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> |
5d1ea48b | 54 | #include <linux/swap_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; | |
79dfdacc MH |
299 | |
300 | bool oom_lock; | |
301 | atomic_t under_oom; | |
3812c8c8 | 302 | atomic_t oom_wakeups; |
79dfdacc | 303 | |
1f4c025b | 304 | int swappiness; |
3c11ecf4 KH |
305 | /* OOM-Killer disable */ |
306 | int oom_kill_disable; | |
a7885eb8 | 307 | |
2e72b634 KS |
308 | /* protect arrays of thresholds */ |
309 | struct mutex thresholds_lock; | |
310 | ||
311 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 312 | struct mem_cgroup_thresholds thresholds; |
907860ed | 313 | |
2e72b634 | 314 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 315 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 316 | |
9490ff27 KH |
317 | /* For oom notifier event fd */ |
318 | struct list_head oom_notify; | |
185efc0f | 319 | |
7dc74be0 DN |
320 | /* |
321 | * Should we move charges of a task when a task is moved into this | |
322 | * mem_cgroup ? And what type of charges should we move ? | |
323 | */ | |
f894ffa8 | 324 | unsigned long move_charge_at_immigrate; |
619d094b KH |
325 | /* |
326 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
327 | */ | |
6de22619 | 328 | atomic_t moving_account; |
312734c0 | 329 | /* taken only while moving_account > 0 */ |
6de22619 JW |
330 | spinlock_t move_lock; |
331 | struct task_struct *move_lock_task; | |
332 | unsigned long move_lock_flags; | |
d52aa412 | 333 | /* |
c62b1a3b | 334 | * percpu counter. |
d52aa412 | 335 | */ |
3a7951b4 | 336 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
337 | /* |
338 | * used when a cpu is offlined or other synchronizations | |
339 | * See mem_cgroup_read_stat(). | |
340 | */ | |
341 | struct mem_cgroup_stat_cpu nocpu_base; | |
342 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 343 | |
4bd2c1ee | 344 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
2e685cad | 345 | struct cg_proto tcp_mem; |
d1a4c0b3 | 346 | #endif |
2633d7a0 | 347 | #if defined(CONFIG_MEMCG_KMEM) |
2633d7a0 GC |
348 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ |
349 | int kmemcg_id; | |
350 | #endif | |
45cf7ebd GC |
351 | |
352 | int last_scanned_node; | |
353 | #if MAX_NUMNODES > 1 | |
354 | nodemask_t scan_nodes; | |
355 | atomic_t numainfo_events; | |
356 | atomic_t numainfo_updating; | |
357 | #endif | |
70ddf637 | 358 | |
fba94807 TH |
359 | /* List of events which userspace want to receive */ |
360 | struct list_head event_list; | |
361 | spinlock_t event_list_lock; | |
362 | ||
54f72fe0 JW |
363 | struct mem_cgroup_per_node *nodeinfo[0]; |
364 | /* WARNING: nodeinfo must be the last member here */ | |
8cdea7c0 BS |
365 | }; |
366 | ||
510fc4e1 | 367 | #ifdef CONFIG_MEMCG_KMEM |
7de37682 GC |
368 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) |
369 | { | |
900a38f0 | 370 | return memcg->kmemcg_id >= 0; |
7de37682 | 371 | } |
510fc4e1 GC |
372 | #endif |
373 | ||
7dc74be0 DN |
374 | /* Stuffs for move charges at task migration. */ |
375 | /* | |
ee5e8472 GC |
376 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
377 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
378 | */ |
379 | enum move_type { | |
4ffef5fe | 380 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 381 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
382 | NR_MOVE_TYPE, |
383 | }; | |
384 | ||
4ffef5fe DN |
385 | /* "mc" and its members are protected by cgroup_mutex */ |
386 | static struct move_charge_struct { | |
b1dd693e | 387 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
388 | struct mem_cgroup *from; |
389 | struct mem_cgroup *to; | |
ee5e8472 | 390 | unsigned long immigrate_flags; |
4ffef5fe | 391 | unsigned long precharge; |
854ffa8d | 392 | unsigned long moved_charge; |
483c30b5 | 393 | unsigned long moved_swap; |
8033b97c DN |
394 | struct task_struct *moving_task; /* a task moving charges */ |
395 | wait_queue_head_t waitq; /* a waitq for other context */ | |
396 | } mc = { | |
2bd9bb20 | 397 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
398 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
399 | }; | |
4ffef5fe | 400 | |
90254a65 DN |
401 | static bool move_anon(void) |
402 | { | |
ee5e8472 | 403 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
404 | } |
405 | ||
87946a72 DN |
406 | static bool move_file(void) |
407 | { | |
ee5e8472 | 408 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
409 | } |
410 | ||
4e416953 BS |
411 | /* |
412 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
413 | * limit reclaim to prevent infinite loops, if they ever occur. | |
414 | */ | |
a0db00fc | 415 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 416 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 417 | |
217bc319 KH |
418 | enum charge_type { |
419 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 420 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 421 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 422 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
423 | NR_CHARGE_TYPE, |
424 | }; | |
425 | ||
8c7c6e34 | 426 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
427 | enum res_type { |
428 | _MEM, | |
429 | _MEMSWAP, | |
430 | _OOM_TYPE, | |
510fc4e1 | 431 | _KMEM, |
86ae53e1 GC |
432 | }; |
433 | ||
a0db00fc KS |
434 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
435 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 436 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
437 | /* Used for OOM nofiier */ |
438 | #define OOM_CONTROL (0) | |
8c7c6e34 | 439 | |
0999821b GC |
440 | /* |
441 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
442 | * As a consequence, any change that needs to protect against new child cgroups | |
443 | * appearing has to hold it as well. | |
444 | */ | |
445 | static DEFINE_MUTEX(memcg_create_mutex); | |
446 | ||
b2145145 WL |
447 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
448 | { | |
a7c6d554 | 449 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
b2145145 WL |
450 | } |
451 | ||
70ddf637 AV |
452 | /* Some nice accessors for the vmpressure. */ |
453 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
454 | { | |
455 | if (!memcg) | |
456 | memcg = root_mem_cgroup; | |
457 | return &memcg->vmpressure; | |
458 | } | |
459 | ||
460 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
461 | { | |
462 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
463 | } | |
464 | ||
7ffc0edc MH |
465 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
466 | { | |
467 | return (memcg == root_mem_cgroup); | |
468 | } | |
469 | ||
4219b2da LZ |
470 | /* |
471 | * We restrict the id in the range of [1, 65535], so it can fit into | |
472 | * an unsigned short. | |
473 | */ | |
474 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
475 | ||
34c00c31 LZ |
476 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
477 | { | |
15a4c835 | 478 | return memcg->css.id; |
34c00c31 LZ |
479 | } |
480 | ||
481 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
482 | { | |
483 | struct cgroup_subsys_state *css; | |
484 | ||
7d699ddb | 485 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
486 | return mem_cgroup_from_css(css); |
487 | } | |
488 | ||
e1aab161 | 489 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 490 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 491 | |
e1aab161 GC |
492 | void sock_update_memcg(struct sock *sk) |
493 | { | |
376be5ff | 494 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 495 | struct mem_cgroup *memcg; |
3f134619 | 496 | struct cg_proto *cg_proto; |
e1aab161 GC |
497 | |
498 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
499 | ||
f3f511e1 GC |
500 | /* Socket cloning can throw us here with sk_cgrp already |
501 | * filled. It won't however, necessarily happen from | |
502 | * process context. So the test for root memcg given | |
503 | * the current task's memcg won't help us in this case. | |
504 | * | |
505 | * Respecting the original socket's memcg is a better | |
506 | * decision in this case. | |
507 | */ | |
508 | if (sk->sk_cgrp) { | |
509 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
5347e5ae | 510 | css_get(&sk->sk_cgrp->memcg->css); |
f3f511e1 GC |
511 | return; |
512 | } | |
513 | ||
e1aab161 GC |
514 | rcu_read_lock(); |
515 | memcg = mem_cgroup_from_task(current); | |
3f134619 | 516 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
5347e5ae | 517 | if (!mem_cgroup_is_root(memcg) && |
ec903c0c TH |
518 | memcg_proto_active(cg_proto) && |
519 | css_tryget_online(&memcg->css)) { | |
3f134619 | 520 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
521 | } |
522 | rcu_read_unlock(); | |
523 | } | |
524 | } | |
525 | EXPORT_SYMBOL(sock_update_memcg); | |
526 | ||
527 | void sock_release_memcg(struct sock *sk) | |
528 | { | |
376be5ff | 529 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
530 | struct mem_cgroup *memcg; |
531 | WARN_ON(!sk->sk_cgrp->memcg); | |
532 | memcg = sk->sk_cgrp->memcg; | |
5347e5ae | 533 | css_put(&sk->sk_cgrp->memcg->css); |
e1aab161 GC |
534 | } |
535 | } | |
d1a4c0b3 GC |
536 | |
537 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
538 | { | |
539 | if (!memcg || mem_cgroup_is_root(memcg)) | |
540 | return NULL; | |
541 | ||
2e685cad | 542 | return &memcg->tcp_mem; |
d1a4c0b3 GC |
543 | } |
544 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 545 | |
3f134619 GC |
546 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
547 | { | |
2e685cad | 548 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
3f134619 GC |
549 | return; |
550 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
551 | } | |
552 | #else | |
553 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
554 | { | |
555 | } | |
556 | #endif | |
557 | ||
a8964b9b | 558 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
559 | /* |
560 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
b8627835 LZ |
561 | * The main reason for not using cgroup id for this: |
562 | * this works better in sparse environments, where we have a lot of memcgs, | |
563 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
564 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
565 | * 200 entry array for that. | |
55007d84 GC |
566 | * |
567 | * The current size of the caches array is stored in | |
568 | * memcg_limited_groups_array_size. It will double each time we have to | |
569 | * increase it. | |
570 | */ | |
571 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
572 | int memcg_limited_groups_array_size; |
573 | ||
55007d84 GC |
574 | /* |
575 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
576 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
577 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
578 | * tunable, but that is strictly not necessary. | |
579 | * | |
b8627835 | 580 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
581 | * this constant directly from cgroup, but it is understandable that this is |
582 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 583 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
584 | * increase ours as well if it increases. |
585 | */ | |
586 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 587 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 588 | |
d7f25f8a GC |
589 | /* |
590 | * A lot of the calls to the cache allocation functions are expected to be | |
591 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
592 | * conditional to this static branch, we'll have to allow modules that does | |
593 | * kmem_cache_alloc and the such to see this symbol as well | |
594 | */ | |
a8964b9b | 595 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 596 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 597 | |
f3bb3043 VD |
598 | static void memcg_free_cache_id(int id); |
599 | ||
a8964b9b GC |
600 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
601 | { | |
55007d84 | 602 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 603 | static_key_slow_dec(&memcg_kmem_enabled_key); |
f3bb3043 | 604 | memcg_free_cache_id(memcg->kmemcg_id); |
55007d84 | 605 | } |
bea207c8 GC |
606 | /* |
607 | * This check can't live in kmem destruction function, | |
608 | * since the charges will outlive the cgroup | |
609 | */ | |
3e32cb2e | 610 | WARN_ON(page_counter_read(&memcg->kmem)); |
a8964b9b GC |
611 | } |
612 | #else | |
613 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
614 | { | |
615 | } | |
616 | #endif /* CONFIG_MEMCG_KMEM */ | |
617 | ||
618 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
619 | { | |
620 | disarm_sock_keys(memcg); | |
621 | disarm_kmem_keys(memcg); | |
622 | } | |
623 | ||
f64c3f54 | 624 | static struct mem_cgroup_per_zone * |
e231875b | 625 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 626 | { |
e231875b JZ |
627 | int nid = zone_to_nid(zone); |
628 | int zid = zone_idx(zone); | |
629 | ||
54f72fe0 | 630 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
631 | } |
632 | ||
c0ff4b85 | 633 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 634 | { |
c0ff4b85 | 635 | return &memcg->css; |
d324236b WF |
636 | } |
637 | ||
f64c3f54 | 638 | static struct mem_cgroup_per_zone * |
e231875b | 639 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 640 | { |
97a6c37b JW |
641 | int nid = page_to_nid(page); |
642 | int zid = page_zonenum(page); | |
f64c3f54 | 643 | |
e231875b | 644 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
645 | } |
646 | ||
bb4cc1a8 AM |
647 | static struct mem_cgroup_tree_per_zone * |
648 | soft_limit_tree_node_zone(int nid, int zid) | |
649 | { | |
650 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
651 | } | |
652 | ||
653 | static struct mem_cgroup_tree_per_zone * | |
654 | soft_limit_tree_from_page(struct page *page) | |
655 | { | |
656 | int nid = page_to_nid(page); | |
657 | int zid = page_zonenum(page); | |
658 | ||
659 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
660 | } | |
661 | ||
cf2c8127 JW |
662 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
663 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 664 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
665 | { |
666 | struct rb_node **p = &mctz->rb_root.rb_node; | |
667 | struct rb_node *parent = NULL; | |
668 | struct mem_cgroup_per_zone *mz_node; | |
669 | ||
670 | if (mz->on_tree) | |
671 | return; | |
672 | ||
673 | mz->usage_in_excess = new_usage_in_excess; | |
674 | if (!mz->usage_in_excess) | |
675 | return; | |
676 | while (*p) { | |
677 | parent = *p; | |
678 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
679 | tree_node); | |
680 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
681 | p = &(*p)->rb_left; | |
682 | /* | |
683 | * We can't avoid mem cgroups that are over their soft | |
684 | * limit by the same amount | |
685 | */ | |
686 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
687 | p = &(*p)->rb_right; | |
688 | } | |
689 | rb_link_node(&mz->tree_node, parent, p); | |
690 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
691 | mz->on_tree = true; | |
692 | } | |
693 | ||
cf2c8127 JW |
694 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
695 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
696 | { |
697 | if (!mz->on_tree) | |
698 | return; | |
699 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
700 | mz->on_tree = false; | |
701 | } | |
702 | ||
cf2c8127 JW |
703 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
704 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 705 | { |
0a31bc97 JW |
706 | unsigned long flags; |
707 | ||
708 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 709 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 710 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
711 | } |
712 | ||
3e32cb2e JW |
713 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
714 | { | |
715 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
716 | unsigned long soft_limit = ACCESS_ONCE(memcg->soft_limit); | |
717 | unsigned long excess = 0; | |
718 | ||
719 | if (nr_pages > soft_limit) | |
720 | excess = nr_pages - soft_limit; | |
721 | ||
722 | return excess; | |
723 | } | |
bb4cc1a8 AM |
724 | |
725 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
726 | { | |
3e32cb2e | 727 | unsigned long excess; |
bb4cc1a8 AM |
728 | struct mem_cgroup_per_zone *mz; |
729 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 730 | |
e231875b | 731 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
732 | /* |
733 | * Necessary to update all ancestors when hierarchy is used. | |
734 | * because their event counter is not touched. | |
735 | */ | |
736 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 737 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 738 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
739 | /* |
740 | * We have to update the tree if mz is on RB-tree or | |
741 | * mem is over its softlimit. | |
742 | */ | |
743 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
744 | unsigned long flags; |
745 | ||
746 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
747 | /* if on-tree, remove it */ |
748 | if (mz->on_tree) | |
cf2c8127 | 749 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
750 | /* |
751 | * Insert again. mz->usage_in_excess will be updated. | |
752 | * If excess is 0, no tree ops. | |
753 | */ | |
cf2c8127 | 754 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 755 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
756 | } |
757 | } | |
758 | } | |
759 | ||
760 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
761 | { | |
bb4cc1a8 | 762 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
763 | struct mem_cgroup_per_zone *mz; |
764 | int nid, zid; | |
bb4cc1a8 | 765 | |
e231875b JZ |
766 | for_each_node(nid) { |
767 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
768 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
769 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 770 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
771 | } |
772 | } | |
773 | } | |
774 | ||
775 | static struct mem_cgroup_per_zone * | |
776 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
777 | { | |
778 | struct rb_node *rightmost = NULL; | |
779 | struct mem_cgroup_per_zone *mz; | |
780 | ||
781 | retry: | |
782 | mz = NULL; | |
783 | rightmost = rb_last(&mctz->rb_root); | |
784 | if (!rightmost) | |
785 | goto done; /* Nothing to reclaim from */ | |
786 | ||
787 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
788 | /* | |
789 | * Remove the node now but someone else can add it back, | |
790 | * we will to add it back at the end of reclaim to its correct | |
791 | * position in the tree. | |
792 | */ | |
cf2c8127 | 793 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 794 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 795 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
796 | goto retry; |
797 | done: | |
798 | return mz; | |
799 | } | |
800 | ||
801 | static struct mem_cgroup_per_zone * | |
802 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
803 | { | |
804 | struct mem_cgroup_per_zone *mz; | |
805 | ||
0a31bc97 | 806 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 807 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 808 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
809 | return mz; |
810 | } | |
811 | ||
711d3d2c KH |
812 | /* |
813 | * Implementation Note: reading percpu statistics for memcg. | |
814 | * | |
815 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
816 | * synchronization to implement "quick" read. There are trade-off between | |
817 | * reading cost and precision of value. Then, we may have a chance to implement | |
818 | * a periodic synchronizion of counter in memcg's counter. | |
819 | * | |
820 | * But this _read() function is used for user interface now. The user accounts | |
821 | * memory usage by memory cgroup and he _always_ requires exact value because | |
822 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
823 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
824 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
825 | * | |
826 | * If there are kernel internal actions which can make use of some not-exact | |
827 | * value, and reading all cpu value can be performance bottleneck in some | |
828 | * common workload, threashold and synchonization as vmstat[] should be | |
829 | * implemented. | |
830 | */ | |
c0ff4b85 | 831 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 832 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 833 | { |
7a159cc9 | 834 | long val = 0; |
c62b1a3b | 835 | int cpu; |
c62b1a3b | 836 | |
711d3d2c KH |
837 | get_online_cpus(); |
838 | for_each_online_cpu(cpu) | |
c0ff4b85 | 839 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 840 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
841 | spin_lock(&memcg->pcp_counter_lock); |
842 | val += memcg->nocpu_base.count[idx]; | |
843 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
844 | #endif |
845 | put_online_cpus(); | |
c62b1a3b KH |
846 | return val; |
847 | } | |
848 | ||
c0ff4b85 | 849 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
850 | enum mem_cgroup_events_index idx) |
851 | { | |
852 | unsigned long val = 0; | |
853 | int cpu; | |
854 | ||
9c567512 | 855 | get_online_cpus(); |
e9f8974f | 856 | for_each_online_cpu(cpu) |
c0ff4b85 | 857 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 858 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
859 | spin_lock(&memcg->pcp_counter_lock); |
860 | val += memcg->nocpu_base.events[idx]; | |
861 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f | 862 | #endif |
9c567512 | 863 | put_online_cpus(); |
e9f8974f JW |
864 | return val; |
865 | } | |
866 | ||
c0ff4b85 | 867 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 868 | struct page *page, |
0a31bc97 | 869 | int nr_pages) |
d52aa412 | 870 | { |
b2402857 KH |
871 | /* |
872 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
873 | * counted as CACHE even if it's on ANON LRU. | |
874 | */ | |
0a31bc97 | 875 | if (PageAnon(page)) |
b2402857 | 876 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 877 | nr_pages); |
d52aa412 | 878 | else |
b2402857 | 879 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 880 | nr_pages); |
55e462b0 | 881 | |
b070e65c DR |
882 | if (PageTransHuge(page)) |
883 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], | |
884 | nr_pages); | |
885 | ||
e401f176 KH |
886 | /* pagein of a big page is an event. So, ignore page size */ |
887 | if (nr_pages > 0) | |
c0ff4b85 | 888 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 889 | else { |
c0ff4b85 | 890 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
891 | nr_pages = -nr_pages; /* for event */ |
892 | } | |
e401f176 | 893 | |
13114716 | 894 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
895 | } |
896 | ||
e231875b | 897 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
898 | { |
899 | struct mem_cgroup_per_zone *mz; | |
900 | ||
901 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
902 | return mz->lru_size[lru]; | |
903 | } | |
904 | ||
e231875b JZ |
905 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
906 | int nid, | |
907 | unsigned int lru_mask) | |
bb2a0de9 | 908 | { |
e231875b | 909 | unsigned long nr = 0; |
889976db YH |
910 | int zid; |
911 | ||
e231875b | 912 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 913 | |
e231875b JZ |
914 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
915 | struct mem_cgroup_per_zone *mz; | |
916 | enum lru_list lru; | |
917 | ||
918 | for_each_lru(lru) { | |
919 | if (!(BIT(lru) & lru_mask)) | |
920 | continue; | |
921 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
922 | nr += mz->lru_size[lru]; | |
923 | } | |
924 | } | |
925 | return nr; | |
889976db | 926 | } |
bb2a0de9 | 927 | |
c0ff4b85 | 928 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 929 | unsigned int lru_mask) |
6d12e2d8 | 930 | { |
e231875b | 931 | unsigned long nr = 0; |
889976db | 932 | int nid; |
6d12e2d8 | 933 | |
31aaea4a | 934 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
935 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
936 | return nr; | |
d52aa412 KH |
937 | } |
938 | ||
f53d7ce3 JW |
939 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
940 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
941 | { |
942 | unsigned long val, next; | |
943 | ||
13114716 | 944 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 945 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 946 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
947 | if ((long)next - (long)val < 0) { |
948 | switch (target) { | |
949 | case MEM_CGROUP_TARGET_THRESH: | |
950 | next = val + THRESHOLDS_EVENTS_TARGET; | |
951 | break; | |
bb4cc1a8 AM |
952 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
953 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
954 | break; | |
f53d7ce3 JW |
955 | case MEM_CGROUP_TARGET_NUMAINFO: |
956 | next = val + NUMAINFO_EVENTS_TARGET; | |
957 | break; | |
958 | default: | |
959 | break; | |
960 | } | |
961 | __this_cpu_write(memcg->stat->targets[target], next); | |
962 | return true; | |
7a159cc9 | 963 | } |
f53d7ce3 | 964 | return false; |
d2265e6f KH |
965 | } |
966 | ||
967 | /* | |
968 | * Check events in order. | |
969 | * | |
970 | */ | |
c0ff4b85 | 971 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
972 | { |
973 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
974 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
975 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 976 | bool do_softlimit; |
82b3f2a7 | 977 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 978 | |
bb4cc1a8 AM |
979 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
980 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
981 | #if MAX_NUMNODES > 1 |
982 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
983 | MEM_CGROUP_TARGET_NUMAINFO); | |
984 | #endif | |
c0ff4b85 | 985 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
986 | if (unlikely(do_softlimit)) |
987 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 988 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 989 | if (unlikely(do_numainfo)) |
c0ff4b85 | 990 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 991 | #endif |
0a31bc97 | 992 | } |
d2265e6f KH |
993 | } |
994 | ||
cf475ad2 | 995 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 996 | { |
31a78f23 BS |
997 | /* |
998 | * mm_update_next_owner() may clear mm->owner to NULL | |
999 | * if it races with swapoff, page migration, etc. | |
1000 | * So this can be called with p == NULL. | |
1001 | */ | |
1002 | if (unlikely(!p)) | |
1003 | return NULL; | |
1004 | ||
073219e9 | 1005 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 PE |
1006 | } |
1007 | ||
df381975 | 1008 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1009 | { |
c0ff4b85 | 1010 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 1011 | |
54595fe2 KH |
1012 | rcu_read_lock(); |
1013 | do { | |
6f6acb00 MH |
1014 | /* |
1015 | * Page cache insertions can happen withou an | |
1016 | * actual mm context, e.g. during disk probing | |
1017 | * on boot, loopback IO, acct() writes etc. | |
1018 | */ | |
1019 | if (unlikely(!mm)) | |
df381975 | 1020 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1021 | else { |
1022 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1023 | if (unlikely(!memcg)) | |
1024 | memcg = root_mem_cgroup; | |
1025 | } | |
ec903c0c | 1026 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 1027 | rcu_read_unlock(); |
c0ff4b85 | 1028 | return memcg; |
54595fe2 KH |
1029 | } |
1030 | ||
5660048c JW |
1031 | /** |
1032 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1033 | * @root: hierarchy root | |
1034 | * @prev: previously returned memcg, NULL on first invocation | |
1035 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1036 | * | |
1037 | * Returns references to children of the hierarchy below @root, or | |
1038 | * @root itself, or %NULL after a full round-trip. | |
1039 | * | |
1040 | * Caller must pass the return value in @prev on subsequent | |
1041 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1042 | * to cancel a hierarchy walk before the round-trip is complete. | |
1043 | * | |
1044 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1045 | * divide up the memcgs in the hierarchy among all concurrent | |
1046 | * reclaimers operating on the same zone and priority. | |
1047 | */ | |
694fbc0f | 1048 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1049 | struct mem_cgroup *prev, |
694fbc0f | 1050 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1051 | { |
5ac8fb31 JW |
1052 | struct reclaim_iter *uninitialized_var(iter); |
1053 | struct cgroup_subsys_state *css = NULL; | |
9f3a0d09 | 1054 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1055 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1056 | |
694fbc0f AM |
1057 | if (mem_cgroup_disabled()) |
1058 | return NULL; | |
5660048c | 1059 | |
9f3a0d09 JW |
1060 | if (!root) |
1061 | root = root_mem_cgroup; | |
7d74b06f | 1062 | |
9f3a0d09 | 1063 | if (prev && !reclaim) |
5ac8fb31 | 1064 | pos = prev; |
14067bb3 | 1065 | |
9f3a0d09 JW |
1066 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1067 | if (prev) | |
5ac8fb31 | 1068 | goto out; |
694fbc0f | 1069 | return root; |
9f3a0d09 | 1070 | } |
14067bb3 | 1071 | |
542f85f9 | 1072 | rcu_read_lock(); |
5f578161 | 1073 | |
5ac8fb31 JW |
1074 | if (reclaim) { |
1075 | struct mem_cgroup_per_zone *mz; | |
1076 | ||
1077 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
1078 | iter = &mz->iter[reclaim->priority]; | |
1079 | ||
1080 | if (prev && reclaim->generation != iter->generation) | |
1081 | goto out_unlock; | |
1082 | ||
1083 | do { | |
1084 | pos = ACCESS_ONCE(iter->position); | |
1085 | /* | |
1086 | * A racing update may change the position and | |
1087 | * put the last reference, hence css_tryget(), | |
1088 | * or retry to see the updated position. | |
1089 | */ | |
1090 | } while (pos && !css_tryget(&pos->css)); | |
1091 | } | |
1092 | ||
1093 | if (pos) | |
1094 | css = &pos->css; | |
1095 | ||
1096 | for (;;) { | |
1097 | css = css_next_descendant_pre(css, &root->css); | |
1098 | if (!css) { | |
1099 | /* | |
1100 | * Reclaimers share the hierarchy walk, and a | |
1101 | * new one might jump in right at the end of | |
1102 | * the hierarchy - make sure they see at least | |
1103 | * one group and restart from the beginning. | |
1104 | */ | |
1105 | if (!prev) | |
1106 | continue; | |
1107 | break; | |
527a5ec9 | 1108 | } |
7d74b06f | 1109 | |
5ac8fb31 JW |
1110 | /* |
1111 | * Verify the css and acquire a reference. The root | |
1112 | * is provided by the caller, so we know it's alive | |
1113 | * and kicking, and don't take an extra reference. | |
1114 | */ | |
1115 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1116 | |
5ac8fb31 JW |
1117 | if (css == &root->css) |
1118 | break; | |
14067bb3 | 1119 | |
b2052564 | 1120 | if (css_tryget(css)) { |
5ac8fb31 JW |
1121 | /* |
1122 | * Make sure the memcg is initialized: | |
1123 | * mem_cgroup_css_online() orders the the | |
1124 | * initialization against setting the flag. | |
1125 | */ | |
1126 | if (smp_load_acquire(&memcg->initialized)) | |
1127 | break; | |
542f85f9 | 1128 | |
5ac8fb31 | 1129 | css_put(css); |
527a5ec9 | 1130 | } |
9f3a0d09 | 1131 | |
5ac8fb31 | 1132 | memcg = NULL; |
9f3a0d09 | 1133 | } |
5ac8fb31 JW |
1134 | |
1135 | if (reclaim) { | |
1136 | if (cmpxchg(&iter->position, pos, memcg) == pos) { | |
1137 | if (memcg) | |
1138 | css_get(&memcg->css); | |
1139 | if (pos) | |
1140 | css_put(&pos->css); | |
1141 | } | |
1142 | ||
1143 | /* | |
1144 | * pairs with css_tryget when dereferencing iter->position | |
1145 | * above. | |
1146 | */ | |
1147 | if (pos) | |
1148 | css_put(&pos->css); | |
1149 | ||
1150 | if (!memcg) | |
1151 | iter->generation++; | |
1152 | else if (!prev) | |
1153 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1154 | } |
5ac8fb31 | 1155 | |
542f85f9 MH |
1156 | out_unlock: |
1157 | rcu_read_unlock(); | |
5ac8fb31 | 1158 | out: |
c40046f3 MH |
1159 | if (prev && prev != root) |
1160 | css_put(&prev->css); | |
1161 | ||
9f3a0d09 | 1162 | return memcg; |
14067bb3 | 1163 | } |
7d74b06f | 1164 | |
5660048c JW |
1165 | /** |
1166 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1167 | * @root: hierarchy root | |
1168 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1169 | */ | |
1170 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1171 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1172 | { |
1173 | if (!root) | |
1174 | root = root_mem_cgroup; | |
1175 | if (prev && prev != root) | |
1176 | css_put(&prev->css); | |
1177 | } | |
7d74b06f | 1178 | |
9f3a0d09 JW |
1179 | /* |
1180 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1181 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1182 | * be used for reference counting. | |
1183 | */ | |
1184 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1185 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1186 | iter != NULL; \ |
527a5ec9 | 1187 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1188 | |
9f3a0d09 | 1189 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1190 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1191 | iter != NULL; \ |
527a5ec9 | 1192 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1193 | |
68ae564b | 1194 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1195 | { |
c0ff4b85 | 1196 | struct mem_cgroup *memcg; |
456f998e | 1197 | |
456f998e | 1198 | rcu_read_lock(); |
c0ff4b85 R |
1199 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1200 | if (unlikely(!memcg)) | |
456f998e YH |
1201 | goto out; |
1202 | ||
1203 | switch (idx) { | |
456f998e | 1204 | case PGFAULT: |
0e574a93 JW |
1205 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1206 | break; | |
1207 | case PGMAJFAULT: | |
1208 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1209 | break; |
1210 | default: | |
1211 | BUG(); | |
1212 | } | |
1213 | out: | |
1214 | rcu_read_unlock(); | |
1215 | } | |
68ae564b | 1216 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1217 | |
925b7673 JW |
1218 | /** |
1219 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1220 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1221 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1222 | * |
1223 | * Returns the lru list vector holding pages for the given @zone and | |
1224 | * @mem. This can be the global zone lruvec, if the memory controller | |
1225 | * is disabled. | |
1226 | */ | |
1227 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1228 | struct mem_cgroup *memcg) | |
1229 | { | |
1230 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1231 | struct lruvec *lruvec; |
925b7673 | 1232 | |
bea8c150 HD |
1233 | if (mem_cgroup_disabled()) { |
1234 | lruvec = &zone->lruvec; | |
1235 | goto out; | |
1236 | } | |
925b7673 | 1237 | |
e231875b | 1238 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1239 | lruvec = &mz->lruvec; |
1240 | out: | |
1241 | /* | |
1242 | * Since a node can be onlined after the mem_cgroup was created, | |
1243 | * we have to be prepared to initialize lruvec->zone here; | |
1244 | * and if offlined then reonlined, we need to reinitialize it. | |
1245 | */ | |
1246 | if (unlikely(lruvec->zone != zone)) | |
1247 | lruvec->zone = zone; | |
1248 | return lruvec; | |
925b7673 JW |
1249 | } |
1250 | ||
925b7673 | 1251 | /** |
dfe0e773 | 1252 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1253 | * @page: the page |
fa9add64 | 1254 | * @zone: zone of the page |
dfe0e773 JW |
1255 | * |
1256 | * This function is only safe when following the LRU page isolation | |
1257 | * and putback protocol: the LRU lock must be held, and the page must | |
1258 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1259 | */ |
fa9add64 | 1260 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1261 | { |
08e552c6 | 1262 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1263 | struct mem_cgroup *memcg; |
bea8c150 | 1264 | struct lruvec *lruvec; |
6d12e2d8 | 1265 | |
bea8c150 HD |
1266 | if (mem_cgroup_disabled()) { |
1267 | lruvec = &zone->lruvec; | |
1268 | goto out; | |
1269 | } | |
925b7673 | 1270 | |
1306a85a | 1271 | memcg = page->mem_cgroup; |
7512102c | 1272 | /* |
dfe0e773 | 1273 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1274 | * possibly migrated - before they are charged. |
7512102c | 1275 | */ |
29833315 JW |
1276 | if (!memcg) |
1277 | memcg = root_mem_cgroup; | |
7512102c | 1278 | |
e231875b | 1279 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1280 | lruvec = &mz->lruvec; |
1281 | out: | |
1282 | /* | |
1283 | * Since a node can be onlined after the mem_cgroup was created, | |
1284 | * we have to be prepared to initialize lruvec->zone here; | |
1285 | * and if offlined then reonlined, we need to reinitialize it. | |
1286 | */ | |
1287 | if (unlikely(lruvec->zone != zone)) | |
1288 | lruvec->zone = zone; | |
1289 | return lruvec; | |
08e552c6 | 1290 | } |
b69408e8 | 1291 | |
925b7673 | 1292 | /** |
fa9add64 HD |
1293 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1294 | * @lruvec: mem_cgroup per zone lru vector | |
1295 | * @lru: index of lru list the page is sitting on | |
1296 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1297 | * |
fa9add64 HD |
1298 | * This function must be called when a page is added to or removed from an |
1299 | * lru list. | |
3f58a829 | 1300 | */ |
fa9add64 HD |
1301 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1302 | int nr_pages) | |
3f58a829 MK |
1303 | { |
1304 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1305 | unsigned long *lru_size; |
3f58a829 MK |
1306 | |
1307 | if (mem_cgroup_disabled()) | |
1308 | return; | |
1309 | ||
fa9add64 HD |
1310 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1311 | lru_size = mz->lru_size + lru; | |
1312 | *lru_size += nr_pages; | |
1313 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1314 | } |
544122e5 | 1315 | |
2314b42d | 1316 | bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root) |
3e92041d | 1317 | { |
2314b42d | 1318 | if (root == memcg) |
91c63734 | 1319 | return true; |
2314b42d | 1320 | if (!root->use_hierarchy) |
91c63734 | 1321 | return false; |
2314b42d | 1322 | return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); |
c3ac9a8a JW |
1323 | } |
1324 | ||
2314b42d | 1325 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1326 | { |
2314b42d | 1327 | struct mem_cgroup *task_memcg; |
158e0a2d | 1328 | struct task_struct *p; |
ffbdccf5 | 1329 | bool ret; |
4c4a2214 | 1330 | |
158e0a2d | 1331 | p = find_lock_task_mm(task); |
de077d22 | 1332 | if (p) { |
2314b42d | 1333 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1334 | task_unlock(p); |
1335 | } else { | |
1336 | /* | |
1337 | * All threads may have already detached their mm's, but the oom | |
1338 | * killer still needs to detect if they have already been oom | |
1339 | * killed to prevent needlessly killing additional tasks. | |
1340 | */ | |
ffbdccf5 | 1341 | rcu_read_lock(); |
2314b42d JW |
1342 | task_memcg = mem_cgroup_from_task(task); |
1343 | css_get(&task_memcg->css); | |
ffbdccf5 | 1344 | rcu_read_unlock(); |
de077d22 | 1345 | } |
2314b42d JW |
1346 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1347 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1348 | return ret; |
1349 | } | |
1350 | ||
c56d5c7d | 1351 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1352 | { |
9b272977 | 1353 | unsigned long inactive_ratio; |
14797e23 | 1354 | unsigned long inactive; |
9b272977 | 1355 | unsigned long active; |
c772be93 | 1356 | unsigned long gb; |
14797e23 | 1357 | |
4d7dcca2 HD |
1358 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1359 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1360 | |
c772be93 KM |
1361 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1362 | if (gb) | |
1363 | inactive_ratio = int_sqrt(10 * gb); | |
1364 | else | |
1365 | inactive_ratio = 1; | |
1366 | ||
9b272977 | 1367 | return inactive * inactive_ratio < active; |
14797e23 KM |
1368 | } |
1369 | ||
3e32cb2e | 1370 | #define mem_cgroup_from_counter(counter, member) \ |
6d61ef40 BS |
1371 | container_of(counter, struct mem_cgroup, member) |
1372 | ||
19942822 | 1373 | /** |
9d11ea9f | 1374 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1375 | * @memcg: the memory cgroup |
19942822 | 1376 | * |
9d11ea9f | 1377 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1378 | * pages. |
19942822 | 1379 | */ |
c0ff4b85 | 1380 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1381 | { |
3e32cb2e JW |
1382 | unsigned long margin = 0; |
1383 | unsigned long count; | |
1384 | unsigned long limit; | |
9d11ea9f | 1385 | |
3e32cb2e JW |
1386 | count = page_counter_read(&memcg->memory); |
1387 | limit = ACCESS_ONCE(memcg->memory.limit); | |
1388 | if (count < limit) | |
1389 | margin = limit - count; | |
1390 | ||
1391 | if (do_swap_account) { | |
1392 | count = page_counter_read(&memcg->memsw); | |
1393 | limit = ACCESS_ONCE(memcg->memsw.limit); | |
1394 | if (count <= limit) | |
1395 | margin = min(margin, limit - count); | |
1396 | } | |
1397 | ||
1398 | return margin; | |
19942822 JW |
1399 | } |
1400 | ||
1f4c025b | 1401 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 | 1402 | { |
a7885eb8 | 1403 | /* root ? */ |
14208b0e | 1404 | if (mem_cgroup_disabled() || !memcg->css.parent) |
a7885eb8 KM |
1405 | return vm_swappiness; |
1406 | ||
bf1ff263 | 1407 | return memcg->swappiness; |
a7885eb8 KM |
1408 | } |
1409 | ||
32047e2a | 1410 | /* |
bdcbb659 | 1411 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1412 | * |
bdcbb659 QH |
1413 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1414 | * moving cgroups. This is for waiting at high-memory pressure | |
1415 | * caused by "move". | |
32047e2a | 1416 | */ |
c0ff4b85 | 1417 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1418 | { |
2bd9bb20 KH |
1419 | struct mem_cgroup *from; |
1420 | struct mem_cgroup *to; | |
4b534334 | 1421 | bool ret = false; |
2bd9bb20 KH |
1422 | /* |
1423 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1424 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1425 | */ | |
1426 | spin_lock(&mc.lock); | |
1427 | from = mc.from; | |
1428 | to = mc.to; | |
1429 | if (!from) | |
1430 | goto unlock; | |
3e92041d | 1431 | |
2314b42d JW |
1432 | ret = mem_cgroup_is_descendant(from, memcg) || |
1433 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1434 | unlock: |
1435 | spin_unlock(&mc.lock); | |
4b534334 KH |
1436 | return ret; |
1437 | } | |
1438 | ||
c0ff4b85 | 1439 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1440 | { |
1441 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1442 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1443 | DEFINE_WAIT(wait); |
1444 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1445 | /* moving charge context might have finished. */ | |
1446 | if (mc.moving_task) | |
1447 | schedule(); | |
1448 | finish_wait(&mc.waitq, &wait); | |
1449 | return true; | |
1450 | } | |
1451 | } | |
1452 | return false; | |
1453 | } | |
1454 | ||
58cf188e | 1455 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1456 | /** |
58cf188e | 1457 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1458 | * @memcg: The memory cgroup that went over limit |
1459 | * @p: Task that is going to be killed | |
1460 | * | |
1461 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1462 | * enabled | |
1463 | */ | |
1464 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1465 | { | |
e61734c5 | 1466 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1467 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1468 | struct mem_cgroup *iter; |
1469 | unsigned int i; | |
e222432b | 1470 | |
58cf188e | 1471 | if (!p) |
e222432b BS |
1472 | return; |
1473 | ||
08088cb9 | 1474 | mutex_lock(&oom_info_lock); |
e222432b BS |
1475 | rcu_read_lock(); |
1476 | ||
e61734c5 TH |
1477 | pr_info("Task in "); |
1478 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
0346dadb | 1479 | pr_cont(" killed as a result of limit of "); |
e61734c5 | 1480 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1481 | pr_cont("\n"); |
e222432b | 1482 | |
e222432b BS |
1483 | rcu_read_unlock(); |
1484 | ||
3e32cb2e JW |
1485 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1486 | K((u64)page_counter_read(&memcg->memory)), | |
1487 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1488 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1489 | K((u64)page_counter_read(&memcg->memsw)), | |
1490 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1491 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1492 | K((u64)page_counter_read(&memcg->kmem)), | |
1493 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1494 | |
1495 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1496 | pr_info("Memory cgroup stats for "); |
1497 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1498 | pr_cont(":"); |
1499 | ||
1500 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1501 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1502 | continue; | |
1503 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1504 | K(mem_cgroup_read_stat(iter, i))); | |
1505 | } | |
1506 | ||
1507 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1508 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1509 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1510 | ||
1511 | pr_cont("\n"); | |
1512 | } | |
08088cb9 | 1513 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1514 | } |
1515 | ||
81d39c20 KH |
1516 | /* |
1517 | * This function returns the number of memcg under hierarchy tree. Returns | |
1518 | * 1(self count) if no children. | |
1519 | */ | |
c0ff4b85 | 1520 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1521 | { |
1522 | int num = 0; | |
7d74b06f KH |
1523 | struct mem_cgroup *iter; |
1524 | ||
c0ff4b85 | 1525 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1526 | num++; |
81d39c20 KH |
1527 | return num; |
1528 | } | |
1529 | ||
a63d83f4 DR |
1530 | /* |
1531 | * Return the memory (and swap, if configured) limit for a memcg. | |
1532 | */ | |
3e32cb2e | 1533 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1534 | { |
3e32cb2e | 1535 | unsigned long limit; |
f3e8eb70 | 1536 | |
3e32cb2e | 1537 | limit = memcg->memory.limit; |
9a5a8f19 | 1538 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1539 | unsigned long memsw_limit; |
9a5a8f19 | 1540 | |
3e32cb2e JW |
1541 | memsw_limit = memcg->memsw.limit; |
1542 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1543 | } |
9a5a8f19 | 1544 | return limit; |
a63d83f4 DR |
1545 | } |
1546 | ||
19965460 DR |
1547 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1548 | int order) | |
9cbb78bb DR |
1549 | { |
1550 | struct mem_cgroup *iter; | |
1551 | unsigned long chosen_points = 0; | |
1552 | unsigned long totalpages; | |
1553 | unsigned int points = 0; | |
1554 | struct task_struct *chosen = NULL; | |
1555 | ||
876aafbf | 1556 | /* |
465adcf1 DR |
1557 | * If current has a pending SIGKILL or is exiting, then automatically |
1558 | * select it. The goal is to allow it to allocate so that it may | |
1559 | * quickly exit and free its memory. | |
876aafbf | 1560 | */ |
d003f371 | 1561 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
876aafbf DR |
1562 | set_thread_flag(TIF_MEMDIE); |
1563 | return; | |
1564 | } | |
1565 | ||
1566 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
3e32cb2e | 1567 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1568 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1569 | struct css_task_iter it; |
9cbb78bb DR |
1570 | struct task_struct *task; |
1571 | ||
72ec7029 TH |
1572 | css_task_iter_start(&iter->css, &it); |
1573 | while ((task = css_task_iter_next(&it))) { | |
9cbb78bb DR |
1574 | switch (oom_scan_process_thread(task, totalpages, NULL, |
1575 | false)) { | |
1576 | case OOM_SCAN_SELECT: | |
1577 | if (chosen) | |
1578 | put_task_struct(chosen); | |
1579 | chosen = task; | |
1580 | chosen_points = ULONG_MAX; | |
1581 | get_task_struct(chosen); | |
1582 | /* fall through */ | |
1583 | case OOM_SCAN_CONTINUE: | |
1584 | continue; | |
1585 | case OOM_SCAN_ABORT: | |
72ec7029 | 1586 | css_task_iter_end(&it); |
9cbb78bb DR |
1587 | mem_cgroup_iter_break(memcg, iter); |
1588 | if (chosen) | |
1589 | put_task_struct(chosen); | |
1590 | return; | |
1591 | case OOM_SCAN_OK: | |
1592 | break; | |
1593 | }; | |
1594 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1595 | if (!points || points < chosen_points) |
1596 | continue; | |
1597 | /* Prefer thread group leaders for display purposes */ | |
1598 | if (points == chosen_points && | |
1599 | thread_group_leader(chosen)) | |
1600 | continue; | |
1601 | ||
1602 | if (chosen) | |
1603 | put_task_struct(chosen); | |
1604 | chosen = task; | |
1605 | chosen_points = points; | |
1606 | get_task_struct(chosen); | |
9cbb78bb | 1607 | } |
72ec7029 | 1608 | css_task_iter_end(&it); |
9cbb78bb DR |
1609 | } |
1610 | ||
1611 | if (!chosen) | |
1612 | return; | |
1613 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1614 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1615 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1616 | } |
1617 | ||
ae6e71d3 MC |
1618 | #if MAX_NUMNODES > 1 |
1619 | ||
4d0c066d KH |
1620 | /** |
1621 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1622 | * @memcg: the target memcg |
4d0c066d KH |
1623 | * @nid: the node ID to be checked. |
1624 | * @noswap : specify true here if the user wants flle only information. | |
1625 | * | |
1626 | * This function returns whether the specified memcg contains any | |
1627 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1628 | * pages in the node. | |
1629 | */ | |
c0ff4b85 | 1630 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1631 | int nid, bool noswap) |
1632 | { | |
c0ff4b85 | 1633 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1634 | return true; |
1635 | if (noswap || !total_swap_pages) | |
1636 | return false; | |
c0ff4b85 | 1637 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1638 | return true; |
1639 | return false; | |
1640 | ||
1641 | } | |
889976db YH |
1642 | |
1643 | /* | |
1644 | * Always updating the nodemask is not very good - even if we have an empty | |
1645 | * list or the wrong list here, we can start from some node and traverse all | |
1646 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1647 | * | |
1648 | */ | |
c0ff4b85 | 1649 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1650 | { |
1651 | int nid; | |
453a9bf3 KH |
1652 | /* |
1653 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1654 | * pagein/pageout changes since the last update. | |
1655 | */ | |
c0ff4b85 | 1656 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1657 | return; |
c0ff4b85 | 1658 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1659 | return; |
1660 | ||
889976db | 1661 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1662 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1663 | |
31aaea4a | 1664 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1665 | |
c0ff4b85 R |
1666 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1667 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1668 | } |
453a9bf3 | 1669 | |
c0ff4b85 R |
1670 | atomic_set(&memcg->numainfo_events, 0); |
1671 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1672 | } |
1673 | ||
1674 | /* | |
1675 | * Selecting a node where we start reclaim from. Because what we need is just | |
1676 | * reducing usage counter, start from anywhere is O,K. Considering | |
1677 | * memory reclaim from current node, there are pros. and cons. | |
1678 | * | |
1679 | * Freeing memory from current node means freeing memory from a node which | |
1680 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1681 | * hit limits, it will see a contention on a node. But freeing from remote | |
1682 | * node means more costs for memory reclaim because of memory latency. | |
1683 | * | |
1684 | * Now, we use round-robin. Better algorithm is welcomed. | |
1685 | */ | |
c0ff4b85 | 1686 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1687 | { |
1688 | int node; | |
1689 | ||
c0ff4b85 R |
1690 | mem_cgroup_may_update_nodemask(memcg); |
1691 | node = memcg->last_scanned_node; | |
889976db | 1692 | |
c0ff4b85 | 1693 | node = next_node(node, memcg->scan_nodes); |
889976db | 1694 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1695 | node = first_node(memcg->scan_nodes); |
889976db YH |
1696 | /* |
1697 | * We call this when we hit limit, not when pages are added to LRU. | |
1698 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1699 | * memcg is too small and all pages are not on LRU. In that case, | |
1700 | * we use curret node. | |
1701 | */ | |
1702 | if (unlikely(node == MAX_NUMNODES)) | |
1703 | node = numa_node_id(); | |
1704 | ||
c0ff4b85 | 1705 | memcg->last_scanned_node = node; |
889976db YH |
1706 | return node; |
1707 | } | |
889976db | 1708 | #else |
c0ff4b85 | 1709 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1710 | { |
1711 | return 0; | |
1712 | } | |
1713 | #endif | |
1714 | ||
0608f43d AM |
1715 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1716 | struct zone *zone, | |
1717 | gfp_t gfp_mask, | |
1718 | unsigned long *total_scanned) | |
1719 | { | |
1720 | struct mem_cgroup *victim = NULL; | |
1721 | int total = 0; | |
1722 | int loop = 0; | |
1723 | unsigned long excess; | |
1724 | unsigned long nr_scanned; | |
1725 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1726 | .zone = zone, | |
1727 | .priority = 0, | |
1728 | }; | |
1729 | ||
3e32cb2e | 1730 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1731 | |
1732 | while (1) { | |
1733 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1734 | if (!victim) { | |
1735 | loop++; | |
1736 | if (loop >= 2) { | |
1737 | /* | |
1738 | * If we have not been able to reclaim | |
1739 | * anything, it might because there are | |
1740 | * no reclaimable pages under this hierarchy | |
1741 | */ | |
1742 | if (!total) | |
1743 | break; | |
1744 | /* | |
1745 | * We want to do more targeted reclaim. | |
1746 | * excess >> 2 is not to excessive so as to | |
1747 | * reclaim too much, nor too less that we keep | |
1748 | * coming back to reclaim from this cgroup | |
1749 | */ | |
1750 | if (total >= (excess >> 2) || | |
1751 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1752 | break; | |
1753 | } | |
1754 | continue; | |
1755 | } | |
0608f43d AM |
1756 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1757 | zone, &nr_scanned); | |
1758 | *total_scanned += nr_scanned; | |
3e32cb2e | 1759 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1760 | break; |
6d61ef40 | 1761 | } |
0608f43d AM |
1762 | mem_cgroup_iter_break(root_memcg, victim); |
1763 | return total; | |
6d61ef40 BS |
1764 | } |
1765 | ||
0056f4e6 JW |
1766 | #ifdef CONFIG_LOCKDEP |
1767 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1768 | .name = "memcg_oom_lock", | |
1769 | }; | |
1770 | #endif | |
1771 | ||
fb2a6fc5 JW |
1772 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1773 | ||
867578cb KH |
1774 | /* |
1775 | * Check OOM-Killer is already running under our hierarchy. | |
1776 | * If someone is running, return false. | |
1777 | */ | |
fb2a6fc5 | 1778 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1779 | { |
79dfdacc | 1780 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1781 | |
fb2a6fc5 JW |
1782 | spin_lock(&memcg_oom_lock); |
1783 | ||
9f3a0d09 | 1784 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1785 | if (iter->oom_lock) { |
79dfdacc MH |
1786 | /* |
1787 | * this subtree of our hierarchy is already locked | |
1788 | * so we cannot give a lock. | |
1789 | */ | |
79dfdacc | 1790 | failed = iter; |
9f3a0d09 JW |
1791 | mem_cgroup_iter_break(memcg, iter); |
1792 | break; | |
23751be0 JW |
1793 | } else |
1794 | iter->oom_lock = true; | |
7d74b06f | 1795 | } |
867578cb | 1796 | |
fb2a6fc5 JW |
1797 | if (failed) { |
1798 | /* | |
1799 | * OK, we failed to lock the whole subtree so we have | |
1800 | * to clean up what we set up to the failing subtree | |
1801 | */ | |
1802 | for_each_mem_cgroup_tree(iter, memcg) { | |
1803 | if (iter == failed) { | |
1804 | mem_cgroup_iter_break(memcg, iter); | |
1805 | break; | |
1806 | } | |
1807 | iter->oom_lock = false; | |
79dfdacc | 1808 | } |
0056f4e6 JW |
1809 | } else |
1810 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1811 | |
1812 | spin_unlock(&memcg_oom_lock); | |
1813 | ||
1814 | return !failed; | |
a636b327 | 1815 | } |
0b7f569e | 1816 | |
fb2a6fc5 | 1817 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1818 | { |
7d74b06f KH |
1819 | struct mem_cgroup *iter; |
1820 | ||
fb2a6fc5 | 1821 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1822 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1823 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1824 | iter->oom_lock = false; |
fb2a6fc5 | 1825 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1826 | } |
1827 | ||
c0ff4b85 | 1828 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1829 | { |
1830 | struct mem_cgroup *iter; | |
1831 | ||
c0ff4b85 | 1832 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1833 | atomic_inc(&iter->under_oom); |
1834 | } | |
1835 | ||
c0ff4b85 | 1836 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1837 | { |
1838 | struct mem_cgroup *iter; | |
1839 | ||
867578cb KH |
1840 | /* |
1841 | * When a new child is created while the hierarchy is under oom, | |
1842 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1843 | * atomic_add_unless() here. | |
1844 | */ | |
c0ff4b85 | 1845 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1846 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1847 | } |
1848 | ||
867578cb KH |
1849 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1850 | ||
dc98df5a | 1851 | struct oom_wait_info { |
d79154bb | 1852 | struct mem_cgroup *memcg; |
dc98df5a KH |
1853 | wait_queue_t wait; |
1854 | }; | |
1855 | ||
1856 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1857 | unsigned mode, int sync, void *arg) | |
1858 | { | |
d79154bb HD |
1859 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1860 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1861 | struct oom_wait_info *oom_wait_info; |
1862 | ||
1863 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1864 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1865 | |
2314b42d JW |
1866 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1867 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1868 | return 0; |
dc98df5a KH |
1869 | return autoremove_wake_function(wait, mode, sync, arg); |
1870 | } | |
1871 | ||
c0ff4b85 | 1872 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1873 | { |
3812c8c8 | 1874 | atomic_inc(&memcg->oom_wakeups); |
c0ff4b85 R |
1875 | /* for filtering, pass "memcg" as argument. */ |
1876 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1877 | } |
1878 | ||
c0ff4b85 | 1879 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1880 | { |
c0ff4b85 R |
1881 | if (memcg && atomic_read(&memcg->under_oom)) |
1882 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1883 | } |
1884 | ||
3812c8c8 | 1885 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1886 | { |
3812c8c8 JW |
1887 | if (!current->memcg_oom.may_oom) |
1888 | return; | |
867578cb | 1889 | /* |
49426420 JW |
1890 | * We are in the middle of the charge context here, so we |
1891 | * don't want to block when potentially sitting on a callstack | |
1892 | * that holds all kinds of filesystem and mm locks. | |
1893 | * | |
1894 | * Also, the caller may handle a failed allocation gracefully | |
1895 | * (like optional page cache readahead) and so an OOM killer | |
1896 | * invocation might not even be necessary. | |
1897 | * | |
1898 | * That's why we don't do anything here except remember the | |
1899 | * OOM context and then deal with it at the end of the page | |
1900 | * fault when the stack is unwound, the locks are released, | |
1901 | * and when we know whether the fault was overall successful. | |
867578cb | 1902 | */ |
49426420 JW |
1903 | css_get(&memcg->css); |
1904 | current->memcg_oom.memcg = memcg; | |
1905 | current->memcg_oom.gfp_mask = mask; | |
1906 | current->memcg_oom.order = order; | |
3812c8c8 JW |
1907 | } |
1908 | ||
1909 | /** | |
1910 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1911 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1912 | * |
49426420 JW |
1913 | * This has to be called at the end of a page fault if the memcg OOM |
1914 | * handler was enabled. | |
3812c8c8 | 1915 | * |
49426420 | 1916 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1917 | * sleep on a waitqueue until the userspace task resolves the |
1918 | * situation. Sleeping directly in the charge context with all kinds | |
1919 | * of locks held is not a good idea, instead we remember an OOM state | |
1920 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1921 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1922 | * |
1923 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1924 | * completed, %false otherwise. |
3812c8c8 | 1925 | */ |
49426420 | 1926 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1927 | { |
49426420 | 1928 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
3812c8c8 | 1929 | struct oom_wait_info owait; |
49426420 | 1930 | bool locked; |
3812c8c8 JW |
1931 | |
1932 | /* OOM is global, do not handle */ | |
3812c8c8 | 1933 | if (!memcg) |
49426420 | 1934 | return false; |
3812c8c8 | 1935 | |
49426420 JW |
1936 | if (!handle) |
1937 | goto cleanup; | |
3812c8c8 JW |
1938 | |
1939 | owait.memcg = memcg; | |
1940 | owait.wait.flags = 0; | |
1941 | owait.wait.func = memcg_oom_wake_function; | |
1942 | owait.wait.private = current; | |
1943 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1944 | |
3812c8c8 | 1945 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1946 | mem_cgroup_mark_under_oom(memcg); |
1947 | ||
1948 | locked = mem_cgroup_oom_trylock(memcg); | |
1949 | ||
1950 | if (locked) | |
1951 | mem_cgroup_oom_notify(memcg); | |
1952 | ||
1953 | if (locked && !memcg->oom_kill_disable) { | |
1954 | mem_cgroup_unmark_under_oom(memcg); | |
1955 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1956 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, | |
1957 | current->memcg_oom.order); | |
1958 | } else { | |
3812c8c8 | 1959 | schedule(); |
49426420 JW |
1960 | mem_cgroup_unmark_under_oom(memcg); |
1961 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1962 | } | |
1963 | ||
1964 | if (locked) { | |
fb2a6fc5 JW |
1965 | mem_cgroup_oom_unlock(memcg); |
1966 | /* | |
1967 | * There is no guarantee that an OOM-lock contender | |
1968 | * sees the wakeups triggered by the OOM kill | |
1969 | * uncharges. Wake any sleepers explicitely. | |
1970 | */ | |
1971 | memcg_oom_recover(memcg); | |
1972 | } | |
49426420 JW |
1973 | cleanup: |
1974 | current->memcg_oom.memcg = NULL; | |
3812c8c8 | 1975 | css_put(&memcg->css); |
867578cb | 1976 | return true; |
0b7f569e KH |
1977 | } |
1978 | ||
d7365e78 JW |
1979 | /** |
1980 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1981 | * @page: page that is going to change accounted state | |
32047e2a | 1982 | * |
d7365e78 JW |
1983 | * This function must mark the beginning of an accounted page state |
1984 | * change to prevent double accounting when the page is concurrently | |
1985 | * being moved to another memcg: | |
32047e2a | 1986 | * |
6de22619 | 1987 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1988 | * if (TestClearPageState(page)) |
1989 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1990 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1991 | */ |
6de22619 | 1992 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1993 | { |
1994 | struct mem_cgroup *memcg; | |
6de22619 | 1995 | unsigned long flags; |
89c06bd5 | 1996 | |
6de22619 JW |
1997 | /* |
1998 | * The RCU lock is held throughout the transaction. The fast | |
1999 | * path can get away without acquiring the memcg->move_lock | |
2000 | * because page moving starts with an RCU grace period. | |
2001 | * | |
2002 | * The RCU lock also protects the memcg from being freed when | |
2003 | * the page state that is going to change is the only thing | |
2004 | * preventing the page from being uncharged. | |
2005 | * E.g. end-writeback clearing PageWriteback(), which allows | |
2006 | * migration to go ahead and uncharge the page before the | |
2007 | * account transaction might be complete. | |
2008 | */ | |
d7365e78 JW |
2009 | rcu_read_lock(); |
2010 | ||
2011 | if (mem_cgroup_disabled()) | |
2012 | return NULL; | |
89c06bd5 | 2013 | again: |
1306a85a | 2014 | memcg = page->mem_cgroup; |
29833315 | 2015 | if (unlikely(!memcg)) |
d7365e78 JW |
2016 | return NULL; |
2017 | ||
bdcbb659 | 2018 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 2019 | return memcg; |
89c06bd5 | 2020 | |
6de22619 | 2021 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 2022 | if (memcg != page->mem_cgroup) { |
6de22619 | 2023 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2024 | goto again; |
2025 | } | |
6de22619 JW |
2026 | |
2027 | /* | |
2028 | * When charge migration first begins, we can have locked and | |
2029 | * unlocked page stat updates happening concurrently. Track | |
2030 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
2031 | */ | |
2032 | memcg->move_lock_task = current; | |
2033 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
2034 | |
2035 | return memcg; | |
89c06bd5 KH |
2036 | } |
2037 | ||
d7365e78 JW |
2038 | /** |
2039 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
2040 | * @memcg: the memcg that was accounted against | |
d7365e78 | 2041 | */ |
6de22619 | 2042 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 2043 | { |
6de22619 JW |
2044 | if (memcg && memcg->move_lock_task == current) { |
2045 | unsigned long flags = memcg->move_lock_flags; | |
2046 | ||
2047 | memcg->move_lock_task = NULL; | |
2048 | memcg->move_lock_flags = 0; | |
2049 | ||
2050 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2051 | } | |
89c06bd5 | 2052 | |
d7365e78 | 2053 | rcu_read_unlock(); |
89c06bd5 KH |
2054 | } |
2055 | ||
d7365e78 JW |
2056 | /** |
2057 | * mem_cgroup_update_page_stat - update page state statistics | |
2058 | * @memcg: memcg to account against | |
2059 | * @idx: page state item to account | |
2060 | * @val: number of pages (positive or negative) | |
2061 | * | |
2062 | * See mem_cgroup_begin_page_stat() for locking requirements. | |
2063 | */ | |
2064 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, | |
68b4876d | 2065 | enum mem_cgroup_stat_index idx, int val) |
d69b042f | 2066 | { |
658b72c5 | 2067 | VM_BUG_ON(!rcu_read_lock_held()); |
26174efd | 2068 | |
d7365e78 JW |
2069 | if (memcg) |
2070 | this_cpu_add(memcg->stat->count[idx], val); | |
d69b042f | 2071 | } |
26174efd | 2072 | |
cdec2e42 KH |
2073 | /* |
2074 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2075 | * TODO: maybe necessary to use big numbers in big irons. | |
2076 | */ | |
7ec99d62 | 2077 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2078 | struct memcg_stock_pcp { |
2079 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2080 | unsigned int nr_pages; |
cdec2e42 | 2081 | struct work_struct work; |
26fe6168 | 2082 | unsigned long flags; |
a0db00fc | 2083 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2084 | }; |
2085 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2086 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2087 | |
a0956d54 SS |
2088 | /** |
2089 | * consume_stock: Try to consume stocked charge on this cpu. | |
2090 | * @memcg: memcg to consume from. | |
2091 | * @nr_pages: how many pages to charge. | |
2092 | * | |
2093 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2094 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2095 | * service an allocation will refill the stock. | |
2096 | * | |
2097 | * returns true if successful, false otherwise. | |
cdec2e42 | 2098 | */ |
a0956d54 | 2099 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2100 | { |
2101 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 2102 | bool ret = false; |
cdec2e42 | 2103 | |
a0956d54 | 2104 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 2105 | return ret; |
a0956d54 | 2106 | |
cdec2e42 | 2107 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 2108 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2109 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2110 | ret = true; |
2111 | } | |
cdec2e42 KH |
2112 | put_cpu_var(memcg_stock); |
2113 | return ret; | |
2114 | } | |
2115 | ||
2116 | /* | |
3e32cb2e | 2117 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2118 | */ |
2119 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2120 | { | |
2121 | struct mem_cgroup *old = stock->cached; | |
2122 | ||
11c9ea4e | 2123 | if (stock->nr_pages) { |
3e32cb2e | 2124 | page_counter_uncharge(&old->memory, stock->nr_pages); |
cdec2e42 | 2125 | if (do_swap_account) |
3e32cb2e | 2126 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2127 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2128 | stock->nr_pages = 0; |
cdec2e42 KH |
2129 | } |
2130 | stock->cached = NULL; | |
cdec2e42 KH |
2131 | } |
2132 | ||
2133 | /* | |
2134 | * This must be called under preempt disabled or must be called by | |
2135 | * a thread which is pinned to local cpu. | |
2136 | */ | |
2137 | static void drain_local_stock(struct work_struct *dummy) | |
2138 | { | |
7c8e0181 | 2139 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 2140 | drain_stock(stock); |
26fe6168 | 2141 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2142 | } |
2143 | ||
e4777496 MH |
2144 | static void __init memcg_stock_init(void) |
2145 | { | |
2146 | int cpu; | |
2147 | ||
2148 | for_each_possible_cpu(cpu) { | |
2149 | struct memcg_stock_pcp *stock = | |
2150 | &per_cpu(memcg_stock, cpu); | |
2151 | INIT_WORK(&stock->work, drain_local_stock); | |
2152 | } | |
2153 | } | |
2154 | ||
cdec2e42 | 2155 | /* |
3e32cb2e | 2156 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2157 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2158 | */ |
c0ff4b85 | 2159 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2160 | { |
2161 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2162 | ||
c0ff4b85 | 2163 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2164 | drain_stock(stock); |
c0ff4b85 | 2165 | stock->cached = memcg; |
cdec2e42 | 2166 | } |
11c9ea4e | 2167 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2168 | put_cpu_var(memcg_stock); |
2169 | } | |
2170 | ||
2171 | /* | |
c0ff4b85 | 2172 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2173 | * of the hierarchy under it. |
cdec2e42 | 2174 | */ |
6d3d6aa2 | 2175 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2176 | { |
26fe6168 | 2177 | int cpu, curcpu; |
d38144b7 | 2178 | |
6d3d6aa2 JW |
2179 | /* If someone's already draining, avoid adding running more workers. */ |
2180 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2181 | return; | |
cdec2e42 | 2182 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2183 | get_online_cpus(); |
5af12d0e | 2184 | curcpu = get_cpu(); |
cdec2e42 KH |
2185 | for_each_online_cpu(cpu) { |
2186 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2187 | struct mem_cgroup *memcg; |
26fe6168 | 2188 | |
c0ff4b85 R |
2189 | memcg = stock->cached; |
2190 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2191 | continue; |
2314b42d | 2192 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 2193 | continue; |
d1a05b69 MH |
2194 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2195 | if (cpu == curcpu) | |
2196 | drain_local_stock(&stock->work); | |
2197 | else | |
2198 | schedule_work_on(cpu, &stock->work); | |
2199 | } | |
cdec2e42 | 2200 | } |
5af12d0e | 2201 | put_cpu(); |
f894ffa8 | 2202 | put_online_cpus(); |
9f50fad6 | 2203 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2204 | } |
2205 | ||
711d3d2c KH |
2206 | /* |
2207 | * This function drains percpu counter value from DEAD cpu and | |
2208 | * move it to local cpu. Note that this function can be preempted. | |
2209 | */ | |
c0ff4b85 | 2210 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2211 | { |
2212 | int i; | |
2213 | ||
c0ff4b85 | 2214 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2215 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2216 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2217 | |
c0ff4b85 R |
2218 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2219 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2220 | } |
e9f8974f | 2221 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2222 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2223 | |
c0ff4b85 R |
2224 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2225 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2226 | } |
c0ff4b85 | 2227 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2228 | } |
2229 | ||
0db0628d | 2230 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
2231 | unsigned long action, |
2232 | void *hcpu) | |
2233 | { | |
2234 | int cpu = (unsigned long)hcpu; | |
2235 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2236 | struct mem_cgroup *iter; |
cdec2e42 | 2237 | |
619d094b | 2238 | if (action == CPU_ONLINE) |
1489ebad | 2239 | return NOTIFY_OK; |
1489ebad | 2240 | |
d833049b | 2241 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2242 | return NOTIFY_OK; |
711d3d2c | 2243 | |
9f3a0d09 | 2244 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2245 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2246 | ||
cdec2e42 KH |
2247 | stock = &per_cpu(memcg_stock, cpu); |
2248 | drain_stock(stock); | |
2249 | return NOTIFY_OK; | |
2250 | } | |
2251 | ||
00501b53 JW |
2252 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2253 | unsigned int nr_pages) | |
8a9f3ccd | 2254 | { |
7ec99d62 | 2255 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 2256 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2257 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2258 | struct page_counter *counter; |
6539cc05 | 2259 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2260 | bool may_swap = true; |
2261 | bool drained = false; | |
05b84301 | 2262 | int ret = 0; |
a636b327 | 2263 | |
ce00a967 JW |
2264 | if (mem_cgroup_is_root(memcg)) |
2265 | goto done; | |
6539cc05 | 2266 | retry: |
b6b6cc72 MH |
2267 | if (consume_stock(memcg, nr_pages)) |
2268 | goto done; | |
8a9f3ccd | 2269 | |
3fbe7244 | 2270 | if (!do_swap_account || |
3e32cb2e JW |
2271 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2272 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2273 | goto done_restock; |
3fbe7244 | 2274 | if (do_swap_account) |
3e32cb2e JW |
2275 | page_counter_uncharge(&memcg->memsw, batch); |
2276 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2277 | } else { |
3e32cb2e | 2278 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2279 | may_swap = false; |
3fbe7244 | 2280 | } |
7a81b88c | 2281 | |
6539cc05 JW |
2282 | if (batch > nr_pages) { |
2283 | batch = nr_pages; | |
2284 | goto retry; | |
2285 | } | |
6d61ef40 | 2286 | |
06b078fc JW |
2287 | /* |
2288 | * Unlike in global OOM situations, memcg is not in a physical | |
2289 | * memory shortage. Allow dying and OOM-killed tasks to | |
2290 | * bypass the last charges so that they can exit quickly and | |
2291 | * free their memory. | |
2292 | */ | |
2293 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2294 | fatal_signal_pending(current) || | |
2295 | current->flags & PF_EXITING)) | |
2296 | goto bypass; | |
2297 | ||
2298 | if (unlikely(task_in_memcg_oom(current))) | |
2299 | goto nomem; | |
2300 | ||
6539cc05 JW |
2301 | if (!(gfp_mask & __GFP_WAIT)) |
2302 | goto nomem; | |
4b534334 | 2303 | |
b70a2a21 JW |
2304 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2305 | gfp_mask, may_swap); | |
6539cc05 | 2306 | |
61e02c74 | 2307 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2308 | goto retry; |
28c34c29 | 2309 | |
b70a2a21 | 2310 | if (!drained) { |
6d3d6aa2 | 2311 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2312 | drained = true; |
2313 | goto retry; | |
2314 | } | |
2315 | ||
28c34c29 JW |
2316 | if (gfp_mask & __GFP_NORETRY) |
2317 | goto nomem; | |
6539cc05 JW |
2318 | /* |
2319 | * Even though the limit is exceeded at this point, reclaim | |
2320 | * may have been able to free some pages. Retry the charge | |
2321 | * before killing the task. | |
2322 | * | |
2323 | * Only for regular pages, though: huge pages are rather | |
2324 | * unlikely to succeed so close to the limit, and we fall back | |
2325 | * to regular pages anyway in case of failure. | |
2326 | */ | |
61e02c74 | 2327 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2328 | goto retry; |
2329 | /* | |
2330 | * At task move, charge accounts can be doubly counted. So, it's | |
2331 | * better to wait until the end of task_move if something is going on. | |
2332 | */ | |
2333 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2334 | goto retry; | |
2335 | ||
9b130619 JW |
2336 | if (nr_retries--) |
2337 | goto retry; | |
2338 | ||
06b078fc JW |
2339 | if (gfp_mask & __GFP_NOFAIL) |
2340 | goto bypass; | |
2341 | ||
6539cc05 JW |
2342 | if (fatal_signal_pending(current)) |
2343 | goto bypass; | |
2344 | ||
61e02c74 | 2345 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
7a81b88c | 2346 | nomem: |
6d1fdc48 | 2347 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2348 | return -ENOMEM; |
867578cb | 2349 | bypass: |
ce00a967 | 2350 | return -EINTR; |
6539cc05 JW |
2351 | |
2352 | done_restock: | |
e8ea14cc | 2353 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2354 | if (batch > nr_pages) |
2355 | refill_stock(memcg, batch - nr_pages); | |
2356 | done: | |
05b84301 | 2357 | return ret; |
7a81b88c | 2358 | } |
8a9f3ccd | 2359 | |
00501b53 | 2360 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2361 | { |
ce00a967 JW |
2362 | if (mem_cgroup_is_root(memcg)) |
2363 | return; | |
2364 | ||
3e32cb2e | 2365 | page_counter_uncharge(&memcg->memory, nr_pages); |
05b84301 | 2366 | if (do_swap_account) |
3e32cb2e | 2367 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2368 | |
e8ea14cc | 2369 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2370 | } |
2371 | ||
a3b2d692 KH |
2372 | /* |
2373 | * A helper function to get mem_cgroup from ID. must be called under | |
ec903c0c TH |
2374 | * rcu_read_lock(). The caller is responsible for calling |
2375 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
2376 | * refcnt from swap can be called against removed memcg.) | |
a3b2d692 KH |
2377 | */ |
2378 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2379 | { | |
a3b2d692 KH |
2380 | /* ID 0 is unused ID */ |
2381 | if (!id) | |
2382 | return NULL; | |
34c00c31 | 2383 | return mem_cgroup_from_id(id); |
a3b2d692 KH |
2384 | } |
2385 | ||
0a31bc97 JW |
2386 | /* |
2387 | * try_get_mem_cgroup_from_page - look up page's memcg association | |
2388 | * @page: the page | |
2389 | * | |
2390 | * Look up, get a css reference, and return the memcg that owns @page. | |
2391 | * | |
2392 | * The page must be locked to prevent racing with swap-in and page | |
2393 | * cache charges. If coming from an unlocked page table, the caller | |
2394 | * must ensure the page is on the LRU or this can race with charging. | |
2395 | */ | |
e42d9d5d | 2396 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2397 | { |
29833315 | 2398 | struct mem_cgroup *memcg; |
a3b2d692 | 2399 | unsigned short id; |
b5a84319 KH |
2400 | swp_entry_t ent; |
2401 | ||
309381fe | 2402 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
3c776e64 | 2403 | |
1306a85a | 2404 | memcg = page->mem_cgroup; |
29833315 JW |
2405 | if (memcg) { |
2406 | if (!css_tryget_online(&memcg->css)) | |
c0ff4b85 | 2407 | memcg = NULL; |
e42d9d5d | 2408 | } else if (PageSwapCache(page)) { |
3c776e64 | 2409 | ent.val = page_private(page); |
9fb4b7cc | 2410 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2411 | rcu_read_lock(); |
c0ff4b85 | 2412 | memcg = mem_cgroup_lookup(id); |
ec903c0c | 2413 | if (memcg && !css_tryget_online(&memcg->css)) |
c0ff4b85 | 2414 | memcg = NULL; |
a3b2d692 | 2415 | rcu_read_unlock(); |
3c776e64 | 2416 | } |
c0ff4b85 | 2417 | return memcg; |
b5a84319 KH |
2418 | } |
2419 | ||
0a31bc97 JW |
2420 | static void lock_page_lru(struct page *page, int *isolated) |
2421 | { | |
2422 | struct zone *zone = page_zone(page); | |
2423 | ||
2424 | spin_lock_irq(&zone->lru_lock); | |
2425 | if (PageLRU(page)) { | |
2426 | struct lruvec *lruvec; | |
2427 | ||
2428 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2429 | ClearPageLRU(page); | |
2430 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2431 | *isolated = 1; | |
2432 | } else | |
2433 | *isolated = 0; | |
2434 | } | |
2435 | ||
2436 | static void unlock_page_lru(struct page *page, int isolated) | |
2437 | { | |
2438 | struct zone *zone = page_zone(page); | |
2439 | ||
2440 | if (isolated) { | |
2441 | struct lruvec *lruvec; | |
2442 | ||
2443 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2444 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2445 | SetPageLRU(page); | |
2446 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2447 | } | |
2448 | spin_unlock_irq(&zone->lru_lock); | |
2449 | } | |
2450 | ||
00501b53 | 2451 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2452 | bool lrucare) |
7a81b88c | 2453 | { |
0a31bc97 | 2454 | int isolated; |
9ce70c02 | 2455 | |
1306a85a | 2456 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2457 | |
2458 | /* | |
2459 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2460 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2461 | */ | |
0a31bc97 JW |
2462 | if (lrucare) |
2463 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2464 | |
0a31bc97 JW |
2465 | /* |
2466 | * Nobody should be changing or seriously looking at | |
1306a85a | 2467 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2468 | * |
2469 | * - the page is uncharged | |
2470 | * | |
2471 | * - the page is off-LRU | |
2472 | * | |
2473 | * - an anonymous fault has exclusive page access, except for | |
2474 | * a locked page table | |
2475 | * | |
2476 | * - a page cache insertion, a swapin fault, or a migration | |
2477 | * have the page locked | |
2478 | */ | |
1306a85a | 2479 | page->mem_cgroup = memcg; |
9ce70c02 | 2480 | |
0a31bc97 JW |
2481 | if (lrucare) |
2482 | unlock_page_lru(page, isolated); | |
7a81b88c | 2483 | } |
66e1707b | 2484 | |
7ae1e1d0 | 2485 | #ifdef CONFIG_MEMCG_KMEM |
dbf22eb6 VD |
2486 | int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
2487 | unsigned long nr_pages) | |
7ae1e1d0 | 2488 | { |
3e32cb2e | 2489 | struct page_counter *counter; |
7ae1e1d0 | 2490 | int ret = 0; |
7ae1e1d0 | 2491 | |
3e32cb2e JW |
2492 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
2493 | if (ret < 0) | |
7ae1e1d0 GC |
2494 | return ret; |
2495 | ||
3e32cb2e | 2496 | ret = try_charge(memcg, gfp, nr_pages); |
7ae1e1d0 GC |
2497 | if (ret == -EINTR) { |
2498 | /* | |
00501b53 JW |
2499 | * try_charge() chose to bypass to root due to OOM kill or |
2500 | * fatal signal. Since our only options are to either fail | |
2501 | * the allocation or charge it to this cgroup, do it as a | |
2502 | * temporary condition. But we can't fail. From a kmem/slab | |
2503 | * perspective, the cache has already been selected, by | |
2504 | * mem_cgroup_kmem_get_cache(), so it is too late to change | |
7ae1e1d0 GC |
2505 | * our minds. |
2506 | * | |
2507 | * This condition will only trigger if the task entered | |
00501b53 JW |
2508 | * memcg_charge_kmem in a sane state, but was OOM-killed |
2509 | * during try_charge() above. Tasks that were already dying | |
2510 | * when the allocation triggers should have been already | |
7ae1e1d0 GC |
2511 | * directed to the root cgroup in memcontrol.h |
2512 | */ | |
3e32cb2e | 2513 | page_counter_charge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2514 | if (do_swap_account) |
3e32cb2e | 2515 | page_counter_charge(&memcg->memsw, nr_pages); |
e8ea14cc | 2516 | css_get_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2517 | ret = 0; |
2518 | } else if (ret) | |
3e32cb2e | 2519 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7ae1e1d0 GC |
2520 | |
2521 | return ret; | |
2522 | } | |
2523 | ||
dbf22eb6 | 2524 | void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages) |
7ae1e1d0 | 2525 | { |
3e32cb2e | 2526 | page_counter_uncharge(&memcg->memory, nr_pages); |
7ae1e1d0 | 2527 | if (do_swap_account) |
3e32cb2e | 2528 | page_counter_uncharge(&memcg->memsw, nr_pages); |
7de37682 | 2529 | |
64f21993 | 2530 | page_counter_uncharge(&memcg->kmem, nr_pages); |
7de37682 | 2531 | |
e8ea14cc | 2532 | css_put_many(&memcg->css, nr_pages); |
7ae1e1d0 GC |
2533 | } |
2534 | ||
2633d7a0 GC |
2535 | /* |
2536 | * helper for acessing a memcg's index. It will be used as an index in the | |
2537 | * child cache array in kmem_cache, and also to derive its name. This function | |
2538 | * will return -1 when this is not a kmem-limited memcg. | |
2539 | */ | |
2540 | int memcg_cache_id(struct mem_cgroup *memcg) | |
2541 | { | |
2542 | return memcg ? memcg->kmemcg_id : -1; | |
2543 | } | |
2544 | ||
f3bb3043 | 2545 | static int memcg_alloc_cache_id(void) |
55007d84 | 2546 | { |
f3bb3043 VD |
2547 | int id, size; |
2548 | int err; | |
2549 | ||
2550 | id = ida_simple_get(&kmem_limited_groups, | |
2551 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
2552 | if (id < 0) | |
2553 | return id; | |
55007d84 | 2554 | |
f3bb3043 VD |
2555 | if (id < memcg_limited_groups_array_size) |
2556 | return id; | |
2557 | ||
2558 | /* | |
2559 | * There's no space for the new id in memcg_caches arrays, | |
2560 | * so we have to grow them. | |
2561 | */ | |
2562 | ||
2563 | size = 2 * (id + 1); | |
55007d84 GC |
2564 | if (size < MEMCG_CACHES_MIN_SIZE) |
2565 | size = MEMCG_CACHES_MIN_SIZE; | |
2566 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2567 | size = MEMCG_CACHES_MAX_SIZE; | |
2568 | ||
f3bb3043 | 2569 | err = memcg_update_all_caches(size); |
f3bb3043 VD |
2570 | if (err) { |
2571 | ida_simple_remove(&kmem_limited_groups, id); | |
2572 | return err; | |
2573 | } | |
2574 | return id; | |
2575 | } | |
2576 | ||
2577 | static void memcg_free_cache_id(int id) | |
2578 | { | |
2579 | ida_simple_remove(&kmem_limited_groups, id); | |
55007d84 GC |
2580 | } |
2581 | ||
2582 | /* | |
2583 | * We should update the current array size iff all caches updates succeed. This | |
2584 | * can only be done from the slab side. The slab mutex needs to be held when | |
2585 | * calling this. | |
2586 | */ | |
2587 | void memcg_update_array_size(int num) | |
2588 | { | |
f3bb3043 | 2589 | memcg_limited_groups_array_size = num; |
55007d84 GC |
2590 | } |
2591 | ||
d5b3cf71 | 2592 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2593 | struct mem_cgroup *memcg; |
2594 | struct kmem_cache *cachep; | |
2595 | struct work_struct work; | |
2596 | }; | |
2597 | ||
d5b3cf71 | 2598 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2599 | { |
d5b3cf71 VD |
2600 | struct memcg_kmem_cache_create_work *cw = |
2601 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2602 | struct mem_cgroup *memcg = cw->memcg; |
2603 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2604 | |
d5b3cf71 | 2605 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2606 | |
5722d094 | 2607 | css_put(&memcg->css); |
d7f25f8a GC |
2608 | kfree(cw); |
2609 | } | |
2610 | ||
2611 | /* | |
2612 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2613 | */ |
d5b3cf71 VD |
2614 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2615 | struct kmem_cache *cachep) | |
d7f25f8a | 2616 | { |
d5b3cf71 | 2617 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2618 | |
776ed0f0 | 2619 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2620 | if (!cw) |
d7f25f8a | 2621 | return; |
8135be5a VD |
2622 | |
2623 | css_get(&memcg->css); | |
d7f25f8a GC |
2624 | |
2625 | cw->memcg = memcg; | |
2626 | cw->cachep = cachep; | |
d5b3cf71 | 2627 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2628 | |
d7f25f8a GC |
2629 | schedule_work(&cw->work); |
2630 | } | |
2631 | ||
d5b3cf71 VD |
2632 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2633 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2634 | { |
2635 | /* | |
2636 | * We need to stop accounting when we kmalloc, because if the | |
2637 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2638 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2639 | * |
2640 | * However, it is better to enclose the whole function. Depending on | |
2641 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2642 | * trigger an allocation. This too, will make us recurse. Because at | |
2643 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2644 | * the safest choice is to do it like this, wrapping the whole function. | |
2645 | */ | |
6f185c29 | 2646 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2647 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2648 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2649 | } |
c67a8a68 | 2650 | |
d7f25f8a GC |
2651 | /* |
2652 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2653 | * We try to use the current memcg's version of the cache. | |
2654 | * | |
2655 | * If the cache does not exist yet, if we are the first user of it, | |
2656 | * we either create it immediately, if possible, or create it asynchronously | |
2657 | * in a workqueue. | |
2658 | * In the latter case, we will let the current allocation go through with | |
2659 | * the original cache. | |
2660 | * | |
2661 | * Can't be called in interrupt context or from kernel threads. | |
2662 | * This function needs to be called with rcu_read_lock() held. | |
2663 | */ | |
056b7cce | 2664 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2665 | { |
2666 | struct mem_cgroup *memcg; | |
959c8963 | 2667 | struct kmem_cache *memcg_cachep; |
d7f25f8a GC |
2668 | |
2669 | VM_BUG_ON(!cachep->memcg_params); | |
2670 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
2671 | ||
9d100c5e | 2672 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2673 | return cachep; |
2674 | ||
8135be5a | 2675 | memcg = get_mem_cgroup_from_mm(current->mm); |
cf2b8fbf | 2676 | if (!memcg_kmem_is_active(memcg)) |
ca0dde97 | 2677 | goto out; |
d7f25f8a | 2678 | |
959c8963 | 2679 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
8135be5a VD |
2680 | if (likely(memcg_cachep)) |
2681 | return memcg_cachep; | |
ca0dde97 LZ |
2682 | |
2683 | /* | |
2684 | * If we are in a safe context (can wait, and not in interrupt | |
2685 | * context), we could be be predictable and return right away. | |
2686 | * This would guarantee that the allocation being performed | |
2687 | * already belongs in the new cache. | |
2688 | * | |
2689 | * However, there are some clashes that can arrive from locking. | |
2690 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2691 | * memcg_create_kmem_cache, this means no further allocation |
2692 | * could happen with the slab_mutex held. So it's better to | |
2693 | * defer everything. | |
ca0dde97 | 2694 | */ |
d5b3cf71 | 2695 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2696 | out: |
8135be5a | 2697 | css_put(&memcg->css); |
ca0dde97 | 2698 | return cachep; |
d7f25f8a | 2699 | } |
d7f25f8a | 2700 | |
8135be5a VD |
2701 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2702 | { | |
2703 | if (!is_root_cache(cachep)) | |
2704 | css_put(&cachep->memcg_params->memcg->css); | |
2705 | } | |
2706 | ||
7ae1e1d0 GC |
2707 | /* |
2708 | * We need to verify if the allocation against current->mm->owner's memcg is | |
2709 | * possible for the given order. But the page is not allocated yet, so we'll | |
2710 | * need a further commit step to do the final arrangements. | |
2711 | * | |
2712 | * It is possible for the task to switch cgroups in this mean time, so at | |
2713 | * commit time, we can't rely on task conversion any longer. We'll then use | |
2714 | * the handle argument to return to the caller which cgroup we should commit | |
2715 | * against. We could also return the memcg directly and avoid the pointer | |
2716 | * passing, but a boolean return value gives better semantics considering | |
2717 | * the compiled-out case as well. | |
2718 | * | |
2719 | * Returning true means the allocation is possible. | |
2720 | */ | |
2721 | bool | |
2722 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
2723 | { | |
2724 | struct mem_cgroup *memcg; | |
2725 | int ret; | |
2726 | ||
2727 | *_memcg = NULL; | |
6d42c232 | 2728 | |
df381975 | 2729 | memcg = get_mem_cgroup_from_mm(current->mm); |
7ae1e1d0 | 2730 | |
cf2b8fbf | 2731 | if (!memcg_kmem_is_active(memcg)) { |
7ae1e1d0 GC |
2732 | css_put(&memcg->css); |
2733 | return true; | |
2734 | } | |
2735 | ||
3e32cb2e | 2736 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
7ae1e1d0 GC |
2737 | if (!ret) |
2738 | *_memcg = memcg; | |
7ae1e1d0 GC |
2739 | |
2740 | css_put(&memcg->css); | |
2741 | return (ret == 0); | |
2742 | } | |
2743 | ||
2744 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
2745 | int order) | |
2746 | { | |
7ae1e1d0 GC |
2747 | VM_BUG_ON(mem_cgroup_is_root(memcg)); |
2748 | ||
2749 | /* The page allocation failed. Revert */ | |
2750 | if (!page) { | |
3e32cb2e | 2751 | memcg_uncharge_kmem(memcg, 1 << order); |
7ae1e1d0 GC |
2752 | return; |
2753 | } | |
1306a85a | 2754 | page->mem_cgroup = memcg; |
7ae1e1d0 GC |
2755 | } |
2756 | ||
2757 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
2758 | { | |
1306a85a | 2759 | struct mem_cgroup *memcg = page->mem_cgroup; |
7ae1e1d0 | 2760 | |
7ae1e1d0 GC |
2761 | if (!memcg) |
2762 | return; | |
2763 | ||
309381fe | 2764 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2765 | |
3e32cb2e | 2766 | memcg_uncharge_kmem(memcg, 1 << order); |
1306a85a | 2767 | page->mem_cgroup = NULL; |
7ae1e1d0 GC |
2768 | } |
2769 | #endif /* CONFIG_MEMCG_KMEM */ | |
2770 | ||
ca3e0214 KH |
2771 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2772 | ||
ca3e0214 KH |
2773 | /* |
2774 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2775 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2776 | * charge/uncharge will be never happen and move_account() is done under | |
2777 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2778 | */ |
e94c8a9c | 2779 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2780 | { |
e94c8a9c | 2781 | int i; |
ca3e0214 | 2782 | |
3d37c4a9 KH |
2783 | if (mem_cgroup_disabled()) |
2784 | return; | |
b070e65c | 2785 | |
29833315 | 2786 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2787 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2788 | |
1306a85a | 2789 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2790 | HPAGE_PMD_NR); |
ca3e0214 | 2791 | } |
12d27107 | 2792 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2793 | |
f817ed48 | 2794 | /** |
de3638d9 | 2795 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2796 | * @page: the page |
7ec99d62 | 2797 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2798 | * @from: mem_cgroup which the page is moved from. |
2799 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2800 | * | |
2801 | * The caller must confirm following. | |
08e552c6 | 2802 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2803 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2804 | * |
2f3479b1 KH |
2805 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2806 | * from old cgroup. | |
f817ed48 | 2807 | */ |
7ec99d62 JW |
2808 | static int mem_cgroup_move_account(struct page *page, |
2809 | unsigned int nr_pages, | |
7ec99d62 | 2810 | struct mem_cgroup *from, |
2f3479b1 | 2811 | struct mem_cgroup *to) |
f817ed48 | 2812 | { |
de3638d9 JW |
2813 | unsigned long flags; |
2814 | int ret; | |
987eba66 | 2815 | |
f817ed48 | 2816 | VM_BUG_ON(from == to); |
309381fe | 2817 | VM_BUG_ON_PAGE(PageLRU(page), page); |
de3638d9 JW |
2818 | /* |
2819 | * The page is isolated from LRU. So, collapse function | |
2820 | * will not handle this page. But page splitting can happen. | |
2821 | * Do this check under compound_page_lock(). The caller should | |
2822 | * hold it. | |
2823 | */ | |
2824 | ret = -EBUSY; | |
7ec99d62 | 2825 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2826 | goto out; |
2827 | ||
0a31bc97 | 2828 | /* |
1306a85a | 2829 | * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup |
0a31bc97 JW |
2830 | * of its source page while we change it: page migration takes |
2831 | * both pages off the LRU, but page cache replacement doesn't. | |
2832 | */ | |
2833 | if (!trylock_page(page)) | |
2834 | goto out; | |
de3638d9 JW |
2835 | |
2836 | ret = -EINVAL; | |
1306a85a | 2837 | if (page->mem_cgroup != from) |
0a31bc97 | 2838 | goto out_unlock; |
de3638d9 | 2839 | |
354a4783 | 2840 | spin_lock_irqsave(&from->move_lock, flags); |
f817ed48 | 2841 | |
0a31bc97 | 2842 | if (!PageAnon(page) && page_mapped(page)) { |
59d1d256 JW |
2843 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
2844 | nr_pages); | |
2845 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
2846 | nr_pages); | |
2847 | } | |
3ea67d06 | 2848 | |
59d1d256 JW |
2849 | if (PageWriteback(page)) { |
2850 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
2851 | nr_pages); | |
2852 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
2853 | nr_pages); | |
2854 | } | |
3ea67d06 | 2855 | |
0a31bc97 | 2856 | /* |
1306a85a | 2857 | * It is safe to change page->mem_cgroup here because the page |
0a31bc97 JW |
2858 | * is referenced, charged, and isolated - we can't race with |
2859 | * uncharging, charging, migration, or LRU putback. | |
2860 | */ | |
d69b042f | 2861 | |
854ffa8d | 2862 | /* caller should have done css_get */ |
1306a85a | 2863 | page->mem_cgroup = to; |
354a4783 JW |
2864 | spin_unlock_irqrestore(&from->move_lock, flags); |
2865 | ||
de3638d9 | 2866 | ret = 0; |
0a31bc97 JW |
2867 | |
2868 | local_irq_disable(); | |
2869 | mem_cgroup_charge_statistics(to, page, nr_pages); | |
5564e88b | 2870 | memcg_check_events(to, page); |
0a31bc97 | 2871 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
5564e88b | 2872 | memcg_check_events(from, page); |
0a31bc97 JW |
2873 | local_irq_enable(); |
2874 | out_unlock: | |
2875 | unlock_page(page); | |
de3638d9 | 2876 | out: |
f817ed48 KH |
2877 | return ret; |
2878 | } | |
2879 | ||
c255a458 | 2880 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2881 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2882 | bool charge) | |
d13d1443 | 2883 | { |
0a31bc97 JW |
2884 | int val = (charge) ? 1 : -1; |
2885 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2886 | } |
02491447 DN |
2887 | |
2888 | /** | |
2889 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2890 | * @entry: swap entry to be moved | |
2891 | * @from: mem_cgroup which the entry is moved from | |
2892 | * @to: mem_cgroup which the entry is moved to | |
2893 | * | |
2894 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2895 | * as the mem_cgroup's id of @from. | |
2896 | * | |
2897 | * Returns 0 on success, -EINVAL on failure. | |
2898 | * | |
3e32cb2e | 2899 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2900 | * both res and memsw, and called css_get(). |
2901 | */ | |
2902 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2903 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2904 | { |
2905 | unsigned short old_id, new_id; | |
2906 | ||
34c00c31 LZ |
2907 | old_id = mem_cgroup_id(from); |
2908 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2909 | |
2910 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2911 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2912 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2913 | return 0; |
2914 | } | |
2915 | return -EINVAL; | |
2916 | } | |
2917 | #else | |
2918 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2919 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2920 | { |
2921 | return -EINVAL; | |
2922 | } | |
8c7c6e34 | 2923 | #endif |
d13d1443 | 2924 | |
3e32cb2e | 2925 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2926 | |
d38d2a75 | 2927 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2928 | unsigned long limit) |
628f4235 | 2929 | { |
3e32cb2e JW |
2930 | unsigned long curusage; |
2931 | unsigned long oldusage; | |
2932 | bool enlarge = false; | |
81d39c20 | 2933 | int retry_count; |
3e32cb2e | 2934 | int ret; |
81d39c20 KH |
2935 | |
2936 | /* | |
2937 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2938 | * is depends on callers. We set our retry-count to be function | |
2939 | * of # of children which we should visit in this loop. | |
2940 | */ | |
3e32cb2e JW |
2941 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2942 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2943 | |
3e32cb2e | 2944 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2945 | |
3e32cb2e | 2946 | do { |
628f4235 KH |
2947 | if (signal_pending(current)) { |
2948 | ret = -EINTR; | |
2949 | break; | |
2950 | } | |
3e32cb2e JW |
2951 | |
2952 | mutex_lock(&memcg_limit_mutex); | |
2953 | if (limit > memcg->memsw.limit) { | |
2954 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2955 | ret = -EINVAL; |
628f4235 KH |
2956 | break; |
2957 | } | |
3e32cb2e JW |
2958 | if (limit > memcg->memory.limit) |
2959 | enlarge = true; | |
2960 | ret = page_counter_limit(&memcg->memory, limit); | |
2961 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2962 | |
2963 | if (!ret) | |
2964 | break; | |
2965 | ||
b70a2a21 JW |
2966 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2967 | ||
3e32cb2e | 2968 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2969 | /* Usage is reduced ? */ |
f894ffa8 | 2970 | if (curusage >= oldusage) |
81d39c20 KH |
2971 | retry_count--; |
2972 | else | |
2973 | oldusage = curusage; | |
3e32cb2e JW |
2974 | } while (retry_count); |
2975 | ||
3c11ecf4 KH |
2976 | if (!ret && enlarge) |
2977 | memcg_oom_recover(memcg); | |
14797e23 | 2978 | |
8c7c6e34 KH |
2979 | return ret; |
2980 | } | |
2981 | ||
338c8431 | 2982 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2983 | unsigned long limit) |
8c7c6e34 | 2984 | { |
3e32cb2e JW |
2985 | unsigned long curusage; |
2986 | unsigned long oldusage; | |
2987 | bool enlarge = false; | |
81d39c20 | 2988 | int retry_count; |
3e32cb2e | 2989 | int ret; |
8c7c6e34 | 2990 | |
81d39c20 | 2991 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2992 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2993 | mem_cgroup_count_children(memcg); | |
2994 | ||
2995 | oldusage = page_counter_read(&memcg->memsw); | |
2996 | ||
2997 | do { | |
8c7c6e34 KH |
2998 | if (signal_pending(current)) { |
2999 | ret = -EINTR; | |
3000 | break; | |
3001 | } | |
3e32cb2e JW |
3002 | |
3003 | mutex_lock(&memcg_limit_mutex); | |
3004 | if (limit < memcg->memory.limit) { | |
3005 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 3006 | ret = -EINVAL; |
8c7c6e34 KH |
3007 | break; |
3008 | } | |
3e32cb2e JW |
3009 | if (limit > memcg->memsw.limit) |
3010 | enlarge = true; | |
3011 | ret = page_counter_limit(&memcg->memsw, limit); | |
3012 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
3013 | |
3014 | if (!ret) | |
3015 | break; | |
3016 | ||
b70a2a21 JW |
3017 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
3018 | ||
3e32cb2e | 3019 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 3020 | /* Usage is reduced ? */ |
8c7c6e34 | 3021 | if (curusage >= oldusage) |
628f4235 | 3022 | retry_count--; |
81d39c20 KH |
3023 | else |
3024 | oldusage = curusage; | |
3e32cb2e JW |
3025 | } while (retry_count); |
3026 | ||
3c11ecf4 KH |
3027 | if (!ret && enlarge) |
3028 | memcg_oom_recover(memcg); | |
3e32cb2e | 3029 | |
628f4235 KH |
3030 | return ret; |
3031 | } | |
3032 | ||
0608f43d AM |
3033 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
3034 | gfp_t gfp_mask, | |
3035 | unsigned long *total_scanned) | |
3036 | { | |
3037 | unsigned long nr_reclaimed = 0; | |
3038 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3039 | unsigned long reclaimed; | |
3040 | int loop = 0; | |
3041 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 3042 | unsigned long excess; |
0608f43d AM |
3043 | unsigned long nr_scanned; |
3044 | ||
3045 | if (order > 0) | |
3046 | return 0; | |
3047 | ||
3048 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
3049 | /* | |
3050 | * This loop can run a while, specially if mem_cgroup's continuously | |
3051 | * keep exceeding their soft limit and putting the system under | |
3052 | * pressure | |
3053 | */ | |
3054 | do { | |
3055 | if (next_mz) | |
3056 | mz = next_mz; | |
3057 | else | |
3058 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3059 | if (!mz) | |
3060 | break; | |
3061 | ||
3062 | nr_scanned = 0; | |
3063 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
3064 | gfp_mask, &nr_scanned); | |
3065 | nr_reclaimed += reclaimed; | |
3066 | *total_scanned += nr_scanned; | |
0a31bc97 | 3067 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3068 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3069 | |
3070 | /* | |
3071 | * If we failed to reclaim anything from this memory cgroup | |
3072 | * it is time to move on to the next cgroup | |
3073 | */ | |
3074 | next_mz = NULL; | |
bc2f2e7f VD |
3075 | if (!reclaimed) |
3076 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3077 | ||
3e32cb2e | 3078 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3079 | /* |
3080 | * One school of thought says that we should not add | |
3081 | * back the node to the tree if reclaim returns 0. | |
3082 | * But our reclaim could return 0, simply because due | |
3083 | * to priority we are exposing a smaller subset of | |
3084 | * memory to reclaim from. Consider this as a longer | |
3085 | * term TODO. | |
3086 | */ | |
3087 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3088 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3089 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3090 | css_put(&mz->memcg->css); |
3091 | loop++; | |
3092 | /* | |
3093 | * Could not reclaim anything and there are no more | |
3094 | * mem cgroups to try or we seem to be looping without | |
3095 | * reclaiming anything. | |
3096 | */ | |
3097 | if (!nr_reclaimed && | |
3098 | (next_mz == NULL || | |
3099 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3100 | break; | |
3101 | } while (!nr_reclaimed); | |
3102 | if (next_mz) | |
3103 | css_put(&next_mz->memcg->css); | |
3104 | return nr_reclaimed; | |
3105 | } | |
3106 | ||
ea280e7b TH |
3107 | /* |
3108 | * Test whether @memcg has children, dead or alive. Note that this | |
3109 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3110 | * returns %true if there are child csses according to the cgroup | |
3111 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3112 | */ | |
b5f99b53 GC |
3113 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3114 | { | |
ea280e7b TH |
3115 | bool ret; |
3116 | ||
696ac172 | 3117 | /* |
ea280e7b TH |
3118 | * The lock does not prevent addition or deletion of children, but |
3119 | * it prevents a new child from being initialized based on this | |
3120 | * parent in css_online(), so it's enough to decide whether | |
3121 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 3122 | */ |
ea280e7b TH |
3123 | lockdep_assert_held(&memcg_create_mutex); |
3124 | ||
3125 | rcu_read_lock(); | |
3126 | ret = css_next_child(NULL, &memcg->css); | |
3127 | rcu_read_unlock(); | |
3128 | return ret; | |
b5f99b53 GC |
3129 | } |
3130 | ||
c26251f9 MH |
3131 | /* |
3132 | * Reclaims as many pages from the given memcg as possible and moves | |
3133 | * the rest to the parent. | |
3134 | * | |
3135 | * Caller is responsible for holding css reference for memcg. | |
3136 | */ | |
3137 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3138 | { | |
3139 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3140 | |
c1e862c1 KH |
3141 | /* we call try-to-free pages for make this cgroup empty */ |
3142 | lru_add_drain_all(); | |
f817ed48 | 3143 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3144 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3145 | int progress; |
c1e862c1 | 3146 | |
c26251f9 MH |
3147 | if (signal_pending(current)) |
3148 | return -EINTR; | |
3149 | ||
b70a2a21 JW |
3150 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3151 | GFP_KERNEL, true); | |
c1e862c1 | 3152 | if (!progress) { |
f817ed48 | 3153 | nr_retries--; |
c1e862c1 | 3154 | /* maybe some writeback is necessary */ |
8aa7e847 | 3155 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3156 | } |
f817ed48 KH |
3157 | |
3158 | } | |
ab5196c2 MH |
3159 | |
3160 | return 0; | |
cc847582 KH |
3161 | } |
3162 | ||
6770c64e TH |
3163 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3164 | char *buf, size_t nbytes, | |
3165 | loff_t off) | |
c1e862c1 | 3166 | { |
6770c64e | 3167 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3168 | |
d8423011 MH |
3169 | if (mem_cgroup_is_root(memcg)) |
3170 | return -EINVAL; | |
6770c64e | 3171 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3172 | } |
3173 | ||
182446d0 TH |
3174 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3175 | struct cftype *cft) | |
18f59ea7 | 3176 | { |
182446d0 | 3177 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3178 | } |
3179 | ||
182446d0 TH |
3180 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3181 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3182 | { |
3183 | int retval = 0; | |
182446d0 | 3184 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3185 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3186 | |
0999821b | 3187 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
3188 | |
3189 | if (memcg->use_hierarchy == val) | |
3190 | goto out; | |
3191 | ||
18f59ea7 | 3192 | /* |
af901ca1 | 3193 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3194 | * in the child subtrees. If it is unset, then the change can |
3195 | * occur, provided the current cgroup has no children. | |
3196 | * | |
3197 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3198 | * set if there are no children. | |
3199 | */ | |
c0ff4b85 | 3200 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3201 | (val == 1 || val == 0)) { |
ea280e7b | 3202 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3203 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3204 | else |
3205 | retval = -EBUSY; | |
3206 | } else | |
3207 | retval = -EINVAL; | |
567fb435 GC |
3208 | |
3209 | out: | |
0999821b | 3210 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
3211 | |
3212 | return retval; | |
3213 | } | |
3214 | ||
3e32cb2e JW |
3215 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
3216 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
3217 | { |
3218 | struct mem_cgroup *iter; | |
3219 | long val = 0; | |
3220 | ||
3221 | /* Per-cpu values can be negative, use a signed accumulator */ | |
3222 | for_each_mem_cgroup_tree(iter, memcg) | |
3223 | val += mem_cgroup_read_stat(iter, idx); | |
3224 | ||
3225 | if (val < 0) /* race ? */ | |
3226 | val = 0; | |
3227 | return val; | |
3228 | } | |
3229 | ||
3230 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) | |
3231 | { | |
3232 | u64 val; | |
3233 | ||
3e32cb2e JW |
3234 | if (mem_cgroup_is_root(memcg)) { |
3235 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
3236 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
3237 | if (swap) | |
3238 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
3239 | } else { | |
ce00a967 | 3240 | if (!swap) |
3e32cb2e | 3241 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3242 | else |
3e32cb2e | 3243 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3244 | } |
ce00a967 JW |
3245 | return val << PAGE_SHIFT; |
3246 | } | |
3247 | ||
3e32cb2e JW |
3248 | enum { |
3249 | RES_USAGE, | |
3250 | RES_LIMIT, | |
3251 | RES_MAX_USAGE, | |
3252 | RES_FAILCNT, | |
3253 | RES_SOFT_LIMIT, | |
3254 | }; | |
ce00a967 | 3255 | |
791badbd | 3256 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3257 | struct cftype *cft) |
8cdea7c0 | 3258 | { |
182446d0 | 3259 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3260 | struct page_counter *counter; |
af36f906 | 3261 | |
3e32cb2e | 3262 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3263 | case _MEM: |
3e32cb2e JW |
3264 | counter = &memcg->memory; |
3265 | break; | |
8c7c6e34 | 3266 | case _MEMSWAP: |
3e32cb2e JW |
3267 | counter = &memcg->memsw; |
3268 | break; | |
510fc4e1 | 3269 | case _KMEM: |
3e32cb2e | 3270 | counter = &memcg->kmem; |
510fc4e1 | 3271 | break; |
8c7c6e34 KH |
3272 | default: |
3273 | BUG(); | |
8c7c6e34 | 3274 | } |
3e32cb2e JW |
3275 | |
3276 | switch (MEMFILE_ATTR(cft->private)) { | |
3277 | case RES_USAGE: | |
3278 | if (counter == &memcg->memory) | |
3279 | return mem_cgroup_usage(memcg, false); | |
3280 | if (counter == &memcg->memsw) | |
3281 | return mem_cgroup_usage(memcg, true); | |
3282 | return (u64)page_counter_read(counter) * PAGE_SIZE; | |
3283 | case RES_LIMIT: | |
3284 | return (u64)counter->limit * PAGE_SIZE; | |
3285 | case RES_MAX_USAGE: | |
3286 | return (u64)counter->watermark * PAGE_SIZE; | |
3287 | case RES_FAILCNT: | |
3288 | return counter->failcnt; | |
3289 | case RES_SOFT_LIMIT: | |
3290 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3291 | default: | |
3292 | BUG(); | |
3293 | } | |
8cdea7c0 | 3294 | } |
510fc4e1 | 3295 | |
510fc4e1 | 3296 | #ifdef CONFIG_MEMCG_KMEM |
8c0145b6 VD |
3297 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
3298 | unsigned long nr_pages) | |
d6441637 VD |
3299 | { |
3300 | int err = 0; | |
3301 | int memcg_id; | |
3302 | ||
3303 | if (memcg_kmem_is_active(memcg)) | |
3304 | return 0; | |
3305 | ||
510fc4e1 GC |
3306 | /* |
3307 | * For simplicity, we won't allow this to be disabled. It also can't | |
3308 | * be changed if the cgroup has children already, or if tasks had | |
3309 | * already joined. | |
3310 | * | |
3311 | * If tasks join before we set the limit, a person looking at | |
3312 | * kmem.usage_in_bytes will have no way to determine when it took | |
3313 | * place, which makes the value quite meaningless. | |
3314 | * | |
3315 | * After it first became limited, changes in the value of the limit are | |
3316 | * of course permitted. | |
510fc4e1 | 3317 | */ |
0999821b | 3318 | mutex_lock(&memcg_create_mutex); |
ea280e7b TH |
3319 | if (cgroup_has_tasks(memcg->css.cgroup) || |
3320 | (memcg->use_hierarchy && memcg_has_children(memcg))) | |
d6441637 VD |
3321 | err = -EBUSY; |
3322 | mutex_unlock(&memcg_create_mutex); | |
3323 | if (err) | |
3324 | goto out; | |
510fc4e1 | 3325 | |
f3bb3043 | 3326 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
3327 | if (memcg_id < 0) { |
3328 | err = memcg_id; | |
3329 | goto out; | |
3330 | } | |
3331 | ||
d6441637 | 3332 | /* |
900a38f0 VD |
3333 | * We couldn't have accounted to this cgroup, because it hasn't got |
3334 | * activated yet, so this should succeed. | |
d6441637 | 3335 | */ |
3e32cb2e | 3336 | err = page_counter_limit(&memcg->kmem, nr_pages); |
d6441637 VD |
3337 | VM_BUG_ON(err); |
3338 | ||
3339 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
3340 | /* | |
900a38f0 VD |
3341 | * A memory cgroup is considered kmem-active as soon as it gets |
3342 | * kmemcg_id. Setting the id after enabling static branching will | |
d6441637 VD |
3343 | * guarantee no one starts accounting before all call sites are |
3344 | * patched. | |
3345 | */ | |
900a38f0 | 3346 | memcg->kmemcg_id = memcg_id; |
510fc4e1 | 3347 | out: |
d6441637 | 3348 | return err; |
d6441637 VD |
3349 | } |
3350 | ||
d6441637 | 3351 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3352 | unsigned long limit) |
d6441637 VD |
3353 | { |
3354 | int ret; | |
3355 | ||
3e32cb2e | 3356 | mutex_lock(&memcg_limit_mutex); |
d6441637 | 3357 | if (!memcg_kmem_is_active(memcg)) |
3e32cb2e | 3358 | ret = memcg_activate_kmem(memcg, limit); |
d6441637 | 3359 | else |
3e32cb2e JW |
3360 | ret = page_counter_limit(&memcg->kmem, limit); |
3361 | mutex_unlock(&memcg_limit_mutex); | |
510fc4e1 GC |
3362 | return ret; |
3363 | } | |
3364 | ||
55007d84 | 3365 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 3366 | { |
55007d84 | 3367 | int ret = 0; |
510fc4e1 | 3368 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 3369 | |
d6441637 VD |
3370 | if (!parent) |
3371 | return 0; | |
55007d84 | 3372 | |
8c0145b6 | 3373 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 3374 | /* |
d6441637 VD |
3375 | * If the parent cgroup is not kmem-active now, it cannot be activated |
3376 | * after this point, because it has at least one child already. | |
55007d84 | 3377 | */ |
d6441637 | 3378 | if (memcg_kmem_is_active(parent)) |
8c0145b6 VD |
3379 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
3380 | mutex_unlock(&memcg_limit_mutex); | |
55007d84 | 3381 | return ret; |
510fc4e1 | 3382 | } |
d6441637 VD |
3383 | #else |
3384 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, | |
3e32cb2e | 3385 | unsigned long limit) |
d6441637 VD |
3386 | { |
3387 | return -EINVAL; | |
3388 | } | |
6d043990 | 3389 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 3390 | |
628f4235 KH |
3391 | /* |
3392 | * The user of this function is... | |
3393 | * RES_LIMIT. | |
3394 | */ | |
451af504 TH |
3395 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3396 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3397 | { |
451af504 | 3398 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3399 | unsigned long nr_pages; |
628f4235 KH |
3400 | int ret; |
3401 | ||
451af504 | 3402 | buf = strstrip(buf); |
3e32cb2e JW |
3403 | ret = page_counter_memparse(buf, &nr_pages); |
3404 | if (ret) | |
3405 | return ret; | |
af36f906 | 3406 | |
3e32cb2e | 3407 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3408 | case RES_LIMIT: |
4b3bde4c BS |
3409 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3410 | ret = -EINVAL; | |
3411 | break; | |
3412 | } | |
3e32cb2e JW |
3413 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3414 | case _MEM: | |
3415 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3416 | break; |
3e32cb2e JW |
3417 | case _MEMSWAP: |
3418 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3419 | break; |
3e32cb2e JW |
3420 | case _KMEM: |
3421 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3422 | break; | |
3423 | } | |
296c81d8 | 3424 | break; |
3e32cb2e JW |
3425 | case RES_SOFT_LIMIT: |
3426 | memcg->soft_limit = nr_pages; | |
3427 | ret = 0; | |
628f4235 KH |
3428 | break; |
3429 | } | |
451af504 | 3430 | return ret ?: nbytes; |
8cdea7c0 BS |
3431 | } |
3432 | ||
6770c64e TH |
3433 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3434 | size_t nbytes, loff_t off) | |
c84872e1 | 3435 | { |
6770c64e | 3436 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3437 | struct page_counter *counter; |
c84872e1 | 3438 | |
3e32cb2e JW |
3439 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3440 | case _MEM: | |
3441 | counter = &memcg->memory; | |
3442 | break; | |
3443 | case _MEMSWAP: | |
3444 | counter = &memcg->memsw; | |
3445 | break; | |
3446 | case _KMEM: | |
3447 | counter = &memcg->kmem; | |
3448 | break; | |
3449 | default: | |
3450 | BUG(); | |
3451 | } | |
af36f906 | 3452 | |
3e32cb2e | 3453 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3454 | case RES_MAX_USAGE: |
3e32cb2e | 3455 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3456 | break; |
3457 | case RES_FAILCNT: | |
3e32cb2e | 3458 | counter->failcnt = 0; |
29f2a4da | 3459 | break; |
3e32cb2e JW |
3460 | default: |
3461 | BUG(); | |
29f2a4da | 3462 | } |
f64c3f54 | 3463 | |
6770c64e | 3464 | return nbytes; |
c84872e1 PE |
3465 | } |
3466 | ||
182446d0 | 3467 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3468 | struct cftype *cft) |
3469 | { | |
182446d0 | 3470 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3471 | } |
3472 | ||
02491447 | 3473 | #ifdef CONFIG_MMU |
182446d0 | 3474 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3475 | struct cftype *cft, u64 val) |
3476 | { | |
182446d0 | 3477 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 DN |
3478 | |
3479 | if (val >= (1 << NR_MOVE_TYPE)) | |
3480 | return -EINVAL; | |
ee5e8472 | 3481 | |
7dc74be0 | 3482 | /* |
ee5e8472 GC |
3483 | * No kind of locking is needed in here, because ->can_attach() will |
3484 | * check this value once in the beginning of the process, and then carry | |
3485 | * on with stale data. This means that changes to this value will only | |
3486 | * affect task migrations starting after the change. | |
7dc74be0 | 3487 | */ |
c0ff4b85 | 3488 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3489 | return 0; |
3490 | } | |
02491447 | 3491 | #else |
182446d0 | 3492 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3493 | struct cftype *cft, u64 val) |
3494 | { | |
3495 | return -ENOSYS; | |
3496 | } | |
3497 | #endif | |
7dc74be0 | 3498 | |
406eb0c9 | 3499 | #ifdef CONFIG_NUMA |
2da8ca82 | 3500 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3501 | { |
25485de6 GT |
3502 | struct numa_stat { |
3503 | const char *name; | |
3504 | unsigned int lru_mask; | |
3505 | }; | |
3506 | ||
3507 | static const struct numa_stat stats[] = { | |
3508 | { "total", LRU_ALL }, | |
3509 | { "file", LRU_ALL_FILE }, | |
3510 | { "anon", LRU_ALL_ANON }, | |
3511 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3512 | }; | |
3513 | const struct numa_stat *stat; | |
406eb0c9 | 3514 | int nid; |
25485de6 | 3515 | unsigned long nr; |
2da8ca82 | 3516 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3517 | |
25485de6 GT |
3518 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3519 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3520 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3521 | for_each_node_state(nid, N_MEMORY) { | |
3522 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3523 | stat->lru_mask); | |
3524 | seq_printf(m, " N%d=%lu", nid, nr); | |
3525 | } | |
3526 | seq_putc(m, '\n'); | |
406eb0c9 | 3527 | } |
406eb0c9 | 3528 | |
071aee13 YH |
3529 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3530 | struct mem_cgroup *iter; | |
3531 | ||
3532 | nr = 0; | |
3533 | for_each_mem_cgroup_tree(iter, memcg) | |
3534 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3535 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3536 | for_each_node_state(nid, N_MEMORY) { | |
3537 | nr = 0; | |
3538 | for_each_mem_cgroup_tree(iter, memcg) | |
3539 | nr += mem_cgroup_node_nr_lru_pages( | |
3540 | iter, nid, stat->lru_mask); | |
3541 | seq_printf(m, " N%d=%lu", nid, nr); | |
3542 | } | |
3543 | seq_putc(m, '\n'); | |
406eb0c9 | 3544 | } |
406eb0c9 | 3545 | |
406eb0c9 YH |
3546 | return 0; |
3547 | } | |
3548 | #endif /* CONFIG_NUMA */ | |
3549 | ||
2da8ca82 | 3550 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3551 | { |
2da8ca82 | 3552 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3553 | unsigned long memory, memsw; |
af7c4b0e JW |
3554 | struct mem_cgroup *mi; |
3555 | unsigned int i; | |
406eb0c9 | 3556 | |
70bc068c RS |
3557 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3558 | ||
af7c4b0e | 3559 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 3560 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3561 | continue; |
af7c4b0e JW |
3562 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
3563 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 3564 | } |
7b854121 | 3565 | |
af7c4b0e JW |
3566 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3567 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3568 | mem_cgroup_read_events(memcg, i)); | |
3569 | ||
3570 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3571 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3572 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3573 | ||
14067bb3 | 3574 | /* Hierarchical information */ |
3e32cb2e JW |
3575 | memory = memsw = PAGE_COUNTER_MAX; |
3576 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3577 | memory = min(memory, mi->memory.limit); | |
3578 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3579 | } |
3e32cb2e JW |
3580 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3581 | (u64)memory * PAGE_SIZE); | |
3582 | if (do_swap_account) | |
3583 | seq_printf(m, "hierarchical_memsw_limit %llu\n", | |
3584 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3585 | |
af7c4b0e JW |
3586 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
3587 | long long val = 0; | |
3588 | ||
bff6bb83 | 3589 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 3590 | continue; |
af7c4b0e JW |
3591 | for_each_mem_cgroup_tree(mi, memcg) |
3592 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
3593 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
3594 | } | |
3595 | ||
3596 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3597 | unsigned long long val = 0; | |
3598 | ||
3599 | for_each_mem_cgroup_tree(mi, memcg) | |
3600 | val += mem_cgroup_read_events(mi, i); | |
3601 | seq_printf(m, "total_%s %llu\n", | |
3602 | mem_cgroup_events_names[i], val); | |
3603 | } | |
3604 | ||
3605 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3606 | unsigned long long val = 0; | |
3607 | ||
3608 | for_each_mem_cgroup_tree(mi, memcg) | |
3609 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3610 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3611 | } |
14067bb3 | 3612 | |
7f016ee8 | 3613 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3614 | { |
3615 | int nid, zid; | |
3616 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3617 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3618 | unsigned long recent_rotated[2] = {0, 0}; |
3619 | unsigned long recent_scanned[2] = {0, 0}; | |
3620 | ||
3621 | for_each_online_node(nid) | |
3622 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3623 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3624 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3625 | |
89abfab1 HD |
3626 | recent_rotated[0] += rstat->recent_rotated[0]; |
3627 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3628 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3629 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3630 | } |
78ccf5b5 JW |
3631 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3632 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3633 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3634 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3635 | } |
3636 | #endif | |
3637 | ||
d2ceb9b7 KH |
3638 | return 0; |
3639 | } | |
3640 | ||
182446d0 TH |
3641 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3642 | struct cftype *cft) | |
a7885eb8 | 3643 | { |
182446d0 | 3644 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3645 | |
1f4c025b | 3646 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3647 | } |
3648 | ||
182446d0 TH |
3649 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3650 | struct cftype *cft, u64 val) | |
a7885eb8 | 3651 | { |
182446d0 | 3652 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3653 | |
3dae7fec | 3654 | if (val > 100) |
a7885eb8 KM |
3655 | return -EINVAL; |
3656 | ||
14208b0e | 3657 | if (css->parent) |
3dae7fec JW |
3658 | memcg->swappiness = val; |
3659 | else | |
3660 | vm_swappiness = val; | |
068b38c1 | 3661 | |
a7885eb8 KM |
3662 | return 0; |
3663 | } | |
3664 | ||
2e72b634 KS |
3665 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3666 | { | |
3667 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3668 | unsigned long usage; |
2e72b634 KS |
3669 | int i; |
3670 | ||
3671 | rcu_read_lock(); | |
3672 | if (!swap) | |
2c488db2 | 3673 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3674 | else |
2c488db2 | 3675 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3676 | |
3677 | if (!t) | |
3678 | goto unlock; | |
3679 | ||
ce00a967 | 3680 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3681 | |
3682 | /* | |
748dad36 | 3683 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3684 | * If it's not true, a threshold was crossed after last |
3685 | * call of __mem_cgroup_threshold(). | |
3686 | */ | |
5407a562 | 3687 | i = t->current_threshold; |
2e72b634 KS |
3688 | |
3689 | /* | |
3690 | * Iterate backward over array of thresholds starting from | |
3691 | * current_threshold and check if a threshold is crossed. | |
3692 | * If none of thresholds below usage is crossed, we read | |
3693 | * only one element of the array here. | |
3694 | */ | |
3695 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3696 | eventfd_signal(t->entries[i].eventfd, 1); | |
3697 | ||
3698 | /* i = current_threshold + 1 */ | |
3699 | i++; | |
3700 | ||
3701 | /* | |
3702 | * Iterate forward over array of thresholds starting from | |
3703 | * current_threshold+1 and check if a threshold is crossed. | |
3704 | * If none of thresholds above usage is crossed, we read | |
3705 | * only one element of the array here. | |
3706 | */ | |
3707 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3708 | eventfd_signal(t->entries[i].eventfd, 1); | |
3709 | ||
3710 | /* Update current_threshold */ | |
5407a562 | 3711 | t->current_threshold = i - 1; |
2e72b634 KS |
3712 | unlock: |
3713 | rcu_read_unlock(); | |
3714 | } | |
3715 | ||
3716 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3717 | { | |
ad4ca5f4 KS |
3718 | while (memcg) { |
3719 | __mem_cgroup_threshold(memcg, false); | |
3720 | if (do_swap_account) | |
3721 | __mem_cgroup_threshold(memcg, true); | |
3722 | ||
3723 | memcg = parent_mem_cgroup(memcg); | |
3724 | } | |
2e72b634 KS |
3725 | } |
3726 | ||
3727 | static int compare_thresholds(const void *a, const void *b) | |
3728 | { | |
3729 | const struct mem_cgroup_threshold *_a = a; | |
3730 | const struct mem_cgroup_threshold *_b = b; | |
3731 | ||
2bff24a3 GT |
3732 | if (_a->threshold > _b->threshold) |
3733 | return 1; | |
3734 | ||
3735 | if (_a->threshold < _b->threshold) | |
3736 | return -1; | |
3737 | ||
3738 | return 0; | |
2e72b634 KS |
3739 | } |
3740 | ||
c0ff4b85 | 3741 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3742 | { |
3743 | struct mem_cgroup_eventfd_list *ev; | |
3744 | ||
2bcf2e92 MH |
3745 | spin_lock(&memcg_oom_lock); |
3746 | ||
c0ff4b85 | 3747 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3748 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3749 | |
3750 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3751 | return 0; |
3752 | } | |
3753 | ||
c0ff4b85 | 3754 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3755 | { |
7d74b06f KH |
3756 | struct mem_cgroup *iter; |
3757 | ||
c0ff4b85 | 3758 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3759 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3760 | } |
3761 | ||
59b6f873 | 3762 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3763 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3764 | { |
2c488db2 KS |
3765 | struct mem_cgroup_thresholds *thresholds; |
3766 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3767 | unsigned long threshold; |
3768 | unsigned long usage; | |
2c488db2 | 3769 | int i, size, ret; |
2e72b634 | 3770 | |
3e32cb2e | 3771 | ret = page_counter_memparse(args, &threshold); |
2e72b634 KS |
3772 | if (ret) |
3773 | return ret; | |
3774 | ||
3775 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3776 | |
05b84301 | 3777 | if (type == _MEM) { |
2c488db2 | 3778 | thresholds = &memcg->thresholds; |
ce00a967 | 3779 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3780 | } else if (type == _MEMSWAP) { |
2c488db2 | 3781 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3782 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3783 | } else |
2e72b634 KS |
3784 | BUG(); |
3785 | ||
2e72b634 | 3786 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3787 | if (thresholds->primary) |
2e72b634 KS |
3788 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3789 | ||
2c488db2 | 3790 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3791 | |
3792 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3793 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3794 | GFP_KERNEL); |
2c488db2 | 3795 | if (!new) { |
2e72b634 KS |
3796 | ret = -ENOMEM; |
3797 | goto unlock; | |
3798 | } | |
2c488db2 | 3799 | new->size = size; |
2e72b634 KS |
3800 | |
3801 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3802 | if (thresholds->primary) { |
3803 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3804 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3805 | } |
3806 | ||
2e72b634 | 3807 | /* Add new threshold */ |
2c488db2 KS |
3808 | new->entries[size - 1].eventfd = eventfd; |
3809 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3810 | |
3811 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3812 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3813 | compare_thresholds, NULL); |
3814 | ||
3815 | /* Find current threshold */ | |
2c488db2 | 3816 | new->current_threshold = -1; |
2e72b634 | 3817 | for (i = 0; i < size; i++) { |
748dad36 | 3818 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3819 | /* |
2c488db2 KS |
3820 | * new->current_threshold will not be used until |
3821 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3822 | * it here. |
3823 | */ | |
2c488db2 | 3824 | ++new->current_threshold; |
748dad36 SZ |
3825 | } else |
3826 | break; | |
2e72b634 KS |
3827 | } |
3828 | ||
2c488db2 KS |
3829 | /* Free old spare buffer and save old primary buffer as spare */ |
3830 | kfree(thresholds->spare); | |
3831 | thresholds->spare = thresholds->primary; | |
3832 | ||
3833 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3834 | |
907860ed | 3835 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3836 | synchronize_rcu(); |
3837 | ||
2e72b634 KS |
3838 | unlock: |
3839 | mutex_unlock(&memcg->thresholds_lock); | |
3840 | ||
3841 | return ret; | |
3842 | } | |
3843 | ||
59b6f873 | 3844 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3845 | struct eventfd_ctx *eventfd, const char *args) |
3846 | { | |
59b6f873 | 3847 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3848 | } |
3849 | ||
59b6f873 | 3850 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3851 | struct eventfd_ctx *eventfd, const char *args) |
3852 | { | |
59b6f873 | 3853 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3854 | } |
3855 | ||
59b6f873 | 3856 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3857 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3858 | { |
2c488db2 KS |
3859 | struct mem_cgroup_thresholds *thresholds; |
3860 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3861 | unsigned long usage; |
2c488db2 | 3862 | int i, j, size; |
2e72b634 KS |
3863 | |
3864 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3865 | |
3866 | if (type == _MEM) { | |
2c488db2 | 3867 | thresholds = &memcg->thresholds; |
ce00a967 | 3868 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3869 | } else if (type == _MEMSWAP) { |
2c488db2 | 3870 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3871 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3872 | } else |
2e72b634 KS |
3873 | BUG(); |
3874 | ||
371528ca AV |
3875 | if (!thresholds->primary) |
3876 | goto unlock; | |
3877 | ||
2e72b634 KS |
3878 | /* Check if a threshold crossed before removing */ |
3879 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3880 | ||
3881 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3882 | size = 0; |
3883 | for (i = 0; i < thresholds->primary->size; i++) { | |
3884 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3885 | size++; |
3886 | } | |
3887 | ||
2c488db2 | 3888 | new = thresholds->spare; |
907860ed | 3889 | |
2e72b634 KS |
3890 | /* Set thresholds array to NULL if we don't have thresholds */ |
3891 | if (!size) { | |
2c488db2 KS |
3892 | kfree(new); |
3893 | new = NULL; | |
907860ed | 3894 | goto swap_buffers; |
2e72b634 KS |
3895 | } |
3896 | ||
2c488db2 | 3897 | new->size = size; |
2e72b634 KS |
3898 | |
3899 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3900 | new->current_threshold = -1; |
3901 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3902 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3903 | continue; |
3904 | ||
2c488db2 | 3905 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3906 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3907 | /* |
2c488db2 | 3908 | * new->current_threshold will not be used |
2e72b634 KS |
3909 | * until rcu_assign_pointer(), so it's safe to increment |
3910 | * it here. | |
3911 | */ | |
2c488db2 | 3912 | ++new->current_threshold; |
2e72b634 KS |
3913 | } |
3914 | j++; | |
3915 | } | |
3916 | ||
907860ed | 3917 | swap_buffers: |
2c488db2 KS |
3918 | /* Swap primary and spare array */ |
3919 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
3920 | /* If all events are unregistered, free the spare array */ |
3921 | if (!new) { | |
3922 | kfree(thresholds->spare); | |
3923 | thresholds->spare = NULL; | |
3924 | } | |
3925 | ||
2c488db2 | 3926 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3927 | |
907860ed | 3928 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3929 | synchronize_rcu(); |
371528ca | 3930 | unlock: |
2e72b634 | 3931 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3932 | } |
c1e862c1 | 3933 | |
59b6f873 | 3934 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3935 | struct eventfd_ctx *eventfd) |
3936 | { | |
59b6f873 | 3937 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3938 | } |
3939 | ||
59b6f873 | 3940 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3941 | struct eventfd_ctx *eventfd) |
3942 | { | |
59b6f873 | 3943 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3944 | } |
3945 | ||
59b6f873 | 3946 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3947 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3948 | { |
9490ff27 | 3949 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3950 | |
9490ff27 KH |
3951 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3952 | if (!event) | |
3953 | return -ENOMEM; | |
3954 | ||
1af8efe9 | 3955 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3956 | |
3957 | event->eventfd = eventfd; | |
3958 | list_add(&event->list, &memcg->oom_notify); | |
3959 | ||
3960 | /* already in OOM ? */ | |
79dfdacc | 3961 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 3962 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3963 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3964 | |
3965 | return 0; | |
3966 | } | |
3967 | ||
59b6f873 | 3968 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3969 | struct eventfd_ctx *eventfd) |
9490ff27 | 3970 | { |
9490ff27 | 3971 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3972 | |
1af8efe9 | 3973 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3974 | |
c0ff4b85 | 3975 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3976 | if (ev->eventfd == eventfd) { |
3977 | list_del(&ev->list); | |
3978 | kfree(ev); | |
3979 | } | |
3980 | } | |
3981 | ||
1af8efe9 | 3982 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3983 | } |
3984 | ||
2da8ca82 | 3985 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3986 | { |
2da8ca82 | 3987 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3988 | |
791badbd TH |
3989 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
3990 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); | |
3c11ecf4 KH |
3991 | return 0; |
3992 | } | |
3993 | ||
182446d0 | 3994 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3995 | struct cftype *cft, u64 val) |
3996 | { | |
182446d0 | 3997 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3998 | |
3999 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4000 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4001 | return -EINVAL; |
4002 | ||
c0ff4b85 | 4003 | memcg->oom_kill_disable = val; |
4d845ebf | 4004 | if (!val) |
c0ff4b85 | 4005 | memcg_oom_recover(memcg); |
3dae7fec | 4006 | |
3c11ecf4 KH |
4007 | return 0; |
4008 | } | |
4009 | ||
c255a458 | 4010 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4011 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4012 | { |
55007d84 GC |
4013 | int ret; |
4014 | ||
55007d84 GC |
4015 | ret = memcg_propagate_kmem(memcg); |
4016 | if (ret) | |
4017 | return ret; | |
2633d7a0 | 4018 | |
1d62e436 | 4019 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 4020 | } |
e5671dfa | 4021 | |
10d5ebf4 | 4022 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
d1a4c0b3 | 4023 | { |
d5b3cf71 | 4024 | memcg_destroy_kmem_caches(memcg); |
1d62e436 | 4025 | mem_cgroup_sockets_destroy(memcg); |
10d5ebf4 | 4026 | } |
e5671dfa | 4027 | #else |
cbe128e3 | 4028 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4029 | { |
4030 | return 0; | |
4031 | } | |
d1a4c0b3 | 4032 | |
10d5ebf4 LZ |
4033 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
4034 | { | |
4035 | } | |
e5671dfa GC |
4036 | #endif |
4037 | ||
3bc942f3 TH |
4038 | /* |
4039 | * DO NOT USE IN NEW FILES. | |
4040 | * | |
4041 | * "cgroup.event_control" implementation. | |
4042 | * | |
4043 | * This is way over-engineered. It tries to support fully configurable | |
4044 | * events for each user. Such level of flexibility is completely | |
4045 | * unnecessary especially in the light of the planned unified hierarchy. | |
4046 | * | |
4047 | * Please deprecate this and replace with something simpler if at all | |
4048 | * possible. | |
4049 | */ | |
4050 | ||
79bd9814 TH |
4051 | /* |
4052 | * Unregister event and free resources. | |
4053 | * | |
4054 | * Gets called from workqueue. | |
4055 | */ | |
3bc942f3 | 4056 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4057 | { |
3bc942f3 TH |
4058 | struct mem_cgroup_event *event = |
4059 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4060 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4061 | |
4062 | remove_wait_queue(event->wqh, &event->wait); | |
4063 | ||
59b6f873 | 4064 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4065 | |
4066 | /* Notify userspace the event is going away. */ | |
4067 | eventfd_signal(event->eventfd, 1); | |
4068 | ||
4069 | eventfd_ctx_put(event->eventfd); | |
4070 | kfree(event); | |
59b6f873 | 4071 | css_put(&memcg->css); |
79bd9814 TH |
4072 | } |
4073 | ||
4074 | /* | |
4075 | * Gets called on POLLHUP on eventfd when user closes it. | |
4076 | * | |
4077 | * Called with wqh->lock held and interrupts disabled. | |
4078 | */ | |
3bc942f3 TH |
4079 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
4080 | int sync, void *key) | |
79bd9814 | 4081 | { |
3bc942f3 TH |
4082 | struct mem_cgroup_event *event = |
4083 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4084 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4085 | unsigned long flags = (unsigned long)key; |
4086 | ||
4087 | if (flags & POLLHUP) { | |
4088 | /* | |
4089 | * If the event has been detached at cgroup removal, we | |
4090 | * can simply return knowing the other side will cleanup | |
4091 | * for us. | |
4092 | * | |
4093 | * We can't race against event freeing since the other | |
4094 | * side will require wqh->lock via remove_wait_queue(), | |
4095 | * which we hold. | |
4096 | */ | |
fba94807 | 4097 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4098 | if (!list_empty(&event->list)) { |
4099 | list_del_init(&event->list); | |
4100 | /* | |
4101 | * We are in atomic context, but cgroup_event_remove() | |
4102 | * may sleep, so we have to call it in workqueue. | |
4103 | */ | |
4104 | schedule_work(&event->remove); | |
4105 | } | |
fba94807 | 4106 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4107 | } |
4108 | ||
4109 | return 0; | |
4110 | } | |
4111 | ||
3bc942f3 | 4112 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4113 | wait_queue_head_t *wqh, poll_table *pt) |
4114 | { | |
3bc942f3 TH |
4115 | struct mem_cgroup_event *event = |
4116 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4117 | |
4118 | event->wqh = wqh; | |
4119 | add_wait_queue(wqh, &event->wait); | |
4120 | } | |
4121 | ||
4122 | /* | |
3bc942f3 TH |
4123 | * DO NOT USE IN NEW FILES. |
4124 | * | |
79bd9814 TH |
4125 | * Parse input and register new cgroup event handler. |
4126 | * | |
4127 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4128 | * Interpretation of args is defined by control file implementation. | |
4129 | */ | |
451af504 TH |
4130 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4131 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4132 | { |
451af504 | 4133 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4134 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4135 | struct mem_cgroup_event *event; |
79bd9814 TH |
4136 | struct cgroup_subsys_state *cfile_css; |
4137 | unsigned int efd, cfd; | |
4138 | struct fd efile; | |
4139 | struct fd cfile; | |
fba94807 | 4140 | const char *name; |
79bd9814 TH |
4141 | char *endp; |
4142 | int ret; | |
4143 | ||
451af504 TH |
4144 | buf = strstrip(buf); |
4145 | ||
4146 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4147 | if (*endp != ' ') |
4148 | return -EINVAL; | |
451af504 | 4149 | buf = endp + 1; |
79bd9814 | 4150 | |
451af504 | 4151 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4152 | if ((*endp != ' ') && (*endp != '\0')) |
4153 | return -EINVAL; | |
451af504 | 4154 | buf = endp + 1; |
79bd9814 TH |
4155 | |
4156 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4157 | if (!event) | |
4158 | return -ENOMEM; | |
4159 | ||
59b6f873 | 4160 | event->memcg = memcg; |
79bd9814 | 4161 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4162 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4163 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4164 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4165 | |
4166 | efile = fdget(efd); | |
4167 | if (!efile.file) { | |
4168 | ret = -EBADF; | |
4169 | goto out_kfree; | |
4170 | } | |
4171 | ||
4172 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4173 | if (IS_ERR(event->eventfd)) { | |
4174 | ret = PTR_ERR(event->eventfd); | |
4175 | goto out_put_efile; | |
4176 | } | |
4177 | ||
4178 | cfile = fdget(cfd); | |
4179 | if (!cfile.file) { | |
4180 | ret = -EBADF; | |
4181 | goto out_put_eventfd; | |
4182 | } | |
4183 | ||
4184 | /* the process need read permission on control file */ | |
4185 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4186 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4187 | if (ret < 0) | |
4188 | goto out_put_cfile; | |
4189 | ||
fba94807 TH |
4190 | /* |
4191 | * Determine the event callbacks and set them in @event. This used | |
4192 | * to be done via struct cftype but cgroup core no longer knows | |
4193 | * about these events. The following is crude but the whole thing | |
4194 | * is for compatibility anyway. | |
3bc942f3 TH |
4195 | * |
4196 | * DO NOT ADD NEW FILES. | |
fba94807 | 4197 | */ |
b583043e | 4198 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4199 | |
4200 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4201 | event->register_event = mem_cgroup_usage_register_event; | |
4202 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4203 | } else if (!strcmp(name, "memory.oom_control")) { | |
4204 | event->register_event = mem_cgroup_oom_register_event; | |
4205 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4206 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4207 | event->register_event = vmpressure_register_event; | |
4208 | event->unregister_event = vmpressure_unregister_event; | |
4209 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4210 | event->register_event = memsw_cgroup_usage_register_event; |
4211 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4212 | } else { |
4213 | ret = -EINVAL; | |
4214 | goto out_put_cfile; | |
4215 | } | |
4216 | ||
79bd9814 | 4217 | /* |
b5557c4c TH |
4218 | * Verify @cfile should belong to @css. Also, remaining events are |
4219 | * automatically removed on cgroup destruction but the removal is | |
4220 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4221 | */ |
b583043e | 4222 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4223 | &memory_cgrp_subsys); |
79bd9814 | 4224 | ret = -EINVAL; |
5a17f543 | 4225 | if (IS_ERR(cfile_css)) |
79bd9814 | 4226 | goto out_put_cfile; |
5a17f543 TH |
4227 | if (cfile_css != css) { |
4228 | css_put(cfile_css); | |
79bd9814 | 4229 | goto out_put_cfile; |
5a17f543 | 4230 | } |
79bd9814 | 4231 | |
451af504 | 4232 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4233 | if (ret) |
4234 | goto out_put_css; | |
4235 | ||
4236 | efile.file->f_op->poll(efile.file, &event->pt); | |
4237 | ||
fba94807 TH |
4238 | spin_lock(&memcg->event_list_lock); |
4239 | list_add(&event->list, &memcg->event_list); | |
4240 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4241 | |
4242 | fdput(cfile); | |
4243 | fdput(efile); | |
4244 | ||
451af504 | 4245 | return nbytes; |
79bd9814 TH |
4246 | |
4247 | out_put_css: | |
b5557c4c | 4248 | css_put(css); |
79bd9814 TH |
4249 | out_put_cfile: |
4250 | fdput(cfile); | |
4251 | out_put_eventfd: | |
4252 | eventfd_ctx_put(event->eventfd); | |
4253 | out_put_efile: | |
4254 | fdput(efile); | |
4255 | out_kfree: | |
4256 | kfree(event); | |
4257 | ||
4258 | return ret; | |
4259 | } | |
4260 | ||
8cdea7c0 BS |
4261 | static struct cftype mem_cgroup_files[] = { |
4262 | { | |
0eea1030 | 4263 | .name = "usage_in_bytes", |
8c7c6e34 | 4264 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4265 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4266 | }, |
c84872e1 PE |
4267 | { |
4268 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4269 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4270 | .write = mem_cgroup_reset, |
791badbd | 4271 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4272 | }, |
8cdea7c0 | 4273 | { |
0eea1030 | 4274 | .name = "limit_in_bytes", |
8c7c6e34 | 4275 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4276 | .write = mem_cgroup_write, |
791badbd | 4277 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4278 | }, |
296c81d8 BS |
4279 | { |
4280 | .name = "soft_limit_in_bytes", | |
4281 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4282 | .write = mem_cgroup_write, |
791badbd | 4283 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4284 | }, |
8cdea7c0 BS |
4285 | { |
4286 | .name = "failcnt", | |
8c7c6e34 | 4287 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4288 | .write = mem_cgroup_reset, |
791badbd | 4289 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4290 | }, |
d2ceb9b7 KH |
4291 | { |
4292 | .name = "stat", | |
2da8ca82 | 4293 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4294 | }, |
c1e862c1 KH |
4295 | { |
4296 | .name = "force_empty", | |
6770c64e | 4297 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4298 | }, |
18f59ea7 BS |
4299 | { |
4300 | .name = "use_hierarchy", | |
4301 | .write_u64 = mem_cgroup_hierarchy_write, | |
4302 | .read_u64 = mem_cgroup_hierarchy_read, | |
4303 | }, | |
79bd9814 | 4304 | { |
3bc942f3 | 4305 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4306 | .write = memcg_write_event_control, |
79bd9814 TH |
4307 | .flags = CFTYPE_NO_PREFIX, |
4308 | .mode = S_IWUGO, | |
4309 | }, | |
a7885eb8 KM |
4310 | { |
4311 | .name = "swappiness", | |
4312 | .read_u64 = mem_cgroup_swappiness_read, | |
4313 | .write_u64 = mem_cgroup_swappiness_write, | |
4314 | }, | |
7dc74be0 DN |
4315 | { |
4316 | .name = "move_charge_at_immigrate", | |
4317 | .read_u64 = mem_cgroup_move_charge_read, | |
4318 | .write_u64 = mem_cgroup_move_charge_write, | |
4319 | }, | |
9490ff27 KH |
4320 | { |
4321 | .name = "oom_control", | |
2da8ca82 | 4322 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4323 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4324 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4325 | }, | |
70ddf637 AV |
4326 | { |
4327 | .name = "pressure_level", | |
70ddf637 | 4328 | }, |
406eb0c9 YH |
4329 | #ifdef CONFIG_NUMA |
4330 | { | |
4331 | .name = "numa_stat", | |
2da8ca82 | 4332 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4333 | }, |
4334 | #endif | |
510fc4e1 GC |
4335 | #ifdef CONFIG_MEMCG_KMEM |
4336 | { | |
4337 | .name = "kmem.limit_in_bytes", | |
4338 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4339 | .write = mem_cgroup_write, |
791badbd | 4340 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4341 | }, |
4342 | { | |
4343 | .name = "kmem.usage_in_bytes", | |
4344 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4345 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4346 | }, |
4347 | { | |
4348 | .name = "kmem.failcnt", | |
4349 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4350 | .write = mem_cgroup_reset, |
791badbd | 4351 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4352 | }, |
4353 | { | |
4354 | .name = "kmem.max_usage_in_bytes", | |
4355 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4356 | .write = mem_cgroup_reset, |
791badbd | 4357 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4358 | }, |
749c5415 GC |
4359 | #ifdef CONFIG_SLABINFO |
4360 | { | |
4361 | .name = "kmem.slabinfo", | |
b047501c VD |
4362 | .seq_start = slab_start, |
4363 | .seq_next = slab_next, | |
4364 | .seq_stop = slab_stop, | |
4365 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4366 | }, |
4367 | #endif | |
8c7c6e34 | 4368 | #endif |
6bc10349 | 4369 | { }, /* terminate */ |
af36f906 | 4370 | }; |
8c7c6e34 | 4371 | |
2d11085e MH |
4372 | #ifdef CONFIG_MEMCG_SWAP |
4373 | static struct cftype memsw_cgroup_files[] = { | |
4374 | { | |
4375 | .name = "memsw.usage_in_bytes", | |
4376 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
791badbd | 4377 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4378 | }, |
4379 | { | |
4380 | .name = "memsw.max_usage_in_bytes", | |
4381 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770c64e | 4382 | .write = mem_cgroup_reset, |
791badbd | 4383 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4384 | }, |
4385 | { | |
4386 | .name = "memsw.limit_in_bytes", | |
4387 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
451af504 | 4388 | .write = mem_cgroup_write, |
791badbd | 4389 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4390 | }, |
4391 | { | |
4392 | .name = "memsw.failcnt", | |
4393 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6770c64e | 4394 | .write = mem_cgroup_reset, |
791badbd | 4395 | .read_u64 = mem_cgroup_read_u64, |
2d11085e MH |
4396 | }, |
4397 | { }, /* terminate */ | |
4398 | }; | |
4399 | #endif | |
c0ff4b85 | 4400 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4401 | { |
4402 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4403 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4404 | int zone, tmp = node; |
1ecaab2b KH |
4405 | /* |
4406 | * This routine is called against possible nodes. | |
4407 | * But it's BUG to call kmalloc() against offline node. | |
4408 | * | |
4409 | * TODO: this routine can waste much memory for nodes which will | |
4410 | * never be onlined. It's better to use memory hotplug callback | |
4411 | * function. | |
4412 | */ | |
41e3355d KH |
4413 | if (!node_state(node, N_NORMAL_MEMORY)) |
4414 | tmp = -1; | |
17295c88 | 4415 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4416 | if (!pn) |
4417 | return 1; | |
1ecaab2b | 4418 | |
1ecaab2b KH |
4419 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4420 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4421 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4422 | mz->usage_in_excess = 0; |
4423 | mz->on_tree = false; | |
d79154bb | 4424 | mz->memcg = memcg; |
1ecaab2b | 4425 | } |
54f72fe0 | 4426 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4427 | return 0; |
4428 | } | |
4429 | ||
c0ff4b85 | 4430 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4431 | { |
54f72fe0 | 4432 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4433 | } |
4434 | ||
33327948 KH |
4435 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4436 | { | |
d79154bb | 4437 | struct mem_cgroup *memcg; |
8ff69e2c | 4438 | size_t size; |
33327948 | 4439 | |
8ff69e2c VD |
4440 | size = sizeof(struct mem_cgroup); |
4441 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4442 | |
8ff69e2c | 4443 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4444 | if (!memcg) |
e7bbcdf3 DC |
4445 | return NULL; |
4446 | ||
d79154bb HD |
4447 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4448 | if (!memcg->stat) | |
d2e61b8d | 4449 | goto out_free; |
d79154bb HD |
4450 | spin_lock_init(&memcg->pcp_counter_lock); |
4451 | return memcg; | |
d2e61b8d DC |
4452 | |
4453 | out_free: | |
8ff69e2c | 4454 | kfree(memcg); |
d2e61b8d | 4455 | return NULL; |
33327948 KH |
4456 | } |
4457 | ||
59927fb9 | 4458 | /* |
c8b2a36f GC |
4459 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4460 | * (scanning all at force_empty is too costly...) | |
4461 | * | |
4462 | * Instead of clearing all references at force_empty, we remember | |
4463 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4464 | * it goes down to 0. | |
4465 | * | |
4466 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4467 | */ |
c8b2a36f GC |
4468 | |
4469 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4470 | { |
c8b2a36f | 4471 | int node; |
59927fb9 | 4472 | |
bb4cc1a8 | 4473 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4474 | |
4475 | for_each_node(node) | |
4476 | free_mem_cgroup_per_zone_info(memcg, node); | |
4477 | ||
4478 | free_percpu(memcg->stat); | |
4479 | ||
a8964b9b | 4480 | disarm_static_keys(memcg); |
8ff69e2c | 4481 | kfree(memcg); |
59927fb9 | 4482 | } |
3afe36b1 | 4483 | |
7bcc1bb1 DN |
4484 | /* |
4485 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4486 | */ | |
e1aab161 | 4487 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4488 | { |
3e32cb2e | 4489 | if (!memcg->memory.parent) |
7bcc1bb1 | 4490 | return NULL; |
3e32cb2e | 4491 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
7bcc1bb1 | 4492 | } |
e1aab161 | 4493 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4494 | |
bb4cc1a8 AM |
4495 | static void __init mem_cgroup_soft_limit_tree_init(void) |
4496 | { | |
4497 | struct mem_cgroup_tree_per_node *rtpn; | |
4498 | struct mem_cgroup_tree_per_zone *rtpz; | |
4499 | int tmp, node, zone; | |
4500 | ||
4501 | for_each_node(node) { | |
4502 | tmp = node; | |
4503 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4504 | tmp = -1; | |
4505 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4506 | BUG_ON(!rtpn); | |
4507 | ||
4508 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4509 | ||
4510 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4511 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4512 | rtpz->rb_root = RB_ROOT; | |
4513 | spin_lock_init(&rtpz->lock); | |
4514 | } | |
4515 | } | |
4516 | } | |
4517 | ||
0eb253e2 | 4518 | static struct cgroup_subsys_state * __ref |
eb95419b | 4519 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4520 | { |
d142e3e6 | 4521 | struct mem_cgroup *memcg; |
04046e1a | 4522 | long error = -ENOMEM; |
6d12e2d8 | 4523 | int node; |
8cdea7c0 | 4524 | |
c0ff4b85 R |
4525 | memcg = mem_cgroup_alloc(); |
4526 | if (!memcg) | |
04046e1a | 4527 | return ERR_PTR(error); |
78fb7466 | 4528 | |
3ed28fa1 | 4529 | for_each_node(node) |
c0ff4b85 | 4530 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4531 | goto free_out; |
f64c3f54 | 4532 | |
c077719b | 4533 | /* root ? */ |
eb95419b | 4534 | if (parent_css == NULL) { |
a41c58a6 | 4535 | root_mem_cgroup = memcg; |
3e32cb2e | 4536 | page_counter_init(&memcg->memory, NULL); |
24d404dc | 4537 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4538 | page_counter_init(&memcg->memsw, NULL); |
4539 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4540 | } |
28dbc4b6 | 4541 | |
d142e3e6 GC |
4542 | memcg->last_scanned_node = MAX_NUMNODES; |
4543 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4544 | memcg->move_charge_at_immigrate = 0; |
4545 | mutex_init(&memcg->thresholds_lock); | |
4546 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4547 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4548 | INIT_LIST_HEAD(&memcg->event_list); |
4549 | spin_lock_init(&memcg->event_list_lock); | |
900a38f0 VD |
4550 | #ifdef CONFIG_MEMCG_KMEM |
4551 | memcg->kmemcg_id = -1; | |
900a38f0 | 4552 | #endif |
d142e3e6 GC |
4553 | |
4554 | return &memcg->css; | |
4555 | ||
4556 | free_out: | |
4557 | __mem_cgroup_free(memcg); | |
4558 | return ERR_PTR(error); | |
4559 | } | |
4560 | ||
4561 | static int | |
eb95419b | 4562 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4563 | { |
eb95419b | 4564 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4565 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4566 | int ret; |
d142e3e6 | 4567 | |
15a4c835 | 4568 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4569 | return -ENOSPC; |
4570 | ||
63876986 | 4571 | if (!parent) |
d142e3e6 GC |
4572 | return 0; |
4573 | ||
0999821b | 4574 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4575 | |
4576 | memcg->use_hierarchy = parent->use_hierarchy; | |
4577 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4578 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4579 | ||
4580 | if (parent->use_hierarchy) { | |
3e32cb2e | 4581 | page_counter_init(&memcg->memory, &parent->memory); |
24d404dc | 4582 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4583 | page_counter_init(&memcg->memsw, &parent->memsw); |
4584 | page_counter_init(&memcg->kmem, &parent->kmem); | |
55007d84 | 4585 | |
7bcc1bb1 | 4586 | /* |
8d76a979 LZ |
4587 | * No need to take a reference to the parent because cgroup |
4588 | * core guarantees its existence. | |
7bcc1bb1 | 4589 | */ |
18f59ea7 | 4590 | } else { |
3e32cb2e | 4591 | page_counter_init(&memcg->memory, NULL); |
24d404dc | 4592 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4593 | page_counter_init(&memcg->memsw, NULL); |
4594 | page_counter_init(&memcg->kmem, NULL); | |
8c7f6edb TH |
4595 | /* |
4596 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4597 | * much sense so let cgroup subsystem know about this | |
4598 | * unfortunate state in our controller. | |
4599 | */ | |
d142e3e6 | 4600 | if (parent != root_mem_cgroup) |
073219e9 | 4601 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4602 | } |
0999821b | 4603 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4604 | |
2f7dd7a4 JW |
4605 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
4606 | if (ret) | |
4607 | return ret; | |
4608 | ||
4609 | /* | |
4610 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4611 | * orders reading memcg->initialized against its callers | |
4612 | * reading the memcg members. | |
4613 | */ | |
4614 | smp_store_release(&memcg->initialized, 1); | |
4615 | ||
4616 | return 0; | |
8cdea7c0 BS |
4617 | } |
4618 | ||
eb95419b | 4619 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4620 | { |
eb95419b | 4621 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4622 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4623 | |
4624 | /* | |
4625 | * Unregister events and notify userspace. | |
4626 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4627 | * directory to avoid race between userspace and kernelspace. | |
4628 | */ | |
fba94807 TH |
4629 | spin_lock(&memcg->event_list_lock); |
4630 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4631 | list_del_init(&event->list); |
4632 | schedule_work(&event->remove); | |
4633 | } | |
fba94807 | 4634 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4635 | |
33cb876e | 4636 | vmpressure_cleanup(&memcg->vmpressure); |
df878fb0 KH |
4637 | } |
4638 | ||
eb95419b | 4639 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4640 | { |
eb95419b | 4641 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4642 | |
10d5ebf4 | 4643 | memcg_destroy_kmem(memcg); |
465939a1 | 4644 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4645 | } |
4646 | ||
1ced953b TH |
4647 | /** |
4648 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4649 | * @css: the target css | |
4650 | * | |
4651 | * Reset the states of the mem_cgroup associated with @css. This is | |
4652 | * invoked when the userland requests disabling on the default hierarchy | |
4653 | * but the memcg is pinned through dependency. The memcg should stop | |
4654 | * applying policies and should revert to the vanilla state as it may be | |
4655 | * made visible again. | |
4656 | * | |
4657 | * The current implementation only resets the essential configurations. | |
4658 | * This needs to be expanded to cover all the visible parts. | |
4659 | */ | |
4660 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4661 | { | |
4662 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4663 | ||
3e32cb2e JW |
4664 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4665 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4666 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
24d404dc | 4667 | memcg->soft_limit = PAGE_COUNTER_MAX; |
1ced953b TH |
4668 | } |
4669 | ||
02491447 | 4670 | #ifdef CONFIG_MMU |
7dc74be0 | 4671 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4672 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4673 | { |
05b84301 | 4674 | int ret; |
9476db97 JW |
4675 | |
4676 | /* Try a single bulk charge without reclaim first */ | |
00501b53 | 4677 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
9476db97 | 4678 | if (!ret) { |
854ffa8d | 4679 | mc.precharge += count; |
854ffa8d DN |
4680 | return ret; |
4681 | } | |
692e7c45 | 4682 | if (ret == -EINTR) { |
00501b53 | 4683 | cancel_charge(root_mem_cgroup, count); |
692e7c45 JW |
4684 | return ret; |
4685 | } | |
9476db97 JW |
4686 | |
4687 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4688 | while (count--) { |
00501b53 | 4689 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
9476db97 JW |
4690 | /* |
4691 | * In case of failure, any residual charges against | |
4692 | * mc.to will be dropped by mem_cgroup_clear_mc() | |
692e7c45 JW |
4693 | * later on. However, cancel any charges that are |
4694 | * bypassed to root right away or they'll be lost. | |
9476db97 | 4695 | */ |
692e7c45 | 4696 | if (ret == -EINTR) |
00501b53 | 4697 | cancel_charge(root_mem_cgroup, 1); |
38c5d72f | 4698 | if (ret) |
38c5d72f | 4699 | return ret; |
854ffa8d | 4700 | mc.precharge++; |
9476db97 | 4701 | cond_resched(); |
854ffa8d | 4702 | } |
9476db97 | 4703 | return 0; |
4ffef5fe DN |
4704 | } |
4705 | ||
4706 | /** | |
8d32ff84 | 4707 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4708 | * @vma: the vma the pte to be checked belongs |
4709 | * @addr: the address corresponding to the pte to be checked | |
4710 | * @ptent: the pte to be checked | |
02491447 | 4711 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4712 | * |
4713 | * Returns | |
4714 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4715 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4716 | * move charge. if @target is not NULL, the page is stored in target->page | |
4717 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4718 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4719 | * target for charge migration. if @target is not NULL, the entry is stored | |
4720 | * in target->ent. | |
4ffef5fe DN |
4721 | * |
4722 | * Called with pte lock held. | |
4723 | */ | |
4ffef5fe DN |
4724 | union mc_target { |
4725 | struct page *page; | |
02491447 | 4726 | swp_entry_t ent; |
4ffef5fe DN |
4727 | }; |
4728 | ||
4ffef5fe | 4729 | enum mc_target_type { |
8d32ff84 | 4730 | MC_TARGET_NONE = 0, |
4ffef5fe | 4731 | MC_TARGET_PAGE, |
02491447 | 4732 | MC_TARGET_SWAP, |
4ffef5fe DN |
4733 | }; |
4734 | ||
90254a65 DN |
4735 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4736 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4737 | { |
90254a65 | 4738 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4739 | |
90254a65 DN |
4740 | if (!page || !page_mapped(page)) |
4741 | return NULL; | |
4742 | if (PageAnon(page)) { | |
4743 | /* we don't move shared anon */ | |
4b91355e | 4744 | if (!move_anon()) |
90254a65 | 4745 | return NULL; |
87946a72 DN |
4746 | } else if (!move_file()) |
4747 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
4748 | return NULL; |
4749 | if (!get_page_unless_zero(page)) | |
4750 | return NULL; | |
4751 | ||
4752 | return page; | |
4753 | } | |
4754 | ||
4b91355e | 4755 | #ifdef CONFIG_SWAP |
90254a65 DN |
4756 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4757 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4758 | { | |
90254a65 DN |
4759 | struct page *page = NULL; |
4760 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4761 | ||
4762 | if (!move_anon() || non_swap_entry(ent)) | |
4763 | return NULL; | |
4b91355e KH |
4764 | /* |
4765 | * Because lookup_swap_cache() updates some statistics counter, | |
4766 | * we call find_get_page() with swapper_space directly. | |
4767 | */ | |
33806f06 | 4768 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
4769 | if (do_swap_account) |
4770 | entry->val = ent.val; | |
4771 | ||
4772 | return page; | |
4773 | } | |
4b91355e KH |
4774 | #else |
4775 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4776 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4777 | { | |
4778 | return NULL; | |
4779 | } | |
4780 | #endif | |
90254a65 | 4781 | |
87946a72 DN |
4782 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4783 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4784 | { | |
4785 | struct page *page = NULL; | |
87946a72 DN |
4786 | struct address_space *mapping; |
4787 | pgoff_t pgoff; | |
4788 | ||
4789 | if (!vma->vm_file) /* anonymous vma */ | |
4790 | return NULL; | |
4791 | if (!move_file()) | |
4792 | return NULL; | |
4793 | ||
87946a72 | 4794 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4795 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4796 | |
4797 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4798 | #ifdef CONFIG_SWAP |
4799 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4800 | if (shmem_mapping(mapping)) { |
4801 | page = find_get_entry(mapping, pgoff); | |
4802 | if (radix_tree_exceptional_entry(page)) { | |
4803 | swp_entry_t swp = radix_to_swp_entry(page); | |
4804 | if (do_swap_account) | |
4805 | *entry = swp; | |
4806 | page = find_get_page(swap_address_space(swp), swp.val); | |
4807 | } | |
4808 | } else | |
4809 | page = find_get_page(mapping, pgoff); | |
4810 | #else | |
4811 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4812 | #endif |
87946a72 DN |
4813 | return page; |
4814 | } | |
4815 | ||
8d32ff84 | 4816 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4817 | unsigned long addr, pte_t ptent, union mc_target *target) |
4818 | { | |
4819 | struct page *page = NULL; | |
8d32ff84 | 4820 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4821 | swp_entry_t ent = { .val = 0 }; |
4822 | ||
4823 | if (pte_present(ptent)) | |
4824 | page = mc_handle_present_pte(vma, addr, ptent); | |
4825 | else if (is_swap_pte(ptent)) | |
4826 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4827 | else if (pte_none(ptent)) |
87946a72 | 4828 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4829 | |
4830 | if (!page && !ent.val) | |
8d32ff84 | 4831 | return ret; |
02491447 | 4832 | if (page) { |
02491447 | 4833 | /* |
0a31bc97 | 4834 | * Do only loose check w/o serialization. |
1306a85a | 4835 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4836 | * not under LRU exclusion. |
02491447 | 4837 | */ |
1306a85a | 4838 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4839 | ret = MC_TARGET_PAGE; |
4840 | if (target) | |
4841 | target->page = page; | |
4842 | } | |
4843 | if (!ret || !target) | |
4844 | put_page(page); | |
4845 | } | |
90254a65 DN |
4846 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4847 | if (ent.val && !ret && | |
34c00c31 | 4848 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4849 | ret = MC_TARGET_SWAP; |
4850 | if (target) | |
4851 | target->ent = ent; | |
4ffef5fe | 4852 | } |
4ffef5fe DN |
4853 | return ret; |
4854 | } | |
4855 | ||
12724850 NH |
4856 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4857 | /* | |
4858 | * We don't consider swapping or file mapped pages because THP does not | |
4859 | * support them for now. | |
4860 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4861 | */ | |
4862 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4863 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4864 | { | |
4865 | struct page *page = NULL; | |
12724850 NH |
4866 | enum mc_target_type ret = MC_TARGET_NONE; |
4867 | ||
4868 | page = pmd_page(pmd); | |
309381fe | 4869 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
12724850 NH |
4870 | if (!move_anon()) |
4871 | return ret; | |
1306a85a | 4872 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4873 | ret = MC_TARGET_PAGE; |
4874 | if (target) { | |
4875 | get_page(page); | |
4876 | target->page = page; | |
4877 | } | |
4878 | } | |
4879 | return ret; | |
4880 | } | |
4881 | #else | |
4882 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4883 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4884 | { | |
4885 | return MC_TARGET_NONE; | |
4886 | } | |
4887 | #endif | |
4888 | ||
4ffef5fe DN |
4889 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4890 | unsigned long addr, unsigned long end, | |
4891 | struct mm_walk *walk) | |
4892 | { | |
4893 | struct vm_area_struct *vma = walk->private; | |
4894 | pte_t *pte; | |
4895 | spinlock_t *ptl; | |
4896 | ||
bf929152 | 4897 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
12724850 NH |
4898 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4899 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4900 | spin_unlock(ptl); |
1a5a9906 | 4901 | return 0; |
12724850 | 4902 | } |
03319327 | 4903 | |
45f83cef AA |
4904 | if (pmd_trans_unstable(pmd)) |
4905 | return 0; | |
4ffef5fe DN |
4906 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4907 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4908 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4909 | mc.precharge++; /* increment precharge temporarily */ |
4910 | pte_unmap_unlock(pte - 1, ptl); | |
4911 | cond_resched(); | |
4912 | ||
7dc74be0 DN |
4913 | return 0; |
4914 | } | |
4915 | ||
4ffef5fe DN |
4916 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4917 | { | |
4918 | unsigned long precharge; | |
4919 | struct vm_area_struct *vma; | |
4920 | ||
dfe076b0 | 4921 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
4922 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
4923 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
4924 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4925 | .mm = mm, | |
4926 | .private = vma, | |
4927 | }; | |
4928 | if (is_vm_hugetlb_page(vma)) | |
4929 | continue; | |
4ffef5fe DN |
4930 | walk_page_range(vma->vm_start, vma->vm_end, |
4931 | &mem_cgroup_count_precharge_walk); | |
4932 | } | |
dfe076b0 | 4933 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4934 | |
4935 | precharge = mc.precharge; | |
4936 | mc.precharge = 0; | |
4937 | ||
4938 | return precharge; | |
4939 | } | |
4940 | ||
4ffef5fe DN |
4941 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4942 | { | |
dfe076b0 DN |
4943 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4944 | ||
4945 | VM_BUG_ON(mc.moving_task); | |
4946 | mc.moving_task = current; | |
4947 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4948 | } |
4949 | ||
dfe076b0 DN |
4950 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4951 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4952 | { |
2bd9bb20 KH |
4953 | struct mem_cgroup *from = mc.from; |
4954 | struct mem_cgroup *to = mc.to; | |
4955 | ||
4ffef5fe | 4956 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4957 | if (mc.precharge) { |
00501b53 | 4958 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4959 | mc.precharge = 0; |
4960 | } | |
4961 | /* | |
4962 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4963 | * we must uncharge here. | |
4964 | */ | |
4965 | if (mc.moved_charge) { | |
00501b53 | 4966 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4967 | mc.moved_charge = 0; |
4ffef5fe | 4968 | } |
483c30b5 DN |
4969 | /* we must fixup refcnts and charges */ |
4970 | if (mc.moved_swap) { | |
483c30b5 | 4971 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4972 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4973 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4974 | |
05b84301 | 4975 | /* |
3e32cb2e JW |
4976 | * we charged both to->memory and to->memsw, so we |
4977 | * should uncharge to->memory. | |
05b84301 | 4978 | */ |
ce00a967 | 4979 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4980 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4981 | ||
e8ea14cc | 4982 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4983 | |
4050377b | 4984 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4985 | mc.moved_swap = 0; |
4986 | } | |
dfe076b0 DN |
4987 | memcg_oom_recover(from); |
4988 | memcg_oom_recover(to); | |
4989 | wake_up_all(&mc.waitq); | |
4990 | } | |
4991 | ||
4992 | static void mem_cgroup_clear_mc(void) | |
4993 | { | |
dfe076b0 DN |
4994 | /* |
4995 | * we must clear moving_task before waking up waiters at the end of | |
4996 | * task migration. | |
4997 | */ | |
4998 | mc.moving_task = NULL; | |
4999 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5000 | spin_lock(&mc.lock); |
4ffef5fe DN |
5001 | mc.from = NULL; |
5002 | mc.to = NULL; | |
2bd9bb20 | 5003 | spin_unlock(&mc.lock); |
4ffef5fe DN |
5004 | } |
5005 | ||
eb95419b | 5006 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5007 | struct cgroup_taskset *tset) |
7dc74be0 | 5008 | { |
2f7ee569 | 5009 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5010 | int ret = 0; |
eb95419b | 5011 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
ee5e8472 | 5012 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 5013 | |
ee5e8472 GC |
5014 | /* |
5015 | * We are now commited to this value whatever it is. Changes in this | |
5016 | * tunable will only affect upcoming migrations, not the current one. | |
5017 | * So we need to save it, and keep it going. | |
5018 | */ | |
5019 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
5020 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
5021 | struct mm_struct *mm; |
5022 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5023 | ||
c0ff4b85 | 5024 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5025 | |
5026 | mm = get_task_mm(p); | |
5027 | if (!mm) | |
5028 | return 0; | |
7dc74be0 | 5029 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5030 | if (mm->owner == p) { |
5031 | VM_BUG_ON(mc.from); | |
5032 | VM_BUG_ON(mc.to); | |
5033 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5034 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5035 | VM_BUG_ON(mc.moved_swap); |
247b1447 | 5036 | |
2bd9bb20 | 5037 | spin_lock(&mc.lock); |
4ffef5fe | 5038 | mc.from = from; |
c0ff4b85 | 5039 | mc.to = memcg; |
ee5e8472 | 5040 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 5041 | spin_unlock(&mc.lock); |
dfe076b0 | 5042 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5043 | |
5044 | ret = mem_cgroup_precharge_mc(mm); | |
5045 | if (ret) | |
5046 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5047 | } |
5048 | mmput(mm); | |
7dc74be0 DN |
5049 | } |
5050 | return ret; | |
5051 | } | |
5052 | ||
eb95419b | 5053 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5054 | struct cgroup_taskset *tset) |
7dc74be0 | 5055 | { |
4e2f245d JW |
5056 | if (mc.to) |
5057 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5058 | } |
5059 | ||
4ffef5fe DN |
5060 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5061 | unsigned long addr, unsigned long end, | |
5062 | struct mm_walk *walk) | |
7dc74be0 | 5063 | { |
4ffef5fe DN |
5064 | int ret = 0; |
5065 | struct vm_area_struct *vma = walk->private; | |
5066 | pte_t *pte; | |
5067 | spinlock_t *ptl; | |
12724850 NH |
5068 | enum mc_target_type target_type; |
5069 | union mc_target target; | |
5070 | struct page *page; | |
4ffef5fe | 5071 | |
12724850 NH |
5072 | /* |
5073 | * We don't take compound_lock() here but no race with splitting thp | |
5074 | * happens because: | |
5075 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5076 | * under splitting, which means there's no concurrent thp split, | |
5077 | * - if another thread runs into split_huge_page() just after we | |
5078 | * entered this if-block, the thread must wait for page table lock | |
5079 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5080 | * part of thp split is not executed yet. | |
5081 | */ | |
bf929152 | 5082 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
62ade86a | 5083 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5084 | spin_unlock(ptl); |
12724850 NH |
5085 | return 0; |
5086 | } | |
5087 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5088 | if (target_type == MC_TARGET_PAGE) { | |
5089 | page = target.page; | |
5090 | if (!isolate_lru_page(page)) { | |
12724850 | 5091 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
1306a85a | 5092 | mc.from, mc.to)) { |
12724850 NH |
5093 | mc.precharge -= HPAGE_PMD_NR; |
5094 | mc.moved_charge += HPAGE_PMD_NR; | |
5095 | } | |
5096 | putback_lru_page(page); | |
5097 | } | |
5098 | put_page(page); | |
5099 | } | |
bf929152 | 5100 | spin_unlock(ptl); |
1a5a9906 | 5101 | return 0; |
12724850 NH |
5102 | } |
5103 | ||
45f83cef AA |
5104 | if (pmd_trans_unstable(pmd)) |
5105 | return 0; | |
4ffef5fe DN |
5106 | retry: |
5107 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5108 | for (; addr != end; addr += PAGE_SIZE) { | |
5109 | pte_t ptent = *(pte++); | |
02491447 | 5110 | swp_entry_t ent; |
4ffef5fe DN |
5111 | |
5112 | if (!mc.precharge) | |
5113 | break; | |
5114 | ||
8d32ff84 | 5115 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5116 | case MC_TARGET_PAGE: |
5117 | page = target.page; | |
5118 | if (isolate_lru_page(page)) | |
5119 | goto put; | |
1306a85a | 5120 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
4ffef5fe | 5121 | mc.precharge--; |
854ffa8d DN |
5122 | /* we uncharge from mc.from later. */ |
5123 | mc.moved_charge++; | |
4ffef5fe DN |
5124 | } |
5125 | putback_lru_page(page); | |
8d32ff84 | 5126 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5127 | put_page(page); |
5128 | break; | |
02491447 DN |
5129 | case MC_TARGET_SWAP: |
5130 | ent = target.ent; | |
e91cbb42 | 5131 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5132 | mc.precharge--; |
483c30b5 DN |
5133 | /* we fixup refcnts and charges later. */ |
5134 | mc.moved_swap++; | |
5135 | } | |
02491447 | 5136 | break; |
4ffef5fe DN |
5137 | default: |
5138 | break; | |
5139 | } | |
5140 | } | |
5141 | pte_unmap_unlock(pte - 1, ptl); | |
5142 | cond_resched(); | |
5143 | ||
5144 | if (addr != end) { | |
5145 | /* | |
5146 | * We have consumed all precharges we got in can_attach(). | |
5147 | * We try charge one by one, but don't do any additional | |
5148 | * charges to mc.to if we have failed in charge once in attach() | |
5149 | * phase. | |
5150 | */ | |
854ffa8d | 5151 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5152 | if (!ret) |
5153 | goto retry; | |
5154 | } | |
5155 | ||
5156 | return ret; | |
5157 | } | |
5158 | ||
5159 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5160 | { | |
5161 | struct vm_area_struct *vma; | |
5162 | ||
5163 | lru_add_drain_all(); | |
312722cb JW |
5164 | /* |
5165 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5166 | * move_lock while we're moving its pages to another memcg. | |
5167 | * Then wait for already started RCU-only updates to finish. | |
5168 | */ | |
5169 | atomic_inc(&mc.from->moving_account); | |
5170 | synchronize_rcu(); | |
dfe076b0 DN |
5171 | retry: |
5172 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5173 | /* | |
5174 | * Someone who are holding the mmap_sem might be waiting in | |
5175 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5176 | * and retry. Because we cancel precharges, we might not be able | |
5177 | * to move enough charges, but moving charge is a best-effort | |
5178 | * feature anyway, so it wouldn't be a big problem. | |
5179 | */ | |
5180 | __mem_cgroup_clear_mc(); | |
5181 | cond_resched(); | |
5182 | goto retry; | |
5183 | } | |
4ffef5fe DN |
5184 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5185 | int ret; | |
5186 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5187 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5188 | .mm = mm, | |
5189 | .private = vma, | |
5190 | }; | |
5191 | if (is_vm_hugetlb_page(vma)) | |
5192 | continue; | |
4ffef5fe DN |
5193 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5194 | &mem_cgroup_move_charge_walk); | |
5195 | if (ret) | |
5196 | /* | |
5197 | * means we have consumed all precharges and failed in | |
5198 | * doing additional charge. Just abandon here. | |
5199 | */ | |
5200 | break; | |
5201 | } | |
dfe076b0 | 5202 | up_read(&mm->mmap_sem); |
312722cb | 5203 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5204 | } |
5205 | ||
eb95419b | 5206 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5207 | struct cgroup_taskset *tset) |
67e465a7 | 5208 | { |
2f7ee569 | 5209 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5210 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5211 | |
dfe076b0 | 5212 | if (mm) { |
a433658c KM |
5213 | if (mc.to) |
5214 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5215 | mmput(mm); |
5216 | } | |
a433658c KM |
5217 | if (mc.to) |
5218 | mem_cgroup_clear_mc(); | |
67e465a7 | 5219 | } |
5cfb80a7 | 5220 | #else /* !CONFIG_MMU */ |
eb95419b | 5221 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5222 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5223 | { |
5224 | return 0; | |
5225 | } | |
eb95419b | 5226 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
761b3ef5 | 5227 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5228 | { |
5229 | } | |
eb95419b | 5230 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
761b3ef5 | 5231 | struct cgroup_taskset *tset) |
5cfb80a7 DN |
5232 | { |
5233 | } | |
5234 | #endif | |
67e465a7 | 5235 | |
f00baae7 TH |
5236 | /* |
5237 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5238 | * to verify whether we're attached to the default hierarchy on each mount |
5239 | * attempt. | |
f00baae7 | 5240 | */ |
eb95419b | 5241 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5242 | { |
5243 | /* | |
aa6ec29b | 5244 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5245 | * guarantees that @root doesn't have any children, so turning it |
5246 | * on for the root memcg is enough. | |
5247 | */ | |
aa6ec29b | 5248 | if (cgroup_on_dfl(root_css->cgroup)) |
eb95419b | 5249 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
f00baae7 TH |
5250 | } |
5251 | ||
073219e9 | 5252 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5253 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5254 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
5255 | .css_offline = mem_cgroup_css_offline, |
5256 | .css_free = mem_cgroup_css_free, | |
1ced953b | 5257 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5258 | .can_attach = mem_cgroup_can_attach, |
5259 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5260 | .attach = mem_cgroup_move_task, |
f00baae7 | 5261 | .bind = mem_cgroup_bind, |
5577964e | 5262 | .legacy_cftypes = mem_cgroup_files, |
6d12e2d8 | 5263 | .early_init = 0, |
8cdea7c0 | 5264 | }; |
c077719b | 5265 | |
c255a458 | 5266 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5267 | static int __init enable_swap_account(char *s) |
5268 | { | |
a2c8990a | 5269 | if (!strcmp(s, "1")) |
a42c390c | 5270 | really_do_swap_account = 1; |
a2c8990a | 5271 | else if (!strcmp(s, "0")) |
a42c390c MH |
5272 | really_do_swap_account = 0; |
5273 | return 1; | |
5274 | } | |
a2c8990a | 5275 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5276 | |
2d11085e MH |
5277 | static void __init memsw_file_init(void) |
5278 | { | |
2cf669a5 TH |
5279 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
5280 | memsw_cgroup_files)); | |
6acc8b02 MH |
5281 | } |
5282 | ||
5283 | static void __init enable_swap_cgroup(void) | |
5284 | { | |
5285 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5286 | do_swap_account = 1; | |
5287 | memsw_file_init(); | |
5288 | } | |
2d11085e | 5289 | } |
6acc8b02 | 5290 | |
2d11085e | 5291 | #else |
6acc8b02 | 5292 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
5293 | { |
5294 | } | |
c077719b | 5295 | #endif |
2d11085e | 5296 | |
0a31bc97 JW |
5297 | #ifdef CONFIG_MEMCG_SWAP |
5298 | /** | |
5299 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5300 | * @page: page whose memsw charge to transfer | |
5301 | * @entry: swap entry to move the charge to | |
5302 | * | |
5303 | * Transfer the memsw charge of @page to @entry. | |
5304 | */ | |
5305 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5306 | { | |
7bdd143c | 5307 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5308 | unsigned short oldid; |
5309 | ||
5310 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5311 | VM_BUG_ON_PAGE(page_count(page), page); | |
5312 | ||
5313 | if (!do_swap_account) | |
5314 | return; | |
5315 | ||
1306a85a | 5316 | memcg = page->mem_cgroup; |
0a31bc97 JW |
5317 | |
5318 | /* Readahead page, never charged */ | |
29833315 | 5319 | if (!memcg) |
0a31bc97 JW |
5320 | return; |
5321 | ||
7bdd143c | 5322 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
0a31bc97 | 5323 | VM_BUG_ON_PAGE(oldid, page); |
7bdd143c JW |
5324 | mem_cgroup_swap_statistics(memcg, true); |
5325 | ||
1306a85a | 5326 | page->mem_cgroup = NULL; |
0a31bc97 | 5327 | |
7bdd143c JW |
5328 | if (!mem_cgroup_is_root(memcg)) |
5329 | page_counter_uncharge(&memcg->memory, 1); | |
5330 | ||
5331 | /* XXX: caller holds IRQ-safe mapping->tree_lock */ | |
5332 | VM_BUG_ON(!irqs_disabled()); | |
0a31bc97 | 5333 | |
7bdd143c JW |
5334 | mem_cgroup_charge_statistics(memcg, page, -1); |
5335 | memcg_check_events(memcg, page); | |
0a31bc97 JW |
5336 | } |
5337 | ||
5338 | /** | |
5339 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5340 | * @entry: swap entry to uncharge | |
5341 | * | |
5342 | * Drop the memsw charge associated with @entry. | |
5343 | */ | |
5344 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5345 | { | |
5346 | struct mem_cgroup *memcg; | |
5347 | unsigned short id; | |
5348 | ||
5349 | if (!do_swap_account) | |
5350 | return; | |
5351 | ||
5352 | id = swap_cgroup_record(entry, 0); | |
5353 | rcu_read_lock(); | |
5354 | memcg = mem_cgroup_lookup(id); | |
5355 | if (memcg) { | |
ce00a967 | 5356 | if (!mem_cgroup_is_root(memcg)) |
3e32cb2e | 5357 | page_counter_uncharge(&memcg->memsw, 1); |
0a31bc97 JW |
5358 | mem_cgroup_swap_statistics(memcg, false); |
5359 | css_put(&memcg->css); | |
5360 | } | |
5361 | rcu_read_unlock(); | |
5362 | } | |
5363 | #endif | |
5364 | ||
00501b53 JW |
5365 | /** |
5366 | * mem_cgroup_try_charge - try charging a page | |
5367 | * @page: page to charge | |
5368 | * @mm: mm context of the victim | |
5369 | * @gfp_mask: reclaim mode | |
5370 | * @memcgp: charged memcg return | |
5371 | * | |
5372 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5373 | * pages according to @gfp_mask if necessary. | |
5374 | * | |
5375 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5376 | * Otherwise, an error code is returned. | |
5377 | * | |
5378 | * After page->mapping has been set up, the caller must finalize the | |
5379 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5380 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5381 | */ | |
5382 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
5383 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
5384 | { | |
5385 | struct mem_cgroup *memcg = NULL; | |
5386 | unsigned int nr_pages = 1; | |
5387 | int ret = 0; | |
5388 | ||
5389 | if (mem_cgroup_disabled()) | |
5390 | goto out; | |
5391 | ||
5392 | if (PageSwapCache(page)) { | |
00501b53 JW |
5393 | /* |
5394 | * Every swap fault against a single page tries to charge the | |
5395 | * page, bail as early as possible. shmem_unuse() encounters | |
5396 | * already charged pages, too. The USED bit is protected by | |
5397 | * the page lock, which serializes swap cache removal, which | |
5398 | * in turn serializes uncharging. | |
5399 | */ | |
1306a85a | 5400 | if (page->mem_cgroup) |
00501b53 JW |
5401 | goto out; |
5402 | } | |
5403 | ||
5404 | if (PageTransHuge(page)) { | |
5405 | nr_pages <<= compound_order(page); | |
5406 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5407 | } | |
5408 | ||
5409 | if (do_swap_account && PageSwapCache(page)) | |
5410 | memcg = try_get_mem_cgroup_from_page(page); | |
5411 | if (!memcg) | |
5412 | memcg = get_mem_cgroup_from_mm(mm); | |
5413 | ||
5414 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5415 | ||
5416 | css_put(&memcg->css); | |
5417 | ||
5418 | if (ret == -EINTR) { | |
5419 | memcg = root_mem_cgroup; | |
5420 | ret = 0; | |
5421 | } | |
5422 | out: | |
5423 | *memcgp = memcg; | |
5424 | return ret; | |
5425 | } | |
5426 | ||
5427 | /** | |
5428 | * mem_cgroup_commit_charge - commit a page charge | |
5429 | * @page: page to charge | |
5430 | * @memcg: memcg to charge the page to | |
5431 | * @lrucare: page might be on LRU already | |
5432 | * | |
5433 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5434 | * after page->mapping has been set up. This must happen atomically | |
5435 | * as part of the page instantiation, i.e. under the page table lock | |
5436 | * for anonymous pages, under the page lock for page and swap cache. | |
5437 | * | |
5438 | * In addition, the page must not be on the LRU during the commit, to | |
5439 | * prevent racing with task migration. If it might be, use @lrucare. | |
5440 | * | |
5441 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5442 | */ | |
5443 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
5444 | bool lrucare) | |
5445 | { | |
5446 | unsigned int nr_pages = 1; | |
5447 | ||
5448 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5449 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5450 | ||
5451 | if (mem_cgroup_disabled()) | |
5452 | return; | |
5453 | /* | |
5454 | * Swap faults will attempt to charge the same page multiple | |
5455 | * times. But reuse_swap_page() might have removed the page | |
5456 | * from swapcache already, so we can't check PageSwapCache(). | |
5457 | */ | |
5458 | if (!memcg) | |
5459 | return; | |
5460 | ||
6abb5a86 JW |
5461 | commit_charge(page, memcg, lrucare); |
5462 | ||
00501b53 JW |
5463 | if (PageTransHuge(page)) { |
5464 | nr_pages <<= compound_order(page); | |
5465 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5466 | } | |
5467 | ||
6abb5a86 JW |
5468 | local_irq_disable(); |
5469 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
5470 | memcg_check_events(memcg, page); | |
5471 | local_irq_enable(); | |
00501b53 JW |
5472 | |
5473 | if (do_swap_account && PageSwapCache(page)) { | |
5474 | swp_entry_t entry = { .val = page_private(page) }; | |
5475 | /* | |
5476 | * The swap entry might not get freed for a long time, | |
5477 | * let's not wait for it. The page already received a | |
5478 | * memory+swap charge, drop the swap entry duplicate. | |
5479 | */ | |
5480 | mem_cgroup_uncharge_swap(entry); | |
5481 | } | |
5482 | } | |
5483 | ||
5484 | /** | |
5485 | * mem_cgroup_cancel_charge - cancel a page charge | |
5486 | * @page: page to charge | |
5487 | * @memcg: memcg to charge the page to | |
5488 | * | |
5489 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5490 | */ | |
5491 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) | |
5492 | { | |
5493 | unsigned int nr_pages = 1; | |
5494 | ||
5495 | if (mem_cgroup_disabled()) | |
5496 | return; | |
5497 | /* | |
5498 | * Swap faults will attempt to charge the same page multiple | |
5499 | * times. But reuse_swap_page() might have removed the page | |
5500 | * from swapcache already, so we can't check PageSwapCache(). | |
5501 | */ | |
5502 | if (!memcg) | |
5503 | return; | |
5504 | ||
5505 | if (PageTransHuge(page)) { | |
5506 | nr_pages <<= compound_order(page); | |
5507 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5508 | } | |
5509 | ||
5510 | cancel_charge(memcg, nr_pages); | |
5511 | } | |
5512 | ||
747db954 | 5513 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5514 | unsigned long nr_anon, unsigned long nr_file, |
5515 | unsigned long nr_huge, struct page *dummy_page) | |
5516 | { | |
18eca2e6 | 5517 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5518 | unsigned long flags; |
5519 | ||
ce00a967 | 5520 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 JW |
5521 | page_counter_uncharge(&memcg->memory, nr_pages); |
5522 | if (do_swap_account) | |
5523 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
ce00a967 JW |
5524 | memcg_oom_recover(memcg); |
5525 | } | |
747db954 JW |
5526 | |
5527 | local_irq_save(flags); | |
5528 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5529 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5530 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5531 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5532 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5533 | memcg_check_events(memcg, dummy_page); |
5534 | local_irq_restore(flags); | |
e8ea14cc JW |
5535 | |
5536 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5537 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5538 | } |
5539 | ||
5540 | static void uncharge_list(struct list_head *page_list) | |
5541 | { | |
5542 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5543 | unsigned long nr_anon = 0; |
5544 | unsigned long nr_file = 0; | |
5545 | unsigned long nr_huge = 0; | |
5546 | unsigned long pgpgout = 0; | |
747db954 JW |
5547 | struct list_head *next; |
5548 | struct page *page; | |
5549 | ||
5550 | next = page_list->next; | |
5551 | do { | |
5552 | unsigned int nr_pages = 1; | |
747db954 JW |
5553 | |
5554 | page = list_entry(next, struct page, lru); | |
5555 | next = page->lru.next; | |
5556 | ||
5557 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5558 | VM_BUG_ON_PAGE(page_count(page), page); | |
5559 | ||
1306a85a | 5560 | if (!page->mem_cgroup) |
747db954 JW |
5561 | continue; |
5562 | ||
5563 | /* | |
5564 | * Nobody should be changing or seriously looking at | |
1306a85a | 5565 | * page->mem_cgroup at this point, we have fully |
29833315 | 5566 | * exclusive access to the page. |
747db954 JW |
5567 | */ |
5568 | ||
1306a85a | 5569 | if (memcg != page->mem_cgroup) { |
747db954 | 5570 | if (memcg) { |
18eca2e6 JW |
5571 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5572 | nr_huge, page); | |
5573 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5574 | } |
1306a85a | 5575 | memcg = page->mem_cgroup; |
747db954 JW |
5576 | } |
5577 | ||
5578 | if (PageTransHuge(page)) { | |
5579 | nr_pages <<= compound_order(page); | |
5580 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5581 | nr_huge += nr_pages; | |
5582 | } | |
5583 | ||
5584 | if (PageAnon(page)) | |
5585 | nr_anon += nr_pages; | |
5586 | else | |
5587 | nr_file += nr_pages; | |
5588 | ||
1306a85a | 5589 | page->mem_cgroup = NULL; |
747db954 JW |
5590 | |
5591 | pgpgout++; | |
5592 | } while (next != page_list); | |
5593 | ||
5594 | if (memcg) | |
18eca2e6 JW |
5595 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5596 | nr_huge, page); | |
747db954 JW |
5597 | } |
5598 | ||
0a31bc97 JW |
5599 | /** |
5600 | * mem_cgroup_uncharge - uncharge a page | |
5601 | * @page: page to uncharge | |
5602 | * | |
5603 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5604 | * mem_cgroup_commit_charge(). | |
5605 | */ | |
5606 | void mem_cgroup_uncharge(struct page *page) | |
5607 | { | |
0a31bc97 JW |
5608 | if (mem_cgroup_disabled()) |
5609 | return; | |
5610 | ||
747db954 | 5611 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5612 | if (!page->mem_cgroup) |
0a31bc97 JW |
5613 | return; |
5614 | ||
747db954 JW |
5615 | INIT_LIST_HEAD(&page->lru); |
5616 | uncharge_list(&page->lru); | |
5617 | } | |
0a31bc97 | 5618 | |
747db954 JW |
5619 | /** |
5620 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5621 | * @page_list: list of pages to uncharge | |
5622 | * | |
5623 | * Uncharge a list of pages previously charged with | |
5624 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5625 | */ | |
5626 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5627 | { | |
5628 | if (mem_cgroup_disabled()) | |
5629 | return; | |
0a31bc97 | 5630 | |
747db954 JW |
5631 | if (!list_empty(page_list)) |
5632 | uncharge_list(page_list); | |
0a31bc97 JW |
5633 | } |
5634 | ||
5635 | /** | |
5636 | * mem_cgroup_migrate - migrate a charge to another page | |
5637 | * @oldpage: currently charged page | |
5638 | * @newpage: page to transfer the charge to | |
f5e03a49 | 5639 | * @lrucare: either or both pages might be on the LRU already |
0a31bc97 JW |
5640 | * |
5641 | * Migrate the charge from @oldpage to @newpage. | |
5642 | * | |
5643 | * Both pages must be locked, @newpage->mapping must be set up. | |
5644 | */ | |
5645 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, | |
5646 | bool lrucare) | |
5647 | { | |
29833315 | 5648 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5649 | int isolated; |
5650 | ||
5651 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5652 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
5653 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); | |
5654 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); | |
5655 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); | |
6abb5a86 JW |
5656 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5657 | newpage); | |
0a31bc97 JW |
5658 | |
5659 | if (mem_cgroup_disabled()) | |
5660 | return; | |
5661 | ||
5662 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5663 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5664 | return; |
5665 | ||
7d5e3245 JW |
5666 | /* |
5667 | * Swapcache readahead pages can get migrated before being | |
5668 | * charged, and migration from compaction can happen to an | |
5669 | * uncharged page when the PFN walker finds a page that | |
5670 | * reclaim just put back on the LRU but has not released yet. | |
5671 | */ | |
1306a85a | 5672 | memcg = oldpage->mem_cgroup; |
29833315 | 5673 | if (!memcg) |
0a31bc97 JW |
5674 | return; |
5675 | ||
0a31bc97 JW |
5676 | if (lrucare) |
5677 | lock_page_lru(oldpage, &isolated); | |
5678 | ||
1306a85a | 5679 | oldpage->mem_cgroup = NULL; |
0a31bc97 JW |
5680 | |
5681 | if (lrucare) | |
5682 | unlock_page_lru(oldpage, isolated); | |
5683 | ||
29833315 | 5684 | commit_charge(newpage, memcg, lrucare); |
0a31bc97 JW |
5685 | } |
5686 | ||
2d11085e | 5687 | /* |
1081312f MH |
5688 | * subsys_initcall() for memory controller. |
5689 | * | |
5690 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5691 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5692 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5693 | * be initialized from here. | |
2d11085e MH |
5694 | */ |
5695 | static int __init mem_cgroup_init(void) | |
5696 | { | |
5697 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 5698 | enable_swap_cgroup(); |
bb4cc1a8 | 5699 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 5700 | memcg_stock_init(); |
2d11085e MH |
5701 | return 0; |
5702 | } | |
5703 | subsys_initcall(mem_cgroup_init); |