dde97 | 2906 | goto out; |
d7f25f8a | 2907 | |
959c8963 VD |
2908 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
2909 | if (likely(memcg_cachep)) { | |
2910 | cachep = memcg_cachep; | |
ca0dde97 | 2911 | goto out; |
d7f25f8a GC |
2912 | } |
2913 | ||
ca0dde97 | 2914 | /* The corresponding put will be done in the workqueue. */ |
ec903c0c | 2915 | if (!css_tryget_online(&memcg->css)) |
ca0dde97 LZ |
2916 | goto out; |
2917 | rcu_read_unlock(); | |
2918 | ||
2919 | /* | |
2920 | * If we are in a safe context (can wait, and not in interrupt | |
2921 | * context), we could be be predictable and return right away. | |
2922 | * This would guarantee that the allocation being performed | |
2923 | * already belongs in the new cache. | |
2924 | * | |
2925 | * However, there are some clashes that can arrive from locking. | |
2926 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2927 | * memcg_create_kmem_cache, this means no further allocation |
2928 | * could happen with the slab_mutex held. So it's better to | |
2929 | * defer everything. | |
ca0dde97 | 2930 | */ |
776ed0f0 | 2931 | memcg_schedule_register_cache(memcg, cachep); |
ca0dde97 LZ |
2932 | return cachep; |
2933 | out: | |
2934 | rcu_read_unlock(); | |
2935 | return cachep; | |
d7f25f8a | 2936 | } |
d7f25f8a | 2937 | |
7ae1e1d0 GC |
2938 | /* |
2939 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2940 | * possible for the given order. But the page is not allocated yet, so we'll | |
2941 | * need a further commit step to do the final arrangements. | |
2942 | * | |
2943 | * It is possible for the task to switch cgroups in this mean time, so at | |
2944 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2945 | * the handle argument to return to the caller which cgroup we should commit | |
2946 | * against. We could also return the memcg directly and avoid the pointer | |
2947 | * passing, but a boolean return value gives better semantics considering | |
2948 | * the compiled-out case as well. | |
2949 | * | |
2950 | * Returning true means the allocation is possible. | |
2951 | */ | |
2952 | bool | |
2953 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2954 | { | |
2955 | struct mem_cgroup *memcg; | |
2956 | int ret; | |
2957 | ||
2958 | *_memcg = NULL; | |
6d42c232 GC |
2959 | |
2960 | /* | |
2961 | * Disabling accounting is only relevant for some specific memcg | |
2962 | * internal allocations. Therefore we would initially not have such | |
52383431 VD |
2963 | * check here, since direct calls to the page allocator that are |
2964 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen | |
2965 | * outside memcg core. We are mostly concerned with cache allocations, | |
2966 | * and by having this test at memcg_kmem_get_cache, we are already able | |
2967 | * to relay the allocation to the root cache and bypass the memcg cache | |
2968 | * altogether. | |
6d42c232 GC |
2969 | * |
2970 | * There is one exception, though: the SLUB allocator does not create | |
2971 | * large order caches, but rather service large kmallocs directly from | |
2972 | * the page allocator. Therefore, the following sequence when backed by | |
2973 | * the SLUB allocator: | |
2974 | * | |
f894ffa8 AM |
2975 | * memcg_stop_kmem_account(); |
2976 | * kmalloc(<large_number>) | |
2977 | * memcg_resume_kmem_account(); | |
6d42c232 GC |
2978 | * |
2979 | * would effectively ignore the fact that we should skip accounting, | |
2980 | * since it will drive us directly to this function without passing | |
2981 | * through the cache selector memcg_kmem_get_cache. Such large | |
2982 | * allocations are extremely rare but can happen, for instance, for the | |
2983 | * cache arrays. We bring this test here. | |
2984 | */ | |
2985 | if (!current->mm || current->memcg_kmem_skip_account) | |
2986 | return true; | |
2987 | ||
df381975 | 2988 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 2989 | |
cf2b8fbf | 2990 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
2991 | css_put(&memcg->css); |
2992 | return true; | |
2993 | } | |
2994 | ||
3e32cb2e | 2995 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
2996 | if (!ret) |
2997 | *_memcg = memcg; | |
7ae1e1d0 GC |
2998 | |
2999 | css_put(&memcg->css); | |
3000 | return (ret == 0); | |
3001 | } | |
3002 | ||
3003 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3004 | int order) | |
3005 | { | |
3006 | struct page_cgroup *pc; | |
3007 | ||
3008 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3009 | ||
3010 | /* The page allocation failed. Revert */ | |
3011 | if (!page) { | |
3e32cb2e | 3012 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
3013 | return; |
3014 | } | |
7ae1e1d0 | 3015 | pc = lookup_page_cgroup(page); |
7ae1e1d0 | 3016 | pc->mem_cgroup = memcg; |
7ae1e1d0 GC |
3017 | } |
3018 | ||
3019 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3020 | { | |
29833315 JW |
3021 | struct page_cgroup *pc = lookup_page_cgroup(page); |
3022 | struct mem_cgroup *memcg = pc->mem_cgroup; | |
7ae1e1d0 | 3023 | |
7ae1e1d0 GC |
3024 | if (!memcg) |
3025 | return; | |
3026 | ||
309381fe | 3027 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 3028 | |
3e32cb2e | 3029 | memcg_uncharge_kmem(memcg, 1 << order); |
29833315 | 3030 | pc->mem_cgroup = NULL; |
7ae1e1d0 | 3031 | } |
1f458cbf | 3032 | #else |
776ed0f0 | 3033 | static inline void memcg_unregister_all_caches(struct mem_cgroup *memcg) |
1f458cbf GC |
3034 | { |
3035 | } | |
7ae1e1d0 GC |
3036 | #endif /* CONFIG_MEMCG_KMEM */ |
3037 | ||
ca3e0214 KH |
3038 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3039 | ||
ca3e0214 KH |
3040 | /* |
3041 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3042 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3043 | * charge/uncharge will be never happen and move_account() is done under | |
3044 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3045 | */ |
e94c8a9c | 3046 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3047 | { |
29833315 | 3048 | struct page_cgroup *pc = lookup_page_cgroup(head); |
e94c8a9c | 3049 | int i; |
ca3e0214 | 3050 | |
3d37c4a9 KH |
3051 | if (mem_cgroup_disabled()) |
3052 | return; | |
b070e65c | 3053 | |
29833315 JW |
3054 | for (i = 1; i < HPAGE_PMD_NR; i++) |
3055 | pc[i].mem_cgroup = pc[0].mem_cgroup; | |
b9982f8d | 3056 | |
29833315 | 3057 | __this_cpu_sub(pc[0].mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 3058 | HPAGE_PMD_NR); |
ca3e0214 | 3059 | } |
12d27107 | 3060 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3061 | |
f817ed48 | 3062 | /** |
de3638d9 | 3063 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3064 | * @page: the page |
7ec99d62 | 3065 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3066 | * @pc: page_cgroup of the page. |
3067 | * @from: mem_cgroup which the page is moved from. | |
3068 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3069 | * | |
3070 | * The caller must confirm following. | |
08e552c6 | 3071 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3072 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3073 | * |
2f3479b1 KH |
3074 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3075 | * from old cgroup. | |
f817ed48 | 3076 | */ |
7ec99d62 JW |
3077 | static int mem_cgroup_move_account(struct page *page, |
3078 | unsigned int nr_pages, | |
3079 | struct page_cgroup *pc, | |
3080 | struct mem_cgroup *from, | |
2f3479b1 | 3081 | struct mem_cgroup *to) |
f817ed48 | 3082 | { |
de3638d9 JW |
3083 | unsigned long flags; |
3084 | int ret; | |
987eba66 | 3085 | |
f817ed48 | 3086 | VM_BUG_ON(from == to); |
309381fe | 3087 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
3088 | /* |
3089 | * The page is isolated from LRU. So, collapse function | |
3090 | * will not handle this page. But page splitting can happen. | |
3091 | * Do this check under compound_page_lock(). The caller should | |
3092 | * hold it. | |
3093 | */ | |
3094 | ret = -EBUSY; | |
7ec99d62 | 3095 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3096 | goto out; |
3097 | ||
0a31bc97 JW |
3098 | /* |
3099 | * Prevent mem_cgroup_migrate() from looking at pc->mem_cgroup | |
3100 | * of its source page while we change it: page migration takes | |
3101 | * both pages off the LRU, but page cache replacement doesn't. | |
3102 | */ | |
3103 | if (!trylock_page(page)) | |
3104 | goto out; | |
de3638d9 JW |
3105 | |
3106 | ret = -EINVAL; | |
29833315 | 3107 | if (pc->mem_cgroup != from) |
0a31bc97 | 3108 | goto out_unlock; |
de3638d9 | 3109 | |
354a4783 | 3110 | spin_lock_irqsave(&from->move_lock, flags); |
f817ed48 | 3111 | |
0a31bc97 | 3112 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
3113 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
3114 | nr_pages); | |
3115 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
3116 | nr_pages); | |
3117 | } | |
3ea67d06 | 3118 | |
59d1d256 JW |
3119 | if (PageWriteback(page)) { |
3120 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3121 | nr_pages); | |
3122 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
3123 | nr_pages); | |
3124 | } | |
3ea67d06 | 3125 | |
0a31bc97 JW |
3126 | /* |
3127 | * It is safe to change pc->mem_cgroup here because the page | |
3128 | * is referenced, charged, and isolated - we can't race with | |
3129 | * uncharging, charging, migration, or LRU putback. | |
3130 | */ | |
d69b042f | 3131 | |
854ffa8d | 3132 | /* caller should have done css_get */ |
08e552c6 | 3133 | pc->mem_cgroup = to; |
354a4783 JW |
3134 | spin_unlock_irqrestore(&from->move_lock, flags); |
3135 | ||
de3638d9 | 3136 | ret = 0; |
0a31bc97 JW |
3137 | |
3138 | local_irq_disable(); | |
3139 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 3140 | memcg_check_events(to, page); |
0a31bc97 | 3141 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 3142 | memcg_check_events(from, page); |
0a31bc97 JW |
3143 | local_irq_enable(); |
3144 | out_unlock: | |
3145 | unlock_page(page); | |
de3638d9 | 3146 | out: |
f817ed48 KH |
3147 | return ret; |
3148 | } | |
3149 | ||
c255a458 | 3150 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
3151 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
3152 | bool charge) | |
d13d1443 | 3153 | { |
0a31bc97 JW |
3154 | int val = (charge) ? 1 : -1; |
3155 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 3156 | } |
02491447 DN |
3157 | |
3158 | /** | |
3159 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3160 | * @entry: swap entry to be moved | |
3161 | * @from: mem_cgroup which the entry is moved from | |
3162 | * @to: mem_cgroup which the entry is moved to | |
3163 | * | |
3164 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3165 | * as the mem_cgroup's id of @from. | |
3166 | * | |
3167 | * Returns 0 on success, -EINVAL on failure. | |
3168 | * | |
3e32cb2e | 3169 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3170 | * both res and memsw, and called css_get(). |
3171 | */ | |
3172 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3173 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3174 | { |
3175 | unsigned short old_id, new_id; | |
3176 | ||
34c00c31 LZ |
3177 | old_id = mem_cgroup_id(from); |
3178 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3179 | |
3180 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3181 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3182 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3183 | /* |
483c30b5 | 3184 | * This function is only called from task migration context now. |
3e32cb2e | 3185 | * It postpones page_counter and refcount handling till the end |
483c30b5 | 3186 | * of task migration(mem_cgroup_clear_mc()) for performance |
4050377b LZ |
3187 | * improvement. But we cannot postpone css_get(to) because if |
3188 | * the process that has been moved to @to does swap-in, the | |
3189 | * refcount of @to might be decreased to 0. | |
3190 | * | |
3191 | * We are in attach() phase, so the cgroup is guaranteed to be | |
3192 | * alive, so we can just call css_get(). | |
02491447 | 3193 | */ |
4050377b | 3194 | css_get(&to->css); |
02491447 DN |
3195 | return 0; |
3196 | } | |
3197 | return -EINVAL; | |
3198 | } | |
3199 | #else | |
3200 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3201 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3202 | { |
3203 | return -EINVAL; | |
3204 | } | |
8c7c6e34 | 3205 | #endif |
d13d1443 | 3206 | |
f212ad7c DN |
3207 | #ifdef CONFIG_DEBUG_VM |
3208 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3209 | { | |
3210 | struct page_cgroup *pc; | |
3211 | ||
3212 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3213 | /* |
3214 | * Can be NULL while feeding pages into the page allocator for | |
3215 | * the first time, i.e. during boot or memory hotplug; | |
3216 | * or when mem_cgroup_disabled(). | |
3217 | */ | |
29833315 | 3218 | if (likely(pc) && pc->mem_cgroup) |
f212ad7c DN |
3219 | return pc; |
3220 | return NULL; | |
3221 | } | |
3222 | ||
3223 | bool mem_cgroup_bad_page_check(struct page *page) | |
3224 | { | |
3225 | if (mem_cgroup_disabled()) | |
3226 | return false; | |
3227 | ||
3228 | return lookup_page_cgroup_used(page) != NULL; | |
3229 | } | |
3230 | ||
3231 | void mem_cgroup_print_bad_page(struct page *page) | |
3232 | { | |
3233 | struct page_cgroup *pc; | |
3234 | ||
3235 | pc = lookup_page_cgroup_used(page); | |
29833315 JW |
3236 | if (pc) |
3237 | pr_alert("pc:%p pc->mem_cgroup:%p\n", pc, pc->mem_cgroup); | |
f212ad7c DN |
3238 | } |
3239 | #endif | |
3240 | ||
3e32cb2e JW |
3241 | static DEFINE_MUTEX(memcg_limit_mutex); |
3242 | ||
d38d2a75 | 3243 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3244 | unsigned long limit) |
628f4235 | 3245 | { |
3e32cb2e JW |
3246 | unsigned long curusage; |
3247 | unsigned long oldusage; | |
3248 | bool enlarge = false; | |
81d39c20 | 3249 | int retry_count; |
3e32cb2e | 3250 | int ret; |
81d39c20 KH |
3251 | |
3252 | /* | |
3253 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3254 | * is depends on callers. We set our retry-count to be function | |
3255 | * of # of children which we should visit in this loop. | |
3256 | */ | |
3e32cb2e JW |
3257 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3258 | mem_cgroup_count_children(memcg); | |
81d39c20 | 3259 | |
3e32cb2e | 3260 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 3261 | |
3e32cb2e | 3262 | do { |
628f4235 KH |
3263 | if (signal_pending(current)) { |
3264 | ret = -EINTR; | |
3265 | break; | |
3266 | } | |
3e32cb2e JW |
3267 | |
3268 | mutex_lock(&memcg_limit_mutex); | |
3269 | if (limit > memcg->memsw.limit) { | |
3270 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3271 | ret = -EINVAL; |
628f4235 KH |
3272 | break; |
3273 | } | |
3e32cb2e JW |
3274 | if (limit > memcg->memory.limit) |
3275 | enlarge = true; | |
3276 | ret = page_counter_limit(&memcg->memory, limit); | |
3277 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3278 | |
3279 | if (!ret) | |
3280 | break; | |
3281 | ||
b70a2a21 JW |
3282 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
3283 | ||
3e32cb2e | 3284 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 3285 | /* Usage is reduced ? */ |
f894ffa8 | 3286 | if (curusage >= oldusage) |
81d39c20 KH |
3287 | retry_count--; |
3288 | else | |
3289 | oldusage = curusage; | |
3e32cb2e JW |
3290 | } while (retry_count); |
3291 | ||
3c11ecf4 KH |
3292 | if (!ret && enlarge) |
3293 | memcg_oom_recover(memcg); | |
14797e23 | 3294 | |
8c7c6e34 KH |
3295 | return ret; |
3296 | } | |
3297 | ||
338c8431 | 3298 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3299 | unsigned long limit) |
8c7c6e34 | 3300 | { |
3e32cb2e JW |
3301 | unsigned long curusage; |
3302 | unsigned long oldusage; | |
3303 | bool enlarge = false; | |
81d39c20 | 3304 | int retry_count; |
3e32cb2e | 3305 | int ret; |
8c7c6e34 | 3306 | |
81d39c20 | 3307 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
3308 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
3309 | mem_cgroup_count_children(memcg); | |
3310 | ||
3311 | oldusage = page_counter_read(&memcg->memsw); | |
3312 | ||
3313 | do { | |
8c7c6e34 KH |
3314 | if (signal_pending(current)) { |
3315 | ret = -EINTR; | |
3316 | break; | |
3317 | } | |
3e32cb2e JW |
3318 | |
3319 | mutex_lock(&memcg_limit_mutex); | |
3320 | if (limit < memcg->memory.limit) { | |
3321 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3322 | ret = -EINVAL; |
8c7c6e34 KH |
3323 | break; |
3324 | } | |
3e32cb2e JW |
3325 | if (limit > memcg->memsw.limit) |
3326 | enlarge = true; | |
3327 | ret = page_counter_limit(&memcg->memsw, limit); | |
3328 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3329 | |
3330 | if (!ret) | |
3331 | break; | |
3332 | ||
b70a2a21 JW |
3333 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3334 | ||
3e32cb2e | 3335 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3336 | /* Usage is reduced ? */ |
8c7c6e34 | 3337 | if (curusage >= oldusage) |
628f4235 | 3338 | retry_count--; |
81d39c20 KH |
3339 | else |
3340 | oldusage = curusage; | |
3e32cb2e JW |
3341 | } while (retry_count); |
3342 | ||
3c11ecf4 KH |
3343 | if (!ret && enlarge) |
3344 | memcg_oom_recover(memcg); | |
3e32cb2e | 3345 | |
628f4235 KH |
3346 | return ret; |
3347 | } | |
3348 | ||
0608f43d AM |
3349 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3350 | gfp_t gfp_mask, | |
3351 | unsigned long *total_scanned) | |
3352 | { | |
3353 | unsigned long nr_reclaimed = 0; | |
3354 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3355 | unsigned long reclaimed; | |
3356 | int loop = 0; | |
3357 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3358 | unsigned long excess; |
0608f43d AM |
3359 | unsigned long nr_scanned; |
3360 | ||
3361 | if (order > 0) | |
3362 | return 0; | |
3363 | ||
3364 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3365 | /* | |
3366 | * This loop can run a while, specially if mem_cgroup's continuously | |
3367 | * keep exceeding their soft limit and putting the system under | |
3368 | * pressure | |
3369 | */ | |
3370 | do { | |
3371 | if (next_mz) | |
3372 | mz = next_mz; | |
3373 | else | |
3374 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3375 | if (!mz) | |
3376 | break; | |
3377 | ||
3378 | nr_scanned = 0; | |
3379 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3380 | gfp_mask, &nr_scanned); | |
3381 | nr_reclaimed += reclaimed; | |
3382 | *total_scanned += nr_scanned; | |
0a31bc97 | 3383 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3384 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3385 | |
3386 | /* | |
3387 | * If we failed to reclaim anything from this memory cgroup | |
3388 | * it is time to move on to the next cgroup | |
3389 | */ | |
3390 | next_mz = NULL; | |
bc2f2e7f VD |
3391 | if (!reclaimed) |
3392 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3393 | ||
3e32cb2e | 3394 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3395 | /* |
3396 | * One school of thought says that we should not add | |
3397 | * back the node to the tree if reclaim returns 0. | |
3398 | * But our reclaim could return 0, simply because due | |
3399 | * to priority we are exposing a smaller subset of | |
3400 | * memory to reclaim from. Consider this as a longer | |
3401 | * term TODO. | |
3402 | */ | |
3403 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3404 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3405 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3406 | css_put(&mz->memcg->css); |
3407 | loop++; | |
3408 | /* | |
3409 | * Could not reclaim anything and there are no more | |
3410 | * mem cgroups to try or we seem to be looping without | |
3411 | * reclaiming anything. | |
3412 | */ | |
3413 | if (!nr_reclaimed && | |
3414 | (next_mz == NULL || | |
3415 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3416 | break; | |
3417 | } while (!nr_reclaimed); | |
3418 | if (next_mz) | |
3419 | css_put(&next_mz->memcg->css); | |
3420 | return nr_reclaimed; | |
3421 | } | |
3422 | ||
ea280e7b TH |
3423 | /* |
3424 | * Test whether @memcg has children, dead or alive. Note that this | |
3425 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3426 | * returns %true if there are child csses according to the cgroup | |
3427 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3428 | */ | |
b5f99b53 GC |
3429 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3430 | { | |
ea280e7b TH |
3431 | bool ret; |
3432 | ||
696ac172 | 3433 | /* |
ea280e7b TH |
3434 | * The lock does not prevent addition or deletion of children, but |
3435 | * it prevents a new child from being initialized based on this | |
3436 | * parent in css_online(), so it's enough to decide whether | |
3437 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3438 | */ |
ea280e7b TH |
3439 | lockdep_assert_held(&memcg_create_mutex); |
3440 | ||
3441 | rcu_read_lock(); | |
3442 | ret = css_next_child(NULL, &memcg->css); | |
3443 | rcu_read_unlock(); | |
3444 | return ret; | |
b5f99b53 GC |
3445 | } |
3446 | ||
c26251f9 MH |
3447 | /* |
3448 | * Reclaims as many pages from the given memcg as possible and moves | |
3449 | * the rest to the parent. | |
3450 | * | |
3451 | * Caller is responsible for holding css reference for memcg. | |
3452 | */ | |
3453 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3454 | { | |
3455 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3456 | |
c1e862c1 KH |
3457 | /* we call try-to-free pages for make this cgroup empty */ |
3458 | lru_add_drain_all(); | |
f817ed48 | 3459 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3460 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3461 | int progress; |
c1e862c1 | 3462 | |
c26251f9 MH |
3463 | if (signal_pending(current)) |
3464 | return -EINTR; | |
3465 | ||
b70a2a21 JW |
3466 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3467 | GFP_KERNEL, true); | |
c1e862c1 | 3468 | if (!progress) { |
f817ed48 | 3469 | nr_retries--; |
c1e862c1 | 3470 | /* maybe some writeback is necessary */ |
8aa7e847 | 3471 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3472 | } |
f817ed48 KH |
3473 | |
3474 | } | |
ab5196c2 MH |
3475 | |
3476 | return 0; | |
cc847582 KH |
3477 | } |
3478 | ||
6770c64e TH |
3479 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3480 | char *buf, size_t nbytes, | |
3481 | loff_t off) | |
c1e862c1 | 3482 | { |
6770c64e | 3483 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3484 | |
d8423011 MH |
3485 | if (mem_cgroup_is_root(memcg)) |
3486 | return -EINVAL; | |
6770c64e | 3487 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3488 | } |
3489 | ||
182446d0 TH |
3490 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3491 | struct cftype *cft) | |
18f59ea7 | 3492 | { |
182446d0 | 3493 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3494 | } |
3495 | ||
182446d0 TH |
3496 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3497 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3498 | { |
3499 | int retval = 0; | |
182446d0 | 3500 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3501 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3502 | |
0999821b | 3503 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3504 | |
3505 | if (memcg->use_hierarchy == val) | |
3506 | goto out; | |
3507 | ||
18f59ea7 | 3508 | /* |
af901ca1 | 3509 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3510 | * in the child subtrees. If it is unset, then the change can |
3511 | * occur, provided the current cgroup has no children. | |
3512 | * | |
3513 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3514 | * set if there are no children. | |
3515 | */ | |
c0ff4b85 | 3516 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3517 | (val == 1 || val == 0)) { |
ea280e7b | 3518 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3519 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3520 | else |
3521 | retval = -EBUSY; | |
3522 | } else | |
3523 | retval = -EINVAL; | |
567fb435 GC |
3524 | |
3525 | out: | |
0999821b | 3526 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3527 | |
3528 | return retval; | |
3529 | } | |
3530 | ||
3e32cb2e JW |
3531 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3532 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3533 | { |
3534 | struct mem_cgroup *iter; | |
3535 | long val = 0; | |
3536 | ||
3537 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3538 | for_each_mem_cgroup_tree(iter, memcg) | |
3539 | val += mem_cgroup_read_stat(iter, idx); | |
3540 | ||
3541 | if (val < 0) /* race ? */ | |
3542 | val = 0; | |
3543 | return val; | |
3544 | } | |
3545 | ||
3546 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3547 | { | |
3548 | u64 val; | |
3549 | ||
3e32cb2e JW |
3550 | if (mem_cgroup_is_root(memcg)) { |
3551 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3552 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3553 | if (swap) | |
3554 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3555 | } else { | |
ce00a967 | 3556 | if (!swap) |
3e32cb2e | 3557 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3558 | else |
3e32cb2e | 3559 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3560 | } |
ce00a967 JW |
3561 | return val << PAGE_SHIFT; |
3562 | } | |
3563 | ||
3e32cb2e JW |
3564 | enum { |
3565 | RES_USAGE, | |
3566 | RES_LIMIT, | |
3567 | RES_MAX_USAGE, | |
3568 | RES_FAILCNT, | |
3569 | RES_SOFT_LIMIT, | |
3570 | }; | |
ce00a967 | 3571 | |
791badbd | 3572 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3573 | struct cftype *cft) |
8cdea7c0 | 3574 | { |
182446d0 | 3575 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3576 | struct page_counter *counter; |
af36f906 | 3577 | |
3e32cb2e | 3578 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3579 | case _MEM: |
3e32cb2e JW |
3580 | counter = &memcg->memory; |
3581 | break; | |
8c7c6e34 | 3582 | case _MEMSWAP: |
3e32cb2e JW |
3583 | counter = &memcg->memsw; |
3584 | break; | |
510fc4e1 | 3585 | case _KMEM: |
3e32cb2e | 3586 | counter = &memcg->kmem; |
510fc4e1 | 3587 | break; |
8c7c6e34 KH |
3588 | default: |
3589 | BUG(); | |
8c7c6e34 | 3590 | } |
3e32cb2e JW |
3591 | |
3592 | switch (MEMFILE_ATTR(cft->private)) { | |
3593 | case RES_USAGE: | |
3594 | if (counter == &memcg->memory) | |
3595 | return mem_cgroup_usage(memcg, false); | |
3596 | if (counter == &memcg->memsw) | |
3597 | return mem_cgroup_usage(memcg, true); | |
3598 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3599 | case RES_LIMIT: | |
3600 | return (u64)counter->limit * PAGE_SIZE; | |
3601 | case RES_MAX_USAGE: | |
3602 | return (u64)counter->watermark * PAGE_SIZE; | |
3603 | case RES_FAILCNT: | |
3604 | return counter->failcnt; | |
3605 | case RES_SOFT_LIMIT: | |
3606 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3607 | default: | |
3608 | BUG(); | |
3609 | } | |
8cdea7c0 | 3610 | } |
510fc4e1 | 3611 | |
510fc4e1 | 3612 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
3613 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
3614 | unsigned long nr_pages) | |
d6441637 VD |
3615 | { |
3616 | int err = 0; | |
3617 | int memcg_id; | |
3618 | ||
3619 | if (memcg_kmem_is_active(memcg)) | |
3620 | return 0; | |
3621 | ||
3622 | /* | |
3623 | * We are going to allocate memory for data shared by all memory | |
3624 | * cgroups so let's stop accounting here. | |
3625 | */ | |
3626 | memcg_stop_kmem_account(); | |
3627 | ||
510fc4e1 GC |
3628 | /* |
3629 | * For simplicity, we won't allow this to be disabled. It also can't | |
3630 | * be changed if the cgroup has children already, or if tasks had | |
3631 | * already joined. | |
3632 | * | |
3633 | * If tasks join before we set the limit, a person looking at | |
3634 | * kmem.usage_in_bytes will have no way to determine when it took | |
3635 | * place, which makes the value quite meaningless. | |
3636 | * | |
3637 | * After it first became limited, changes in the value of the limit are | |
3638 | * of course permitted. | |
510fc4e1 | 3639 | */ |
0999821b | 3640 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
3641 | if (cgroup_has_tasks(memcg->css.cgroup) || |
3642 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
3643 | err = -EBUSY; |
3644 | mutex_unlock(&memcg_create_mutex); | |
3645 | if (err) | |
3646 | goto out; | |
510fc4e1 | 3647 | |
f3bb3043 | 3648 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
3649 | if (memcg_id < 0) { |
3650 | err = memcg_id; | |
3651 | goto out; | |
3652 | } | |
3653 | ||
d6441637 VD |
3654 | memcg->kmemcg_id = memcg_id; |
3655 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
d6441637 VD |
3656 | |
3657 | /* | |
3658 | * We couldn't have accounted to this cgroup, because it hasn't got the | |
3659 | * active bit set yet, so this should succeed. | |
3660 | */ | |
3e32cb2e | 3661 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
3662 | VM_BUG_ON(err); |
3663 | ||
3664 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
3665 | /* | |
3666 | * Setting the active bit after enabling static branching will | |
3667 | * guarantee no one starts accounting before all call sites are | |
3668 | * patched. | |
3669 | */ | |
3670 | memcg_kmem_set_active(memcg); | |
510fc4e1 | 3671 | out: |
d6441637 VD |
3672 | memcg_resume_kmem_account(); |
3673 | return err; | |
d6441637 VD |
3674 | } |
3675 | ||
d6441637 | 3676 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3677 | unsigned long limit) |
d6441637 VD |
3678 | { |
3679 | int ret; | |
3680 | ||
3e32cb2e | 3681 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 3682 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 3683 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 3684 | else |
3e32cb2e JW |
3685 | ret = page_counter_limit(&memcg->kmem, limit); |
3686 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
3687 | return ret; |
3688 | } | |
3689 | ||
55007d84 | 3690 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 3691 | { |
55007d84 | 3692 | int ret = 0; |
510fc4e1 | 3693 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 3694 | |
d6441637 VD |
3695 | if (!parent) |
3696 | return 0; | |
55007d84 | 3697 | |
8c0145b6 | 3698 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 3699 | /* |
d6441637 VD |
3700 | * If the parent cgroup is not kmem-active now, it cannot be activated |
3701 | * after this point, because it has at least one child already. | |
55007d84 | 3702 | */ |
d6441637 | 3703 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3704 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3705 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3706 | return ret; |
510fc4e1 | 3707 | } |
d6441637 VD |
3708 | #else |
3709 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3710 | unsigned long limit) |
d6441637 VD |
3711 | { |
3712 | return -EINVAL; | |
3713 | } | |
6d043990 | 3714 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3715 | |
628f4235 KH |
3716 | /* |
3717 | * The user of this function is... | |
3718 | * RES_LIMIT. | |
3719 | */ | |
451af504 TH |
3720 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3721 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3722 | { |
451af504 | 3723 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3724 | unsigned long nr_pages; |
628f4235 KH |
3725 | int ret; |
3726 | ||
451af504 | 3727 | buf = strstrip(buf); |
3e32cb2e JW |
3728 | ret = page_counter_memparse(buf, &nr_pages); |
3729 | if (ret) | |
3730 | return ret; | |
af36f906 | 3731 | |
3e32cb2e | 3732 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3733 | case RES_LIMIT: |
4b3bde4c BS |
3734 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3735 | ret = -EINVAL; | |
3736 | break; | |
3737 | } | |
3e32cb2e JW |
3738 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3739 | case _MEM: | |
3740 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3741 | break; |
3e32cb2e JW |
3742 | case _MEMSWAP: |
3743 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3744 | break; |
3e32cb2e JW |
3745 | case _KMEM: |
3746 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3747 | break; | |
3748 | } | |
296c81d8 | 3749 | break; |
3e32cb2e JW |
3750 | case RES_SOFT_LIMIT: |
3751 | memcg->soft_limit = nr_pages; | |
3752 | ret = 0; | |
628f4235 KH |
3753 | break; |
3754 | } | |
451af504 | 3755 | return ret ?: nbytes; |
8cdea7c0 BS |
3756 | } |
3757 | ||
6770c64e TH |
3758 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3759 | size_t nbytes, loff_t off) | |
c84872e1 | 3760 | { |
6770c64e | 3761 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3762 | struct page_counter *counter; |
c84872e1 | 3763 | |
3e32cb2e JW |
3764 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3765 | case _MEM: | |
3766 | counter = &memcg->memory; | |
3767 | break; | |
3768 | case _MEMSWAP: | |
3769 | counter = &memcg->memsw; | |
3770 | break; | |
3771 | case _KMEM: | |
3772 | counter = &memcg->kmem; | |
3773 | break; | |
3774 | default: | |
3775 | BUG(); | |
3776 | } | |
af36f906 | 3777 | |
3e32cb2e | 3778 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3779 | case RES_MAX_USAGE: |
3e32cb2e | 3780 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3781 | break; |
3782 | case RES_FAILCNT: | |
3e32cb2e | 3783 | counter->failcnt = 0; |
29f2a4da | 3784 | break; |
3e32cb2e JW |
3785 | default: |
3786 | BUG(); | |
29f2a4da | 3787 | } |
f64c3f54 | 3788 | |
6770c64e | 3789 | return nbytes; |
c84872e1 PE |
3790 | } |
3791 | ||
182446d0 | 3792 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3793 | struct cftype *cft) |
3794 | { | |
182446d0 | 3795 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3796 | } |
3797 | ||
02491447 | 3798 | #ifdef CONFIG_MMU |
182446d0 | 3799 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3800 | struct cftype *cft, u64 val) |
3801 | { | |
182446d0 | 3802 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
3803 | |
3804 | if (val >= (1 << NR_MOVE_TYPE)) | |
3805 | return -EINVAL; | |
ee5e8472 | 3806 | |
7dc74be0 | 3807 | /* |
ee5e8472 GC |
3808 | * No kind of locking is needed in here, because ->can_attach() will |
3809 | * check this value once in the beginning of the process, and then carry | |
3810 | * on with stale data. This means that changes to this value will only | |
3811 | * affect task migrations starting after the change. | |
7dc74be0 | 3812 | */ |
c0ff4b85 | 3813 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3814 | return 0; |
3815 | } | |
02491447 | 3816 | #else |
182446d0 | 3817 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3818 | struct cftype *cft, u64 val) |
3819 | { | |
3820 | return -ENOSYS; | |
3821 | } | |
3822 | #endif | |
7dc74be0 | 3823 | |
406eb0c9 | 3824 | #ifdef CONFIG_NUMA |
2da8ca82 | 3825 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3826 | { |
25485de6 GT |
3827 | struct numa_stat { |
3828 | const char *name; | |
3829 | unsigned int lru_mask; | |
3830 | }; | |
3831 | ||
3832 | static const struct numa_stat stats[] = { | |
3833 | { "total", LRU_ALL }, | |
3834 | { "file", LRU_ALL_FILE }, | |
3835 | { "anon", LRU_ALL_ANON }, | |
3836 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3837 | }; | |
3838 | const struct numa_stat *stat; | |
406eb0c9 | 3839 | int nid; |
25485de6 | 3840 | unsigned long nr; |
2da8ca82 | 3841 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3842 | |
25485de6 GT |
3843 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3844 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3845 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3846 | for_each_node_state(nid, N_MEMORY) { | |
3847 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3848 | stat->lru_mask); | |
3849 | seq_printf(m, " N%d=%lu", nid, nr); | |
3850 | } | |
3851 | seq_putc(m, '\n'); | |
406eb0c9 | 3852 | } |
406eb0c9 | 3853 | |
071aee13 YH |
3854 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3855 | struct mem_cgroup *iter; | |
3856 | ||
3857 | nr = 0; | |
3858 | for_each_mem_cgroup_tree(iter, memcg) | |
3859 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3860 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3861 | for_each_node_state(nid, N_MEMORY) { | |
3862 | nr = 0; | |
3863 | for_each_mem_cgroup_tree(iter, memcg) | |
3864 | nr += mem_cgroup_node_nr_lru_pages( | |
3865 | iter, nid, stat->lru_mask); | |
3866 | seq_printf(m, " N%d=%lu", nid, nr); | |
3867 | } | |
3868 | seq_putc(m, '\n'); | |
406eb0c9 | 3869 | } |
406eb0c9 | 3870 | |
406eb0c9 YH |
3871 | return 0; |
3872 | } | |
3873 | #endif /* CONFIG_NUMA */ | |
3874 | ||
af7c4b0e JW |
3875 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
3876 | { | |
3877 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
3878 | } | |
3879 | ||
2da8ca82 | 3880 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3881 | { |
2da8ca82 | 3882 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3883 | unsigned long memory, memsw; |
af7c4b0e JW |
3884 | struct mem_cgroup *mi; |
3885 | unsigned int i; | |
406eb0c9 | 3886 | |
af7c4b0e | 3887 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3888 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3889 | continue; |
af7c4b0e JW |
3890 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
3891 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 3892 | } |
7b854121 | 3893 | |
af7c4b0e JW |
3894 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3895 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3896 | mem_cgroup_read_events(memcg, i)); | |
3897 | ||
3898 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3899 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3900 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3901 | ||
14067bb3 | 3902 | /* Hierarchical information */ |
3e32cb2e JW |
3903 | memory = memsw = PAGE_COUNTER_MAX; |
3904 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3905 | memory = min(memory, mi->memory.limit); | |
3906 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3907 | } |
3e32cb2e JW |
3908 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3909 | (u64)memory * PAGE_SIZE); | |
3910 | if (do_swap_account) | |
3911 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3912 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3913 | |
af7c4b0e JW |
3914 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
3915 | long long val = 0; | |
3916 | ||
bff6bb83 | 3917 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3918 | continue; |
af7c4b0e JW |
3919 | for_each_mem_cgroup_tree(mi, memcg) |
3920 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
3921 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
3922 | } | |
3923 | ||
3924 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3925 | unsigned long long val = 0; | |
3926 | ||
3927 | for_each_mem_cgroup_tree(mi, memcg) | |
3928 | val += mem_cgroup_read_events(mi, i); | |
3929 | seq_printf(m, "total_%s %llu\n", | |
3930 | mem_cgroup_events_names[i], val); | |
3931 | } | |
3932 | ||
3933 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3934 | unsigned long long val = 0; | |
3935 | ||
3936 | for_each_mem_cgroup_tree(mi, memcg) | |
3937 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3938 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3939 | } |
14067bb3 | 3940 | |
7f016ee8 | 3941 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3942 | { |
3943 | int nid, zid; | |
3944 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3945 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3946 | unsigned long recent_rotated[2] = {0, 0}; |
3947 | unsigned long recent_scanned[2] = {0, 0}; | |
3948 | ||
3949 | for_each_online_node(nid) | |
3950 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3951 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3952 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3953 | |
89abfab1 HD |
3954 | recent_rotated[0] += rstat->recent_rotated[0]; |
3955 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3956 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3957 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3958 | } |
78ccf5b5 JW |
3959 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3960 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3961 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3962 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3963 | } |
3964 | #endif | |
3965 | ||
d2ceb9b7 KH |
3966 | return 0; |
3967 | } | |
3968 | ||
182446d0 TH |
3969 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3970 | struct cftype *cft) | |
a7885eb8 | 3971 | { |
182446d0 | 3972 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3973 | |
1f4c025b | 3974 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3975 | } |
3976 | ||
182446d0 TH |
3977 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3978 | struct cftype *cft, u64 val) | |
a7885eb8 | 3979 | { |
182446d0 | 3980 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3981 | |
3dae7fec | 3982 | if (val > 100) |
a7885eb8 KM |
3983 | return -EINVAL; |
3984 | ||
14208b0e | 3985 | if (css->parent) |
3dae7fec JW |
3986 | memcg->swappiness = val; |
3987 | else | |
3988 | vm_swappiness = val; | |
068b38c1 | 3989 | |
a7885eb8 KM |
3990 | return 0; |
3991 | } | |
3992 | ||
2e72b634 KS |
3993 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3994 | { | |
3995 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3996 | unsigned long usage; |
2e72b634 KS |
3997 | int i; |
3998 | ||
3999 | rcu_read_lock(); | |
4000 | if (!swap) | |
2c488db2 | 4001 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4002 | else |
2c488db2 | 4003 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4004 | |
4005 | if (!t) | |
4006 | goto unlock; | |
4007 | ||
ce00a967 | 4008 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4009 | |
4010 | /* | |
748dad36 | 4011 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4012 | * If it's not true, a threshold was crossed after last |
4013 | * call of __mem_cgroup_threshold(). | |
4014 | */ | |
5407a562 | 4015 | i = t->current_threshold; |
2e72b634 KS |
4016 | |
4017 | /* | |
4018 | * Iterate backward over array of thresholds starting from | |
4019 | * current_threshold and check if a threshold is crossed. | |
4020 | * If none of thresholds below usage is crossed, we read | |
4021 | * only one element of the array here. | |
4022 | */ | |
4023 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4024 | eventfd_signal(t->entries[i].eventfd, 1); | |
4025 | ||
4026 | /* i = current_threshold + 1 */ | |
4027 | i++; | |
4028 | ||
4029 | /* | |
4030 | * Iterate forward over array of thresholds starting from | |
4031 | * current_threshold+1 and check if a threshold is crossed. | |
4032 | * If none of thresholds above usage is crossed, we read | |
4033 | * only one element of the array here. | |
4034 | */ | |
4035 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4036 | eventfd_signal(t->entries[i].eventfd, 1); | |
4037 | ||
4038 | /* Update current_threshold */ | |
5407a562 | 4039 | t->current_threshold = i - 1; |
2e72b634 KS |
4040 | unlock: |
4041 | rcu_read_unlock(); | |
4042 | } | |
4043 | ||
4044 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4045 | { | |
ad4ca5f4 KS |
4046 | while (memcg) { |
4047 | __mem_cgroup_threshold(memcg, false); | |
4048 | if (do_swap_account) | |
4049 | __mem_cgroup_threshold(memcg, true); | |
4050 | ||
4051 | memcg = parent_mem_cgroup(memcg); | |
4052 | } | |
2e72b634 KS |
4053 | } |
4054 | ||
4055 | static int compare_thresholds(const void *a, const void *b) | |
4056 | { | |
4057 | const struct mem_cgroup_threshold *_a = a; | |
4058 | const struct mem_cgroup_threshold *_b = b; | |
4059 | ||
2bff24a3 GT |
4060 | if (_a->threshold > _b->threshold) |
4061 | return 1; | |
4062 | ||
4063 | if (_a->threshold < _b->threshold) | |
4064 | return -1; | |
4065 | ||
4066 | return 0; | |
2e72b634 KS |
4067 | } |
4068 | ||
c0ff4b85 | 4069 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4070 | { |
4071 | struct mem_cgroup_eventfd_list *ev; | |
4072 | ||
2bcf2e92 MH |
4073 | spin_lock(&memcg_oom_lock); |
4074 | ||
c0ff4b85 | 4075 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4076 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4077 | |
4078 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4079 | return 0; |
4080 | } | |
4081 | ||
c0ff4b85 | 4082 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4083 | { |
7d74b06f KH |
4084 | struct mem_cgroup *iter; |
4085 | ||
c0ff4b85 | 4086 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4087 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4088 | } |
4089 | ||
59b6f873 | 4090 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4091 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4092 | { |
2c488db2 KS |
4093 | struct mem_cgroup_thresholds *thresholds; |
4094 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4095 | unsigned long threshold; |
4096 | unsigned long usage; | |
2c488db2 | 4097 | int i, size, ret; |
2e72b634 | 4098 | |
3e32cb2e | 4099 | ret = page_counter_memparse(args, &threshold); |
2e72b634 KS |
4100 | if (ret) |
4101 | return ret; | |
4102 | ||
4103 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4104 | |
05b84301 | 4105 | if (type == _MEM) { |
2c488db2 | 4106 | thresholds = &memcg->thresholds; |
ce00a967 | 4107 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4108 | } else if (type == _MEMSWAP) { |
2c488db2 | 4109 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4110 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4111 | } else |
2e72b634 KS |
4112 | BUG(); |
4113 | ||
2e72b634 | 4114 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4115 | if (thresholds->primary) |
2e72b634 KS |
4116 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4117 | ||
2c488db2 | 4118 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4119 | |
4120 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4121 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4122 | GFP_KERNEL); |
2c488db2 | 4123 | if (!new) { |
2e72b634 KS |
4124 | ret = -ENOMEM; |
4125 | goto unlock; | |
4126 | } | |
2c488db2 | 4127 | new->size = size; |
2e72b634 KS |
4128 | |
4129 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4130 | if (thresholds->primary) { |
4131 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4132 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4133 | } |
4134 | ||
2e72b634 | 4135 | /* Add new threshold */ |
2c488db2 KS |
4136 | new->entries[size - 1].eventfd = eventfd; |
4137 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4138 | |
4139 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4140 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4141 | compare_thresholds, NULL); |
4142 | ||
4143 | /* Find current threshold */ | |
2c488db2 | 4144 | new->current_threshold = -1; |
2e72b634 | 4145 | for (i = 0; i < size; i++) { |
748dad36 | 4146 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4147 | /* |
2c488db2 KS |
4148 | * new->current_threshold will not be used until |
4149 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4150 | * it here. |
4151 | */ | |
2c488db2 | 4152 | ++new->current_threshold; |
748dad36 SZ |
4153 | } else |
4154 | break; | |
2e72b634 KS |
4155 | } |
4156 | ||
2c488db2 KS |
4157 | /* Free old spare buffer and save old primary buffer as spare */ |
4158 | kfree(thresholds->spare); | |
4159 | thresholds->spare = thresholds->primary; | |
4160 | ||
4161 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4162 | |
907860ed | 4163 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4164 | synchronize_rcu(); |
4165 | ||
2e72b634 KS |
4166 | unlock: |
4167 | mutex_unlock(&memcg->thresholds_lock); | |
4168 | ||
4169 | return ret; | |
4170 | } | |
4171 | ||
59b6f873 | 4172 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4173 | struct eventfd_ctx *eventfd, const char *args) |
4174 | { | |
59b6f873 | 4175 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4176 | } |
4177 | ||
59b6f873 | 4178 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4179 | struct eventfd_ctx *eventfd, const char *args) |
4180 | { | |
59b6f873 | 4181 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4182 | } |
4183 | ||
59b6f873 | 4184 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4185 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4186 | { |
2c488db2 KS |
4187 | struct mem_cgroup_thresholds *thresholds; |
4188 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4189 | unsigned long usage; |
2c488db2 | 4190 | int i, j, size; |
2e72b634 KS |
4191 | |
4192 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4193 | |
4194 | if (type == _MEM) { | |
2c488db2 | 4195 | thresholds = &memcg->thresholds; |
ce00a967 | 4196 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4197 | } else if (type == _MEMSWAP) { |
2c488db2 | 4198 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4199 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4200 | } else |
2e72b634 KS |
4201 | BUG(); |
4202 | ||
371528ca AV |
4203 | if (!thresholds->primary) |
4204 | goto unlock; | |
4205 | ||
2e72b634 KS |
4206 | /* Check if a threshold crossed before removing */ |
4207 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4208 | ||
4209 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4210 | size = 0; |
4211 | for (i = 0; i < thresholds->primary->size; i++) { | |
4212 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4213 | size++; |
4214 | } | |
4215 | ||
2c488db2 | 4216 | new = thresholds->spare; |
907860ed | 4217 | |
2e72b634 KS |
4218 | /* Set thresholds array to NULL if we don't have thresholds */ |
4219 | if (!size) { | |
2c488db2 KS |
4220 | kfree(new); |
4221 | new = NULL; | |
907860ed | 4222 | goto swap_buffers; |
2e72b634 KS |
4223 | } |
4224 | ||
2c488db2 | 4225 | new->size = size; |
2e72b634 KS |
4226 | |
4227 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4228 | new->current_threshold = -1; |
4229 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4230 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4231 | continue; |
4232 | ||
2c488db2 | 4233 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4234 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4235 | /* |
2c488db2 | 4236 | * new->current_threshold will not be used |
2e72b634 KS |
4237 | * until rcu_assign_pointer(), so it's safe to increment |
4238 | * it here. | |
4239 | */ | |
2c488db2 | 4240 | ++new->current_threshold; |
2e72b634 KS |
4241 | } |
4242 | j++; | |
4243 | } | |
4244 | ||
907860ed | 4245 | swap_buffers: |
2c488db2 KS |
4246 | /* Swap primary and spare array */ |
4247 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4248 | /* If all events are unregistered, free the spare array */ |
4249 | if (!new) { | |
4250 | kfree(thresholds->spare); | |
4251 | thresholds->spare = NULL; | |
4252 | } | |
4253 | ||
2c488db2 | 4254 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4255 | |
907860ed | 4256 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4257 | synchronize_rcu(); |
371528ca | 4258 | unlock: |
2e72b634 | 4259 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4260 | } |
c1e862c1 | 4261 | |
59b6f873 | 4262 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4263 | struct eventfd_ctx *eventfd) |
4264 | { | |
59b6f873 | 4265 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4266 | } |
4267 | ||
59b6f873 | 4268 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4269 | struct eventfd_ctx *eventfd) |
4270 | { | |
59b6f873 | 4271 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4272 | } |
4273 | ||
59b6f873 | 4274 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4275 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4276 | { |
9490ff27 | 4277 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4278 | |
9490ff27 KH |
4279 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4280 | if (!event) | |
4281 | return -ENOMEM; | |
4282 | ||
1af8efe9 | 4283 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4284 | |
4285 | event->eventfd = eventfd; | |
4286 | list_add(&event->list, &memcg->oom_notify); | |
4287 | ||
4288 | /* already in OOM ? */ | |
79dfdacc | 4289 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4290 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4291 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4292 | |
4293 | return 0; | |
4294 | } | |
4295 | ||
59b6f873 | 4296 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4297 | struct eventfd_ctx *eventfd) |
9490ff27 | 4298 | { |
9490ff27 | 4299 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4300 | |
1af8efe9 | 4301 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4302 | |
c0ff4b85 | 4303 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4304 | if (ev->eventfd == eventfd) { |
4305 | list_del(&ev->list); | |
4306 | kfree(ev); | |
4307 | } | |
4308 | } | |
4309 | ||
1af8efe9 | 4310 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4311 | } |
4312 | ||
2da8ca82 | 4313 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4314 | { |
2da8ca82 | 4315 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 4316 | |
791badbd TH |
4317 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
4318 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
4319 | return 0; |
4320 | } | |
4321 | ||
182446d0 | 4322 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4323 | struct cftype *cft, u64 val) |
4324 | { | |
182446d0 | 4325 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4326 | |
4327 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4328 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4329 | return -EINVAL; |
4330 | ||
c0ff4b85 | 4331 | memcg->oom_kill_disable = val; |
4d845ebf | 4332 | if (!val) |
c0ff4b85 | 4333 | memcg_oom_recover(memcg); |
3dae7fec | 4334 | |
3c11ecf4 KH |
4335 | return 0; |
4336 | } | |
4337 | ||
c255a458 | 4338 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4339 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4340 | { |
55007d84 GC |
4341 | int ret; |
4342 | ||
2633d7a0 | 4343 | memcg->kmemcg_id = -1; |
55007d84 GC |
4344 | ret = memcg_propagate_kmem(memcg); |
4345 | if (ret) | |
4346 | return ret; | |
2633d7a0 | 4347 | |
1d62e436 | 4348 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4349 | } |
e5671dfa | 4350 | |
10d5ebf4 | 4351 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4352 | { |
1d62e436 | 4353 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 4354 | } |
e5671dfa | 4355 | #else |
cbe128e3 | 4356 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4357 | { |
4358 | return 0; | |
4359 | } | |
d1a4c0b3 | 4360 | |
10d5ebf4 LZ |
4361 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4362 | { | |
4363 | } | |
e5671dfa GC |
4364 | #endif |
4365 | ||
3bc942f3 TH |
4366 | /* |
4367 | * DO NOT USE IN NEW FILES. | |
4368 | * | |
4369 | * "cgroup.event_control" implementation. | |
4370 | * | |
4371 | * This is way over-engineered. It tries to support fully configurable | |
4372 | * events for each user. Such level of flexibility is completely | |
4373 | * unnecessary especially in the light of the planned unified hierarchy. | |
4374 | * | |
4375 | * Please deprecate this and replace with something simpler if at all | |
4376 | * possible. | |
4377 | */ | |
4378 | ||
79bd9814 TH |
4379 | /* |
4380 | * Unregister event and free resources. | |
4381 | * | |
4382 | * Gets called from workqueue. | |
4383 | */ | |
3bc942f3 | 4384 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4385 | { |
3bc942f3 TH |
4386 | struct mem_cgroup_event *event = |
4387 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4388 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4389 | |
4390 | remove_wait_queue(event->wqh, &event->wait); | |
4391 | ||
59b6f873 | 4392 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4393 | |
4394 | /* Notify userspace the event is going away. */ | |
4395 | eventfd_signal(event->eventfd, 1); | |
4396 | ||
4397 | eventfd_ctx_put(event->eventfd); | |
4398 | kfree(event); | |
59b6f873 | 4399 | css_put(&memcg->css); |
79bd9814 TH |
4400 | } |
4401 | ||
4402 | /* | |
4403 | * Gets called on POLLHUP on eventfd when user closes it. | |
4404 | * | |
4405 | * Called with wqh->lock held and interrupts disabled. | |
4406 | */ | |
3bc942f3 TH |
4407 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4408 | int sync, void *key) | |
79bd9814 | 4409 | { |
3bc942f3 TH |
4410 | struct mem_cgroup_event *event = |
4411 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4412 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4413 | unsigned long flags = (unsigned long)key; |
4414 | ||
4415 | if (flags & POLLHUP) { | |
4416 | /* | |
4417 | * If the event has been detached at cgroup removal, we | |
4418 | * can simply return knowing the other side will cleanup | |
4419 | * for us. | |
4420 | * | |
4421 | * We can't race against event freeing since the other | |
4422 | * side will require wqh->lock via remove_wait_queue(), | |
4423 | * which we hold. | |
4424 | */ | |
fba94807 | 4425 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4426 | if (!list_empty(&event->list)) { |
4427 | list_del_init(&event->list); | |
4428 | /* | |
4429 | * We are in atomic context, but cgroup_event_remove() | |
4430 | * may sleep, so we have to call it in workqueue. | |
4431 | */ | |
4432 | schedule_work(&event->remove); | |
4433 | } | |
fba94807 | 4434 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4435 | } |
4436 | ||
4437 | return 0; | |
4438 | } | |
4439 | ||
3bc942f3 | 4440 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4441 | wait_queue_head_t *wqh, poll_table *pt) |
4442 | { | |
3bc942f3 TH |
4443 | struct mem_cgroup_event *event = |
4444 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4445 | |
4446 | event->wqh = wqh; | |
4447 | add_wait_queue(wqh, &event->wait); | |
4448 | } | |
4449 | ||
4450 | /* | |
3bc942f3 TH |
4451 | * DO NOT USE IN NEW FILES. |
4452 | * | |
79bd9814 TH |
4453 | * Parse input and register new cgroup event handler. |
4454 | * | |
4455 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4456 | * Interpretation of args is defined by control file implementation. | |
4457 | */ | |
451af504 TH |
4458 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4459 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4460 | { |
451af504 | 4461 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4462 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4463 | struct mem_cgroup_event *event; |
79bd9814 TH |
4464 | struct cgroup_subsys_state *cfile_css; |
4465 | unsigned int efd, cfd; | |
4466 | struct fd efile; | |
4467 | struct fd cfile; | |
fba94807 | 4468 | const char *name; |
79bd9814 TH |
4469 | char *endp; |
4470 | int ret; | |
4471 | ||
451af504 TH |
4472 | buf = strstrip(buf); |
4473 | ||
4474 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4475 | if (*endp != ' ') |
4476 | return -EINVAL; | |
451af504 | 4477 | buf = endp + 1; |
79bd9814 | 4478 | |
451af504 | 4479 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4480 | if ((*endp != ' ') && (*endp != '\0')) |
4481 | return -EINVAL; | |
451af504 | 4482 | buf = endp + 1; |
79bd9814 TH |
4483 | |
4484 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4485 | if (!event) | |
4486 | return -ENOMEM; | |
4487 | ||
59b6f873 | 4488 | event->memcg = memcg; |
79bd9814 | 4489 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4490 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4491 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4492 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4493 | |
4494 | efile = fdget(efd); | |
4495 | if (!efile.file) { | |
4496 | ret = -EBADF; | |
4497 | goto out_kfree; | |
4498 | } | |
4499 | ||
4500 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4501 | if (IS_ERR(event->eventfd)) { | |
4502 | ret = PTR_ERR(event->eventfd); | |
4503 | goto out_put_efile; | |
4504 | } | |
4505 | ||
4506 | cfile = fdget(cfd); | |
4507 | if (!cfile.file) { | |
4508 | ret = -EBADF; | |
4509 | goto out_put_eventfd; | |
4510 | } | |
4511 | ||
4512 | /* the process need read permission on control file */ | |
4513 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4514 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4515 | if (ret < 0) | |
4516 | goto out_put_cfile; | |
4517 | ||
fba94807 TH |
4518 | /* |
4519 | * Determine the event callbacks and set them in @event. This used | |
4520 | * to be done via struct cftype but cgroup core no longer knows | |
4521 | * about these events. The following is crude but the whole thing | |
4522 | * is for compatibility anyway. | |
3bc942f3 TH |
4523 | * |
4524 | * DO NOT ADD NEW FILES. | |
fba94807 TH |
4525 | */ |
4526 | name = cfile.file->f_dentry->d_name.name; | |
4527 | ||
4528 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4529 | event->register_event = mem_cgroup_usage_register_event; | |
4530 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4531 | } else if (!strcmp(name, "memory.oom_control")) { | |
4532 | event->register_event = mem_cgroup_oom_register_event; | |
4533 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4534 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4535 | event->register_event = vmpressure_register_event; | |
4536 | event->unregister_event = vmpressure_unregister_event; | |
4537 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4538 | event->register_event = memsw_cgroup_usage_register_event; |
4539 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4540 | } else { |
4541 | ret = -EINVAL; | |
4542 | goto out_put_cfile; | |
4543 | } | |
4544 | ||
79bd9814 | 4545 | /* |
b5557c4c TH |
4546 | * Verify @cfile should belong to @css. Also, remaining events are |
4547 | * automatically removed on cgroup destruction but the removal is | |
4548 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4549 | */ |
ec903c0c TH |
4550 | cfile_css = css_tryget_online_from_dir(cfile.file->f_dentry->d_parent, |
4551 | &memory_cgrp_subsys); | |
79bd9814 | 4552 | ret = -EINVAL; |
5a17f543 | 4553 | if (IS_ERR(cfile_css)) |
79bd9814 | 4554 | goto out_put_cfile; |
5a17f543 TH |
4555 | if (cfile_css != css) { |
4556 | css_put(cfile_css); | |
79bd9814 | 4557 | goto out_put_cfile; |
5a17f543 | 4558 | } |
79bd9814 | 4559 | |
451af504 | 4560 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4561 | if (ret) |
4562 | goto out_put_css; | |
4563 | ||
4564 | efile.file->f_op->poll(efile.file, &event->pt); | |
4565 | ||
fba94807 TH |
4566 | spin_lock(&memcg->event_list_lock); |
4567 | list_add(&event->list, &memcg->event_list); | |
4568 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4569 | |
4570 | fdput(cfile); | |
4571 | fdput(efile); | |
4572 | ||
451af504 | 4573 | return nbytes; |
79bd9814 TH |
4574 | |
4575 | out_put_css: | |
b5557c4c | 4576 | css_put(css); |
79bd9814 TH |
4577 | out_put_cfile: |
4578 | fdput(cfile); | |
4579 | out_put_eventfd: | |
4580 | eventfd_ctx_put(event->eventfd); | |
4581 | out_put_efile: | |
4582 | fdput(efile); | |
4583 | out_kfree: | |
4584 | kfree(event); | |
4585 | ||
4586 | return ret; | |
4587 | } | |
4588 | ||
8cdea7c0 BS |
4589 | static struct cftype mem_cgroup_files[] = { |
4590 | { | |
0eea1030 | 4591 | .name = "usage_in_bytes", |
8c7c6e34 | 4592 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4593 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4594 | }, |
c84872e1 PE |
4595 | { |
4596 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4597 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4598 | .write = mem_cgroup_reset, |
791badbd | 4599 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4600 | }, |
8cdea7c0 | 4601 | { |
0eea1030 | 4602 | .name = "limit_in_bytes", |
8c7c6e34 | 4603 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4604 | .write = mem_cgroup_write, |
791badbd | 4605 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4606 | }, |
296c81d8 BS |
4607 | { |
4608 | .name = "soft_limit_in_bytes", | |
4609 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4610 | .write = mem_cgroup_write, |
791badbd | 4611 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4612 | }, |
8cdea7c0 BS |
4613 | { |
4614 | .name = "failcnt", | |
8c7c6e34 | 4615 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4616 | .write = mem_cgroup_reset, |
791badbd | 4617 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4618 | }, |
d2ceb9b7 KH |
4619 | { |
4620 | .name = "stat", | |
2da8ca82 | 4621 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4622 | }, |
c1e862c1 KH |
4623 | { |
4624 | .name = "force_empty", | |
6770c64e | 4625 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4626 | }, |
18f59ea7 BS |
4627 | { |
4628 | .name = "use_hierarchy", | |
4629 | .write_u64 = mem_cgroup_hierarchy_write, | |
4630 | .read_u64 = mem_cgroup_hierarchy_read, | |
4631 | }, | |
79bd9814 | 4632 | { |
3bc942f3 | 4633 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4634 | .write = memcg_write_event_control, |
79bd9814 TH |
4635 | .flags = CFTYPE_NO_PREFIX, |
4636 | .mode = S_IWUGO, | |
4637 | }, | |
a7885eb8 KM |
4638 | { |
4639 | .name = "swappiness", | |
4640 | .read_u64 = mem_cgroup_swappiness_read, | |
4641 | .write_u64 = mem_cgroup_swappiness_write, | |
4642 | }, | |
7dc74be0 DN |
4643 | { |
4644 | .name = "move_charge_at_immigrate", | |
4645 | .read_u64 = mem_cgroup_move_charge_read, | |
4646 | .write_u64 = mem_cgroup_move_charge_write, | |
4647 | }, | |
9490ff27 KH |
4648 | { |
4649 | .name = "oom_control", | |
2da8ca82 | 4650 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4651 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4652 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4653 | }, | |
70ddf637 AV |
4654 | { |
4655 | .name = "pressure_level", | |
70ddf637 | 4656 | }, |
406eb0c9 YH |
4657 | #ifdef CONFIG_NUMA |
4658 | { | |
4659 | .name = "numa_stat", | |
2da8ca82 | 4660 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4661 | }, |
4662 | #endif | |
510fc4e1 GC |
4663 | #ifdef CONFIG_MEMCG_KMEM |
4664 | { | |
4665 | .name = "kmem.limit_in_bytes", | |
4666 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4667 | .write = mem_cgroup_write, |
791badbd | 4668 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4669 | }, |
4670 | { | |
4671 | .name = "kmem.usage_in_bytes", | |
4672 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4673 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4674 | }, |
4675 | { | |
4676 | .name = "kmem.failcnt", | |
4677 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4678 | .write = mem_cgroup_reset, |
791badbd | 4679 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4680 | }, |
4681 | { | |
4682 | .name = "kmem.max_usage_in_bytes", | |
4683 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4684 | .write = mem_cgroup_reset, |
791badbd | 4685 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4686 | }, |
749c5415 GC |
4687 | #ifdef CONFIG_SLABINFO |
4688 | { | |
4689 | .name = "kmem.slabinfo", | |
b047501c VD |
4690 | .seq_start = slab_start, |
4691 | .seq_next = slab_next, | |
4692 | .seq_stop = slab_stop, | |
4693 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4694 | }, |
4695 | #endif | |
8c7c6e34 | 4696 | #endif |
6bc10349 | 4697 | { }, /* terminate */ |
af36f906 | 4698 | }; |
8c7c6e34 | 4699 | |
2d11085e MH |
4700 | #ifdef CONFIG_MEMCG_SWAP |
4701 | static struct cftype memsw_cgroup_files[] = { | |
4702 | { | |
4703 | .name = "memsw.usage_in_bytes", | |
4704 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 4705 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4706 | }, |
4707 | { | |
4708 | .name = "memsw.max_usage_in_bytes", | |
4709 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 4710 | .write = mem_cgroup_reset, |
791badbd | 4711 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4712 | }, |
4713 | { | |
4714 | .name = "memsw.limit_in_bytes", | |
4715 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 4716 | .write = mem_cgroup_write, |
791badbd | 4717 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4718 | }, |
4719 | { | |
4720 | .name = "memsw.failcnt", | |
4721 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 4722 | .write = mem_cgroup_reset, |
791badbd | 4723 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4724 | }, |
4725 | { }, /* terminate */ | |
4726 | }; | |
4727 | #endif | |
c0ff4b85 | 4728 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4729 | { |
4730 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4731 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4732 | int zone, tmp = node; |
1ecaab2b KH |
4733 | /* |
4734 | * This routine is called against possible nodes. | |
4735 | * But it's BUG to call kmalloc() against offline node. | |
4736 | * | |
4737 | * TODO: this routine can waste much memory for nodes which will | |
4738 | * never be onlined. It's better to use memory hotplug callback | |
4739 | * function. | |
4740 | */ | |
41e3355d KH |
4741 | if (!node_state(node, N_NORMAL_MEMORY)) |
4742 | tmp = -1; | |
17295c88 | 4743 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4744 | if (!pn) |
4745 | return 1; | |
1ecaab2b | 4746 | |
1ecaab2b KH |
4747 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4748 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4749 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4750 | mz->usage_in_excess = 0; |
4751 | mz->on_tree = false; | |
d79154bb | 4752 | mz->memcg = memcg; |
1ecaab2b | 4753 | } |
54f72fe0 | 4754 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4755 | return 0; |
4756 | } | |
4757 | ||
c0ff4b85 | 4758 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4759 | { |
54f72fe0 | 4760 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4761 | } |
4762 | ||
33327948 KH |
4763 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4764 | { | |
d79154bb | 4765 | struct mem_cgroup *memcg; |
8ff69e2c | 4766 | size_t size; |
33327948 | 4767 | |
8ff69e2c VD |
4768 | size = sizeof(struct mem_cgroup); |
4769 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4770 | |
8ff69e2c | 4771 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4772 | if (!memcg) |
e7bbcdf3 DC |
4773 | return NULL; |
4774 | ||
d79154bb HD |
4775 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4776 | if (!memcg->stat) | |
d2e61b8d | 4777 | goto out_free; |
d79154bb HD |
4778 | spin_lock_init(&memcg->pcp_counter_lock); |
4779 | return memcg; | |
d2e61b8d DC |
4780 | |
4781 | out_free: | |
8ff69e2c | 4782 | kfree(memcg); |
d2e61b8d | 4783 | return NULL; |
33327948 KH |
4784 | } |
4785 | ||
59927fb9 | 4786 | /* |
c8b2a36f GC |
4787 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4788 | * (scanning all at force_empty is too costly...) | |
4789 | * | |
4790 | * Instead of clearing all references at force_empty, we remember | |
4791 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4792 | * it goes down to 0. | |
4793 | * | |
4794 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4795 | */ |
c8b2a36f GC |
4796 | |
4797 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4798 | { |
c8b2a36f | 4799 | int node; |
59927fb9 | 4800 | |
bb4cc1a8 | 4801 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4802 | |
4803 | for_each_node(node) | |
4804 | free_mem_cgroup_per_zone_info(memcg, node); | |
4805 | ||
4806 | free_percpu(memcg->stat); | |
4807 | ||
3f134619 GC |
4808 | /* |
4809 | * We need to make sure that (at least for now), the jump label | |
4810 | * destruction code runs outside of the cgroup lock. This is because | |
4811 | * get_online_cpus(), which is called from the static_branch update, | |
4812 | * can't be called inside the cgroup_lock. cpusets are the ones | |
4813 | * enforcing this dependency, so if they ever change, we might as well. | |
4814 | * | |
4815 | * schedule_work() will guarantee this happens. Be careful if you need | |
4816 | * to move this code around, and make sure it is outside | |
4817 | * the cgroup_lock. | |
4818 | */ | |
a8964b9b | 4819 | disarm_static_keys(memcg); |
8ff69e2c | 4820 | kfree(memcg); |
59927fb9 | 4821 | } |
3afe36b1 | 4822 | |
7bcc1bb1 DN |
4823 | /* |
4824 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4825 | */ | |
e1aab161 | 4826 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4827 | { |
3e32cb2e | 4828 | if (!memcg->memory.parent) |
7bcc1bb1 | 4829 | return NULL; |
3e32cb2e | 4830 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4831 | } |
e1aab161 | 4832 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4833 | |
bb4cc1a8 AM |
4834 | static void __init mem_cgroup_soft_limit_tree_init(void) |
4835 | { | |
4836 | struct mem_cgroup_tree_per_node *rtpn; | |
4837 | struct mem_cgroup_tree_per_zone *rtpz; | |
4838 | int tmp, node, zone; | |
4839 | ||
4840 | for_each_node(node) { | |
4841 | tmp = node; | |
4842 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4843 | tmp = -1; | |
4844 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4845 | BUG_ON(!rtpn); | |
4846 | ||
4847 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4848 | ||
4849 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4850 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4851 | rtpz->rb_root = RB_ROOT; | |
4852 | spin_lock_init(&rtpz->lock); | |
4853 | } | |
4854 | } | |
4855 | } | |
4856 | ||
0eb253e2 | 4857 | static struct cgroup_subsys_state * __ref |
eb95419b | 4858 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4859 | { |
d142e3e6 | 4860 | struct mem_cgroup *memcg; |
04046e1a | 4861 | long error = -ENOMEM; |
6d12e2d8 | 4862 | int node; |
8cdea7c0 | 4863 | |
c0ff4b85 R |
4864 | memcg = mem_cgroup_alloc(); |
4865 | if (!memcg) | |
04046e1a | 4866 | return ERR_PTR(error); |
78fb7466 | 4867 | |
3ed28fa1 | 4868 | for_each_node(node) |
c0ff4b85 | 4869 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4870 | goto free_out; |
f64c3f54 | 4871 | |
c077719b | 4872 | /* root ? */ |
eb95419b | 4873 | if (parent_css == NULL) { |
a41c58a6 | 4874 | root_mem_cgroup = memcg; |
3e32cb2e JW |
4875 | page_counter_init(&memcg->memory, NULL); |
4876 | page_counter_init(&memcg->memsw, NULL); | |
4877 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4878 | } |
28dbc4b6 | 4879 | |
d142e3e6 GC |
4880 | memcg->last_scanned_node = MAX_NUMNODES; |
4881 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4882 | memcg->move_charge_at_immigrate = 0; |
4883 | mutex_init(&memcg->thresholds_lock); | |
4884 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4885 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4886 | INIT_LIST_HEAD(&memcg->event_list); |
4887 | spin_lock_init(&memcg->event_list_lock); | |
d142e3e6 GC |
4888 | |
4889 | return &memcg->css; | |
4890 | ||
4891 | free_out: | |
4892 | __mem_cgroup_free(memcg); | |
4893 | return ERR_PTR(error); | |
4894 | } | |
4895 | ||
4896 | static int | |
eb95419b | 4897 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4898 | { |
eb95419b | 4899 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4900 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4901 | int ret; |
d142e3e6 | 4902 | |
15a4c835 | 4903 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4904 | return -ENOSPC; |
4905 | ||
63876986 | 4906 | if (!parent) |
d142e3e6 GC |
4907 | return 0; |
4908 | ||
0999821b | 4909 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4910 | |
4911 | memcg->use_hierarchy = parent->use_hierarchy; | |
4912 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4913 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4914 | ||
4915 | if (parent->use_hierarchy) { | |
3e32cb2e JW |
4916 | page_counter_init(&memcg->memory, &parent->memory); |
4917 | page_counter_init(&memcg->memsw, &parent->memsw); | |
4918 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4919 | |
7bcc1bb1 | 4920 | /* |
8d76a979 LZ |
4921 | * No need to take a reference to the parent because cgroup |
4922 | * core guarantees its existence. | |
7bcc1bb1 | 4923 | */ |
18f59ea7 | 4924 | } else { |
3e32cb2e JW |
4925 | page_counter_init(&memcg->memory, NULL); |
4926 | page_counter_init(&memcg->memsw, NULL); | |
4927 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4928 | /* |
4929 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4930 | * much sense so let cgroup subsystem know about this | |
4931 | * unfortunate state in our controller. | |
4932 | */ | |
d142e3e6 | 4933 | if (parent != root_mem_cgroup) |
073219e9 | 4934 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4935 | } |
0999821b | 4936 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4937 | |
2f7dd7a4 JW |
4938 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4939 | if (ret) | |
4940 | return ret; | |
4941 | ||
4942 | /* | |
4943 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4944 | * orders reading memcg->initialized against its callers | |
4945 | * reading the memcg members. | |
4946 | */ | |
4947 | smp_store_release(&memcg->initialized, 1); | |
4948 | ||
4949 | return 0; | |
8cdea7c0 BS |
4950 | } |
4951 | ||
eb95419b | 4952 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4953 | { |
eb95419b | 4954 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4955 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4956 | |
4957 | /* | |
4958 | * Unregister events and notify userspace. | |
4959 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4960 | * directory to avoid race between userspace and kernelspace. | |
4961 | */ | |
fba94807 TH |
4962 | spin_lock(&memcg->event_list_lock); |
4963 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4964 | list_del_init(&event->list); |
4965 | schedule_work(&event->remove); | |
4966 | } | |
fba94807 | 4967 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4968 | |
776ed0f0 | 4969 | memcg_unregister_all_caches(memcg); |
33cb876e | 4970 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
4971 | } |
4972 | ||
eb95419b | 4973 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4974 | { |
eb95419b | 4975 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4976 | |
10d5ebf4 | 4977 | memcg_destroy_kmem(memcg); |
465939a1 | 4978 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4979 | } |
4980 | ||
1ced953b TH |
4981 | /** |
4982 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4983 | * @css: the target css | |
4984 | * | |
4985 | * Reset the states of the mem_cgroup associated with @css. This is | |
4986 | * invoked when the userland requests disabling on the default hierarchy | |
4987 | * but the memcg is pinned through dependency. The memcg should stop | |
4988 | * applying policies and should revert to the vanilla state as it may be | |
4989 | * made visible again. | |
4990 | * | |
4991 | * The current implementation only resets the essential configurations. | |
4992 | * This needs to be expanded to cover all the visible parts. | |
4993 | */ | |
4994 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4995 | { | |
4996 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4997 | ||
3e32cb2e JW |
4998 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4999 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
5000 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
5001 | memcg->soft_limit = 0; | |
1ced953b TH |
5002 | } |
5003 | ||
02491447 | 5004 | #ifdef CONFIG_MMU |
7dc74be0 | 5005 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5006 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5007 | { |
05b84301 | 5008 | int ret; |
9476db97 JW |
5009 | |
5010 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 5011 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 5012 | if (!ret) { |
854ffa8d | 5013 | mc.precharge += count; |
854ffa8d DN |
5014 | return ret; |
5015 | } | |
692e7c45 | 5016 | if (ret == -EINTR) { |
00501b53 | 5017 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
5018 | return ret; |
5019 | } | |
9476db97 JW |
5020 | |
5021 | /* Try charges one by one with reclaim */ | |
854ffa8d | 5022 | while (count--) { |
00501b53 | 5023 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
5024 | /* |
5025 | * In case of failure, any residual charges against | |
5026 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
5027 | * later on. However, cancel any charges that are |
5028 | * bypassed to root right away or they'll be lost. | |
9476db97 | 5029 | */ |
692e7c45 | 5030 | if (ret == -EINTR) |
00501b53 | 5031 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 5032 | if (ret) |
38c5d72f | 5033 | return ret; |
854ffa8d | 5034 | mc.precharge++; |
9476db97 | 5035 | cond_resched(); |
854ffa8d | 5036 | } |
9476db97 | 5037 | return 0; |
4ffef5fe DN |
5038 | } |
5039 | ||
5040 | /** | |
8d32ff84 | 5041 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5042 | * @vma: the vma the pte to be checked belongs |
5043 | * @addr: the address corresponding to the pte to be checked | |
5044 | * @ptent: the pte to be checked | |
02491447 | 5045 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5046 | * |
5047 | * Returns | |
5048 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5049 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5050 | * move charge. if @target is not NULL, the page is stored in target->page | |
5051 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5052 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5053 | * target for charge migration. if @target is not NULL, the entry is stored | |
5054 | * in target->ent. | |
4ffef5fe DN |
5055 | * |
5056 | * Called with pte lock held. | |
5057 | */ | |
4ffef5fe DN |
5058 | union mc_target { |
5059 | struct page *page; | |
02491447 | 5060 | swp_entry_t ent; |
4ffef5fe DN |
5061 | }; |
5062 | ||
4ffef5fe | 5063 | enum mc_target_type { |
8d32ff84 | 5064 | MC_TARGET_NONE = 0, |
4ffef5fe | 5065 | MC_TARGET_PAGE, |
02491447 | 5066 | MC_TARGET_SWAP, |
4ffef5fe DN |
5067 | }; |
5068 | ||
90254a65 DN |
5069 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5070 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5071 | { |
90254a65 | 5072 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5073 | |
90254a65 DN |
5074 | if (!page || !page_mapped(page)) |
5075 | return NULL; | |
5076 | if (PageAnon(page)) { | |
5077 | /* we don't move shared anon */ | |
4b91355e | 5078 | if (!move_anon()) |
90254a65 | 5079 | return NULL; |
87946a72 DN |
5080 | } else if (!move_file()) |
5081 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5082 | return NULL; |
5083 | if (!get_page_unless_zero(page)) | |
5084 | return NULL; | |
5085 | ||
5086 | return page; | |
5087 | } | |
5088 | ||
4b91355e | 5089 | #ifdef CONFIG_SWAP |
90254a65 DN |
5090 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5091 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5092 | { | |
90254a65 DN |
5093 | struct page *page = NULL; |
5094 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5095 | ||
5096 | if (!move_anon() || non_swap_entry(ent)) | |
5097 | return NULL; | |
4b91355e KH |
5098 | /* |
5099 | * Because lookup_swap_cache() updates some statistics counter, | |
5100 | * we call find_get_page() with swapper_space directly. | |
5101 | */ | |
33806f06 | 5102 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
5103 | if (do_swap_account) |
5104 | entry->val = ent.val; | |
5105 | ||
5106 | return page; | |
5107 | } | |
4b91355e KH |
5108 | #else |
5109 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5110 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5111 | { | |
5112 | return NULL; | |
5113 | } | |
5114 | #endif | |
90254a65 | 5115 | |
87946a72 DN |
5116 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5117 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5118 | { | |
5119 | struct page *page = NULL; | |
87946a72 DN |
5120 | struct address_space *mapping; |
5121 | pgoff_t pgoff; | |
5122 | ||
5123 | if (!vma->vm_file) /* anonymous vma */ | |
5124 | return NULL; | |
5125 | if (!move_file()) | |
5126 | return NULL; | |
5127 | ||
87946a72 DN |
5128 | mapping = vma->vm_file->f_mapping; |
5129 | if (pte_none(ptent)) | |
5130 | pgoff = linear_page_index(vma, addr); | |
5131 | else /* pte_file(ptent) is true */ | |
5132 | pgoff = pte_to_pgoff(ptent); | |
5133 | ||
5134 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5135 | #ifdef CONFIG_SWAP |
5136 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5137 | if (shmem_mapping(mapping)) { |
5138 | page = find_get_entry(mapping, pgoff); | |
5139 | if (radix_tree_exceptional_entry(page)) { | |
5140 | swp_entry_t swp = radix_to_swp_entry(page); | |
5141 | if (do_swap_account) | |
5142 | *entry = swp; | |
5143 | page = find_get_page(swap_address_space(swp), swp.val); | |
5144 | } | |
5145 | } else | |
5146 | page = find_get_page(mapping, pgoff); | |
5147 | #else | |
5148 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5149 | #endif |
87946a72 DN |
5150 | return page; |
5151 | } | |
5152 | ||
8d32ff84 | 5153 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5154 | unsigned long addr, pte_t ptent, union mc_target *target) |
5155 | { | |
5156 | struct page *page = NULL; | |
5157 | struct page_cgroup *pc; | |
8d32ff84 | 5158 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5159 | swp_entry_t ent = { .val = 0 }; |
5160 | ||
5161 | if (pte_present(ptent)) | |
5162 | page = mc_handle_present_pte(vma, addr, ptent); | |
5163 | else if (is_swap_pte(ptent)) | |
5164 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5165 | else if (pte_none(ptent) || pte_file(ptent)) |
5166 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5167 | |
5168 | if (!page && !ent.val) | |
8d32ff84 | 5169 | return ret; |
02491447 DN |
5170 | if (page) { |
5171 | pc = lookup_page_cgroup(page); | |
5172 | /* | |
0a31bc97 JW |
5173 | * Do only loose check w/o serialization. |
5174 | * mem_cgroup_move_account() checks the pc is valid or | |
5175 | * not under LRU exclusion. | |
02491447 | 5176 | */ |
29833315 | 5177 | if (pc->mem_cgroup == mc.from) { |
02491447 DN |
5178 | ret = MC_TARGET_PAGE; |
5179 | if (target) | |
5180 | target->page = page; | |
5181 | } | |
5182 | if (!ret || !target) | |
5183 | put_page(page); | |
5184 | } | |
90254a65 DN |
5185 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5186 | if (ent.val && !ret && | |
34c00c31 | 5187 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5188 | ret = MC_TARGET_SWAP; |
5189 | if (target) | |
5190 | target->ent = ent; | |
4ffef5fe | 5191 | } |
4ffef5fe DN |
5192 | return ret; |
5193 | } | |
5194 | ||
12724850 NH |
5195 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5196 | /* | |
5197 | * We don't consider swapping or file mapped pages because THP does not | |
5198 | * support them for now. | |
5199 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5200 | */ | |
5201 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5202 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5203 | { | |
5204 | struct page *page = NULL; | |
5205 | struct page_cgroup *pc; | |
5206 | enum mc_target_type ret = MC_TARGET_NONE; | |
5207 | ||
5208 | page = pmd_page(pmd); | |
309381fe | 5209 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
5210 | if (!move_anon()) |
5211 | return ret; | |
5212 | pc = lookup_page_cgroup(page); | |
29833315 | 5213 | if (pc->mem_cgroup == mc.from) { |
12724850 NH |
5214 | ret = MC_TARGET_PAGE; |
5215 | if (target) { | |
5216 | get_page(page); | |
5217 | target->page = page; | |
5218 | } | |
5219 | } | |
5220 | return ret; | |
5221 | } | |
5222 | #else | |
5223 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5224 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5225 | { | |
5226 | return MC_TARGET_NONE; | |
5227 | } | |
5228 | #endif | |
5229 | ||
4ffef5fe DN |
5230 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5231 | unsigned long addr, unsigned long end, | |
5232 | struct mm_walk *walk) | |
5233 | { | |
5234 | struct vm_area_struct *vma = walk->private; | |
5235 | pte_t *pte; | |
5236 | spinlock_t *ptl; | |
5237 | ||
bf929152 | 5238 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
5239 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5240 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5241 | spin_unlock(ptl); |
1a5a9906 | 5242 | return 0; |
12724850 | 5243 | } |
03319327 | 5244 | |
45f83cef AA |
5245 | if (pmd_trans_unstable(pmd)) |
5246 | return 0; | |
4ffef5fe DN |
5247 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5248 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5249 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5250 | mc.precharge++; /* increment precharge temporarily */ |
5251 | pte_unmap_unlock(pte - 1, ptl); | |
5252 | cond_resched(); | |
5253 | ||
7dc74be0 DN |
5254 | return 0; |
5255 | } | |
5256 | ||
4ffef5fe DN |
5257 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5258 | { | |
5259 | unsigned long precharge; | |
5260 | struct vm_area_struct *vma; | |
5261 | ||
dfe076b0 | 5262 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5263 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5264 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5265 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5266 | .mm = mm, | |
5267 | .private = vma, | |
5268 | }; | |
5269 | if (is_vm_hugetlb_page(vma)) | |
5270 | continue; | |
4ffef5fe DN |
5271 | walk_page_range(vma->vm_start, vma->vm_end, |
5272 | &mem_cgroup_count_precharge_walk); | |
5273 | } | |
dfe076b0 | 5274 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5275 | |
5276 | precharge = mc.precharge; | |
5277 | mc.precharge = 0; | |
5278 | ||
5279 | return precharge; | |
5280 | } | |
5281 | ||
4ffef5fe DN |
5282 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5283 | { | |
dfe076b0 DN |
5284 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5285 | ||
5286 | VM_BUG_ON(mc.moving_task); | |
5287 | mc.moving_task = current; | |
5288 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5289 | } |
5290 | ||
dfe076b0 DN |
5291 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5292 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5293 | { |
2bd9bb20 KH |
5294 | struct mem_cgroup *from = mc.from; |
5295 | struct mem_cgroup *to = mc.to; | |
5296 | ||
4ffef5fe | 5297 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5298 | if (mc.precharge) { |
00501b53 | 5299 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5300 | mc.precharge = 0; |
5301 | } | |
5302 | /* | |
5303 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5304 | * we must uncharge here. | |
5305 | */ | |
5306 | if (mc.moved_charge) { | |
00501b53 | 5307 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5308 | mc.moved_charge = 0; |
4ffef5fe | 5309 | } |
483c30b5 DN |
5310 | /* we must fixup refcnts and charges */ |
5311 | if (mc.moved_swap) { | |
483c30b5 | 5312 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5313 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5314 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5315 | |
05b84301 | 5316 | /* |
3e32cb2e JW |
5317 | * we charged both to->memory and to->memsw, so we |
5318 | * should uncharge to->memory. | |
05b84301 | 5319 | */ |
ce00a967 | 5320 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5321 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5322 | ||
e8ea14cc | 5323 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 5324 | |
4050377b | 5325 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
5326 | mc.moved_swap = 0; |
5327 | } | |
dfe076b0 DN |
5328 | memcg_oom_recover(from); |
5329 | memcg_oom_recover(to); | |
5330 | wake_up_all(&mc.waitq); | |
5331 | } | |
5332 | ||
5333 | static void mem_cgroup_clear_mc(void) | |
5334 | { | |
dfe076b0 DN |
5335 | /* |
5336 | * we must clear moving_task before waking up waiters at the end of | |
5337 | * task migration. | |
5338 | */ | |
5339 | mc.moving_task = NULL; | |
5340 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5341 | spin_lock(&mc.lock); |
4ffef5fe DN |
5342 | mc.from = NULL; |
5343 | mc.to = NULL; | |
2bd9bb20 | 5344 | spin_unlock(&mc.lock); |
4ffef5fe DN |
5345 | } |
5346 | ||
eb95419b | 5347 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5348 | struct cgroup_taskset *tset) |
7dc74be0 | 5349 | { |
2f7ee569 | 5350 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5351 | int ret = 0; |
eb95419b | 5352 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 5353 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 5354 | |
ee5e8472 GC |
5355 | /* |
5356 | * We are now commited to this value whatever it is. Changes in this | |
5357 | * tunable will only affect upcoming migrations, not the current one. | |
5358 | * So we need to save it, and keep it going. | |
5359 | */ | |
5360 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
5361 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
5362 | struct mm_struct *mm; |
5363 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5364 | ||
c0ff4b85 | 5365 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5366 | |
5367 | mm = get_task_mm(p); | |
5368 | if (!mm) | |
5369 | return 0; | |
7dc74be0 | 5370 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5371 | if (mm->owner == p) { |
5372 | VM_BUG_ON(mc.from); | |
5373 | VM_BUG_ON(mc.to); | |
5374 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5375 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5376 | VM_BUG_ON(mc.moved_swap); |
247b1447 | 5377 | |
2bd9bb20 | 5378 | spin_lock(&mc.lock); |
4ffef5fe | 5379 | mc.from = from; |
c0ff4b85 | 5380 | mc.to = memcg; |
ee5e8472 | 5381 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 5382 | spin_unlock(&mc.lock); |
dfe076b0 | 5383 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5384 | |
5385 | ret = mem_cgroup_precharge_mc(mm); | |
5386 | if (ret) | |
5387 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5388 | } |
5389 | mmput(mm); | |
7dc74be0 DN |
5390 | } |
5391 | return ret; | |
5392 | } | |
5393 | ||
eb95419b | 5394 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5395 | struct cgroup_taskset *tset) |
7dc74be0 | 5396 | { |
4e2f245d JW |
5397 | if (mc.to) |
5398 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5399 | } |
5400 | ||
4ffef5fe DN |
5401 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5402 | unsigned long addr, unsigned long end, | |
5403 | struct mm_walk *walk) | |
7dc74be0 | 5404 | { |
4ffef5fe DN |
5405 | int ret = 0; |
5406 | struct vm_area_struct *vma = walk->private; | |
5407 | pte_t *pte; | |
5408 | spinlock_t *ptl; | |
12724850 NH |
5409 | enum mc_target_type target_type; |
5410 | union mc_target target; | |
5411 | struct page *page; | |
5412 | struct page_cgroup *pc; | |
4ffef5fe | 5413 | |
12724850 NH |
5414 | /* |
5415 | * We don't take compound_lock() here but no race with splitting thp | |
5416 | * happens because: | |
5417 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5418 | * under splitting, which means there's no concurrent thp split, | |
5419 | * - if another thread runs into split_huge_page() just after we | |
5420 | * entered this if-block, the thread must wait for page table lock | |
5421 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5422 | * part of thp split is not executed yet. | |
5423 | */ | |
bf929152 | 5424 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5425 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5426 | spin_unlock(ptl); |
12724850 NH |
5427 | return 0; |
5428 | } | |
5429 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5430 | if (target_type == MC_TARGET_PAGE) { | |
5431 | page = target.page; | |
5432 | if (!isolate_lru_page(page)) { | |
5433 | pc = lookup_page_cgroup(page); | |
5434 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5435 | pc, mc.from, mc.to)) { |
12724850 NH |
5436 | mc.precharge -= HPAGE_PMD_NR; |
5437 | mc.moved_charge += HPAGE_PMD_NR; | |
5438 | } | |
5439 | putback_lru_page(page); | |
5440 | } | |
5441 | put_page(page); | |
5442 | } | |
bf929152 | 5443 | spin_unlock(ptl); |
1a5a9906 | 5444 | return 0; |
12724850 NH |
5445 | } |
5446 | ||
45f83cef AA |
5447 | if (pmd_trans_unstable(pmd)) |
5448 | return 0; | |
4ffef5fe DN |
5449 | retry: |
5450 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5451 | for (; addr != end; addr += PAGE_SIZE) { | |
5452 | pte_t ptent = *(pte++); | |
02491447 | 5453 | swp_entry_t ent; |
4ffef5fe DN |
5454 | |
5455 | if (!mc.precharge) | |
5456 | break; | |
5457 | ||
8d32ff84 | 5458 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5459 | case MC_TARGET_PAGE: |
5460 | page = target.page; | |
5461 | if (isolate_lru_page(page)) | |
5462 | goto put; | |
5463 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5464 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5465 | mc.from, mc.to)) { |
4ffef5fe | 5466 | mc.precharge--; |
854ffa8d DN |
5467 | /* we uncharge from mc.from later. */ |
5468 | mc.moved_charge++; | |
4ffef5fe DN |
5469 | } |
5470 | putback_lru_page(page); | |
8d32ff84 | 5471 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5472 | put_page(page); |
5473 | break; | |
02491447 DN |
5474 | case MC_TARGET_SWAP: |
5475 | ent = target.ent; | |
e91cbb42 | 5476 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5477 | mc.precharge--; |
483c30b5 DN |
5478 | /* we fixup refcnts and charges later. */ |
5479 | mc.moved_swap++; | |
5480 | } | |
02491447 | 5481 | break; |
4ffef5fe DN |
5482 | default: |
5483 | break; | |
5484 | } | |
5485 | } | |
5486 | pte_unmap_unlock(pte - 1, ptl); | |
5487 | cond_resched(); | |
5488 | ||
5489 | if (addr != end) { | |
5490 | /* | |
5491 | * We have consumed all precharges we got in can_attach(). | |
5492 | * We try charge one by one, but don't do any additional | |
5493 | * charges to mc.to if we have failed in charge once in attach() | |
5494 | * phase. | |
5495 | */ | |
854ffa8d | 5496 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5497 | if (!ret) |
5498 | goto retry; | |
5499 | } | |
5500 | ||
5501 | return ret; | |
5502 | } | |
5503 | ||
5504 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5505 | { | |
5506 | struct vm_area_struct *vma; | |
5507 | ||
5508 | lru_add_drain_all(); | |
312722cb JW |
5509 | /* |
5510 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5511 | * move_lock while we're moving its pages to another memcg. | |
5512 | * Then wait for already started RCU-only updates to finish. | |
5513 | */ | |
5514 | atomic_inc(&mc.from->moving_account); | |
5515 | synchronize_rcu(); | |
dfe076b0 DN |
5516 | retry: |
5517 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5518 | /* | |
5519 | * Someone who are holding the mmap_sem might be waiting in | |
5520 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5521 | * and retry. Because we cancel precharges, we might not be able | |
5522 | * to move enough charges, but moving charge is a best-effort | |
5523 | * feature anyway, so it wouldn't be a big problem. | |
5524 | */ | |
5525 | __mem_cgroup_clear_mc(); | |
5526 | cond_resched(); | |
5527 | goto retry; | |
5528 | } | |
4ffef5fe DN |
5529 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5530 | int ret; | |
5531 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5532 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5533 | .mm = mm, | |
5534 | .private = vma, | |
5535 | }; | |
5536 | if (is_vm_hugetlb_page(vma)) | |
5537 | continue; | |
4ffef5fe DN |
5538 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5539 | &mem_cgroup_move_charge_walk); | |
5540 | if (ret) | |
5541 | /* | |
5542 | * means we have consumed all precharges and failed in | |
5543 | * doing additional charge. Just abandon here. | |
5544 | */ | |
5545 | break; | |
5546 | } | |
dfe076b0 | 5547 | up_read(&mm->mmap_sem); |
312722cb | 5548 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5549 | } |
5550 | ||
eb95419b | 5551 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5552 | struct cgroup_taskset *tset) |
67e465a7 | 5553 | { |
2f7ee569 | 5554 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5555 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5556 | |
dfe076b0 | 5557 | if (mm) { |
a433658c KM |
5558 | if (mc.to) |
5559 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5560 | mmput(mm); |
5561 | } | |
a433658c KM |
5562 | if (mc.to) |
5563 | mem_cgroup_clear_mc(); | |
67e465a7 | 5564 | } |
5cfb80a7 | 5565 | #else /* !CONFIG_MMU */ |
eb95419b | 5566 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5567 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5568 | { |
5569 | return 0; | |
5570 | } | |
eb95419b | 5571 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5572 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5573 | { |
5574 | } | |
eb95419b | 5575 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5576 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5577 | { |
5578 | } | |
5579 | #endif | |
67e465a7 | 5580 | |
f00baae7 TH |
5581 | /* |
5582 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5583 | * to verify whether we're attached to the default hierarchy on each mount |
5584 | * attempt. | |
f00baae7 | 5585 | */ |
eb95419b | 5586 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5587 | { |
5588 | /* | |
aa6ec29b | 5589 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5590 | * guarantees that @root doesn't have any children, so turning it |
5591 | * on for the root memcg is enough. | |
5592 | */ | |
aa6ec29b | 5593 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 5594 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
5595 | } |
5596 | ||
073219e9 | 5597 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5598 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5599 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5600 | .css_offline = mem_cgroup_css_offline, |
5601 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5602 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5603 | .can_attach = mem_cgroup_can_attach, |
5604 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5605 | .attach = mem_cgroup_move_task, |
f00baae7 | 5606 | .bind = mem_cgroup_bind, |
5577964e | 5607 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 5608 | .early_init = 0, |
8cdea7c0 | 5609 | }; |
c077719b | 5610 | |
c255a458 | 5611 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5612 | static int __init enable_swap_account(char *s) |
5613 | { | |
a2c8990a | 5614 | if (!strcmp(s, "1")) |
a42c390c | 5615 | really_do_swap_account = 1; |
a2c8990a | 5616 | else if (!strcmp(s, "0")) |
a42c390c MH |
5617 | really_do_swap_account = 0; |
5618 | return 1; | |
5619 | } | |
a2c8990a | 5620 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5621 | |
2d11085e MH |
5622 | static void __init memsw_file_init(void) |
5623 | { | |
2cf669a5 TH |
5624 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
5625 | memsw_cgroup_files)); | |
6acc8b02 MH |
5626 | } |
5627 | ||
5628 | static void __init enable_swap_cgroup(void) | |
5629 | { | |
5630 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5631 | do_swap_account = 1; | |
5632 | memsw_file_init(); | |
5633 | } | |
2d11085e | 5634 | } |
6acc8b02 | 5635 | |
2d11085e | 5636 | #else |
6acc8b02 | 5637 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
5638 | { |
5639 | } | |
c077719b | 5640 | #endif |
2d11085e | 5641 | |
0a31bc97 JW |
5642 | #ifdef CONFIG_MEMCG_SWAP |
5643 | /** | |
5644 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5645 | * @page: page whose memsw charge to transfer | |
5646 | * @entry: swap entry to move the charge to | |
5647 | * | |
5648 | * Transfer the memsw charge of @page to @entry. | |
5649 | */ | |
5650 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5651 | { | |
7bdd143c | 5652 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5653 | struct page_cgroup *pc; |
5654 | unsigned short oldid; | |
5655 | ||
5656 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5657 | VM_BUG_ON_PAGE(page_count(page), page); | |
5658 | ||
5659 | if (!do_swap_account) | |
5660 | return; | |
5661 | ||
5662 | pc = lookup_page_cgroup(page); | |
29833315 | 5663 | memcg = pc->mem_cgroup; |
0a31bc97 JW |
5664 | |
5665 | /* Readahead page, never charged */ | |
29833315 | 5666 | if (!memcg) |
0a31bc97 JW |
5667 | return; |
5668 | ||
7bdd143c | 5669 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
0a31bc97 | 5670 | VM_BUG_ON_PAGE(oldid, page); |
7bdd143c JW |
5671 | mem_cgroup_swap_statistics(memcg, true); |
5672 | ||
29833315 | 5673 | pc->mem_cgroup = NULL; |
7bdd143c JW |
5674 | |
5675 | if (!mem_cgroup_is_root(memcg)) | |
5676 | page_counter_uncharge(&memcg->memory, 1); | |
5677 | ||
5678 | /* XXX: caller holds IRQ-safe mapping->tree_lock */ | |
5679 | VM_BUG_ON(!irqs_disabled()); | |
0a31bc97 | 5680 | |
7bdd143c JW |
5681 | mem_cgroup_charge_statistics(memcg, page, -1); |
5682 | memcg_check_events(memcg, page); | |
0a31bc97 JW |
5683 | } |
5684 | ||
5685 | /** | |
5686 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5687 | * @entry: swap entry to uncharge | |
5688 | * | |
5689 | * Drop the memsw charge associated with @entry. | |
5690 | */ | |
5691 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5692 | { | |
5693 | struct mem_cgroup *memcg; | |
5694 | unsigned short id; | |
5695 | ||
5696 | if (!do_swap_account) | |
5697 | return; | |
5698 | ||
5699 | id = swap_cgroup_record(entry, 0); | |
5700 | rcu_read_lock(); | |
5701 | memcg = mem_cgroup_lookup(id); | |
5702 | if (memcg) { | |
ce00a967 | 5703 | if (!mem_cgroup_is_root(memcg)) |
3e32cb2e | 5704 | page_counter_uncharge(&memcg->memsw, 1); |
0a31bc97 JW |
5705 | mem_cgroup_swap_statistics(memcg, false); |
5706 | css_put(&memcg->css); | |
5707 | } | |
5708 | rcu_read_unlock(); | |
5709 | } | |
5710 | #endif | |
5711 | ||
00501b53 JW |
5712 | /** |
5713 | * mem_cgroup_try_charge - try charging a page | |
5714 | * @page: page to charge | |
5715 | * @mm: mm context of the victim | |
5716 | * @gfp_mask: reclaim mode | |
5717 | * @memcgp: charged memcg return | |
5718 | * | |
5719 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5720 | * pages according to @gfp_mask if necessary. | |
5721 | * | |
5722 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5723 | * Otherwise, an error code is returned. | |
5724 | * | |
5725 | * After page->mapping has been set up, the caller must finalize the | |
5726 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5727 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5728 | */ | |
5729 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5730 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5731 | { | |
5732 | struct mem_cgroup *memcg = NULL; | |
5733 | unsigned int nr_pages = 1; | |
5734 | int ret = 0; | |
5735 | ||
5736 | if (mem_cgroup_disabled()) | |
5737 | goto out; | |
5738 | ||
5739 | if (PageSwapCache(page)) { | |
5740 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
5741 | /* | |
5742 | * Every swap fault against a single page tries to charge the | |
5743 | * page, bail as early as possible. shmem_unuse() encounters | |
5744 | * already charged pages, too. The USED bit is protected by | |
5745 | * the page lock, which serializes swap cache removal, which | |
5746 | * in turn serializes uncharging. | |
5747 | */ | |
29833315 | 5748 | if (pc->mem_cgroup) |
00501b53 JW |
5749 | goto out; |
5750 | } | |
5751 | ||
5752 | if (PageTransHuge(page)) { | |
5753 | nr_pages <<= compound_order(page); | |
5754 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5755 | } | |
5756 | ||
5757 | if (do_swap_account && PageSwapCache(page)) | |
5758 | memcg = try_get_mem_cgroup_from_page(page); | |
5759 | if (!memcg) | |
5760 | memcg = get_mem_cgroup_from_mm(mm); | |
5761 | ||
5762 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5763 | ||
5764 | css_put(&memcg->css); | |
5765 | ||
5766 | if (ret == -EINTR) { | |
5767 | memcg = root_mem_cgroup; | |
5768 | ret = 0; | |
5769 | } | |
5770 | out: | |
5771 | *memcgp = memcg; | |
5772 | return ret; | |
5773 | } | |
5774 | ||
5775 | /** | |
5776 | * mem_cgroup_commit_charge - commit a page charge | |
5777 | * @page: page to charge | |
5778 | * @memcg: memcg to charge the page to | |
5779 | * @lrucare: page might be on LRU already | |
5780 | * | |
5781 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5782 | * after page->mapping has been set up. This must happen atomically | |
5783 | * as part of the page instantiation, i.e. under the page table lock | |
5784 | * for anonymous pages, under the page lock for page and swap cache. | |
5785 | * | |
5786 | * In addition, the page must not be on the LRU during the commit, to | |
5787 | * prevent racing with task migration. If it might be, use @lrucare. | |
5788 | * | |
5789 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5790 | */ | |
5791 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5792 | bool lrucare) | |
5793 | { | |
5794 | unsigned int nr_pages = 1; | |
5795 | ||
5796 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5797 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5798 | ||
5799 | if (mem_cgroup_disabled()) | |
5800 | return; | |
5801 | /* | |
5802 | * Swap faults will attempt to charge the same page multiple | |
5803 | * times. But reuse_swap_page() might have removed the page | |
5804 | * from swapcache already, so we can't check PageSwapCache(). | |
5805 | */ | |
5806 | if (!memcg) | |
5807 | return; | |
5808 | ||
6abb5a86 JW |
5809 | commit_charge(page, memcg, lrucare); |
5810 | ||
00501b53 JW |
5811 | if (PageTransHuge(page)) { |
5812 | nr_pages <<= compound_order(page); | |
5813 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5814 | } | |
5815 | ||
6abb5a86 JW |
5816 | local_irq_disable(); |
5817 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5818 | memcg_check_events(memcg, page); | |
5819 | local_irq_enable(); | |
00501b53 JW |
5820 | |
5821 | if (do_swap_account && PageSwapCache(page)) { | |
5822 | swp_entry_t entry = { .val = page_private(page) }; | |
5823 | /* | |
5824 | * The swap entry might not get freed for a long time, | |
5825 | * let's not wait for it. The page already received a | |
5826 | * memory+swap charge, drop the swap entry duplicate. | |
5827 | */ | |
5828 | mem_cgroup_uncharge_swap(entry); | |
5829 | } | |
5830 | } | |
5831 | ||
5832 | /** | |
5833 | * mem_cgroup_cancel_charge - cancel a page charge | |
5834 | * @page: page to charge | |
5835 | * @memcg: memcg to charge the page to | |
5836 | * | |
5837 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5838 | */ | |
5839 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5840 | { | |
5841 | unsigned int nr_pages = 1; | |
5842 | ||
5843 | if (mem_cgroup_disabled()) | |
5844 | return; | |
5845 | /* | |
5846 | * Swap faults will attempt to charge the same page multiple | |
5847 | * times. But reuse_swap_page() might have removed the page | |
5848 | * from swapcache already, so we can't check PageSwapCache(). | |
5849 | */ | |
5850 | if (!memcg) | |
5851 | return; | |
5852 | ||
5853 | if (PageTransHuge(page)) { | |
5854 | nr_pages <<= compound_order(page); | |
5855 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5856 | } | |
5857 | ||
5858 | cancel_charge(memcg, nr_pages); | |
5859 | } | |
5860 | ||
747db954 | 5861 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5862 | unsigned long nr_anon, unsigned long nr_file, |
5863 | unsigned long nr_huge, struct page *dummy_page) | |
5864 | { | |
18eca2e6 | 5865 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5866 | unsigned long flags; |
5867 | ||
ce00a967 | 5868 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5869 | page_counter_uncharge(&memcg->memory, nr_pages); |
5870 | if (do_swap_account) | |
5871 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5872 | memcg_oom_recover(memcg); |
5873 | } | |
747db954 JW |
5874 | |
5875 | local_irq_save(flags); | |
5876 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5877 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5878 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5879 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5880 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5881 | memcg_check_events(memcg, dummy_page); |
5882 | local_irq_restore(flags); | |
e8ea14cc JW |
5883 | |
5884 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5885 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5886 | } |
5887 | ||
5888 | static void uncharge_list(struct list_head *page_list) | |
5889 | { | |
5890 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5891 | unsigned long nr_anon = 0; |
5892 | unsigned long nr_file = 0; | |
5893 | unsigned long nr_huge = 0; | |
5894 | unsigned long pgpgout = 0; | |
747db954 JW |
5895 | struct list_head *next; |
5896 | struct page *page; | |
5897 | ||
5898 | next = page_list->next; | |
5899 | do { | |
5900 | unsigned int nr_pages = 1; | |
5901 | struct page_cgroup *pc; | |
5902 | ||
5903 | page = list_entry(next, struct page, lru); | |
5904 | next = page->lru.next; | |
5905 | ||
5906 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5907 | VM_BUG_ON_PAGE(page_count(page), page); | |
5908 | ||
5909 | pc = lookup_page_cgroup(page); | |
29833315 | 5910 | if (!pc->mem_cgroup) |
747db954 JW |
5911 | continue; |
5912 | ||
5913 | /* | |
5914 | * Nobody should be changing or seriously looking at | |
29833315 JW |
5915 | * pc->mem_cgroup at this point, we have fully |
5916 | * exclusive access to the page. | |
747db954 JW |
5917 | */ |
5918 | ||
5919 | if (memcg != pc->mem_cgroup) { | |
5920 | if (memcg) { | |
18eca2e6 JW |
5921 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5922 | nr_huge, page); | |
5923 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 JW |
5924 | } |
5925 | memcg = pc->mem_cgroup; | |
5926 | } | |
5927 | ||
5928 | if (PageTransHuge(page)) { | |
5929 | nr_pages <<= compound_order(page); | |
5930 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5931 | nr_huge += nr_pages; | |
5932 | } | |
5933 | ||
5934 | if (PageAnon(page)) | |
5935 | nr_anon += nr_pages; | |
5936 | else | |
5937 | nr_file += nr_pages; | |
5938 | ||
29833315 | 5939 | pc->mem_cgroup = NULL; |
747db954 JW |
5940 | |
5941 | pgpgout++; | |
5942 | } while (next != page_list); | |
5943 | ||
5944 | if (memcg) | |
18eca2e6 JW |
5945 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5946 | nr_huge, page); | |
747db954 JW |
5947 | } |
5948 | ||
0a31bc97 JW |
5949 | /** |
5950 | * mem_cgroup_uncharge - uncharge a page | |
5951 | * @page: page to uncharge | |
5952 | * | |
5953 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5954 | * mem_cgroup_commit_charge(). | |
5955 | */ | |
5956 | void mem_cgroup_uncharge(struct page *page) | |
5957 | { | |
0a31bc97 | 5958 | struct page_cgroup *pc; |
0a31bc97 JW |
5959 | |
5960 | if (mem_cgroup_disabled()) | |
5961 | return; | |
5962 | ||
747db954 | 5963 | /* Don't touch page->lru of any random page, pre-check: */ |
0a31bc97 | 5964 | pc = lookup_page_cgroup(page); |
29833315 | 5965 | if (!pc->mem_cgroup) |
0a31bc97 JW |
5966 | return; |
5967 | ||
747db954 JW |
5968 | INIT_LIST_HEAD(&page->lru); |
5969 | uncharge_list(&page->lru); | |
5970 | } | |
0a31bc97 | 5971 | |
747db954 JW |
5972 | /** |
5973 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5974 | * @page_list: list of pages to uncharge | |
5975 | * | |
5976 | * Uncharge a list of pages previously charged with | |
5977 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5978 | */ | |
5979 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5980 | { | |
5981 | if (mem_cgroup_disabled()) | |
5982 | return; | |
0a31bc97 | 5983 | |
747db954 JW |
5984 | if (!list_empty(page_list)) |
5985 | uncharge_list(page_list); | |
0a31bc97 JW |
5986 | } |
5987 | ||
5988 | /** | |
5989 | * mem_cgroup_migrate - migrate a charge to another page | |
5990 | * @oldpage: currently charged page | |
5991 | * @newpage: page to transfer the charge to | |
5992 | * @lrucare: both pages might be on the LRU already | |
5993 | * | |
5994 | * Migrate the charge from @oldpage to @newpage. | |
5995 | * | |
5996 | * Both pages must be locked, @newpage->mapping must be set up. | |
5997 | */ | |
5998 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
5999 | bool lrucare) | |
6000 | { | |
29833315 | 6001 | struct mem_cgroup *memcg; |
0a31bc97 JW |
6002 | struct page_cgroup *pc; |
6003 | int isolated; | |
6004 | ||
6005 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6006 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
6007 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
6008 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
6009 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
6010 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6011 | newpage); | |
0a31bc97 JW |
6012 | |
6013 | if (mem_cgroup_disabled()) | |
6014 | return; | |
6015 | ||
6016 | /* Page cache replacement: new page already charged? */ | |
6017 | pc = lookup_page_cgroup(newpage); | |
29833315 | 6018 | if (pc->mem_cgroup) |
0a31bc97 JW |
6019 | return; |
6020 | ||
7d5e3245 JW |
6021 | /* |
6022 | * Swapcache readahead pages can get migrated before being | |
6023 | * charged, and migration from compaction can happen to an | |
6024 | * uncharged page when the PFN walker finds a page that | |
6025 | * reclaim just put back on the LRU but has not released yet. | |
6026 | */ | |
0a31bc97 | 6027 | pc = lookup_page_cgroup(oldpage); |
29833315 JW |
6028 | memcg = pc->mem_cgroup; |
6029 | if (!memcg) | |
0a31bc97 JW |
6030 | return; |
6031 | ||
0a31bc97 JW |
6032 | if (lrucare) |
6033 | lock_page_lru(oldpage, &isolated); | |
6034 | ||
29833315 | 6035 | pc->mem_cgroup = NULL; |
0a31bc97 JW |
6036 | |
6037 | if (lrucare) | |
6038 | unlock_page_lru(oldpage, isolated); | |
6039 | ||
29833315 | 6040 | commit_charge(newpage, memcg, lrucare); |
0a31bc97 JW |
6041 | } |
6042 | ||
2d11085e | 6043 | /* |
1081312f MH |
6044 | * subsys_initcall() for memory controller. |
6045 | * | |
6046 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6047 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6048 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6049 | * be initialized from here. | |
2d11085e MH |
6050 | */ |
6051 | static int __init mem_cgroup_init(void) | |
6052 | { | |
6053 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6054 | enable_swap_cgroup(); |
bb4cc1a8 | 6055 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6056 | memcg_stock_init(); |
2d11085e MH |
6057 | return 0; |
6058 | } | |
6059 | subsys_initcall(mem_cgroup_init); |