]>
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 | * | |
8cdea7c0 BS |
13 | * This program is free software; you can redistribute it and/or modify |
14 | * it under the terms of the GNU General Public License as published by | |
15 | * the Free Software Foundation; either version 2 of the License, or | |
16 | * (at your option) any later version. | |
17 | * | |
18 | * This program is distributed in the hope that it will be useful, | |
19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | * GNU General Public License for more details. | |
22 | */ | |
23 | ||
24 | #include <linux/res_counter.h> | |
25 | #include <linux/memcontrol.h> | |
26 | #include <linux/cgroup.h> | |
78fb7466 | 27 | #include <linux/mm.h> |
4ffef5fe | 28 | #include <linux/hugetlb.h> |
d13d1443 | 29 | #include <linux/pagemap.h> |
d52aa412 | 30 | #include <linux/smp.h> |
8a9f3ccd | 31 | #include <linux/page-flags.h> |
66e1707b | 32 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
33 | #include <linux/bit_spinlock.h> |
34 | #include <linux/rcupdate.h> | |
e222432b | 35 | #include <linux/limits.h> |
b9e15baf | 36 | #include <linux/export.h> |
8c7c6e34 | 37 | #include <linux/mutex.h> |
f64c3f54 | 38 | #include <linux/rbtree.h> |
b6ac57d5 | 39 | #include <linux/slab.h> |
66e1707b | 40 | #include <linux/swap.h> |
02491447 | 41 | #include <linux/swapops.h> |
66e1707b | 42 | #include <linux/spinlock.h> |
2e72b634 KS |
43 | #include <linux/eventfd.h> |
44 | #include <linux/sort.h> | |
66e1707b | 45 | #include <linux/fs.h> |
d2ceb9b7 | 46 | #include <linux/seq_file.h> |
33327948 | 47 | #include <linux/vmalloc.h> |
b69408e8 | 48 | #include <linux/mm_inline.h> |
52d4b9ac | 49 | #include <linux/page_cgroup.h> |
cdec2e42 | 50 | #include <linux/cpu.h> |
158e0a2d | 51 | #include <linux/oom.h> |
08e552c6 | 52 | #include "internal.h" |
d1a4c0b3 GC |
53 | #include <net/sock.h> |
54 | #include <net/tcp_memcontrol.h> | |
8cdea7c0 | 55 | |
8697d331 BS |
56 | #include <asm/uaccess.h> |
57 | ||
cc8e970c KM |
58 | #include <trace/events/vmscan.h> |
59 | ||
a181b0e8 | 60 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 61 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 62 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 63 | |
c077719b | 64 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 65 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 66 | int do_swap_account __read_mostly; |
a42c390c MH |
67 | |
68 | /* for remember boot option*/ | |
69 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED | |
70 | static int really_do_swap_account __initdata = 1; | |
71 | #else | |
72 | static int really_do_swap_account __initdata = 0; | |
73 | #endif | |
74 | ||
c077719b | 75 | #else |
a0db00fc | 76 | #define do_swap_account 0 |
c077719b KH |
77 | #endif |
78 | ||
79 | ||
d52aa412 KH |
80 | /* |
81 | * Statistics for memory cgroup. | |
82 | */ | |
83 | enum mem_cgroup_stat_index { | |
84 | /* | |
85 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
86 | */ | |
87 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 88 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 89 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
0c3e73e8 | 90 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
d52aa412 KH |
91 | MEM_CGROUP_STAT_NSTATS, |
92 | }; | |
93 | ||
af7c4b0e JW |
94 | static const char * const mem_cgroup_stat_names[] = { |
95 | "cache", | |
96 | "rss", | |
97 | "mapped_file", | |
98 | "swap", | |
99 | }; | |
100 | ||
e9f8974f JW |
101 | enum mem_cgroup_events_index { |
102 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
103 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
104 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
105 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
106 | MEM_CGROUP_EVENTS_NSTATS, |
107 | }; | |
af7c4b0e JW |
108 | |
109 | static const char * const mem_cgroup_events_names[] = { | |
110 | "pgpgin", | |
111 | "pgpgout", | |
112 | "pgfault", | |
113 | "pgmajfault", | |
114 | }; | |
115 | ||
7a159cc9 JW |
116 | /* |
117 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
118 | * it will be incremated by the number of pages. This counter is used for | |
119 | * for trigger some periodic events. This is straightforward and better | |
120 | * than using jiffies etc. to handle periodic memcg event. | |
121 | */ | |
122 | enum mem_cgroup_events_target { | |
123 | MEM_CGROUP_TARGET_THRESH, | |
124 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 125 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
126 | MEM_CGROUP_NTARGETS, |
127 | }; | |
a0db00fc KS |
128 | #define THRESHOLDS_EVENTS_TARGET 128 |
129 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
130 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 131 | |
d52aa412 | 132 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 133 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 134 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 135 | unsigned long nr_page_events; |
7a159cc9 | 136 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
137 | }; |
138 | ||
527a5ec9 JW |
139 | struct mem_cgroup_reclaim_iter { |
140 | /* css_id of the last scanned hierarchy member */ | |
141 | int position; | |
142 | /* scan generation, increased every round-trip */ | |
143 | unsigned int generation; | |
144 | }; | |
145 | ||
6d12e2d8 KH |
146 | /* |
147 | * per-zone information in memory controller. | |
148 | */ | |
6d12e2d8 | 149 | struct mem_cgroup_per_zone { |
6290df54 | 150 | struct lruvec lruvec; |
1eb49272 | 151 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 152 | |
527a5ec9 JW |
153 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
154 | ||
f64c3f54 BS |
155 | struct rb_node tree_node; /* RB tree node */ |
156 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
157 | /* the soft limit is exceeded*/ | |
158 | bool on_tree; | |
d79154bb | 159 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 160 | /* use container_of */ |
6d12e2d8 | 161 | }; |
6d12e2d8 KH |
162 | |
163 | struct mem_cgroup_per_node { | |
164 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
165 | }; | |
166 | ||
167 | struct mem_cgroup_lru_info { | |
168 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
169 | }; | |
170 | ||
f64c3f54 BS |
171 | /* |
172 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
173 | * their hierarchy representation | |
174 | */ | |
175 | ||
176 | struct mem_cgroup_tree_per_zone { | |
177 | struct rb_root rb_root; | |
178 | spinlock_t lock; | |
179 | }; | |
180 | ||
181 | struct mem_cgroup_tree_per_node { | |
182 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
183 | }; | |
184 | ||
185 | struct mem_cgroup_tree { | |
186 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
187 | }; | |
188 | ||
189 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
190 | ||
2e72b634 KS |
191 | struct mem_cgroup_threshold { |
192 | struct eventfd_ctx *eventfd; | |
193 | u64 threshold; | |
194 | }; | |
195 | ||
9490ff27 | 196 | /* For threshold */ |
2e72b634 | 197 | struct mem_cgroup_threshold_ary { |
748dad36 | 198 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 199 | int current_threshold; |
2e72b634 KS |
200 | /* Size of entries[] */ |
201 | unsigned int size; | |
202 | /* Array of thresholds */ | |
203 | struct mem_cgroup_threshold entries[0]; | |
204 | }; | |
2c488db2 KS |
205 | |
206 | struct mem_cgroup_thresholds { | |
207 | /* Primary thresholds array */ | |
208 | struct mem_cgroup_threshold_ary *primary; | |
209 | /* | |
210 | * Spare threshold array. | |
211 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
212 | * It must be able to store at least primary->size - 1 entries. | |
213 | */ | |
214 | struct mem_cgroup_threshold_ary *spare; | |
215 | }; | |
216 | ||
9490ff27 KH |
217 | /* for OOM */ |
218 | struct mem_cgroup_eventfd_list { | |
219 | struct list_head list; | |
220 | struct eventfd_ctx *eventfd; | |
221 | }; | |
2e72b634 | 222 | |
c0ff4b85 R |
223 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
224 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 225 | |
8cdea7c0 BS |
226 | /* |
227 | * The memory controller data structure. The memory controller controls both | |
228 | * page cache and RSS per cgroup. We would eventually like to provide | |
229 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
230 | * to help the administrator determine what knobs to tune. | |
231 | * | |
232 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
233 | * we hit the water mark. May be even add a low water mark, such that |
234 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
235 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
236 | */ |
237 | struct mem_cgroup { | |
238 | struct cgroup_subsys_state css; | |
239 | /* | |
240 | * the counter to account for memory usage | |
241 | */ | |
242 | struct res_counter res; | |
59927fb9 HD |
243 | |
244 | union { | |
245 | /* | |
246 | * the counter to account for mem+swap usage. | |
247 | */ | |
248 | struct res_counter memsw; | |
249 | ||
250 | /* | |
251 | * rcu_freeing is used only when freeing struct mem_cgroup, | |
252 | * so put it into a union to avoid wasting more memory. | |
253 | * It must be disjoint from the css field. It could be | |
254 | * in a union with the res field, but res plays a much | |
255 | * larger part in mem_cgroup life than memsw, and might | |
256 | * be of interest, even at time of free, when debugging. | |
257 | * So share rcu_head with the less interesting memsw. | |
258 | */ | |
259 | struct rcu_head rcu_freeing; | |
260 | /* | |
261 | * But when using vfree(), that cannot be done at | |
262 | * interrupt time, so we must then queue the work. | |
263 | */ | |
264 | struct work_struct work_freeing; | |
265 | }; | |
266 | ||
78fb7466 PE |
267 | /* |
268 | * Per cgroup active and inactive list, similar to the | |
269 | * per zone LRU lists. | |
78fb7466 | 270 | */ |
6d12e2d8 | 271 | struct mem_cgroup_lru_info info; |
889976db YH |
272 | int last_scanned_node; |
273 | #if MAX_NUMNODES > 1 | |
274 | nodemask_t scan_nodes; | |
453a9bf3 KH |
275 | atomic_t numainfo_events; |
276 | atomic_t numainfo_updating; | |
889976db | 277 | #endif |
18f59ea7 BS |
278 | /* |
279 | * Should the accounting and control be hierarchical, per subtree? | |
280 | */ | |
281 | bool use_hierarchy; | |
79dfdacc MH |
282 | |
283 | bool oom_lock; | |
284 | atomic_t under_oom; | |
285 | ||
8c7c6e34 | 286 | atomic_t refcnt; |
14797e23 | 287 | |
1f4c025b | 288 | int swappiness; |
3c11ecf4 KH |
289 | /* OOM-Killer disable */ |
290 | int oom_kill_disable; | |
a7885eb8 | 291 | |
22a668d7 KH |
292 | /* set when res.limit == memsw.limit */ |
293 | bool memsw_is_minimum; | |
294 | ||
2e72b634 KS |
295 | /* protect arrays of thresholds */ |
296 | struct mutex thresholds_lock; | |
297 | ||
298 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 299 | struct mem_cgroup_thresholds thresholds; |
907860ed | 300 | |
2e72b634 | 301 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 302 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 303 | |
9490ff27 KH |
304 | /* For oom notifier event fd */ |
305 | struct list_head oom_notify; | |
185efc0f | 306 | |
7dc74be0 DN |
307 | /* |
308 | * Should we move charges of a task when a task is moved into this | |
309 | * mem_cgroup ? And what type of charges should we move ? | |
310 | */ | |
311 | unsigned long move_charge_at_immigrate; | |
619d094b KH |
312 | /* |
313 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
314 | */ | |
315 | atomic_t moving_account; | |
312734c0 KH |
316 | /* taken only while moving_account > 0 */ |
317 | spinlock_t move_lock; | |
d52aa412 | 318 | /* |
c62b1a3b | 319 | * percpu counter. |
d52aa412 | 320 | */ |
3a7951b4 | 321 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
322 | /* |
323 | * used when a cpu is offlined or other synchronizations | |
324 | * See mem_cgroup_read_stat(). | |
325 | */ | |
326 | struct mem_cgroup_stat_cpu nocpu_base; | |
327 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 GC |
328 | |
329 | #ifdef CONFIG_INET | |
330 | struct tcp_memcontrol tcp_mem; | |
331 | #endif | |
8cdea7c0 BS |
332 | }; |
333 | ||
7dc74be0 DN |
334 | /* Stuffs for move charges at task migration. */ |
335 | /* | |
336 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
337 | * left-shifted bitmap of these types. | |
338 | */ | |
339 | enum move_type { | |
4ffef5fe | 340 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 341 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
342 | NR_MOVE_TYPE, |
343 | }; | |
344 | ||
4ffef5fe DN |
345 | /* "mc" and its members are protected by cgroup_mutex */ |
346 | static struct move_charge_struct { | |
b1dd693e | 347 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
348 | struct mem_cgroup *from; |
349 | struct mem_cgroup *to; | |
350 | unsigned long precharge; | |
854ffa8d | 351 | unsigned long moved_charge; |
483c30b5 | 352 | unsigned long moved_swap; |
8033b97c DN |
353 | struct task_struct *moving_task; /* a task moving charges */ |
354 | wait_queue_head_t waitq; /* a waitq for other context */ | |
355 | } mc = { | |
2bd9bb20 | 356 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
357 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
358 | }; | |
4ffef5fe | 359 | |
90254a65 DN |
360 | static bool move_anon(void) |
361 | { | |
362 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
363 | &mc.to->move_charge_at_immigrate); | |
364 | } | |
365 | ||
87946a72 DN |
366 | static bool move_file(void) |
367 | { | |
368 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
369 | &mc.to->move_charge_at_immigrate); | |
370 | } | |
371 | ||
4e416953 BS |
372 | /* |
373 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
374 | * limit reclaim to prevent infinite loops, if they ever occur. | |
375 | */ | |
a0db00fc KS |
376 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
377 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 | |
4e416953 | 378 | |
217bc319 KH |
379 | enum charge_type { |
380 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
381 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 382 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 383 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 384 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 385 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
386 | NR_CHARGE_TYPE, |
387 | }; | |
388 | ||
8c7c6e34 | 389 | /* for encoding cft->private value on file */ |
65c64ce8 GC |
390 | #define _MEM (0) |
391 | #define _MEMSWAP (1) | |
392 | #define _OOM_TYPE (2) | |
a0db00fc KS |
393 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
394 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 395 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
396 | /* Used for OOM nofiier */ |
397 | #define OOM_CONTROL (0) | |
8c7c6e34 | 398 | |
75822b44 BS |
399 | /* |
400 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
401 | */ | |
402 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
403 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
404 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
405 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
406 | ||
c0ff4b85 R |
407 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
408 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 GC |
409 | |
410 | /* Writing them here to avoid exposing memcg's inner layout */ | |
411 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM | |
e1aab161 | 412 | #include <net/sock.h> |
d1a4c0b3 | 413 | #include <net/ip.h> |
e1aab161 GC |
414 | |
415 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); | |
416 | void sock_update_memcg(struct sock *sk) | |
417 | { | |
376be5ff | 418 | if (mem_cgroup_sockets_enabled) { |
e1aab161 GC |
419 | struct mem_cgroup *memcg; |
420 | ||
421 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
422 | ||
f3f511e1 GC |
423 | /* Socket cloning can throw us here with sk_cgrp already |
424 | * filled. It won't however, necessarily happen from | |
425 | * process context. So the test for root memcg given | |
426 | * the current task's memcg won't help us in this case. | |
427 | * | |
428 | * Respecting the original socket's memcg is a better | |
429 | * decision in this case. | |
430 | */ | |
431 | if (sk->sk_cgrp) { | |
432 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
433 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
434 | return; | |
435 | } | |
436 | ||
e1aab161 GC |
437 | rcu_read_lock(); |
438 | memcg = mem_cgroup_from_task(current); | |
439 | if (!mem_cgroup_is_root(memcg)) { | |
440 | mem_cgroup_get(memcg); | |
441 | sk->sk_cgrp = sk->sk_prot->proto_cgroup(memcg); | |
442 | } | |
443 | rcu_read_unlock(); | |
444 | } | |
445 | } | |
446 | EXPORT_SYMBOL(sock_update_memcg); | |
447 | ||
448 | void sock_release_memcg(struct sock *sk) | |
449 | { | |
376be5ff | 450 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
451 | struct mem_cgroup *memcg; |
452 | WARN_ON(!sk->sk_cgrp->memcg); | |
453 | memcg = sk->sk_cgrp->memcg; | |
454 | mem_cgroup_put(memcg); | |
455 | } | |
456 | } | |
d1a4c0b3 | 457 | |
319d3b9c | 458 | #ifdef CONFIG_INET |
d1a4c0b3 GC |
459 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
460 | { | |
461 | if (!memcg || mem_cgroup_is_root(memcg)) | |
462 | return NULL; | |
463 | ||
464 | return &memcg->tcp_mem.cg_proto; | |
465 | } | |
466 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 GC |
467 | #endif /* CONFIG_INET */ |
468 | #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */ | |
469 | ||
c0ff4b85 | 470 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 471 | |
f64c3f54 | 472 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 473 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 474 | { |
c0ff4b85 | 475 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
476 | } |
477 | ||
c0ff4b85 | 478 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 479 | { |
c0ff4b85 | 480 | return &memcg->css; |
d324236b WF |
481 | } |
482 | ||
f64c3f54 | 483 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 484 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 485 | { |
97a6c37b JW |
486 | int nid = page_to_nid(page); |
487 | int zid = page_zonenum(page); | |
f64c3f54 | 488 | |
c0ff4b85 | 489 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
490 | } |
491 | ||
492 | static struct mem_cgroup_tree_per_zone * | |
493 | soft_limit_tree_node_zone(int nid, int zid) | |
494 | { | |
495 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
496 | } | |
497 | ||
498 | static struct mem_cgroup_tree_per_zone * | |
499 | soft_limit_tree_from_page(struct page *page) | |
500 | { | |
501 | int nid = page_to_nid(page); | |
502 | int zid = page_zonenum(page); | |
503 | ||
504 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
505 | } | |
506 | ||
507 | static void | |
c0ff4b85 | 508 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 509 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
510 | struct mem_cgroup_tree_per_zone *mctz, |
511 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
512 | { |
513 | struct rb_node **p = &mctz->rb_root.rb_node; | |
514 | struct rb_node *parent = NULL; | |
515 | struct mem_cgroup_per_zone *mz_node; | |
516 | ||
517 | if (mz->on_tree) | |
518 | return; | |
519 | ||
ef8745c1 KH |
520 | mz->usage_in_excess = new_usage_in_excess; |
521 | if (!mz->usage_in_excess) | |
522 | return; | |
f64c3f54 BS |
523 | while (*p) { |
524 | parent = *p; | |
525 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
526 | tree_node); | |
527 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
528 | p = &(*p)->rb_left; | |
529 | /* | |
530 | * We can't avoid mem cgroups that are over their soft | |
531 | * limit by the same amount | |
532 | */ | |
533 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
534 | p = &(*p)->rb_right; | |
535 | } | |
536 | rb_link_node(&mz->tree_node, parent, p); | |
537 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
538 | mz->on_tree = true; | |
4e416953 BS |
539 | } |
540 | ||
541 | static void | |
c0ff4b85 | 542 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
543 | struct mem_cgroup_per_zone *mz, |
544 | struct mem_cgroup_tree_per_zone *mctz) | |
545 | { | |
546 | if (!mz->on_tree) | |
547 | return; | |
548 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
549 | mz->on_tree = false; | |
550 | } | |
551 | ||
f64c3f54 | 552 | static void |
c0ff4b85 | 553 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
554 | struct mem_cgroup_per_zone *mz, |
555 | struct mem_cgroup_tree_per_zone *mctz) | |
556 | { | |
557 | spin_lock(&mctz->lock); | |
c0ff4b85 | 558 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
559 | spin_unlock(&mctz->lock); |
560 | } | |
561 | ||
f64c3f54 | 562 | |
c0ff4b85 | 563 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 564 | { |
ef8745c1 | 565 | unsigned long long excess; |
f64c3f54 BS |
566 | struct mem_cgroup_per_zone *mz; |
567 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
568 | int nid = page_to_nid(page); |
569 | int zid = page_zonenum(page); | |
f64c3f54 BS |
570 | mctz = soft_limit_tree_from_page(page); |
571 | ||
572 | /* | |
4e649152 KH |
573 | * Necessary to update all ancestors when hierarchy is used. |
574 | * because their event counter is not touched. | |
f64c3f54 | 575 | */ |
c0ff4b85 R |
576 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
577 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
578 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
579 | /* |
580 | * We have to update the tree if mz is on RB-tree or | |
581 | * mem is over its softlimit. | |
582 | */ | |
ef8745c1 | 583 | if (excess || mz->on_tree) { |
4e649152 KH |
584 | spin_lock(&mctz->lock); |
585 | /* if on-tree, remove it */ | |
586 | if (mz->on_tree) | |
c0ff4b85 | 587 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 588 | /* |
ef8745c1 KH |
589 | * Insert again. mz->usage_in_excess will be updated. |
590 | * If excess is 0, no tree ops. | |
4e649152 | 591 | */ |
c0ff4b85 | 592 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
593 | spin_unlock(&mctz->lock); |
594 | } | |
f64c3f54 BS |
595 | } |
596 | } | |
597 | ||
c0ff4b85 | 598 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
599 | { |
600 | int node, zone; | |
601 | struct mem_cgroup_per_zone *mz; | |
602 | struct mem_cgroup_tree_per_zone *mctz; | |
603 | ||
3ed28fa1 | 604 | for_each_node(node) { |
f64c3f54 | 605 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 606 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 607 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 608 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
609 | } |
610 | } | |
611 | } | |
612 | ||
4e416953 BS |
613 | static struct mem_cgroup_per_zone * |
614 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
615 | { | |
616 | struct rb_node *rightmost = NULL; | |
26251eaf | 617 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
618 | |
619 | retry: | |
26251eaf | 620 | mz = NULL; |
4e416953 BS |
621 | rightmost = rb_last(&mctz->rb_root); |
622 | if (!rightmost) | |
623 | goto done; /* Nothing to reclaim from */ | |
624 | ||
625 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
626 | /* | |
627 | * Remove the node now but someone else can add it back, | |
628 | * we will to add it back at the end of reclaim to its correct | |
629 | * position in the tree. | |
630 | */ | |
d79154bb HD |
631 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
632 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
633 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
634 | goto retry; |
635 | done: | |
636 | return mz; | |
637 | } | |
638 | ||
639 | static struct mem_cgroup_per_zone * | |
640 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
641 | { | |
642 | struct mem_cgroup_per_zone *mz; | |
643 | ||
644 | spin_lock(&mctz->lock); | |
645 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
646 | spin_unlock(&mctz->lock); | |
647 | return mz; | |
648 | } | |
649 | ||
711d3d2c KH |
650 | /* |
651 | * Implementation Note: reading percpu statistics for memcg. | |
652 | * | |
653 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
654 | * synchronization to implement "quick" read. There are trade-off between | |
655 | * reading cost and precision of value. Then, we may have a chance to implement | |
656 | * a periodic synchronizion of counter in memcg's counter. | |
657 | * | |
658 | * But this _read() function is used for user interface now. The user accounts | |
659 | * memory usage by memory cgroup and he _always_ requires exact value because | |
660 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
661 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
662 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
663 | * | |
664 | * If there are kernel internal actions which can make use of some not-exact | |
665 | * value, and reading all cpu value can be performance bottleneck in some | |
666 | * common workload, threashold and synchonization as vmstat[] should be | |
667 | * implemented. | |
668 | */ | |
c0ff4b85 | 669 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 670 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 671 | { |
7a159cc9 | 672 | long val = 0; |
c62b1a3b | 673 | int cpu; |
c62b1a3b | 674 | |
711d3d2c KH |
675 | get_online_cpus(); |
676 | for_each_online_cpu(cpu) | |
c0ff4b85 | 677 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 678 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
679 | spin_lock(&memcg->pcp_counter_lock); |
680 | val += memcg->nocpu_base.count[idx]; | |
681 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
682 | #endif |
683 | put_online_cpus(); | |
c62b1a3b KH |
684 | return val; |
685 | } | |
686 | ||
c0ff4b85 | 687 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
688 | bool charge) |
689 | { | |
690 | int val = (charge) ? 1 : -1; | |
c0ff4b85 | 691 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
692 | } |
693 | ||
c0ff4b85 | 694 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
695 | enum mem_cgroup_events_index idx) |
696 | { | |
697 | unsigned long val = 0; | |
698 | int cpu; | |
699 | ||
700 | for_each_online_cpu(cpu) | |
c0ff4b85 | 701 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 702 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
703 | spin_lock(&memcg->pcp_counter_lock); |
704 | val += memcg->nocpu_base.events[idx]; | |
705 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
706 | #endif |
707 | return val; | |
708 | } | |
709 | ||
c0ff4b85 | 710 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 711 | bool anon, int nr_pages) |
d52aa412 | 712 | { |
c62b1a3b KH |
713 | preempt_disable(); |
714 | ||
b2402857 KH |
715 | /* |
716 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
717 | * counted as CACHE even if it's on ANON LRU. | |
718 | */ | |
719 | if (anon) | |
720 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 721 | nr_pages); |
d52aa412 | 722 | else |
b2402857 | 723 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 724 | nr_pages); |
55e462b0 | 725 | |
e401f176 KH |
726 | /* pagein of a big page is an event. So, ignore page size */ |
727 | if (nr_pages > 0) | |
c0ff4b85 | 728 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 729 | else { |
c0ff4b85 | 730 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
731 | nr_pages = -nr_pages; /* for event */ |
732 | } | |
e401f176 | 733 | |
13114716 | 734 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 735 | |
c62b1a3b | 736 | preempt_enable(); |
6d12e2d8 KH |
737 | } |
738 | ||
bb2a0de9 | 739 | unsigned long |
4d7dcca2 | 740 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
741 | { |
742 | struct mem_cgroup_per_zone *mz; | |
743 | ||
744 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
745 | return mz->lru_size[lru]; | |
746 | } | |
747 | ||
748 | static unsigned long | |
c0ff4b85 | 749 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 750 | unsigned int lru_mask) |
889976db YH |
751 | { |
752 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 753 | enum lru_list lru; |
bb2a0de9 KH |
754 | unsigned long ret = 0; |
755 | ||
c0ff4b85 | 756 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 757 | |
f156ab93 HD |
758 | for_each_lru(lru) { |
759 | if (BIT(lru) & lru_mask) | |
760 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
761 | } |
762 | return ret; | |
763 | } | |
764 | ||
765 | static unsigned long | |
c0ff4b85 | 766 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
767 | int nid, unsigned int lru_mask) |
768 | { | |
889976db YH |
769 | u64 total = 0; |
770 | int zid; | |
771 | ||
bb2a0de9 | 772 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
773 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
774 | nid, zid, lru_mask); | |
bb2a0de9 | 775 | |
889976db YH |
776 | return total; |
777 | } | |
bb2a0de9 | 778 | |
c0ff4b85 | 779 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 780 | unsigned int lru_mask) |
6d12e2d8 | 781 | { |
889976db | 782 | int nid; |
6d12e2d8 KH |
783 | u64 total = 0; |
784 | ||
bb2a0de9 | 785 | for_each_node_state(nid, N_HIGH_MEMORY) |
c0ff4b85 | 786 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 787 | return total; |
d52aa412 KH |
788 | } |
789 | ||
f53d7ce3 JW |
790 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
791 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
792 | { |
793 | unsigned long val, next; | |
794 | ||
13114716 | 795 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 796 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 797 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
798 | if ((long)next - (long)val < 0) { |
799 | switch (target) { | |
800 | case MEM_CGROUP_TARGET_THRESH: | |
801 | next = val + THRESHOLDS_EVENTS_TARGET; | |
802 | break; | |
803 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
804 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
805 | break; | |
806 | case MEM_CGROUP_TARGET_NUMAINFO: | |
807 | next = val + NUMAINFO_EVENTS_TARGET; | |
808 | break; | |
809 | default: | |
810 | break; | |
811 | } | |
812 | __this_cpu_write(memcg->stat->targets[target], next); | |
813 | return true; | |
7a159cc9 | 814 | } |
f53d7ce3 | 815 | return false; |
d2265e6f KH |
816 | } |
817 | ||
818 | /* | |
819 | * Check events in order. | |
820 | * | |
821 | */ | |
c0ff4b85 | 822 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 823 | { |
4799401f | 824 | preempt_disable(); |
d2265e6f | 825 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
826 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
827 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
828 | bool do_softlimit; |
829 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
830 | |
831 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
832 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
833 | #if MAX_NUMNODES > 1 | |
834 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
835 | MEM_CGROUP_TARGET_NUMAINFO); | |
836 | #endif | |
837 | preempt_enable(); | |
838 | ||
c0ff4b85 | 839 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 840 | if (unlikely(do_softlimit)) |
c0ff4b85 | 841 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 842 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 843 | if (unlikely(do_numainfo)) |
c0ff4b85 | 844 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 845 | #endif |
f53d7ce3 JW |
846 | } else |
847 | preempt_enable(); | |
d2265e6f KH |
848 | } |
849 | ||
d1a4c0b3 | 850 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
851 | { |
852 | return container_of(cgroup_subsys_state(cont, | |
853 | mem_cgroup_subsys_id), struct mem_cgroup, | |
854 | css); | |
855 | } | |
856 | ||
cf475ad2 | 857 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 858 | { |
31a78f23 BS |
859 | /* |
860 | * mm_update_next_owner() may clear mm->owner to NULL | |
861 | * if it races with swapoff, page migration, etc. | |
862 | * So this can be called with p == NULL. | |
863 | */ | |
864 | if (unlikely(!p)) | |
865 | return NULL; | |
866 | ||
78fb7466 PE |
867 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
868 | struct mem_cgroup, css); | |
869 | } | |
870 | ||
a433658c | 871 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 872 | { |
c0ff4b85 | 873 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
874 | |
875 | if (!mm) | |
876 | return NULL; | |
54595fe2 KH |
877 | /* |
878 | * Because we have no locks, mm->owner's may be being moved to other | |
879 | * cgroup. We use css_tryget() here even if this looks | |
880 | * pessimistic (rather than adding locks here). | |
881 | */ | |
882 | rcu_read_lock(); | |
883 | do { | |
c0ff4b85 R |
884 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
885 | if (unlikely(!memcg)) | |
54595fe2 | 886 | break; |
c0ff4b85 | 887 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 888 | rcu_read_unlock(); |
c0ff4b85 | 889 | return memcg; |
54595fe2 KH |
890 | } |
891 | ||
5660048c JW |
892 | /** |
893 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
894 | * @root: hierarchy root | |
895 | * @prev: previously returned memcg, NULL on first invocation | |
896 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
897 | * | |
898 | * Returns references to children of the hierarchy below @root, or | |
899 | * @root itself, or %NULL after a full round-trip. | |
900 | * | |
901 | * Caller must pass the return value in @prev on subsequent | |
902 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
903 | * to cancel a hierarchy walk before the round-trip is complete. | |
904 | * | |
905 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
906 | * divide up the memcgs in the hierarchy among all concurrent | |
907 | * reclaimers operating on the same zone and priority. | |
908 | */ | |
909 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
910 | struct mem_cgroup *prev, | |
911 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 912 | { |
9f3a0d09 JW |
913 | struct mem_cgroup *memcg = NULL; |
914 | int id = 0; | |
711d3d2c | 915 | |
5660048c JW |
916 | if (mem_cgroup_disabled()) |
917 | return NULL; | |
918 | ||
9f3a0d09 JW |
919 | if (!root) |
920 | root = root_mem_cgroup; | |
7d74b06f | 921 | |
9f3a0d09 JW |
922 | if (prev && !reclaim) |
923 | id = css_id(&prev->css); | |
14067bb3 | 924 | |
9f3a0d09 JW |
925 | if (prev && prev != root) |
926 | css_put(&prev->css); | |
14067bb3 | 927 | |
9f3a0d09 JW |
928 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
929 | if (prev) | |
930 | return NULL; | |
931 | return root; | |
932 | } | |
14067bb3 | 933 | |
9f3a0d09 | 934 | while (!memcg) { |
527a5ec9 | 935 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 936 | struct cgroup_subsys_state *css; |
711d3d2c | 937 | |
527a5ec9 JW |
938 | if (reclaim) { |
939 | int nid = zone_to_nid(reclaim->zone); | |
940 | int zid = zone_idx(reclaim->zone); | |
941 | struct mem_cgroup_per_zone *mz; | |
942 | ||
943 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
944 | iter = &mz->reclaim_iter[reclaim->priority]; | |
945 | if (prev && reclaim->generation != iter->generation) | |
946 | return NULL; | |
947 | id = iter->position; | |
948 | } | |
7d74b06f | 949 | |
9f3a0d09 JW |
950 | rcu_read_lock(); |
951 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
952 | if (css) { | |
953 | if (css == &root->css || css_tryget(css)) | |
954 | memcg = container_of(css, | |
955 | struct mem_cgroup, css); | |
956 | } else | |
957 | id = 0; | |
14067bb3 | 958 | rcu_read_unlock(); |
14067bb3 | 959 | |
527a5ec9 JW |
960 | if (reclaim) { |
961 | iter->position = id; | |
962 | if (!css) | |
963 | iter->generation++; | |
964 | else if (!prev && memcg) | |
965 | reclaim->generation = iter->generation; | |
966 | } | |
9f3a0d09 JW |
967 | |
968 | if (prev && !css) | |
969 | return NULL; | |
970 | } | |
971 | return memcg; | |
14067bb3 | 972 | } |
7d74b06f | 973 | |
5660048c JW |
974 | /** |
975 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
976 | * @root: hierarchy root | |
977 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
978 | */ | |
979 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
980 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
981 | { |
982 | if (!root) | |
983 | root = root_mem_cgroup; | |
984 | if (prev && prev != root) | |
985 | css_put(&prev->css); | |
986 | } | |
7d74b06f | 987 | |
9f3a0d09 JW |
988 | /* |
989 | * Iteration constructs for visiting all cgroups (under a tree). If | |
990 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
991 | * be used for reference counting. | |
992 | */ | |
993 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 994 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 995 | iter != NULL; \ |
527a5ec9 | 996 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 997 | |
9f3a0d09 | 998 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 999 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1000 | iter != NULL; \ |
527a5ec9 | 1001 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1002 | |
c0ff4b85 | 1003 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
4b3bde4c | 1004 | { |
c0ff4b85 | 1005 | return (memcg == root_mem_cgroup); |
4b3bde4c BS |
1006 | } |
1007 | ||
456f998e YH |
1008 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
1009 | { | |
c0ff4b85 | 1010 | struct mem_cgroup *memcg; |
456f998e YH |
1011 | |
1012 | if (!mm) | |
1013 | return; | |
1014 | ||
1015 | rcu_read_lock(); | |
c0ff4b85 R |
1016 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1017 | if (unlikely(!memcg)) | |
456f998e YH |
1018 | goto out; |
1019 | ||
1020 | switch (idx) { | |
456f998e | 1021 | case PGFAULT: |
0e574a93 JW |
1022 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1023 | break; | |
1024 | case PGMAJFAULT: | |
1025 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1026 | break; |
1027 | default: | |
1028 | BUG(); | |
1029 | } | |
1030 | out: | |
1031 | rcu_read_unlock(); | |
1032 | } | |
1033 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
1034 | ||
925b7673 JW |
1035 | /** |
1036 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1037 | * @zone: zone of the wanted lruvec | |
1038 | * @mem: memcg of the wanted lruvec | |
1039 | * | |
1040 | * Returns the lru list vector holding pages for the given @zone and | |
1041 | * @mem. This can be the global zone lruvec, if the memory controller | |
1042 | * is disabled. | |
1043 | */ | |
1044 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1045 | struct mem_cgroup *memcg) | |
1046 | { | |
1047 | struct mem_cgroup_per_zone *mz; | |
1048 | ||
1049 | if (mem_cgroup_disabled()) | |
1050 | return &zone->lruvec; | |
1051 | ||
1052 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
1053 | return &mz->lruvec; | |
1054 | } | |
1055 | ||
08e552c6 KH |
1056 | /* |
1057 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1058 | * Operations are called by routine of global LRU independently from memcg. | |
1059 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1060 | * | |
1061 | * Changes to pc->mem_cgroup happens when | |
1062 | * 1. charge | |
1063 | * 2. moving account | |
1064 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1065 | * It is added to LRU before charge. | |
1066 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1067 | * When moving account, the page is not on LRU. It's isolated. | |
1068 | */ | |
4f98a2fe | 1069 | |
925b7673 JW |
1070 | /** |
1071 | * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec | |
1072 | * @zone: zone of the page | |
1073 | * @page: the page | |
1074 | * @lru: current lru | |
1075 | * | |
1076 | * This function accounts for @page being added to @lru, and returns | |
1077 | * the lruvec for the given @zone and the memcg @page is charged to. | |
1078 | * | |
1079 | * The callsite is then responsible for physically linking the page to | |
1080 | * the returned lruvec->lists[@lru]. | |
1081 | */ | |
1082 | struct lruvec *mem_cgroup_lru_add_list(struct zone *zone, struct page *page, | |
1083 | enum lru_list lru) | |
08e552c6 | 1084 | { |
08e552c6 | 1085 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1086 | struct mem_cgroup *memcg; |
1087 | struct page_cgroup *pc; | |
6d12e2d8 | 1088 | |
f8d66542 | 1089 | if (mem_cgroup_disabled()) |
925b7673 JW |
1090 | return &zone->lruvec; |
1091 | ||
08e552c6 | 1092 | pc = lookup_page_cgroup(page); |
38c5d72f | 1093 | memcg = pc->mem_cgroup; |
7512102c HD |
1094 | |
1095 | /* | |
1096 | * Surreptitiously switch any uncharged page to root: | |
1097 | * an uncharged page off lru does nothing to secure | |
1098 | * its former mem_cgroup from sudden removal. | |
1099 | * | |
1100 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1101 | * under page_cgroup lock: between them, they make all uses | |
1102 | * of pc->mem_cgroup safe. | |
1103 | */ | |
1104 | if (!PageCgroupUsed(pc) && memcg != root_mem_cgroup) | |
1105 | pc->mem_cgroup = memcg = root_mem_cgroup; | |
1106 | ||
925b7673 JW |
1107 | mz = page_cgroup_zoneinfo(memcg, page); |
1108 | /* compound_order() is stabilized through lru_lock */ | |
1eb49272 | 1109 | mz->lru_size[lru] += 1 << compound_order(page); |
925b7673 | 1110 | return &mz->lruvec; |
08e552c6 | 1111 | } |
b69408e8 | 1112 | |
925b7673 JW |
1113 | /** |
1114 | * mem_cgroup_lru_del_list - account for removing an lru page | |
1115 | * @page: the page | |
1116 | * @lru: target lru | |
1117 | * | |
1118 | * This function accounts for @page being removed from @lru. | |
1119 | * | |
1120 | * The callsite is then responsible for physically unlinking | |
1121 | * @page->lru. | |
3f58a829 | 1122 | */ |
925b7673 | 1123 | void mem_cgroup_lru_del_list(struct page *page, enum lru_list lru) |
3f58a829 MK |
1124 | { |
1125 | struct mem_cgroup_per_zone *mz; | |
925b7673 | 1126 | struct mem_cgroup *memcg; |
3f58a829 | 1127 | struct page_cgroup *pc; |
3f58a829 MK |
1128 | |
1129 | if (mem_cgroup_disabled()) | |
1130 | return; | |
1131 | ||
1132 | pc = lookup_page_cgroup(page); | |
38c5d72f KH |
1133 | memcg = pc->mem_cgroup; |
1134 | VM_BUG_ON(!memcg); | |
925b7673 JW |
1135 | mz = page_cgroup_zoneinfo(memcg, page); |
1136 | /* huge page split is done under lru_lock. so, we have no races. */ | |
1eb49272 HD |
1137 | VM_BUG_ON(mz->lru_size[lru] < (1 << compound_order(page))); |
1138 | mz->lru_size[lru] -= 1 << compound_order(page); | |
3f58a829 MK |
1139 | } |
1140 | ||
925b7673 JW |
1141 | /** |
1142 | * mem_cgroup_lru_move_lists - account for moving a page between lrus | |
1143 | * @zone: zone of the page | |
1144 | * @page: the page | |
1145 | * @from: current lru | |
1146 | * @to: target lru | |
1147 | * | |
1148 | * This function accounts for @page being moved between the lrus @from | |
1149 | * and @to, and returns the lruvec for the given @zone and the memcg | |
1150 | * @page is charged to. | |
1151 | * | |
1152 | * The callsite is then responsible for physically relinking | |
1153 | * @page->lru to the returned lruvec->lists[@to]. | |
1154 | */ | |
1155 | struct lruvec *mem_cgroup_lru_move_lists(struct zone *zone, | |
1156 | struct page *page, | |
1157 | enum lru_list from, | |
1158 | enum lru_list to) | |
66e1707b | 1159 | { |
925b7673 JW |
1160 | /* XXX: Optimize this, especially for @from == @to */ |
1161 | mem_cgroup_lru_del_list(page, from); | |
1162 | return mem_cgroup_lru_add_list(zone, page, to); | |
08e552c6 | 1163 | } |
544122e5 | 1164 | |
3e92041d | 1165 | /* |
c0ff4b85 | 1166 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1167 | * hierarchy subtree |
1168 | */ | |
c3ac9a8a JW |
1169 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1170 | struct mem_cgroup *memcg) | |
3e92041d | 1171 | { |
91c63734 JW |
1172 | if (root_memcg == memcg) |
1173 | return true; | |
1174 | if (!root_memcg->use_hierarchy) | |
1175 | return false; | |
c3ac9a8a JW |
1176 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1177 | } | |
1178 | ||
1179 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1180 | struct mem_cgroup *memcg) | |
1181 | { | |
1182 | bool ret; | |
1183 | ||
91c63734 | 1184 | rcu_read_lock(); |
c3ac9a8a | 1185 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1186 | rcu_read_unlock(); |
1187 | return ret; | |
3e92041d MH |
1188 | } |
1189 | ||
c0ff4b85 | 1190 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1191 | { |
1192 | int ret; | |
0b7f569e | 1193 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1194 | struct task_struct *p; |
4c4a2214 | 1195 | |
158e0a2d | 1196 | p = find_lock_task_mm(task); |
de077d22 DR |
1197 | if (p) { |
1198 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1199 | task_unlock(p); | |
1200 | } else { | |
1201 | /* | |
1202 | * All threads may have already detached their mm's, but the oom | |
1203 | * killer still needs to detect if they have already been oom | |
1204 | * killed to prevent needlessly killing additional tasks. | |
1205 | */ | |
1206 | task_lock(task); | |
1207 | curr = mem_cgroup_from_task(task); | |
1208 | if (curr) | |
1209 | css_get(&curr->css); | |
1210 | task_unlock(task); | |
1211 | } | |
0b7f569e KH |
1212 | if (!curr) |
1213 | return 0; | |
d31f56db | 1214 | /* |
c0ff4b85 | 1215 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1216 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1217 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1218 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1219 | */ |
c0ff4b85 | 1220 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1221 | css_put(&curr->css); |
4c4a2214 DR |
1222 | return ret; |
1223 | } | |
1224 | ||
c56d5c7d | 1225 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1226 | { |
9b272977 | 1227 | unsigned long inactive_ratio; |
14797e23 | 1228 | unsigned long inactive; |
9b272977 | 1229 | unsigned long active; |
c772be93 | 1230 | unsigned long gb; |
14797e23 | 1231 | |
4d7dcca2 HD |
1232 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1233 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1234 | |
c772be93 KM |
1235 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1236 | if (gb) | |
1237 | inactive_ratio = int_sqrt(10 * gb); | |
1238 | else | |
1239 | inactive_ratio = 1; | |
1240 | ||
9b272977 | 1241 | return inactive * inactive_ratio < active; |
14797e23 KM |
1242 | } |
1243 | ||
c56d5c7d | 1244 | int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 RR |
1245 | { |
1246 | unsigned long active; | |
1247 | unsigned long inactive; | |
1248 | ||
4d7dcca2 HD |
1249 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE); |
1250 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 RR |
1251 | |
1252 | return (active > inactive); | |
1253 | } | |
1254 | ||
3e2f41f1 KM |
1255 | struct zone_reclaim_stat * |
1256 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1257 | { | |
1258 | struct page_cgroup *pc; | |
1259 | struct mem_cgroup_per_zone *mz; | |
1260 | ||
1261 | if (mem_cgroup_disabled()) | |
1262 | return NULL; | |
1263 | ||
1264 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1265 | if (!PageCgroupUsed(pc)) |
1266 | return NULL; | |
713735b4 JW |
1267 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1268 | smp_rmb(); | |
97a6c37b | 1269 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
89abfab1 | 1270 | return &mz->lruvec.reclaim_stat; |
3e2f41f1 KM |
1271 | } |
1272 | ||
6d61ef40 BS |
1273 | #define mem_cgroup_from_res_counter(counter, member) \ |
1274 | container_of(counter, struct mem_cgroup, member) | |
1275 | ||
19942822 | 1276 | /** |
9d11ea9f JW |
1277 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1278 | * @mem: the memory cgroup | |
19942822 | 1279 | * |
9d11ea9f | 1280 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1281 | * pages. |
19942822 | 1282 | */ |
c0ff4b85 | 1283 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1284 | { |
9d11ea9f JW |
1285 | unsigned long long margin; |
1286 | ||
c0ff4b85 | 1287 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1288 | if (do_swap_account) |
c0ff4b85 | 1289 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1290 | return margin >> PAGE_SHIFT; |
19942822 JW |
1291 | } |
1292 | ||
1f4c025b | 1293 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1294 | { |
1295 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1296 | |
1297 | /* root ? */ | |
1298 | if (cgrp->parent == NULL) | |
1299 | return vm_swappiness; | |
1300 | ||
bf1ff263 | 1301 | return memcg->swappiness; |
a7885eb8 KM |
1302 | } |
1303 | ||
619d094b KH |
1304 | /* |
1305 | * memcg->moving_account is used for checking possibility that some thread is | |
1306 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1307 | * a memcg, other threads should check memcg->moving_account under | |
1308 | * rcu_read_lock(), like this: | |
1309 | * | |
1310 | * CPU-A CPU-B | |
1311 | * rcu_read_lock() | |
1312 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1313 | * take heavy locks. | |
1314 | * synchronize_rcu() update something. | |
1315 | * rcu_read_unlock() | |
1316 | * start move here. | |
1317 | */ | |
4331f7d3 KH |
1318 | |
1319 | /* for quick checking without looking up memcg */ | |
1320 | atomic_t memcg_moving __read_mostly; | |
1321 | ||
c0ff4b85 | 1322 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1323 | { |
4331f7d3 | 1324 | atomic_inc(&memcg_moving); |
619d094b | 1325 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1326 | synchronize_rcu(); |
1327 | } | |
1328 | ||
c0ff4b85 | 1329 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1330 | { |
619d094b KH |
1331 | /* |
1332 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1333 | * We check NULL in callee rather than caller. | |
1334 | */ | |
4331f7d3 KH |
1335 | if (memcg) { |
1336 | atomic_dec(&memcg_moving); | |
619d094b | 1337 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1338 | } |
32047e2a | 1339 | } |
619d094b | 1340 | |
32047e2a KH |
1341 | /* |
1342 | * 2 routines for checking "mem" is under move_account() or not. | |
1343 | * | |
13fd1dd9 AM |
1344 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1345 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1346 | * pc->mem_cgroup may be overwritten. |
1347 | * | |
1348 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1349 | * under hierarchy of moving cgroups. This is for | |
1350 | * waiting at hith-memory prressure caused by "move". | |
1351 | */ | |
1352 | ||
13fd1dd9 | 1353 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1354 | { |
1355 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1356 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1357 | } |
4b534334 | 1358 | |
c0ff4b85 | 1359 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1360 | { |
2bd9bb20 KH |
1361 | struct mem_cgroup *from; |
1362 | struct mem_cgroup *to; | |
4b534334 | 1363 | bool ret = false; |
2bd9bb20 KH |
1364 | /* |
1365 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1366 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1367 | */ | |
1368 | spin_lock(&mc.lock); | |
1369 | from = mc.from; | |
1370 | to = mc.to; | |
1371 | if (!from) | |
1372 | goto unlock; | |
3e92041d | 1373 | |
c0ff4b85 R |
1374 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1375 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1376 | unlock: |
1377 | spin_unlock(&mc.lock); | |
4b534334 KH |
1378 | return ret; |
1379 | } | |
1380 | ||
c0ff4b85 | 1381 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1382 | { |
1383 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1384 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1385 | DEFINE_WAIT(wait); |
1386 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1387 | /* moving charge context might have finished. */ | |
1388 | if (mc.moving_task) | |
1389 | schedule(); | |
1390 | finish_wait(&mc.waitq, &wait); | |
1391 | return true; | |
1392 | } | |
1393 | } | |
1394 | return false; | |
1395 | } | |
1396 | ||
312734c0 KH |
1397 | /* |
1398 | * Take this lock when | |
1399 | * - a code tries to modify page's memcg while it's USED. | |
1400 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1401 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1402 | */ |
1403 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1404 | unsigned long *flags) | |
1405 | { | |
1406 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1407 | } | |
1408 | ||
1409 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1410 | unsigned long *flags) | |
1411 | { | |
1412 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1413 | } | |
1414 | ||
e222432b | 1415 | /** |
6a6135b6 | 1416 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1417 | * @memcg: The memory cgroup that went over limit |
1418 | * @p: Task that is going to be killed | |
1419 | * | |
1420 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1421 | * enabled | |
1422 | */ | |
1423 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1424 | { | |
1425 | struct cgroup *task_cgrp; | |
1426 | struct cgroup *mem_cgrp; | |
1427 | /* | |
1428 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1429 | * on the assumption that OOM is serialized for memory controller. | |
1430 | * If this assumption is broken, revisit this code. | |
1431 | */ | |
1432 | static char memcg_name[PATH_MAX]; | |
1433 | int ret; | |
1434 | ||
d31f56db | 1435 | if (!memcg || !p) |
e222432b BS |
1436 | return; |
1437 | ||
e222432b BS |
1438 | rcu_read_lock(); |
1439 | ||
1440 | mem_cgrp = memcg->css.cgroup; | |
1441 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1442 | ||
1443 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1444 | if (ret < 0) { | |
1445 | /* | |
1446 | * Unfortunately, we are unable to convert to a useful name | |
1447 | * But we'll still print out the usage information | |
1448 | */ | |
1449 | rcu_read_unlock(); | |
1450 | goto done; | |
1451 | } | |
1452 | rcu_read_unlock(); | |
1453 | ||
1454 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1455 | ||
1456 | rcu_read_lock(); | |
1457 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1458 | if (ret < 0) { | |
1459 | rcu_read_unlock(); | |
1460 | goto done; | |
1461 | } | |
1462 | rcu_read_unlock(); | |
1463 | ||
1464 | /* | |
1465 | * Continues from above, so we don't need an KERN_ level | |
1466 | */ | |
1467 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1468 | done: | |
1469 | ||
1470 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1471 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1472 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1473 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1474 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1475 | "failcnt %llu\n", | |
1476 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1477 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1478 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1479 | } | |
1480 | ||
81d39c20 KH |
1481 | /* |
1482 | * This function returns the number of memcg under hierarchy tree. Returns | |
1483 | * 1(self count) if no children. | |
1484 | */ | |
c0ff4b85 | 1485 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1486 | { |
1487 | int num = 0; | |
7d74b06f KH |
1488 | struct mem_cgroup *iter; |
1489 | ||
c0ff4b85 | 1490 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1491 | num++; |
81d39c20 KH |
1492 | return num; |
1493 | } | |
1494 | ||
a63d83f4 DR |
1495 | /* |
1496 | * Return the memory (and swap, if configured) limit for a memcg. | |
1497 | */ | |
1498 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1499 | { | |
1500 | u64 limit; | |
1501 | u64 memsw; | |
1502 | ||
f3e8eb70 JW |
1503 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1504 | limit += total_swap_pages << PAGE_SHIFT; | |
1505 | ||
a63d83f4 DR |
1506 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1507 | /* | |
1508 | * If memsw is finite and limits the amount of swap space available | |
1509 | * to this memcg, return that limit. | |
1510 | */ | |
1511 | return min(limit, memsw); | |
1512 | } | |
1513 | ||
5660048c JW |
1514 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1515 | gfp_t gfp_mask, | |
1516 | unsigned long flags) | |
1517 | { | |
1518 | unsigned long total = 0; | |
1519 | bool noswap = false; | |
1520 | int loop; | |
1521 | ||
1522 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1523 | noswap = true; | |
1524 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1525 | noswap = true; | |
1526 | ||
1527 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1528 | if (loop) | |
1529 | drain_all_stock_async(memcg); | |
1530 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1531 | /* | |
1532 | * Allow limit shrinkers, which are triggered directly | |
1533 | * by userspace, to catch signals and stop reclaim | |
1534 | * after minimal progress, regardless of the margin. | |
1535 | */ | |
1536 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1537 | break; | |
1538 | if (mem_cgroup_margin(memcg)) | |
1539 | break; | |
1540 | /* | |
1541 | * If nothing was reclaimed after two attempts, there | |
1542 | * may be no reclaimable pages in this hierarchy. | |
1543 | */ | |
1544 | if (loop && !total) | |
1545 | break; | |
1546 | } | |
1547 | return total; | |
1548 | } | |
1549 | ||
4d0c066d KH |
1550 | /** |
1551 | * test_mem_cgroup_node_reclaimable | |
1552 | * @mem: the target memcg | |
1553 | * @nid: the node ID to be checked. | |
1554 | * @noswap : specify true here if the user wants flle only information. | |
1555 | * | |
1556 | * This function returns whether the specified memcg contains any | |
1557 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1558 | * pages in the node. | |
1559 | */ | |
c0ff4b85 | 1560 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1561 | int nid, bool noswap) |
1562 | { | |
c0ff4b85 | 1563 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1564 | return true; |
1565 | if (noswap || !total_swap_pages) | |
1566 | return false; | |
c0ff4b85 | 1567 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1568 | return true; |
1569 | return false; | |
1570 | ||
1571 | } | |
889976db YH |
1572 | #if MAX_NUMNODES > 1 |
1573 | ||
1574 | /* | |
1575 | * Always updating the nodemask is not very good - even if we have an empty | |
1576 | * list or the wrong list here, we can start from some node and traverse all | |
1577 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1578 | * | |
1579 | */ | |
c0ff4b85 | 1580 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1581 | { |
1582 | int nid; | |
453a9bf3 KH |
1583 | /* |
1584 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1585 | * pagein/pageout changes since the last update. | |
1586 | */ | |
c0ff4b85 | 1587 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1588 | return; |
c0ff4b85 | 1589 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1590 | return; |
1591 | ||
889976db | 1592 | /* make a nodemask where this memcg uses memory from */ |
c0ff4b85 | 1593 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
889976db YH |
1594 | |
1595 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1596 | ||
c0ff4b85 R |
1597 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1598 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1599 | } |
453a9bf3 | 1600 | |
c0ff4b85 R |
1601 | atomic_set(&memcg->numainfo_events, 0); |
1602 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1603 | } |
1604 | ||
1605 | /* | |
1606 | * Selecting a node where we start reclaim from. Because what we need is just | |
1607 | * reducing usage counter, start from anywhere is O,K. Considering | |
1608 | * memory reclaim from current node, there are pros. and cons. | |
1609 | * | |
1610 | * Freeing memory from current node means freeing memory from a node which | |
1611 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1612 | * hit limits, it will see a contention on a node. But freeing from remote | |
1613 | * node means more costs for memory reclaim because of memory latency. | |
1614 | * | |
1615 | * Now, we use round-robin. Better algorithm is welcomed. | |
1616 | */ | |
c0ff4b85 | 1617 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1618 | { |
1619 | int node; | |
1620 | ||
c0ff4b85 R |
1621 | mem_cgroup_may_update_nodemask(memcg); |
1622 | node = memcg->last_scanned_node; | |
889976db | 1623 | |
c0ff4b85 | 1624 | node = next_node(node, memcg->scan_nodes); |
889976db | 1625 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1626 | node = first_node(memcg->scan_nodes); |
889976db YH |
1627 | /* |
1628 | * We call this when we hit limit, not when pages are added to LRU. | |
1629 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1630 | * memcg is too small and all pages are not on LRU. In that case, | |
1631 | * we use curret node. | |
1632 | */ | |
1633 | if (unlikely(node == MAX_NUMNODES)) | |
1634 | node = numa_node_id(); | |
1635 | ||
c0ff4b85 | 1636 | memcg->last_scanned_node = node; |
889976db YH |
1637 | return node; |
1638 | } | |
1639 | ||
4d0c066d KH |
1640 | /* |
1641 | * Check all nodes whether it contains reclaimable pages or not. | |
1642 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1643 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1644 | * enough new information. We need to do double check. | |
1645 | */ | |
6bbda35c | 1646 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1647 | { |
1648 | int nid; | |
1649 | ||
1650 | /* | |
1651 | * quick check...making use of scan_node. | |
1652 | * We can skip unused nodes. | |
1653 | */ | |
c0ff4b85 R |
1654 | if (!nodes_empty(memcg->scan_nodes)) { |
1655 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1656 | nid < MAX_NUMNODES; |
c0ff4b85 | 1657 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1658 | |
c0ff4b85 | 1659 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1660 | return true; |
1661 | } | |
1662 | } | |
1663 | /* | |
1664 | * Check rest of nodes. | |
1665 | */ | |
1666 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
c0ff4b85 | 1667 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1668 | continue; |
c0ff4b85 | 1669 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1670 | return true; |
1671 | } | |
1672 | return false; | |
1673 | } | |
1674 | ||
889976db | 1675 | #else |
c0ff4b85 | 1676 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1677 | { |
1678 | return 0; | |
1679 | } | |
4d0c066d | 1680 | |
6bbda35c | 1681 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1682 | { |
c0ff4b85 | 1683 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1684 | } |
889976db YH |
1685 | #endif |
1686 | ||
5660048c JW |
1687 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1688 | struct zone *zone, | |
1689 | gfp_t gfp_mask, | |
1690 | unsigned long *total_scanned) | |
6d61ef40 | 1691 | { |
9f3a0d09 | 1692 | struct mem_cgroup *victim = NULL; |
5660048c | 1693 | int total = 0; |
04046e1a | 1694 | int loop = 0; |
9d11ea9f | 1695 | unsigned long excess; |
185efc0f | 1696 | unsigned long nr_scanned; |
527a5ec9 JW |
1697 | struct mem_cgroup_reclaim_cookie reclaim = { |
1698 | .zone = zone, | |
1699 | .priority = 0, | |
1700 | }; | |
9d11ea9f | 1701 | |
c0ff4b85 | 1702 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1703 | |
4e416953 | 1704 | while (1) { |
527a5ec9 | 1705 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1706 | if (!victim) { |
04046e1a | 1707 | loop++; |
4e416953 BS |
1708 | if (loop >= 2) { |
1709 | /* | |
1710 | * If we have not been able to reclaim | |
1711 | * anything, it might because there are | |
1712 | * no reclaimable pages under this hierarchy | |
1713 | */ | |
5660048c | 1714 | if (!total) |
4e416953 | 1715 | break; |
4e416953 | 1716 | /* |
25985edc | 1717 | * We want to do more targeted reclaim. |
4e416953 BS |
1718 | * excess >> 2 is not to excessive so as to |
1719 | * reclaim too much, nor too less that we keep | |
1720 | * coming back to reclaim from this cgroup | |
1721 | */ | |
1722 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1723 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1724 | break; |
4e416953 | 1725 | } |
9f3a0d09 | 1726 | continue; |
4e416953 | 1727 | } |
5660048c | 1728 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1729 | continue; |
5660048c JW |
1730 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1731 | zone, &nr_scanned); | |
1732 | *total_scanned += nr_scanned; | |
1733 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1734 | break; |
6d61ef40 | 1735 | } |
9f3a0d09 | 1736 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1737 | return total; |
6d61ef40 BS |
1738 | } |
1739 | ||
867578cb KH |
1740 | /* |
1741 | * Check OOM-Killer is already running under our hierarchy. | |
1742 | * If someone is running, return false. | |
1af8efe9 | 1743 | * Has to be called with memcg_oom_lock |
867578cb | 1744 | */ |
c0ff4b85 | 1745 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1746 | { |
79dfdacc | 1747 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1748 | |
9f3a0d09 | 1749 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1750 | if (iter->oom_lock) { |
79dfdacc MH |
1751 | /* |
1752 | * this subtree of our hierarchy is already locked | |
1753 | * so we cannot give a lock. | |
1754 | */ | |
79dfdacc | 1755 | failed = iter; |
9f3a0d09 JW |
1756 | mem_cgroup_iter_break(memcg, iter); |
1757 | break; | |
23751be0 JW |
1758 | } else |
1759 | iter->oom_lock = true; | |
7d74b06f | 1760 | } |
867578cb | 1761 | |
79dfdacc | 1762 | if (!failed) |
23751be0 | 1763 | return true; |
79dfdacc MH |
1764 | |
1765 | /* | |
1766 | * OK, we failed to lock the whole subtree so we have to clean up | |
1767 | * what we set up to the failing subtree | |
1768 | */ | |
9f3a0d09 | 1769 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1770 | if (iter == failed) { |
9f3a0d09 JW |
1771 | mem_cgroup_iter_break(memcg, iter); |
1772 | break; | |
79dfdacc MH |
1773 | } |
1774 | iter->oom_lock = false; | |
1775 | } | |
23751be0 | 1776 | return false; |
a636b327 | 1777 | } |
0b7f569e | 1778 | |
79dfdacc | 1779 | /* |
1af8efe9 | 1780 | * Has to be called with memcg_oom_lock |
79dfdacc | 1781 | */ |
c0ff4b85 | 1782 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1783 | { |
7d74b06f KH |
1784 | struct mem_cgroup *iter; |
1785 | ||
c0ff4b85 | 1786 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1787 | iter->oom_lock = false; |
1788 | return 0; | |
1789 | } | |
1790 | ||
c0ff4b85 | 1791 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1792 | { |
1793 | struct mem_cgroup *iter; | |
1794 | ||
c0ff4b85 | 1795 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1796 | atomic_inc(&iter->under_oom); |
1797 | } | |
1798 | ||
c0ff4b85 | 1799 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1800 | { |
1801 | struct mem_cgroup *iter; | |
1802 | ||
867578cb KH |
1803 | /* |
1804 | * When a new child is created while the hierarchy is under oom, | |
1805 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1806 | * atomic_add_unless() here. | |
1807 | */ | |
c0ff4b85 | 1808 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1809 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1810 | } |
1811 | ||
1af8efe9 | 1812 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1813 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1814 | ||
dc98df5a | 1815 | struct oom_wait_info { |
d79154bb | 1816 | struct mem_cgroup *memcg; |
dc98df5a KH |
1817 | wait_queue_t wait; |
1818 | }; | |
1819 | ||
1820 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1821 | unsigned mode, int sync, void *arg) | |
1822 | { | |
d79154bb HD |
1823 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1824 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1825 | struct oom_wait_info *oom_wait_info; |
1826 | ||
1827 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1828 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1829 | |
dc98df5a | 1830 | /* |
d79154bb | 1831 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1832 | * Then we can use css_is_ancestor without taking care of RCU. |
1833 | */ | |
c0ff4b85 R |
1834 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1835 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1836 | return 0; |
dc98df5a KH |
1837 | return autoremove_wake_function(wait, mode, sync, arg); |
1838 | } | |
1839 | ||
c0ff4b85 | 1840 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1841 | { |
c0ff4b85 R |
1842 | /* for filtering, pass "memcg" as argument. */ |
1843 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1844 | } |
1845 | ||
c0ff4b85 | 1846 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1847 | { |
c0ff4b85 R |
1848 | if (memcg && atomic_read(&memcg->under_oom)) |
1849 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1850 | } |
1851 | ||
867578cb KH |
1852 | /* |
1853 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1854 | */ | |
6bbda35c KS |
1855 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
1856 | int order) | |
0b7f569e | 1857 | { |
dc98df5a | 1858 | struct oom_wait_info owait; |
3c11ecf4 | 1859 | bool locked, need_to_kill; |
867578cb | 1860 | |
d79154bb | 1861 | owait.memcg = memcg; |
dc98df5a KH |
1862 | owait.wait.flags = 0; |
1863 | owait.wait.func = memcg_oom_wake_function; | |
1864 | owait.wait.private = current; | |
1865 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1866 | need_to_kill = true; |
c0ff4b85 | 1867 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1868 | |
c0ff4b85 | 1869 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1870 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1871 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1872 | /* |
1873 | * Even if signal_pending(), we can't quit charge() loop without | |
1874 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1875 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1876 | */ | |
3c11ecf4 | 1877 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1878 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1879 | need_to_kill = false; |
1880 | if (locked) | |
c0ff4b85 | 1881 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1882 | spin_unlock(&memcg_oom_lock); |
867578cb | 1883 | |
3c11ecf4 KH |
1884 | if (need_to_kill) { |
1885 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 1886 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 1887 | } else { |
867578cb | 1888 | schedule(); |
dc98df5a | 1889 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1890 | } |
1af8efe9 | 1891 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1892 | if (locked) |
c0ff4b85 R |
1893 | mem_cgroup_oom_unlock(memcg); |
1894 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1895 | spin_unlock(&memcg_oom_lock); |
867578cb | 1896 | |
c0ff4b85 | 1897 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1898 | |
867578cb KH |
1899 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1900 | return false; | |
1901 | /* Give chance to dying process */ | |
715a5ee8 | 1902 | schedule_timeout_uninterruptible(1); |
867578cb | 1903 | return true; |
0b7f569e KH |
1904 | } |
1905 | ||
d69b042f BS |
1906 | /* |
1907 | * Currently used to update mapped file statistics, but the routine can be | |
1908 | * generalized to update other statistics as well. | |
32047e2a KH |
1909 | * |
1910 | * Notes: Race condition | |
1911 | * | |
1912 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1913 | * it tends to be costly. But considering some conditions, we doesn't need | |
1914 | * to do so _always_. | |
1915 | * | |
1916 | * Considering "charge", lock_page_cgroup() is not required because all | |
1917 | * file-stat operations happen after a page is attached to radix-tree. There | |
1918 | * are no race with "charge". | |
1919 | * | |
1920 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1921 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1922 | * if there are race with "uncharge". Statistics itself is properly handled | |
1923 | * by flags. | |
1924 | * | |
1925 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
1926 | * small, we check mm->moving_account and detect there are possibility of race |
1927 | * If there is, we take a lock. | |
d69b042f | 1928 | */ |
26174efd | 1929 | |
89c06bd5 KH |
1930 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
1931 | bool *locked, unsigned long *flags) | |
1932 | { | |
1933 | struct mem_cgroup *memcg; | |
1934 | struct page_cgroup *pc; | |
1935 | ||
1936 | pc = lookup_page_cgroup(page); | |
1937 | again: | |
1938 | memcg = pc->mem_cgroup; | |
1939 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
1940 | return; | |
1941 | /* | |
1942 | * If this memory cgroup is not under account moving, we don't | |
1943 | * need to take move_lock_page_cgroup(). Because we already hold | |
1944 | * rcu_read_lock(), any calls to move_account will be delayed until | |
13fd1dd9 | 1945 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 1946 | */ |
13fd1dd9 | 1947 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
1948 | return; |
1949 | ||
1950 | move_lock_mem_cgroup(memcg, flags); | |
1951 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
1952 | move_unlock_mem_cgroup(memcg, flags); | |
1953 | goto again; | |
1954 | } | |
1955 | *locked = true; | |
1956 | } | |
1957 | ||
1958 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
1959 | { | |
1960 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1961 | ||
1962 | /* | |
1963 | * It's guaranteed that pc->mem_cgroup never changes while | |
1964 | * lock is held because a routine modifies pc->mem_cgroup | |
1965 | * should take move_lock_page_cgroup(). | |
1966 | */ | |
1967 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
1968 | } | |
1969 | ||
2a7106f2 GT |
1970 | void mem_cgroup_update_page_stat(struct page *page, |
1971 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 1972 | { |
c0ff4b85 | 1973 | struct mem_cgroup *memcg; |
32047e2a | 1974 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 1975 | unsigned long uninitialized_var(flags); |
d69b042f | 1976 | |
cfa44946 | 1977 | if (mem_cgroup_disabled()) |
d69b042f | 1978 | return; |
89c06bd5 | 1979 | |
c0ff4b85 R |
1980 | memcg = pc->mem_cgroup; |
1981 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 1982 | return; |
26174efd | 1983 | |
26174efd | 1984 | switch (idx) { |
2a7106f2 | 1985 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 1986 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
1987 | break; |
1988 | default: | |
1989 | BUG(); | |
8725d541 | 1990 | } |
d69b042f | 1991 | |
c0ff4b85 | 1992 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 1993 | } |
26174efd | 1994 | |
cdec2e42 KH |
1995 | /* |
1996 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1997 | * TODO: maybe necessary to use big numbers in big irons. | |
1998 | */ | |
7ec99d62 | 1999 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2000 | struct memcg_stock_pcp { |
2001 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2002 | unsigned int nr_pages; |
cdec2e42 | 2003 | struct work_struct work; |
26fe6168 | 2004 | unsigned long flags; |
a0db00fc | 2005 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2006 | }; |
2007 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2008 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
2009 | |
2010 | /* | |
11c9ea4e | 2011 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
2012 | * from local stock and true is returned. If the stock is 0 or charges from a |
2013 | * cgroup which is not current target, returns false. This stock will be | |
2014 | * refilled. | |
2015 | */ | |
c0ff4b85 | 2016 | static bool consume_stock(struct mem_cgroup *memcg) |
cdec2e42 KH |
2017 | { |
2018 | struct memcg_stock_pcp *stock; | |
2019 | bool ret = true; | |
2020 | ||
2021 | stock = &get_cpu_var(memcg_stock); | |
c0ff4b85 | 2022 | if (memcg == stock->cached && stock->nr_pages) |
11c9ea4e | 2023 | stock->nr_pages--; |
cdec2e42 KH |
2024 | else /* need to call res_counter_charge */ |
2025 | ret = false; | |
2026 | put_cpu_var(memcg_stock); | |
2027 | return ret; | |
2028 | } | |
2029 | ||
2030 | /* | |
2031 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2032 | */ | |
2033 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2034 | { | |
2035 | struct mem_cgroup *old = stock->cached; | |
2036 | ||
11c9ea4e JW |
2037 | if (stock->nr_pages) { |
2038 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2039 | ||
2040 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2041 | if (do_swap_account) |
11c9ea4e JW |
2042 | res_counter_uncharge(&old->memsw, bytes); |
2043 | stock->nr_pages = 0; | |
cdec2e42 KH |
2044 | } |
2045 | stock->cached = NULL; | |
cdec2e42 KH |
2046 | } |
2047 | ||
2048 | /* | |
2049 | * This must be called under preempt disabled or must be called by | |
2050 | * a thread which is pinned to local cpu. | |
2051 | */ | |
2052 | static void drain_local_stock(struct work_struct *dummy) | |
2053 | { | |
2054 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2055 | drain_stock(stock); | |
26fe6168 | 2056 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2057 | } |
2058 | ||
2059 | /* | |
2060 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2061 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2062 | */ |
c0ff4b85 | 2063 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2064 | { |
2065 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2066 | ||
c0ff4b85 | 2067 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2068 | drain_stock(stock); |
c0ff4b85 | 2069 | stock->cached = memcg; |
cdec2e42 | 2070 | } |
11c9ea4e | 2071 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2072 | put_cpu_var(memcg_stock); |
2073 | } | |
2074 | ||
2075 | /* | |
c0ff4b85 | 2076 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2077 | * of the hierarchy under it. sync flag says whether we should block |
2078 | * until the work is done. | |
cdec2e42 | 2079 | */ |
c0ff4b85 | 2080 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2081 | { |
26fe6168 | 2082 | int cpu, curcpu; |
d38144b7 | 2083 | |
cdec2e42 | 2084 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2085 | get_online_cpus(); |
5af12d0e | 2086 | curcpu = get_cpu(); |
cdec2e42 KH |
2087 | for_each_online_cpu(cpu) { |
2088 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2089 | struct mem_cgroup *memcg; |
26fe6168 | 2090 | |
c0ff4b85 R |
2091 | memcg = stock->cached; |
2092 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2093 | continue; |
c0ff4b85 | 2094 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2095 | continue; |
d1a05b69 MH |
2096 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2097 | if (cpu == curcpu) | |
2098 | drain_local_stock(&stock->work); | |
2099 | else | |
2100 | schedule_work_on(cpu, &stock->work); | |
2101 | } | |
cdec2e42 | 2102 | } |
5af12d0e | 2103 | put_cpu(); |
d38144b7 MH |
2104 | |
2105 | if (!sync) | |
2106 | goto out; | |
2107 | ||
2108 | for_each_online_cpu(cpu) { | |
2109 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2110 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2111 | flush_work(&stock->work); |
2112 | } | |
2113 | out: | |
cdec2e42 | 2114 | put_online_cpus(); |
d38144b7 MH |
2115 | } |
2116 | ||
2117 | /* | |
2118 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2119 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2120 | * expects some charges will be back to res_counter later but cannot wait for | |
2121 | * it. | |
2122 | */ | |
c0ff4b85 | 2123 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2124 | { |
9f50fad6 MH |
2125 | /* |
2126 | * If someone calls draining, avoid adding more kworker runs. | |
2127 | */ | |
2128 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2129 | return; | |
c0ff4b85 | 2130 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2131 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2132 | } |
2133 | ||
2134 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2135 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2136 | { |
2137 | /* called when force_empty is called */ | |
9f50fad6 | 2138 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2139 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2140 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2141 | } |
2142 | ||
711d3d2c KH |
2143 | /* |
2144 | * This function drains percpu counter value from DEAD cpu and | |
2145 | * move it to local cpu. Note that this function can be preempted. | |
2146 | */ | |
c0ff4b85 | 2147 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2148 | { |
2149 | int i; | |
2150 | ||
c0ff4b85 | 2151 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2152 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2153 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2154 | |
c0ff4b85 R |
2155 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2156 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2157 | } |
e9f8974f | 2158 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2159 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2160 | |
c0ff4b85 R |
2161 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2162 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2163 | } |
c0ff4b85 | 2164 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2165 | } |
2166 | ||
2167 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2168 | unsigned long action, |
2169 | void *hcpu) | |
2170 | { | |
2171 | int cpu = (unsigned long)hcpu; | |
2172 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2173 | struct mem_cgroup *iter; |
cdec2e42 | 2174 | |
619d094b | 2175 | if (action == CPU_ONLINE) |
1489ebad | 2176 | return NOTIFY_OK; |
1489ebad | 2177 | |
d833049b | 2178 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2179 | return NOTIFY_OK; |
711d3d2c | 2180 | |
9f3a0d09 | 2181 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2182 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2183 | ||
cdec2e42 KH |
2184 | stock = &per_cpu(memcg_stock, cpu); |
2185 | drain_stock(stock); | |
2186 | return NOTIFY_OK; | |
2187 | } | |
2188 | ||
4b534334 KH |
2189 | |
2190 | /* See __mem_cgroup_try_charge() for details */ | |
2191 | enum { | |
2192 | CHARGE_OK, /* success */ | |
2193 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2194 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2195 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2196 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2197 | }; | |
2198 | ||
c0ff4b85 | 2199 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
7ec99d62 | 2200 | unsigned int nr_pages, bool oom_check) |
4b534334 | 2201 | { |
7ec99d62 | 2202 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2203 | struct mem_cgroup *mem_over_limit; |
2204 | struct res_counter *fail_res; | |
2205 | unsigned long flags = 0; | |
2206 | int ret; | |
2207 | ||
c0ff4b85 | 2208 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2209 | |
2210 | if (likely(!ret)) { | |
2211 | if (!do_swap_account) | |
2212 | return CHARGE_OK; | |
c0ff4b85 | 2213 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2214 | if (likely(!ret)) |
2215 | return CHARGE_OK; | |
2216 | ||
c0ff4b85 | 2217 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2218 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2219 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2220 | } else | |
2221 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2222 | /* |
7ec99d62 JW |
2223 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2224 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2225 | * |
2226 | * Never reclaim on behalf of optional batching, retry with a | |
2227 | * single page instead. | |
2228 | */ | |
7ec99d62 | 2229 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2230 | return CHARGE_RETRY; |
2231 | ||
2232 | if (!(gfp_mask & __GFP_WAIT)) | |
2233 | return CHARGE_WOULDBLOCK; | |
2234 | ||
5660048c | 2235 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2236 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2237 | return CHARGE_RETRY; |
4b534334 | 2238 | /* |
19942822 JW |
2239 | * Even though the limit is exceeded at this point, reclaim |
2240 | * may have been able to free some pages. Retry the charge | |
2241 | * before killing the task. | |
2242 | * | |
2243 | * Only for regular pages, though: huge pages are rather | |
2244 | * unlikely to succeed so close to the limit, and we fall back | |
2245 | * to regular pages anyway in case of failure. | |
4b534334 | 2246 | */ |
7ec99d62 | 2247 | if (nr_pages == 1 && ret) |
4b534334 KH |
2248 | return CHARGE_RETRY; |
2249 | ||
2250 | /* | |
2251 | * At task move, charge accounts can be doubly counted. So, it's | |
2252 | * better to wait until the end of task_move if something is going on. | |
2253 | */ | |
2254 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2255 | return CHARGE_RETRY; | |
2256 | ||
2257 | /* If we don't need to call oom-killer at el, return immediately */ | |
2258 | if (!oom_check) | |
2259 | return CHARGE_NOMEM; | |
2260 | /* check OOM */ | |
e845e199 | 2261 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2262 | return CHARGE_OOM_DIE; |
2263 | ||
2264 | return CHARGE_RETRY; | |
2265 | } | |
2266 | ||
f817ed48 | 2267 | /* |
38c5d72f KH |
2268 | * __mem_cgroup_try_charge() does |
2269 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2270 | * 2. update res_counter | |
2271 | * 3. call memory reclaim if necessary. | |
2272 | * | |
2273 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2274 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2275 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2276 | * as possible without any hazards. 2: all pages should have a valid | |
2277 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2278 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2279 | * | |
2280 | * So __mem_cgroup_try_charge() will return | |
2281 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2282 | * -ENOMEM ... charge failure because of resource limits. | |
2283 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2284 | * | |
2285 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2286 | * the oom-killer can be invoked. | |
8a9f3ccd | 2287 | */ |
f817ed48 | 2288 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2289 | gfp_t gfp_mask, |
7ec99d62 | 2290 | unsigned int nr_pages, |
c0ff4b85 | 2291 | struct mem_cgroup **ptr, |
7ec99d62 | 2292 | bool oom) |
8a9f3ccd | 2293 | { |
7ec99d62 | 2294 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2295 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2296 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2297 | int ret; |
a636b327 | 2298 | |
867578cb KH |
2299 | /* |
2300 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2301 | * in system level. So, allow to go ahead dying process in addition to | |
2302 | * MEMDIE process. | |
2303 | */ | |
2304 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2305 | || fatal_signal_pending(current))) | |
2306 | goto bypass; | |
a636b327 | 2307 | |
8a9f3ccd | 2308 | /* |
3be91277 HD |
2309 | * We always charge the cgroup the mm_struct belongs to. |
2310 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2311 | * thread group leader migrates. It's possible that mm is not |
2312 | * set, if so charge the init_mm (happens for pagecache usage). | |
2313 | */ | |
c0ff4b85 | 2314 | if (!*ptr && !mm) |
38c5d72f | 2315 | *ptr = root_mem_cgroup; |
f75ca962 | 2316 | again: |
c0ff4b85 R |
2317 | if (*ptr) { /* css should be a valid one */ |
2318 | memcg = *ptr; | |
2319 | VM_BUG_ON(css_is_removed(&memcg->css)); | |
2320 | if (mem_cgroup_is_root(memcg)) | |
f75ca962 | 2321 | goto done; |
c0ff4b85 | 2322 | if (nr_pages == 1 && consume_stock(memcg)) |
f75ca962 | 2323 | goto done; |
c0ff4b85 | 2324 | css_get(&memcg->css); |
4b534334 | 2325 | } else { |
f75ca962 | 2326 | struct task_struct *p; |
54595fe2 | 2327 | |
f75ca962 KH |
2328 | rcu_read_lock(); |
2329 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2330 | /* |
ebb76ce1 | 2331 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2332 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2333 | * race with swapoff. Then, we have small risk of mis-accouning. |
2334 | * But such kind of mis-account by race always happens because | |
2335 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2336 | * small race, here. | |
2337 | * (*) swapoff at el will charge against mm-struct not against | |
2338 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2339 | */ |
c0ff4b85 | 2340 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2341 | if (!memcg) |
2342 | memcg = root_mem_cgroup; | |
2343 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2344 | rcu_read_unlock(); |
2345 | goto done; | |
2346 | } | |
c0ff4b85 | 2347 | if (nr_pages == 1 && consume_stock(memcg)) { |
f75ca962 KH |
2348 | /* |
2349 | * It seems dagerous to access memcg without css_get(). | |
2350 | * But considering how consume_stok works, it's not | |
2351 | * necessary. If consume_stock success, some charges | |
2352 | * from this memcg are cached on this cpu. So, we | |
2353 | * don't need to call css_get()/css_tryget() before | |
2354 | * calling consume_stock(). | |
2355 | */ | |
2356 | rcu_read_unlock(); | |
2357 | goto done; | |
2358 | } | |
2359 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2360 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2361 | rcu_read_unlock(); |
2362 | goto again; | |
2363 | } | |
2364 | rcu_read_unlock(); | |
2365 | } | |
8a9f3ccd | 2366 | |
4b534334 KH |
2367 | do { |
2368 | bool oom_check; | |
7a81b88c | 2369 | |
4b534334 | 2370 | /* If killed, bypass charge */ |
f75ca962 | 2371 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2372 | css_put(&memcg->css); |
4b534334 | 2373 | goto bypass; |
f75ca962 | 2374 | } |
6d61ef40 | 2375 | |
4b534334 KH |
2376 | oom_check = false; |
2377 | if (oom && !nr_oom_retries) { | |
2378 | oom_check = true; | |
2379 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2380 | } |
66e1707b | 2381 | |
c0ff4b85 | 2382 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
4b534334 KH |
2383 | switch (ret) { |
2384 | case CHARGE_OK: | |
2385 | break; | |
2386 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2387 | batch = nr_pages; |
c0ff4b85 R |
2388 | css_put(&memcg->css); |
2389 | memcg = NULL; | |
f75ca962 | 2390 | goto again; |
4b534334 | 2391 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2392 | css_put(&memcg->css); |
4b534334 KH |
2393 | goto nomem; |
2394 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2395 | if (!oom) { |
c0ff4b85 | 2396 | css_put(&memcg->css); |
867578cb | 2397 | goto nomem; |
f75ca962 | 2398 | } |
4b534334 KH |
2399 | /* If oom, we never return -ENOMEM */ |
2400 | nr_oom_retries--; | |
2401 | break; | |
2402 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2403 | css_put(&memcg->css); |
867578cb | 2404 | goto bypass; |
66e1707b | 2405 | } |
4b534334 KH |
2406 | } while (ret != CHARGE_OK); |
2407 | ||
7ec99d62 | 2408 | if (batch > nr_pages) |
c0ff4b85 R |
2409 | refill_stock(memcg, batch - nr_pages); |
2410 | css_put(&memcg->css); | |
0c3e73e8 | 2411 | done: |
c0ff4b85 | 2412 | *ptr = memcg; |
7a81b88c KH |
2413 | return 0; |
2414 | nomem: | |
c0ff4b85 | 2415 | *ptr = NULL; |
7a81b88c | 2416 | return -ENOMEM; |
867578cb | 2417 | bypass: |
38c5d72f KH |
2418 | *ptr = root_mem_cgroup; |
2419 | return -EINTR; | |
7a81b88c | 2420 | } |
8a9f3ccd | 2421 | |
a3032a2c DN |
2422 | /* |
2423 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2424 | * This function is for that and do uncharge, put css's refcnt. | |
2425 | * gotten by try_charge(). | |
2426 | */ | |
c0ff4b85 | 2427 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2428 | unsigned int nr_pages) |
a3032a2c | 2429 | { |
c0ff4b85 | 2430 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2431 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2432 | ||
c0ff4b85 | 2433 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2434 | if (do_swap_account) |
c0ff4b85 | 2435 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2436 | } |
854ffa8d DN |
2437 | } |
2438 | ||
d01dd17f KH |
2439 | /* |
2440 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2441 | * This is useful when moving usage to parent cgroup. | |
2442 | */ | |
2443 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2444 | unsigned int nr_pages) | |
2445 | { | |
2446 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2447 | ||
2448 | if (mem_cgroup_is_root(memcg)) | |
2449 | return; | |
2450 | ||
2451 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2452 | if (do_swap_account) | |
2453 | res_counter_uncharge_until(&memcg->memsw, | |
2454 | memcg->memsw.parent, bytes); | |
2455 | } | |
2456 | ||
a3b2d692 KH |
2457 | /* |
2458 | * A helper function to get mem_cgroup from ID. must be called under | |
2459 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2460 | * it's concern. (dropping refcnt from swap can be called against removed | |
2461 | * memcg.) | |
2462 | */ | |
2463 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2464 | { | |
2465 | struct cgroup_subsys_state *css; | |
2466 | ||
2467 | /* ID 0 is unused ID */ | |
2468 | if (!id) | |
2469 | return NULL; | |
2470 | css = css_lookup(&mem_cgroup_subsys, id); | |
2471 | if (!css) | |
2472 | return NULL; | |
2473 | return container_of(css, struct mem_cgroup, css); | |
2474 | } | |
2475 | ||
e42d9d5d | 2476 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2477 | { |
c0ff4b85 | 2478 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2479 | struct page_cgroup *pc; |
a3b2d692 | 2480 | unsigned short id; |
b5a84319 KH |
2481 | swp_entry_t ent; |
2482 | ||
3c776e64 DN |
2483 | VM_BUG_ON(!PageLocked(page)); |
2484 | ||
3c776e64 | 2485 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2486 | lock_page_cgroup(pc); |
a3b2d692 | 2487 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2488 | memcg = pc->mem_cgroup; |
2489 | if (memcg && !css_tryget(&memcg->css)) | |
2490 | memcg = NULL; | |
e42d9d5d | 2491 | } else if (PageSwapCache(page)) { |
3c776e64 | 2492 | ent.val = page_private(page); |
9fb4b7cc | 2493 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2494 | rcu_read_lock(); |
c0ff4b85 R |
2495 | memcg = mem_cgroup_lookup(id); |
2496 | if (memcg && !css_tryget(&memcg->css)) | |
2497 | memcg = NULL; | |
a3b2d692 | 2498 | rcu_read_unlock(); |
3c776e64 | 2499 | } |
c0bd3f63 | 2500 | unlock_page_cgroup(pc); |
c0ff4b85 | 2501 | return memcg; |
b5a84319 KH |
2502 | } |
2503 | ||
c0ff4b85 | 2504 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2505 | struct page *page, |
7ec99d62 | 2506 | unsigned int nr_pages, |
9ce70c02 HD |
2507 | enum charge_type ctype, |
2508 | bool lrucare) | |
7a81b88c | 2509 | { |
ce587e65 | 2510 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 HD |
2511 | struct zone *uninitialized_var(zone); |
2512 | bool was_on_lru = false; | |
b2402857 | 2513 | bool anon; |
9ce70c02 | 2514 | |
ca3e0214 KH |
2515 | lock_page_cgroup(pc); |
2516 | if (unlikely(PageCgroupUsed(pc))) { | |
2517 | unlock_page_cgroup(pc); | |
c0ff4b85 | 2518 | __mem_cgroup_cancel_charge(memcg, nr_pages); |
ca3e0214 KH |
2519 | return; |
2520 | } | |
2521 | /* | |
2522 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2523 | * accessed by any other context at this point. | |
2524 | */ | |
9ce70c02 HD |
2525 | |
2526 | /* | |
2527 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2528 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2529 | */ | |
2530 | if (lrucare) { | |
2531 | zone = page_zone(page); | |
2532 | spin_lock_irq(&zone->lru_lock); | |
2533 | if (PageLRU(page)) { | |
2534 | ClearPageLRU(page); | |
2535 | del_page_from_lru_list(zone, page, page_lru(page)); | |
2536 | was_on_lru = true; | |
2537 | } | |
2538 | } | |
2539 | ||
c0ff4b85 | 2540 | pc->mem_cgroup = memcg; |
261fb61a KH |
2541 | /* |
2542 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2543 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2544 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2545 | * before USED bit, we need memory barrier here. | |
2546 | * See mem_cgroup_add_lru_list(), etc. | |
2547 | */ | |
08e552c6 | 2548 | smp_wmb(); |
b2402857 | 2549 | SetPageCgroupUsed(pc); |
3be91277 | 2550 | |
9ce70c02 HD |
2551 | if (lrucare) { |
2552 | if (was_on_lru) { | |
2553 | VM_BUG_ON(PageLRU(page)); | |
2554 | SetPageLRU(page); | |
2555 | add_page_to_lru_list(zone, page, page_lru(page)); | |
2556 | } | |
2557 | spin_unlock_irq(&zone->lru_lock); | |
2558 | } | |
2559 | ||
b2402857 KH |
2560 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
2561 | anon = true; | |
2562 | else | |
2563 | anon = false; | |
2564 | ||
2565 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2566 | unlock_page_cgroup(pc); |
9ce70c02 | 2567 | |
430e4863 KH |
2568 | /* |
2569 | * "charge_statistics" updated event counter. Then, check it. | |
2570 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2571 | * if they exceeds softlimit. | |
2572 | */ | |
c0ff4b85 | 2573 | memcg_check_events(memcg, page); |
7a81b88c | 2574 | } |
66e1707b | 2575 | |
ca3e0214 KH |
2576 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2577 | ||
a0db00fc | 2578 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
2579 | /* |
2580 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2581 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2582 | * charge/uncharge will be never happen and move_account() is done under | |
2583 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2584 | */ |
e94c8a9c | 2585 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
2586 | { |
2587 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
2588 | struct page_cgroup *pc; |
2589 | int i; | |
ca3e0214 | 2590 | |
3d37c4a9 KH |
2591 | if (mem_cgroup_disabled()) |
2592 | return; | |
e94c8a9c KH |
2593 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
2594 | pc = head_pc + i; | |
2595 | pc->mem_cgroup = head_pc->mem_cgroup; | |
2596 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
2597 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2598 | } | |
ca3e0214 | 2599 | } |
12d27107 | 2600 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2601 | |
f817ed48 | 2602 | /** |
de3638d9 | 2603 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2604 | * @page: the page |
7ec99d62 | 2605 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2606 | * @pc: page_cgroup of the page. |
2607 | * @from: mem_cgroup which the page is moved from. | |
2608 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2609 | * | |
2610 | * The caller must confirm following. | |
08e552c6 | 2611 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2612 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2613 | * |
2f3479b1 KH |
2614 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2615 | * from old cgroup. | |
f817ed48 | 2616 | */ |
7ec99d62 JW |
2617 | static int mem_cgroup_move_account(struct page *page, |
2618 | unsigned int nr_pages, | |
2619 | struct page_cgroup *pc, | |
2620 | struct mem_cgroup *from, | |
2f3479b1 | 2621 | struct mem_cgroup *to) |
f817ed48 | 2622 | { |
de3638d9 JW |
2623 | unsigned long flags; |
2624 | int ret; | |
b2402857 | 2625 | bool anon = PageAnon(page); |
987eba66 | 2626 | |
f817ed48 | 2627 | VM_BUG_ON(from == to); |
5564e88b | 2628 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2629 | /* |
2630 | * The page is isolated from LRU. So, collapse function | |
2631 | * will not handle this page. But page splitting can happen. | |
2632 | * Do this check under compound_page_lock(). The caller should | |
2633 | * hold it. | |
2634 | */ | |
2635 | ret = -EBUSY; | |
7ec99d62 | 2636 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2637 | goto out; |
2638 | ||
2639 | lock_page_cgroup(pc); | |
2640 | ||
2641 | ret = -EINVAL; | |
2642 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2643 | goto unlock; | |
2644 | ||
312734c0 | 2645 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 2646 | |
2ff76f11 | 2647 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
2648 | /* Update mapped_file data for mem_cgroup */ |
2649 | preempt_disable(); | |
2650 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2651 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2652 | preempt_enable(); | |
d69b042f | 2653 | } |
b2402857 | 2654 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
d69b042f | 2655 | |
854ffa8d | 2656 | /* caller should have done css_get */ |
08e552c6 | 2657 | pc->mem_cgroup = to; |
b2402857 | 2658 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
88703267 KH |
2659 | /* |
2660 | * We charges against "to" which may not have any tasks. Then, "to" | |
2661 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2662 | * this function is just force_empty() and move charge, so it's |
25985edc | 2663 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2664 | * status here. |
88703267 | 2665 | */ |
312734c0 | 2666 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
2667 | ret = 0; |
2668 | unlock: | |
57f9fd7d | 2669 | unlock_page_cgroup(pc); |
d2265e6f KH |
2670 | /* |
2671 | * check events | |
2672 | */ | |
5564e88b JW |
2673 | memcg_check_events(to, page); |
2674 | memcg_check_events(from, page); | |
de3638d9 | 2675 | out: |
f817ed48 KH |
2676 | return ret; |
2677 | } | |
2678 | ||
2679 | /* | |
2680 | * move charges to its parent. | |
2681 | */ | |
2682 | ||
5564e88b JW |
2683 | static int mem_cgroup_move_parent(struct page *page, |
2684 | struct page_cgroup *pc, | |
f817ed48 KH |
2685 | struct mem_cgroup *child, |
2686 | gfp_t gfp_mask) | |
2687 | { | |
f817ed48 | 2688 | struct mem_cgroup *parent; |
7ec99d62 | 2689 | unsigned int nr_pages; |
4be4489f | 2690 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2691 | int ret; |
2692 | ||
2693 | /* Is ROOT ? */ | |
cc926f78 | 2694 | if (mem_cgroup_is_root(child)) |
f817ed48 KH |
2695 | return -EINVAL; |
2696 | ||
57f9fd7d DN |
2697 | ret = -EBUSY; |
2698 | if (!get_page_unless_zero(page)) | |
2699 | goto out; | |
2700 | if (isolate_lru_page(page)) | |
2701 | goto put; | |
52dbb905 | 2702 | |
7ec99d62 | 2703 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2704 | |
cc926f78 KH |
2705 | parent = parent_mem_cgroup(child); |
2706 | /* | |
2707 | * If no parent, move charges to root cgroup. | |
2708 | */ | |
2709 | if (!parent) | |
2710 | parent = root_mem_cgroup; | |
f817ed48 | 2711 | |
7ec99d62 | 2712 | if (nr_pages > 1) |
987eba66 KH |
2713 | flags = compound_lock_irqsave(page); |
2714 | ||
cc926f78 | 2715 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 2716 | pc, child, parent); |
cc926f78 KH |
2717 | if (!ret) |
2718 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 2719 | |
7ec99d62 | 2720 | if (nr_pages > 1) |
987eba66 | 2721 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 2722 | putback_lru_page(page); |
57f9fd7d | 2723 | put: |
40d58138 | 2724 | put_page(page); |
57f9fd7d | 2725 | out: |
f817ed48 KH |
2726 | return ret; |
2727 | } | |
2728 | ||
7a81b88c KH |
2729 | /* |
2730 | * Charge the memory controller for page usage. | |
2731 | * Return | |
2732 | * 0 if the charge was successful | |
2733 | * < 0 if the cgroup is over its limit | |
2734 | */ | |
2735 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2736 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2737 | { |
c0ff4b85 | 2738 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2739 | unsigned int nr_pages = 1; |
8493ae43 | 2740 | bool oom = true; |
7a81b88c | 2741 | int ret; |
ec168510 | 2742 | |
37c2ac78 | 2743 | if (PageTransHuge(page)) { |
7ec99d62 | 2744 | nr_pages <<= compound_order(page); |
37c2ac78 | 2745 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2746 | /* |
2747 | * Never OOM-kill a process for a huge page. The | |
2748 | * fault handler will fall back to regular pages. | |
2749 | */ | |
2750 | oom = false; | |
37c2ac78 | 2751 | } |
7a81b88c | 2752 | |
c0ff4b85 | 2753 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 2754 | if (ret == -ENOMEM) |
7a81b88c | 2755 | return ret; |
ce587e65 | 2756 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 2757 | return 0; |
8a9f3ccd BS |
2758 | } |
2759 | ||
7a81b88c KH |
2760 | int mem_cgroup_newpage_charge(struct page *page, |
2761 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2762 | { |
f8d66542 | 2763 | if (mem_cgroup_disabled()) |
cede86ac | 2764 | return 0; |
7a0524cf JW |
2765 | VM_BUG_ON(page_mapped(page)); |
2766 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
2767 | VM_BUG_ON(!mm); | |
217bc319 | 2768 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
7a0524cf | 2769 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2770 | } |
2771 | ||
83aae4c7 DN |
2772 | static void |
2773 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2774 | enum charge_type ctype); | |
2775 | ||
e1a1cd59 BS |
2776 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2777 | gfp_t gfp_mask) | |
8697d331 | 2778 | { |
c0ff4b85 | 2779 | struct mem_cgroup *memcg = NULL; |
dc67d504 | 2780 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
b5a84319 KH |
2781 | int ret; |
2782 | ||
f8d66542 | 2783 | if (mem_cgroup_disabled()) |
cede86ac | 2784 | return 0; |
52d4b9ac KH |
2785 | if (PageCompound(page)) |
2786 | return 0; | |
accf163e | 2787 | |
73045c47 | 2788 | if (unlikely(!mm)) |
8697d331 | 2789 | mm = &init_mm; |
dc67d504 KH |
2790 | if (!page_is_file_cache(page)) |
2791 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
accf163e | 2792 | |
38c5d72f | 2793 | if (!PageSwapCache(page)) |
dc67d504 | 2794 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); |
38c5d72f | 2795 | else { /* page is swapcache/shmem */ |
c0ff4b85 | 2796 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg); |
83aae4c7 | 2797 | if (!ret) |
dc67d504 KH |
2798 | __mem_cgroup_commit_charge_swapin(page, memcg, type); |
2799 | } | |
b5a84319 | 2800 | return ret; |
e8589cc1 KH |
2801 | } |
2802 | ||
54595fe2 KH |
2803 | /* |
2804 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2805 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2806 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2807 | * "commit()" or removed by "cancel()" |
2808 | */ | |
8c7c6e34 KH |
2809 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2810 | struct page *page, | |
72835c86 | 2811 | gfp_t mask, struct mem_cgroup **memcgp) |
8c7c6e34 | 2812 | { |
c0ff4b85 | 2813 | struct mem_cgroup *memcg; |
54595fe2 | 2814 | int ret; |
8c7c6e34 | 2815 | |
72835c86 | 2816 | *memcgp = NULL; |
56039efa | 2817 | |
f8d66542 | 2818 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2819 | return 0; |
2820 | ||
2821 | if (!do_swap_account) | |
2822 | goto charge_cur_mm; | |
8c7c6e34 KH |
2823 | /* |
2824 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2825 | * the pte, and even removed page from swap cache: in those cases |
2826 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2827 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2828 | */ |
2829 | if (!PageSwapCache(page)) | |
407f9c8b | 2830 | goto charge_cur_mm; |
c0ff4b85 R |
2831 | memcg = try_get_mem_cgroup_from_page(page); |
2832 | if (!memcg) | |
54595fe2 | 2833 | goto charge_cur_mm; |
72835c86 JW |
2834 | *memcgp = memcg; |
2835 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 2836 | css_put(&memcg->css); |
38c5d72f KH |
2837 | if (ret == -EINTR) |
2838 | ret = 0; | |
54595fe2 | 2839 | return ret; |
8c7c6e34 KH |
2840 | charge_cur_mm: |
2841 | if (unlikely(!mm)) | |
2842 | mm = &init_mm; | |
38c5d72f KH |
2843 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
2844 | if (ret == -EINTR) | |
2845 | ret = 0; | |
2846 | return ret; | |
8c7c6e34 KH |
2847 | } |
2848 | ||
83aae4c7 | 2849 | static void |
72835c86 | 2850 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 2851 | enum charge_type ctype) |
7a81b88c | 2852 | { |
f8d66542 | 2853 | if (mem_cgroup_disabled()) |
7a81b88c | 2854 | return; |
72835c86 | 2855 | if (!memcg) |
7a81b88c | 2856 | return; |
72835c86 | 2857 | cgroup_exclude_rmdir(&memcg->css); |
5a6475a4 | 2858 | |
ce587e65 | 2859 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
2860 | /* |
2861 | * Now swap is on-memory. This means this page may be | |
2862 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2863 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2864 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2865 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2866 | */ |
03f3c433 | 2867 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2868 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 2869 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 2870 | } |
88703267 KH |
2871 | /* |
2872 | * At swapin, we may charge account against cgroup which has no tasks. | |
2873 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2874 | * In that case, we need to call pre_destroy() again. check it here. | |
2875 | */ | |
72835c86 | 2876 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
7a81b88c KH |
2877 | } |
2878 | ||
72835c86 JW |
2879 | void mem_cgroup_commit_charge_swapin(struct page *page, |
2880 | struct mem_cgroup *memcg) | |
83aae4c7 | 2881 | { |
72835c86 JW |
2882 | __mem_cgroup_commit_charge_swapin(page, memcg, |
2883 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
83aae4c7 DN |
2884 | } |
2885 | ||
c0ff4b85 | 2886 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
7a81b88c | 2887 | { |
f8d66542 | 2888 | if (mem_cgroup_disabled()) |
7a81b88c | 2889 | return; |
c0ff4b85 | 2890 | if (!memcg) |
7a81b88c | 2891 | return; |
c0ff4b85 | 2892 | __mem_cgroup_cancel_charge(memcg, 1); |
7a81b88c KH |
2893 | } |
2894 | ||
c0ff4b85 | 2895 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2896 | unsigned int nr_pages, |
2897 | const enum charge_type ctype) | |
569b846d KH |
2898 | { |
2899 | struct memcg_batch_info *batch = NULL; | |
2900 | bool uncharge_memsw = true; | |
7ec99d62 | 2901 | |
569b846d KH |
2902 | /* If swapout, usage of swap doesn't decrease */ |
2903 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2904 | uncharge_memsw = false; | |
569b846d KH |
2905 | |
2906 | batch = ¤t->memcg_batch; | |
2907 | /* | |
2908 | * In usual, we do css_get() when we remember memcg pointer. | |
2909 | * But in this case, we keep res->usage until end of a series of | |
2910 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2911 | */ | |
2912 | if (!batch->memcg) | |
c0ff4b85 | 2913 | batch->memcg = memcg; |
3c11ecf4 KH |
2914 | /* |
2915 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2916 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2917 | * the same cgroup and we have chance to coalesce uncharges. |
2918 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2919 | * because we want to do uncharge as soon as possible. | |
2920 | */ | |
2921 | ||
2922 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2923 | goto direct_uncharge; | |
2924 | ||
7ec99d62 | 2925 | if (nr_pages > 1) |
ec168510 AA |
2926 | goto direct_uncharge; |
2927 | ||
569b846d KH |
2928 | /* |
2929 | * In typical case, batch->memcg == mem. This means we can | |
2930 | * merge a series of uncharges to an uncharge of res_counter. | |
2931 | * If not, we uncharge res_counter ony by one. | |
2932 | */ | |
c0ff4b85 | 2933 | if (batch->memcg != memcg) |
569b846d KH |
2934 | goto direct_uncharge; |
2935 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2936 | batch->nr_pages++; |
569b846d | 2937 | if (uncharge_memsw) |
7ffd4ca7 | 2938 | batch->memsw_nr_pages++; |
569b846d KH |
2939 | return; |
2940 | direct_uncharge: | |
c0ff4b85 | 2941 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 2942 | if (uncharge_memsw) |
c0ff4b85 R |
2943 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
2944 | if (unlikely(batch->memcg != memcg)) | |
2945 | memcg_oom_recover(memcg); | |
569b846d | 2946 | } |
7a81b88c | 2947 | |
8a9f3ccd | 2948 | /* |
69029cd5 | 2949 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2950 | */ |
8c7c6e34 | 2951 | static struct mem_cgroup * |
69029cd5 | 2952 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2953 | { |
c0ff4b85 | 2954 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
2955 | unsigned int nr_pages = 1; |
2956 | struct page_cgroup *pc; | |
b2402857 | 2957 | bool anon; |
8a9f3ccd | 2958 | |
f8d66542 | 2959 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2960 | return NULL; |
4077960e | 2961 | |
d13d1443 | 2962 | if (PageSwapCache(page)) |
8c7c6e34 | 2963 | return NULL; |
d13d1443 | 2964 | |
37c2ac78 | 2965 | if (PageTransHuge(page)) { |
7ec99d62 | 2966 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
2967 | VM_BUG_ON(!PageTransHuge(page)); |
2968 | } | |
8697d331 | 2969 | /* |
3c541e14 | 2970 | * Check if our page_cgroup is valid |
8697d331 | 2971 | */ |
52d4b9ac | 2972 | pc = lookup_page_cgroup(page); |
cfa44946 | 2973 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 2974 | return NULL; |
b9c565d5 | 2975 | |
52d4b9ac | 2976 | lock_page_cgroup(pc); |
d13d1443 | 2977 | |
c0ff4b85 | 2978 | memcg = pc->mem_cgroup; |
8c7c6e34 | 2979 | |
d13d1443 KH |
2980 | if (!PageCgroupUsed(pc)) |
2981 | goto unlock_out; | |
2982 | ||
b2402857 KH |
2983 | anon = PageAnon(page); |
2984 | ||
d13d1443 KH |
2985 | switch (ctype) { |
2986 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2ff76f11 KH |
2987 | /* |
2988 | * Generally PageAnon tells if it's the anon statistics to be | |
2989 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
2990 | * used before page reached the stage of being marked PageAnon. | |
2991 | */ | |
b2402857 KH |
2992 | anon = true; |
2993 | /* fallthrough */ | |
8a9478ca | 2994 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 2995 | /* See mem_cgroup_prepare_migration() */ |
2996 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2997 | goto unlock_out; |
2998 | break; | |
2999 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
3000 | if (!PageAnon(page)) { /* Shared memory */ | |
3001 | if (page->mapping && !page_is_file_cache(page)) | |
3002 | goto unlock_out; | |
3003 | } else if (page_mapped(page)) /* Anon */ | |
3004 | goto unlock_out; | |
3005 | break; | |
3006 | default: | |
3007 | break; | |
52d4b9ac | 3008 | } |
d13d1443 | 3009 | |
b2402857 | 3010 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 3011 | |
52d4b9ac | 3012 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3013 | /* |
3014 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3015 | * freed from LRU. This is safe because uncharged page is expected not | |
3016 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3017 | * special functions. | |
3018 | */ | |
b9c565d5 | 3019 | |
52d4b9ac | 3020 | unlock_page_cgroup(pc); |
f75ca962 | 3021 | /* |
c0ff4b85 | 3022 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
3023 | * will never be freed. |
3024 | */ | |
c0ff4b85 | 3025 | memcg_check_events(memcg, page); |
f75ca962 | 3026 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
3027 | mem_cgroup_swap_statistics(memcg, true); |
3028 | mem_cgroup_get(memcg); | |
f75ca962 | 3029 | } |
c0ff4b85 R |
3030 | if (!mem_cgroup_is_root(memcg)) |
3031 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); | |
6d12e2d8 | 3032 | |
c0ff4b85 | 3033 | return memcg; |
d13d1443 KH |
3034 | |
3035 | unlock_out: | |
3036 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3037 | return NULL; |
3c541e14 BS |
3038 | } |
3039 | ||
69029cd5 KH |
3040 | void mem_cgroup_uncharge_page(struct page *page) |
3041 | { | |
52d4b9ac KH |
3042 | /* early check. */ |
3043 | if (page_mapped(page)) | |
3044 | return; | |
40f23a21 | 3045 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
69029cd5 KH |
3046 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
3047 | } | |
3048 | ||
3049 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3050 | { | |
3051 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3052 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
3053 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
3054 | } | |
3055 | ||
569b846d KH |
3056 | /* |
3057 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3058 | * In that cases, pages are freed continuously and we can expect pages | |
3059 | * are in the same memcg. All these calls itself limits the number of | |
3060 | * pages freed at once, then uncharge_start/end() is called properly. | |
3061 | * This may be called prural(2) times in a context, | |
3062 | */ | |
3063 | ||
3064 | void mem_cgroup_uncharge_start(void) | |
3065 | { | |
3066 | current->memcg_batch.do_batch++; | |
3067 | /* We can do nest. */ | |
3068 | if (current->memcg_batch.do_batch == 1) { | |
3069 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3070 | current->memcg_batch.nr_pages = 0; |
3071 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3072 | } |
3073 | } | |
3074 | ||
3075 | void mem_cgroup_uncharge_end(void) | |
3076 | { | |
3077 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3078 | ||
3079 | if (!batch->do_batch) | |
3080 | return; | |
3081 | ||
3082 | batch->do_batch--; | |
3083 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3084 | return; | |
3085 | ||
3086 | if (!batch->memcg) | |
3087 | return; | |
3088 | /* | |
3089 | * This "batch->memcg" is valid without any css_get/put etc... | |
3090 | * bacause we hide charges behind us. | |
3091 | */ | |
7ffd4ca7 JW |
3092 | if (batch->nr_pages) |
3093 | res_counter_uncharge(&batch->memcg->res, | |
3094 | batch->nr_pages * PAGE_SIZE); | |
3095 | if (batch->memsw_nr_pages) | |
3096 | res_counter_uncharge(&batch->memcg->memsw, | |
3097 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3098 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3099 | /* forget this pointer (for sanity check) */ |
3100 | batch->memcg = NULL; | |
3101 | } | |
3102 | ||
e767e056 | 3103 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3104 | /* |
e767e056 | 3105 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3106 | * memcg information is recorded to swap_cgroup of "ent" |
3107 | */ | |
8a9478ca KH |
3108 | void |
3109 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3110 | { |
3111 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3112 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3113 | ||
3114 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3115 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3116 | ||
3117 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3118 | |
f75ca962 KH |
3119 | /* |
3120 | * record memcg information, if swapout && memcg != NULL, | |
3121 | * mem_cgroup_get() was called in uncharge(). | |
3122 | */ | |
3123 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3124 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3125 | } |
e767e056 | 3126 | #endif |
8c7c6e34 KH |
3127 | |
3128 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3129 | /* | |
3130 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3131 | * uncharge "memsw" account. | |
3132 | */ | |
3133 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3134 | { |
8c7c6e34 | 3135 | struct mem_cgroup *memcg; |
a3b2d692 | 3136 | unsigned short id; |
8c7c6e34 KH |
3137 | |
3138 | if (!do_swap_account) | |
3139 | return; | |
3140 | ||
a3b2d692 KH |
3141 | id = swap_cgroup_record(ent, 0); |
3142 | rcu_read_lock(); | |
3143 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3144 | if (memcg) { |
a3b2d692 KH |
3145 | /* |
3146 | * We uncharge this because swap is freed. | |
3147 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3148 | */ | |
0c3e73e8 | 3149 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3150 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3151 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3152 | mem_cgroup_put(memcg); |
3153 | } | |
a3b2d692 | 3154 | rcu_read_unlock(); |
d13d1443 | 3155 | } |
02491447 DN |
3156 | |
3157 | /** | |
3158 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3159 | * @entry: swap entry to be moved | |
3160 | * @from: mem_cgroup which the entry is moved from | |
3161 | * @to: mem_cgroup which the entry is moved to | |
3162 | * | |
3163 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3164 | * as the mem_cgroup's id of @from. | |
3165 | * | |
3166 | * Returns 0 on success, -EINVAL on failure. | |
3167 | * | |
3168 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3169 | * both res and memsw, and called css_get(). | |
3170 | */ | |
3171 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3172 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3173 | { |
3174 | unsigned short old_id, new_id; | |
3175 | ||
3176 | old_id = css_id(&from->css); | |
3177 | new_id = css_id(&to->css); | |
3178 | ||
3179 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3180 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3181 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3182 | /* |
483c30b5 DN |
3183 | * This function is only called from task migration context now. |
3184 | * It postpones res_counter and refcount handling till the end | |
3185 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3186 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3187 | * because if the process that has been moved to @to does | |
3188 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3189 | */ |
02491447 | 3190 | mem_cgroup_get(to); |
02491447 DN |
3191 | return 0; |
3192 | } | |
3193 | return -EINVAL; | |
3194 | } | |
3195 | #else | |
3196 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3197 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3198 | { |
3199 | return -EINVAL; | |
3200 | } | |
8c7c6e34 | 3201 | #endif |
d13d1443 | 3202 | |
ae41be37 | 3203 | /* |
01b1ae63 KH |
3204 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3205 | * page belongs to. | |
ae41be37 | 3206 | */ |
ac39cf8c | 3207 | int mem_cgroup_prepare_migration(struct page *page, |
72835c86 | 3208 | struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask) |
ae41be37 | 3209 | { |
c0ff4b85 | 3210 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3211 | struct page_cgroup *pc; |
ac39cf8c | 3212 | enum charge_type ctype; |
e8589cc1 | 3213 | int ret = 0; |
8869b8f6 | 3214 | |
72835c86 | 3215 | *memcgp = NULL; |
56039efa | 3216 | |
ec168510 | 3217 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3218 | if (mem_cgroup_disabled()) |
4077960e BS |
3219 | return 0; |
3220 | ||
52d4b9ac KH |
3221 | pc = lookup_page_cgroup(page); |
3222 | lock_page_cgroup(pc); | |
3223 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3224 | memcg = pc->mem_cgroup; |
3225 | css_get(&memcg->css); | |
ac39cf8c | 3226 | /* |
3227 | * At migrating an anonymous page, its mapcount goes down | |
3228 | * to 0 and uncharge() will be called. But, even if it's fully | |
3229 | * unmapped, migration may fail and this page has to be | |
3230 | * charged again. We set MIGRATION flag here and delay uncharge | |
3231 | * until end_migration() is called | |
3232 | * | |
3233 | * Corner Case Thinking | |
3234 | * A) | |
3235 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3236 | * while migration was ongoing. | |
3237 | * If unmap finds the old page, uncharge() of it will be delayed | |
3238 | * until end_migration(). If unmap finds a new page, it's | |
3239 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3240 | * finds swap_migration_entry, the new page will not be mapped | |
3241 | * and end_migration() will find it(mapcount==0). | |
3242 | * | |
3243 | * B) | |
3244 | * When the old page was mapped but migraion fails, the kernel | |
3245 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3246 | * if mapcount goes down to 0. We can do remap successfully | |
3247 | * without charging it again. | |
3248 | * | |
3249 | * C) | |
3250 | * The "old" page is under lock_page() until the end of | |
3251 | * migration, so, the old page itself will not be swapped-out. | |
3252 | * If the new page is swapped out before end_migraton, our | |
3253 | * hook to usual swap-out path will catch the event. | |
3254 | */ | |
3255 | if (PageAnon(page)) | |
3256 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3257 | } |
52d4b9ac | 3258 | unlock_page_cgroup(pc); |
ac39cf8c | 3259 | /* |
3260 | * If the page is not charged at this point, | |
3261 | * we return here. | |
3262 | */ | |
c0ff4b85 | 3263 | if (!memcg) |
ac39cf8c | 3264 | return 0; |
01b1ae63 | 3265 | |
72835c86 JW |
3266 | *memcgp = memcg; |
3267 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false); | |
c0ff4b85 | 3268 | css_put(&memcg->css);/* drop extra refcnt */ |
38c5d72f | 3269 | if (ret) { |
ac39cf8c | 3270 | if (PageAnon(page)) { |
3271 | lock_page_cgroup(pc); | |
3272 | ClearPageCgroupMigration(pc); | |
3273 | unlock_page_cgroup(pc); | |
3274 | /* | |
3275 | * The old page may be fully unmapped while we kept it. | |
3276 | */ | |
3277 | mem_cgroup_uncharge_page(page); | |
3278 | } | |
38c5d72f | 3279 | /* we'll need to revisit this error code (we have -EINTR) */ |
ac39cf8c | 3280 | return -ENOMEM; |
e8589cc1 | 3281 | } |
ac39cf8c | 3282 | /* |
3283 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3284 | * is called before end_migration, we can catch all events on this new | |
3285 | * page. In the case new page is migrated but not remapped, new page's | |
3286 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3287 | */ | |
ac39cf8c | 3288 | if (PageAnon(page)) |
3289 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
3290 | else if (page_is_file_cache(page)) | |
3291 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3292 | else | |
3293 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
ce587e65 | 3294 | __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false); |
e8589cc1 | 3295 | return ret; |
ae41be37 | 3296 | } |
8869b8f6 | 3297 | |
69029cd5 | 3298 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3299 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3300 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3301 | { |
ac39cf8c | 3302 | struct page *used, *unused; |
01b1ae63 | 3303 | struct page_cgroup *pc; |
b2402857 | 3304 | bool anon; |
01b1ae63 | 3305 | |
c0ff4b85 | 3306 | if (!memcg) |
01b1ae63 | 3307 | return; |
ac39cf8c | 3308 | /* blocks rmdir() */ |
c0ff4b85 | 3309 | cgroup_exclude_rmdir(&memcg->css); |
50de1dd9 | 3310 | if (!migration_ok) { |
ac39cf8c | 3311 | used = oldpage; |
3312 | unused = newpage; | |
01b1ae63 | 3313 | } else { |
ac39cf8c | 3314 | used = newpage; |
01b1ae63 KH |
3315 | unused = oldpage; |
3316 | } | |
69029cd5 | 3317 | /* |
ac39cf8c | 3318 | * We disallowed uncharge of pages under migration because mapcount |
3319 | * of the page goes down to zero, temporarly. | |
3320 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3321 | */ |
ac39cf8c | 3322 | pc = lookup_page_cgroup(oldpage); |
3323 | lock_page_cgroup(pc); | |
3324 | ClearPageCgroupMigration(pc); | |
3325 | unlock_page_cgroup(pc); | |
b2402857 KH |
3326 | anon = PageAnon(used); |
3327 | __mem_cgroup_uncharge_common(unused, | |
3328 | anon ? MEM_CGROUP_CHARGE_TYPE_MAPPED | |
3329 | : MEM_CGROUP_CHARGE_TYPE_CACHE); | |
ac39cf8c | 3330 | |
01b1ae63 | 3331 | /* |
ac39cf8c | 3332 | * If a page is a file cache, radix-tree replacement is very atomic |
3333 | * and we can skip this check. When it was an Anon page, its mapcount | |
3334 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3335 | * uncharged yet. There are several case but page->mapcount check | |
3336 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3337 | * check. (see prepare_charge() also) | |
69029cd5 | 3338 | */ |
b2402857 | 3339 | if (anon) |
ac39cf8c | 3340 | mem_cgroup_uncharge_page(used); |
88703267 | 3341 | /* |
ac39cf8c | 3342 | * At migration, we may charge account against cgroup which has no |
3343 | * tasks. | |
88703267 KH |
3344 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3345 | * In that case, we need to call pre_destroy() again. check it here. | |
3346 | */ | |
c0ff4b85 | 3347 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
ae41be37 | 3348 | } |
78fb7466 | 3349 | |
ab936cbc KH |
3350 | /* |
3351 | * At replace page cache, newpage is not under any memcg but it's on | |
3352 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
3353 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
3354 | */ | |
3355 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
3356 | struct page *newpage) | |
3357 | { | |
bde05d1c | 3358 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 3359 | struct page_cgroup *pc; |
ab936cbc | 3360 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
3361 | |
3362 | if (mem_cgroup_disabled()) | |
3363 | return; | |
3364 | ||
3365 | pc = lookup_page_cgroup(oldpage); | |
3366 | /* fix accounting on old pages */ | |
3367 | lock_page_cgroup(pc); | |
bde05d1c HD |
3368 | if (PageCgroupUsed(pc)) { |
3369 | memcg = pc->mem_cgroup; | |
3370 | mem_cgroup_charge_statistics(memcg, false, -1); | |
3371 | ClearPageCgroupUsed(pc); | |
3372 | } | |
ab936cbc KH |
3373 | unlock_page_cgroup(pc); |
3374 | ||
bde05d1c HD |
3375 | /* |
3376 | * When called from shmem_replace_page(), in some cases the | |
3377 | * oldpage has already been charged, and in some cases not. | |
3378 | */ | |
3379 | if (!memcg) | |
3380 | return; | |
3381 | ||
ab936cbc KH |
3382 | if (PageSwapBacked(oldpage)) |
3383 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
3384 | ||
ab936cbc KH |
3385 | /* |
3386 | * Even if newpage->mapping was NULL before starting replacement, | |
3387 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
3388 | * LRU while we overwrite pc->mem_cgroup. | |
3389 | */ | |
ce587e65 | 3390 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
3391 | } |
3392 | ||
f212ad7c DN |
3393 | #ifdef CONFIG_DEBUG_VM |
3394 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3395 | { | |
3396 | struct page_cgroup *pc; | |
3397 | ||
3398 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3399 | /* |
3400 | * Can be NULL while feeding pages into the page allocator for | |
3401 | * the first time, i.e. during boot or memory hotplug; | |
3402 | * or when mem_cgroup_disabled(). | |
3403 | */ | |
f212ad7c DN |
3404 | if (likely(pc) && PageCgroupUsed(pc)) |
3405 | return pc; | |
3406 | return NULL; | |
3407 | } | |
3408 | ||
3409 | bool mem_cgroup_bad_page_check(struct page *page) | |
3410 | { | |
3411 | if (mem_cgroup_disabled()) | |
3412 | return false; | |
3413 | ||
3414 | return lookup_page_cgroup_used(page) != NULL; | |
3415 | } | |
3416 | ||
3417 | void mem_cgroup_print_bad_page(struct page *page) | |
3418 | { | |
3419 | struct page_cgroup *pc; | |
3420 | ||
3421 | pc = lookup_page_cgroup_used(page); | |
3422 | if (pc) { | |
90b3feae | 3423 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
f212ad7c | 3424 | pc, pc->flags, pc->mem_cgroup); |
f212ad7c DN |
3425 | } |
3426 | } | |
3427 | #endif | |
3428 | ||
8c7c6e34 KH |
3429 | static DEFINE_MUTEX(set_limit_mutex); |
3430 | ||
d38d2a75 | 3431 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3432 | unsigned long long val) |
628f4235 | 3433 | { |
81d39c20 | 3434 | int retry_count; |
3c11ecf4 | 3435 | u64 memswlimit, memlimit; |
628f4235 | 3436 | int ret = 0; |
81d39c20 KH |
3437 | int children = mem_cgroup_count_children(memcg); |
3438 | u64 curusage, oldusage; | |
3c11ecf4 | 3439 | int enlarge; |
81d39c20 KH |
3440 | |
3441 | /* | |
3442 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3443 | * is depends on callers. We set our retry-count to be function | |
3444 | * of # of children which we should visit in this loop. | |
3445 | */ | |
3446 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3447 | ||
3448 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3449 | |
3c11ecf4 | 3450 | enlarge = 0; |
8c7c6e34 | 3451 | while (retry_count) { |
628f4235 KH |
3452 | if (signal_pending(current)) { |
3453 | ret = -EINTR; | |
3454 | break; | |
3455 | } | |
8c7c6e34 KH |
3456 | /* |
3457 | * Rather than hide all in some function, I do this in | |
3458 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3459 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3460 | */ |
3461 | mutex_lock(&set_limit_mutex); | |
3462 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3463 | if (memswlimit < val) { | |
3464 | ret = -EINVAL; | |
3465 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3466 | break; |
3467 | } | |
3c11ecf4 KH |
3468 | |
3469 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3470 | if (memlimit < val) | |
3471 | enlarge = 1; | |
3472 | ||
8c7c6e34 | 3473 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3474 | if (!ret) { |
3475 | if (memswlimit == val) | |
3476 | memcg->memsw_is_minimum = true; | |
3477 | else | |
3478 | memcg->memsw_is_minimum = false; | |
3479 | } | |
8c7c6e34 KH |
3480 | mutex_unlock(&set_limit_mutex); |
3481 | ||
3482 | if (!ret) | |
3483 | break; | |
3484 | ||
5660048c JW |
3485 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3486 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3487 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3488 | /* Usage is reduced ? */ | |
3489 | if (curusage >= oldusage) | |
3490 | retry_count--; | |
3491 | else | |
3492 | oldusage = curusage; | |
8c7c6e34 | 3493 | } |
3c11ecf4 KH |
3494 | if (!ret && enlarge) |
3495 | memcg_oom_recover(memcg); | |
14797e23 | 3496 | |
8c7c6e34 KH |
3497 | return ret; |
3498 | } | |
3499 | ||
338c8431 LZ |
3500 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3501 | unsigned long long val) | |
8c7c6e34 | 3502 | { |
81d39c20 | 3503 | int retry_count; |
3c11ecf4 | 3504 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3505 | int children = mem_cgroup_count_children(memcg); |
3506 | int ret = -EBUSY; | |
3c11ecf4 | 3507 | int enlarge = 0; |
8c7c6e34 | 3508 | |
81d39c20 KH |
3509 | /* see mem_cgroup_resize_res_limit */ |
3510 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3511 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3512 | while (retry_count) { |
3513 | if (signal_pending(current)) { | |
3514 | ret = -EINTR; | |
3515 | break; | |
3516 | } | |
3517 | /* | |
3518 | * Rather than hide all in some function, I do this in | |
3519 | * open coded manner. You see what this really does. | |
c0ff4b85 | 3520 | * We have to guarantee memcg->res.limit < memcg->memsw.limit. |
8c7c6e34 KH |
3521 | */ |
3522 | mutex_lock(&set_limit_mutex); | |
3523 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3524 | if (memlimit > val) { | |
3525 | ret = -EINVAL; | |
3526 | mutex_unlock(&set_limit_mutex); | |
3527 | break; | |
3528 | } | |
3c11ecf4 KH |
3529 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3530 | if (memswlimit < val) | |
3531 | enlarge = 1; | |
8c7c6e34 | 3532 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3533 | if (!ret) { |
3534 | if (memlimit == val) | |
3535 | memcg->memsw_is_minimum = true; | |
3536 | else | |
3537 | memcg->memsw_is_minimum = false; | |
3538 | } | |
8c7c6e34 KH |
3539 | mutex_unlock(&set_limit_mutex); |
3540 | ||
3541 | if (!ret) | |
3542 | break; | |
3543 | ||
5660048c JW |
3544 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3545 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3546 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3547 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3548 | /* Usage is reduced ? */ |
8c7c6e34 | 3549 | if (curusage >= oldusage) |
628f4235 | 3550 | retry_count--; |
81d39c20 KH |
3551 | else |
3552 | oldusage = curusage; | |
628f4235 | 3553 | } |
3c11ecf4 KH |
3554 | if (!ret && enlarge) |
3555 | memcg_oom_recover(memcg); | |
628f4235 KH |
3556 | return ret; |
3557 | } | |
3558 | ||
4e416953 | 3559 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3560 | gfp_t gfp_mask, |
3561 | unsigned long *total_scanned) | |
4e416953 BS |
3562 | { |
3563 | unsigned long nr_reclaimed = 0; | |
3564 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3565 | unsigned long reclaimed; | |
3566 | int loop = 0; | |
3567 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3568 | unsigned long long excess; |
0ae5e89c | 3569 | unsigned long nr_scanned; |
4e416953 BS |
3570 | |
3571 | if (order > 0) | |
3572 | return 0; | |
3573 | ||
00918b6a | 3574 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3575 | /* |
3576 | * This loop can run a while, specially if mem_cgroup's continuously | |
3577 | * keep exceeding their soft limit and putting the system under | |
3578 | * pressure | |
3579 | */ | |
3580 | do { | |
3581 | if (next_mz) | |
3582 | mz = next_mz; | |
3583 | else | |
3584 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3585 | if (!mz) | |
3586 | break; | |
3587 | ||
0ae5e89c | 3588 | nr_scanned = 0; |
d79154bb | 3589 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 3590 | gfp_mask, &nr_scanned); |
4e416953 | 3591 | nr_reclaimed += reclaimed; |
0ae5e89c | 3592 | *total_scanned += nr_scanned; |
4e416953 BS |
3593 | spin_lock(&mctz->lock); |
3594 | ||
3595 | /* | |
3596 | * If we failed to reclaim anything from this memory cgroup | |
3597 | * it is time to move on to the next cgroup | |
3598 | */ | |
3599 | next_mz = NULL; | |
3600 | if (!reclaimed) { | |
3601 | do { | |
3602 | /* | |
3603 | * Loop until we find yet another one. | |
3604 | * | |
3605 | * By the time we get the soft_limit lock | |
3606 | * again, someone might have aded the | |
3607 | * group back on the RB tree. Iterate to | |
3608 | * make sure we get a different mem. | |
3609 | * mem_cgroup_largest_soft_limit_node returns | |
3610 | * NULL if no other cgroup is present on | |
3611 | * the tree | |
3612 | */ | |
3613 | next_mz = | |
3614 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3615 | if (next_mz == mz) |
d79154bb | 3616 | css_put(&next_mz->memcg->css); |
39cc98f1 | 3617 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3618 | break; |
3619 | } while (1); | |
3620 | } | |
d79154bb HD |
3621 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
3622 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
3623 | /* |
3624 | * One school of thought says that we should not add | |
3625 | * back the node to the tree if reclaim returns 0. | |
3626 | * But our reclaim could return 0, simply because due | |
3627 | * to priority we are exposing a smaller subset of | |
3628 | * memory to reclaim from. Consider this as a longer | |
3629 | * term TODO. | |
3630 | */ | |
ef8745c1 | 3631 | /* If excess == 0, no tree ops */ |
d79154bb | 3632 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 3633 | spin_unlock(&mctz->lock); |
d79154bb | 3634 | css_put(&mz->memcg->css); |
4e416953 BS |
3635 | loop++; |
3636 | /* | |
3637 | * Could not reclaim anything and there are no more | |
3638 | * mem cgroups to try or we seem to be looping without | |
3639 | * reclaiming anything. | |
3640 | */ | |
3641 | if (!nr_reclaimed && | |
3642 | (next_mz == NULL || | |
3643 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3644 | break; | |
3645 | } while (!nr_reclaimed); | |
3646 | if (next_mz) | |
d79154bb | 3647 | css_put(&next_mz->memcg->css); |
4e416953 BS |
3648 | return nr_reclaimed; |
3649 | } | |
3650 | ||
cc847582 KH |
3651 | /* |
3652 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3653 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3654 | */ | |
c0ff4b85 | 3655 | static int mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3656 | int node, int zid, enum lru_list lru) |
cc847582 | 3657 | { |
08e552c6 | 3658 | struct mem_cgroup_per_zone *mz; |
08e552c6 | 3659 | unsigned long flags, loop; |
072c56c1 | 3660 | struct list_head *list; |
925b7673 JW |
3661 | struct page *busy; |
3662 | struct zone *zone; | |
f817ed48 | 3663 | int ret = 0; |
072c56c1 | 3664 | |
08e552c6 | 3665 | zone = &NODE_DATA(node)->node_zones[zid]; |
c0ff4b85 | 3666 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
6290df54 | 3667 | list = &mz->lruvec.lists[lru]; |
cc847582 | 3668 | |
1eb49272 | 3669 | loop = mz->lru_size[lru]; |
f817ed48 KH |
3670 | /* give some margin against EBUSY etc...*/ |
3671 | loop += 256; | |
3672 | busy = NULL; | |
3673 | while (loop--) { | |
925b7673 | 3674 | struct page_cgroup *pc; |
5564e88b JW |
3675 | struct page *page; |
3676 | ||
f817ed48 | 3677 | ret = 0; |
08e552c6 | 3678 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3679 | if (list_empty(list)) { |
08e552c6 | 3680 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3681 | break; |
f817ed48 | 3682 | } |
925b7673 JW |
3683 | page = list_entry(list->prev, struct page, lru); |
3684 | if (busy == page) { | |
3685 | list_move(&page->lru, list); | |
648bcc77 | 3686 | busy = NULL; |
08e552c6 | 3687 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3688 | continue; |
3689 | } | |
08e552c6 | 3690 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3691 | |
925b7673 | 3692 | pc = lookup_page_cgroup(page); |
5564e88b | 3693 | |
c0ff4b85 | 3694 | ret = mem_cgroup_move_parent(page, pc, memcg, GFP_KERNEL); |
38c5d72f | 3695 | if (ret == -ENOMEM || ret == -EINTR) |
52d4b9ac | 3696 | break; |
f817ed48 KH |
3697 | |
3698 | if (ret == -EBUSY || ret == -EINVAL) { | |
3699 | /* found lock contention or "pc" is obsolete. */ | |
925b7673 | 3700 | busy = page; |
f817ed48 KH |
3701 | cond_resched(); |
3702 | } else | |
3703 | busy = NULL; | |
cc847582 | 3704 | } |
08e552c6 | 3705 | |
f817ed48 KH |
3706 | if (!ret && !list_empty(list)) |
3707 | return -EBUSY; | |
3708 | return ret; | |
cc847582 KH |
3709 | } |
3710 | ||
3711 | /* | |
3712 | * make mem_cgroup's charge to be 0 if there is no task. | |
3713 | * This enables deleting this mem_cgroup. | |
3714 | */ | |
c0ff4b85 | 3715 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all) |
cc847582 | 3716 | { |
f817ed48 KH |
3717 | int ret; |
3718 | int node, zid, shrink; | |
3719 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c0ff4b85 | 3720 | struct cgroup *cgrp = memcg->css.cgroup; |
8869b8f6 | 3721 | |
c0ff4b85 | 3722 | css_get(&memcg->css); |
f817ed48 KH |
3723 | |
3724 | shrink = 0; | |
c1e862c1 KH |
3725 | /* should free all ? */ |
3726 | if (free_all) | |
3727 | goto try_to_free; | |
f817ed48 | 3728 | move_account: |
fce66477 | 3729 | do { |
f817ed48 | 3730 | ret = -EBUSY; |
c1e862c1 KH |
3731 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3732 | goto out; | |
3733 | ret = -EINTR; | |
3734 | if (signal_pending(current)) | |
cc847582 | 3735 | goto out; |
52d4b9ac KH |
3736 | /* This is for making all *used* pages to be on LRU. */ |
3737 | lru_add_drain_all(); | |
c0ff4b85 | 3738 | drain_all_stock_sync(memcg); |
f817ed48 | 3739 | ret = 0; |
c0ff4b85 | 3740 | mem_cgroup_start_move(memcg); |
299b4eaa | 3741 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3742 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3743 | enum lru_list lru; |
3744 | for_each_lru(lru) { | |
c0ff4b85 | 3745 | ret = mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3746 | node, zid, lru); |
f817ed48 KH |
3747 | if (ret) |
3748 | break; | |
3749 | } | |
1ecaab2b | 3750 | } |
f817ed48 KH |
3751 | if (ret) |
3752 | break; | |
3753 | } | |
c0ff4b85 R |
3754 | mem_cgroup_end_move(memcg); |
3755 | memcg_oom_recover(memcg); | |
f817ed48 KH |
3756 | /* it seems parent cgroup doesn't have enough mem */ |
3757 | if (ret == -ENOMEM) | |
3758 | goto try_to_free; | |
52d4b9ac | 3759 | cond_resched(); |
fce66477 | 3760 | /* "ret" should also be checked to ensure all lists are empty. */ |
569530fb | 3761 | } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret); |
cc847582 | 3762 | out: |
c0ff4b85 | 3763 | css_put(&memcg->css); |
cc847582 | 3764 | return ret; |
f817ed48 KH |
3765 | |
3766 | try_to_free: | |
c1e862c1 KH |
3767 | /* returns EBUSY if there is a task or if we come here twice. */ |
3768 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3769 | ret = -EBUSY; |
3770 | goto out; | |
3771 | } | |
c1e862c1 KH |
3772 | /* we call try-to-free pages for make this cgroup empty */ |
3773 | lru_add_drain_all(); | |
f817ed48 KH |
3774 | /* try to free all pages in this cgroup */ |
3775 | shrink = 1; | |
569530fb | 3776 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 3777 | int progress; |
c1e862c1 KH |
3778 | |
3779 | if (signal_pending(current)) { | |
3780 | ret = -EINTR; | |
3781 | goto out; | |
3782 | } | |
c0ff4b85 | 3783 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3784 | false); |
c1e862c1 | 3785 | if (!progress) { |
f817ed48 | 3786 | nr_retries--; |
c1e862c1 | 3787 | /* maybe some writeback is necessary */ |
8aa7e847 | 3788 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3789 | } |
f817ed48 KH |
3790 | |
3791 | } | |
08e552c6 | 3792 | lru_add_drain(); |
f817ed48 | 3793 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3794 | goto move_account; |
cc847582 KH |
3795 | } |
3796 | ||
6bbda35c | 3797 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 KH |
3798 | { |
3799 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3800 | } | |
3801 | ||
3802 | ||
18f59ea7 BS |
3803 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3804 | { | |
3805 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3806 | } | |
3807 | ||
3808 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3809 | u64 val) | |
3810 | { | |
3811 | int retval = 0; | |
c0ff4b85 | 3812 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3813 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3814 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3815 | |
3816 | if (parent) | |
c0ff4b85 | 3817 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3818 | |
3819 | cgroup_lock(); | |
3820 | /* | |
af901ca1 | 3821 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3822 | * in the child subtrees. If it is unset, then the change can |
3823 | * occur, provided the current cgroup has no children. | |
3824 | * | |
3825 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3826 | * set if there are no children. | |
3827 | */ | |
c0ff4b85 | 3828 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3829 | (val == 1 || val == 0)) { |
3830 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3831 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3832 | else |
3833 | retval = -EBUSY; | |
3834 | } else | |
3835 | retval = -EINVAL; | |
3836 | cgroup_unlock(); | |
3837 | ||
3838 | return retval; | |
3839 | } | |
3840 | ||
0c3e73e8 | 3841 | |
c0ff4b85 | 3842 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3843 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3844 | { |
7d74b06f | 3845 | struct mem_cgroup *iter; |
7a159cc9 | 3846 | long val = 0; |
0c3e73e8 | 3847 | |
7a159cc9 | 3848 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3849 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3850 | val += mem_cgroup_read_stat(iter, idx); |
3851 | ||
3852 | if (val < 0) /* race ? */ | |
3853 | val = 0; | |
3854 | return val; | |
0c3e73e8 BS |
3855 | } |
3856 | ||
c0ff4b85 | 3857 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3858 | { |
7d74b06f | 3859 | u64 val; |
104f3928 | 3860 | |
c0ff4b85 | 3861 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 3862 | if (!swap) |
65c64ce8 | 3863 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3864 | else |
65c64ce8 | 3865 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
3866 | } |
3867 | ||
c0ff4b85 R |
3868 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3869 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3870 | |
7d74b06f | 3871 | if (swap) |
c0ff4b85 | 3872 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAPOUT); |
104f3928 KS |
3873 | |
3874 | return val << PAGE_SHIFT; | |
3875 | } | |
3876 | ||
af36f906 TH |
3877 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
3878 | struct file *file, char __user *buf, | |
3879 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 3880 | { |
c0ff4b85 | 3881 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 3882 | char str[64]; |
104f3928 | 3883 | u64 val; |
af36f906 | 3884 | int type, name, len; |
8c7c6e34 KH |
3885 | |
3886 | type = MEMFILE_TYPE(cft->private); | |
3887 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3888 | |
3889 | if (!do_swap_account && type == _MEMSWAP) | |
3890 | return -EOPNOTSUPP; | |
3891 | ||
8c7c6e34 KH |
3892 | switch (type) { |
3893 | case _MEM: | |
104f3928 | 3894 | if (name == RES_USAGE) |
c0ff4b85 | 3895 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3896 | else |
c0ff4b85 | 3897 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3898 | break; |
3899 | case _MEMSWAP: | |
104f3928 | 3900 | if (name == RES_USAGE) |
c0ff4b85 | 3901 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 3902 | else |
c0ff4b85 | 3903 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 KH |
3904 | break; |
3905 | default: | |
3906 | BUG(); | |
8c7c6e34 | 3907 | } |
af36f906 TH |
3908 | |
3909 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
3910 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 3911 | } |
628f4235 KH |
3912 | /* |
3913 | * The user of this function is... | |
3914 | * RES_LIMIT. | |
3915 | */ | |
856c13aa PM |
3916 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3917 | const char *buffer) | |
8cdea7c0 | 3918 | { |
628f4235 | 3919 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3920 | int type, name; |
628f4235 KH |
3921 | unsigned long long val; |
3922 | int ret; | |
3923 | ||
8c7c6e34 KH |
3924 | type = MEMFILE_TYPE(cft->private); |
3925 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3926 | |
3927 | if (!do_swap_account && type == _MEMSWAP) | |
3928 | return -EOPNOTSUPP; | |
3929 | ||
8c7c6e34 | 3930 | switch (name) { |
628f4235 | 3931 | case RES_LIMIT: |
4b3bde4c BS |
3932 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3933 | ret = -EINVAL; | |
3934 | break; | |
3935 | } | |
628f4235 KH |
3936 | /* This function does all necessary parse...reuse it */ |
3937 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3938 | if (ret) |
3939 | break; | |
3940 | if (type == _MEM) | |
628f4235 | 3941 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3942 | else |
3943 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3944 | break; |
296c81d8 BS |
3945 | case RES_SOFT_LIMIT: |
3946 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3947 | if (ret) | |
3948 | break; | |
3949 | /* | |
3950 | * For memsw, soft limits are hard to implement in terms | |
3951 | * of semantics, for now, we support soft limits for | |
3952 | * control without swap | |
3953 | */ | |
3954 | if (type == _MEM) | |
3955 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3956 | else | |
3957 | ret = -EINVAL; | |
3958 | break; | |
628f4235 KH |
3959 | default: |
3960 | ret = -EINVAL; /* should be BUG() ? */ | |
3961 | break; | |
3962 | } | |
3963 | return ret; | |
8cdea7c0 BS |
3964 | } |
3965 | ||
fee7b548 KH |
3966 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3967 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3968 | { | |
3969 | struct cgroup *cgroup; | |
3970 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3971 | ||
3972 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3973 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3974 | cgroup = memcg->css.cgroup; | |
3975 | if (!memcg->use_hierarchy) | |
3976 | goto out; | |
3977 | ||
3978 | while (cgroup->parent) { | |
3979 | cgroup = cgroup->parent; | |
3980 | memcg = mem_cgroup_from_cont(cgroup); | |
3981 | if (!memcg->use_hierarchy) | |
3982 | break; | |
3983 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3984 | min_limit = min(min_limit, tmp); | |
3985 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3986 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3987 | } | |
3988 | out: | |
3989 | *mem_limit = min_limit; | |
3990 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
3991 | } |
3992 | ||
29f2a4da | 3993 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 3994 | { |
af36f906 | 3995 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3996 | int type, name; |
c84872e1 | 3997 | |
8c7c6e34 KH |
3998 | type = MEMFILE_TYPE(event); |
3999 | name = MEMFILE_ATTR(event); | |
af36f906 TH |
4000 | |
4001 | if (!do_swap_account && type == _MEMSWAP) | |
4002 | return -EOPNOTSUPP; | |
4003 | ||
8c7c6e34 | 4004 | switch (name) { |
29f2a4da | 4005 | case RES_MAX_USAGE: |
8c7c6e34 | 4006 | if (type == _MEM) |
c0ff4b85 | 4007 | res_counter_reset_max(&memcg->res); |
8c7c6e34 | 4008 | else |
c0ff4b85 | 4009 | res_counter_reset_max(&memcg->memsw); |
29f2a4da PE |
4010 | break; |
4011 | case RES_FAILCNT: | |
8c7c6e34 | 4012 | if (type == _MEM) |
c0ff4b85 | 4013 | res_counter_reset_failcnt(&memcg->res); |
8c7c6e34 | 4014 | else |
c0ff4b85 | 4015 | res_counter_reset_failcnt(&memcg->memsw); |
29f2a4da PE |
4016 | break; |
4017 | } | |
f64c3f54 | 4018 | |
85cc59db | 4019 | return 0; |
c84872e1 PE |
4020 | } |
4021 | ||
7dc74be0 DN |
4022 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
4023 | struct cftype *cft) | |
4024 | { | |
4025 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
4026 | } | |
4027 | ||
02491447 | 4028 | #ifdef CONFIG_MMU |
7dc74be0 DN |
4029 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
4030 | struct cftype *cft, u64 val) | |
4031 | { | |
c0ff4b85 | 4032 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
4033 | |
4034 | if (val >= (1 << NR_MOVE_TYPE)) | |
4035 | return -EINVAL; | |
4036 | /* | |
4037 | * We check this value several times in both in can_attach() and | |
4038 | * attach(), so we need cgroup lock to prevent this value from being | |
4039 | * inconsistent. | |
4040 | */ | |
4041 | cgroup_lock(); | |
c0ff4b85 | 4042 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4043 | cgroup_unlock(); |
4044 | ||
4045 | return 0; | |
4046 | } | |
02491447 DN |
4047 | #else |
4048 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4049 | struct cftype *cft, u64 val) | |
4050 | { | |
4051 | return -ENOSYS; | |
4052 | } | |
4053 | #endif | |
7dc74be0 | 4054 | |
406eb0c9 | 4055 | #ifdef CONFIG_NUMA |
fada52ca JW |
4056 | static int mem_control_numa_stat_show(struct cgroup *cont, struct cftype *cft, |
4057 | struct seq_file *m) | |
406eb0c9 YH |
4058 | { |
4059 | int nid; | |
4060 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4061 | unsigned long node_nr; | |
d79154bb | 4062 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 4063 | |
d79154bb | 4064 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 YH |
4065 | seq_printf(m, "total=%lu", total_nr); |
4066 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4067 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
4068 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4069 | } | |
4070 | seq_putc(m, '\n'); | |
4071 | ||
d79154bb | 4072 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 YH |
4073 | seq_printf(m, "file=%lu", file_nr); |
4074 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4075 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4076 | LRU_ALL_FILE); |
406eb0c9 YH |
4077 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4078 | } | |
4079 | seq_putc(m, '\n'); | |
4080 | ||
d79154bb | 4081 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 YH |
4082 | seq_printf(m, "anon=%lu", anon_nr); |
4083 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4084 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4085 | LRU_ALL_ANON); |
406eb0c9 YH |
4086 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4087 | } | |
4088 | seq_putc(m, '\n'); | |
4089 | ||
d79154bb | 4090 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4091 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4092 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4093 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4094 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4095 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4096 | } | |
4097 | seq_putc(m, '\n'); | |
4098 | return 0; | |
4099 | } | |
4100 | #endif /* CONFIG_NUMA */ | |
4101 | ||
af7c4b0e JW |
4102 | static const char * const mem_cgroup_lru_names[] = { |
4103 | "inactive_anon", | |
4104 | "active_anon", | |
4105 | "inactive_file", | |
4106 | "active_file", | |
4107 | "unevictable", | |
4108 | }; | |
4109 | ||
4110 | static inline void mem_cgroup_lru_names_not_uptodate(void) | |
4111 | { | |
4112 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4113 | } | |
4114 | ||
c64745cf | 4115 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
78ccf5b5 | 4116 | struct seq_file *m) |
d2ceb9b7 | 4117 | { |
d79154bb | 4118 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af7c4b0e JW |
4119 | struct mem_cgroup *mi; |
4120 | unsigned int i; | |
406eb0c9 | 4121 | |
af7c4b0e JW |
4122 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4123 | if (i == MEM_CGROUP_STAT_SWAPOUT && !do_swap_account) | |
1dd3a273 | 4124 | continue; |
af7c4b0e JW |
4125 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4126 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4127 | } |
7b854121 | 4128 | |
af7c4b0e JW |
4129 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4130 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4131 | mem_cgroup_read_events(memcg, i)); | |
4132 | ||
4133 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4134 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4135 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4136 | ||
14067bb3 | 4137 | /* Hierarchical information */ |
fee7b548 KH |
4138 | { |
4139 | unsigned long long limit, memsw_limit; | |
d79154bb | 4140 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 4141 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 4142 | if (do_swap_account) |
78ccf5b5 JW |
4143 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4144 | memsw_limit); | |
fee7b548 | 4145 | } |
7f016ee8 | 4146 | |
af7c4b0e JW |
4147 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4148 | long long val = 0; | |
4149 | ||
4150 | if (i == MEM_CGROUP_STAT_SWAPOUT && !do_swap_account) | |
1dd3a273 | 4151 | continue; |
af7c4b0e JW |
4152 | for_each_mem_cgroup_tree(mi, memcg) |
4153 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4154 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4155 | } | |
4156 | ||
4157 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4158 | unsigned long long val = 0; | |
4159 | ||
4160 | for_each_mem_cgroup_tree(mi, memcg) | |
4161 | val += mem_cgroup_read_events(mi, i); | |
4162 | seq_printf(m, "total_%s %llu\n", | |
4163 | mem_cgroup_events_names[i], val); | |
4164 | } | |
4165 | ||
4166 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4167 | unsigned long long val = 0; | |
4168 | ||
4169 | for_each_mem_cgroup_tree(mi, memcg) | |
4170 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4171 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4172 | } |
14067bb3 | 4173 | |
7f016ee8 | 4174 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4175 | { |
4176 | int nid, zid; | |
4177 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4178 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4179 | unsigned long recent_rotated[2] = {0, 0}; |
4180 | unsigned long recent_scanned[2] = {0, 0}; | |
4181 | ||
4182 | for_each_online_node(nid) | |
4183 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 4184 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 4185 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4186 | |
89abfab1 HD |
4187 | recent_rotated[0] += rstat->recent_rotated[0]; |
4188 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4189 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4190 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4191 | } |
78ccf5b5 JW |
4192 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4193 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4194 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4195 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4196 | } |
4197 | #endif | |
4198 | ||
d2ceb9b7 KH |
4199 | return 0; |
4200 | } | |
4201 | ||
a7885eb8 KM |
4202 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4203 | { | |
4204 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4205 | ||
1f4c025b | 4206 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4207 | } |
4208 | ||
4209 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4210 | u64 val) | |
4211 | { | |
4212 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4213 | struct mem_cgroup *parent; | |
068b38c1 | 4214 | |
a7885eb8 KM |
4215 | if (val > 100) |
4216 | return -EINVAL; | |
4217 | ||
4218 | if (cgrp->parent == NULL) | |
4219 | return -EINVAL; | |
4220 | ||
4221 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4222 | |
4223 | cgroup_lock(); | |
4224 | ||
a7885eb8 KM |
4225 | /* If under hierarchy, only empty-root can set this value */ |
4226 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4227 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4228 | cgroup_unlock(); | |
a7885eb8 | 4229 | return -EINVAL; |
068b38c1 | 4230 | } |
a7885eb8 | 4231 | |
a7885eb8 | 4232 | memcg->swappiness = val; |
a7885eb8 | 4233 | |
068b38c1 LZ |
4234 | cgroup_unlock(); |
4235 | ||
a7885eb8 KM |
4236 | return 0; |
4237 | } | |
4238 | ||
2e72b634 KS |
4239 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4240 | { | |
4241 | struct mem_cgroup_threshold_ary *t; | |
4242 | u64 usage; | |
4243 | int i; | |
4244 | ||
4245 | rcu_read_lock(); | |
4246 | if (!swap) | |
2c488db2 | 4247 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4248 | else |
2c488db2 | 4249 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4250 | |
4251 | if (!t) | |
4252 | goto unlock; | |
4253 | ||
4254 | usage = mem_cgroup_usage(memcg, swap); | |
4255 | ||
4256 | /* | |
748dad36 | 4257 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4258 | * If it's not true, a threshold was crossed after last |
4259 | * call of __mem_cgroup_threshold(). | |
4260 | */ | |
5407a562 | 4261 | i = t->current_threshold; |
2e72b634 KS |
4262 | |
4263 | /* | |
4264 | * Iterate backward over array of thresholds starting from | |
4265 | * current_threshold and check if a threshold is crossed. | |
4266 | * If none of thresholds below usage is crossed, we read | |
4267 | * only one element of the array here. | |
4268 | */ | |
4269 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4270 | eventfd_signal(t->entries[i].eventfd, 1); | |
4271 | ||
4272 | /* i = current_threshold + 1 */ | |
4273 | i++; | |
4274 | ||
4275 | /* | |
4276 | * Iterate forward over array of thresholds starting from | |
4277 | * current_threshold+1 and check if a threshold is crossed. | |
4278 | * If none of thresholds above usage is crossed, we read | |
4279 | * only one element of the array here. | |
4280 | */ | |
4281 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4282 | eventfd_signal(t->entries[i].eventfd, 1); | |
4283 | ||
4284 | /* Update current_threshold */ | |
5407a562 | 4285 | t->current_threshold = i - 1; |
2e72b634 KS |
4286 | unlock: |
4287 | rcu_read_unlock(); | |
4288 | } | |
4289 | ||
4290 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4291 | { | |
ad4ca5f4 KS |
4292 | while (memcg) { |
4293 | __mem_cgroup_threshold(memcg, false); | |
4294 | if (do_swap_account) | |
4295 | __mem_cgroup_threshold(memcg, true); | |
4296 | ||
4297 | memcg = parent_mem_cgroup(memcg); | |
4298 | } | |
2e72b634 KS |
4299 | } |
4300 | ||
4301 | static int compare_thresholds(const void *a, const void *b) | |
4302 | { | |
4303 | const struct mem_cgroup_threshold *_a = a; | |
4304 | const struct mem_cgroup_threshold *_b = b; | |
4305 | ||
4306 | return _a->threshold - _b->threshold; | |
4307 | } | |
4308 | ||
c0ff4b85 | 4309 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4310 | { |
4311 | struct mem_cgroup_eventfd_list *ev; | |
4312 | ||
c0ff4b85 | 4313 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4314 | eventfd_signal(ev->eventfd, 1); |
4315 | return 0; | |
4316 | } | |
4317 | ||
c0ff4b85 | 4318 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4319 | { |
7d74b06f KH |
4320 | struct mem_cgroup *iter; |
4321 | ||
c0ff4b85 | 4322 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4323 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4324 | } |
4325 | ||
4326 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4327 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4328 | { |
4329 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4330 | struct mem_cgroup_thresholds *thresholds; |
4331 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4332 | int type = MEMFILE_TYPE(cft->private); |
4333 | u64 threshold, usage; | |
2c488db2 | 4334 | int i, size, ret; |
2e72b634 KS |
4335 | |
4336 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4337 | if (ret) | |
4338 | return ret; | |
4339 | ||
4340 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4341 | |
2e72b634 | 4342 | if (type == _MEM) |
2c488db2 | 4343 | thresholds = &memcg->thresholds; |
2e72b634 | 4344 | else if (type == _MEMSWAP) |
2c488db2 | 4345 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4346 | else |
4347 | BUG(); | |
4348 | ||
4349 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4350 | ||
4351 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4352 | if (thresholds->primary) |
2e72b634 KS |
4353 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4354 | ||
2c488db2 | 4355 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4356 | |
4357 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4358 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4359 | GFP_KERNEL); |
2c488db2 | 4360 | if (!new) { |
2e72b634 KS |
4361 | ret = -ENOMEM; |
4362 | goto unlock; | |
4363 | } | |
2c488db2 | 4364 | new->size = size; |
2e72b634 KS |
4365 | |
4366 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4367 | if (thresholds->primary) { |
4368 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4369 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4370 | } |
4371 | ||
2e72b634 | 4372 | /* Add new threshold */ |
2c488db2 KS |
4373 | new->entries[size - 1].eventfd = eventfd; |
4374 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4375 | |
4376 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4377 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4378 | compare_thresholds, NULL); |
4379 | ||
4380 | /* Find current threshold */ | |
2c488db2 | 4381 | new->current_threshold = -1; |
2e72b634 | 4382 | for (i = 0; i < size; i++) { |
748dad36 | 4383 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4384 | /* |
2c488db2 KS |
4385 | * new->current_threshold will not be used until |
4386 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4387 | * it here. |
4388 | */ | |
2c488db2 | 4389 | ++new->current_threshold; |
748dad36 SZ |
4390 | } else |
4391 | break; | |
2e72b634 KS |
4392 | } |
4393 | ||
2c488db2 KS |
4394 | /* Free old spare buffer and save old primary buffer as spare */ |
4395 | kfree(thresholds->spare); | |
4396 | thresholds->spare = thresholds->primary; | |
4397 | ||
4398 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4399 | |
907860ed | 4400 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4401 | synchronize_rcu(); |
4402 | ||
2e72b634 KS |
4403 | unlock: |
4404 | mutex_unlock(&memcg->thresholds_lock); | |
4405 | ||
4406 | return ret; | |
4407 | } | |
4408 | ||
907860ed | 4409 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4410 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4411 | { |
4412 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4413 | struct mem_cgroup_thresholds *thresholds; |
4414 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4415 | int type = MEMFILE_TYPE(cft->private); |
4416 | u64 usage; | |
2c488db2 | 4417 | int i, j, size; |
2e72b634 KS |
4418 | |
4419 | mutex_lock(&memcg->thresholds_lock); | |
4420 | if (type == _MEM) | |
2c488db2 | 4421 | thresholds = &memcg->thresholds; |
2e72b634 | 4422 | else if (type == _MEMSWAP) |
2c488db2 | 4423 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4424 | else |
4425 | BUG(); | |
4426 | ||
371528ca AV |
4427 | if (!thresholds->primary) |
4428 | goto unlock; | |
4429 | ||
2e72b634 KS |
4430 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
4431 | ||
4432 | /* Check if a threshold crossed before removing */ | |
4433 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4434 | ||
4435 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4436 | size = 0; |
4437 | for (i = 0; i < thresholds->primary->size; i++) { | |
4438 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4439 | size++; |
4440 | } | |
4441 | ||
2c488db2 | 4442 | new = thresholds->spare; |
907860ed | 4443 | |
2e72b634 KS |
4444 | /* Set thresholds array to NULL if we don't have thresholds */ |
4445 | if (!size) { | |
2c488db2 KS |
4446 | kfree(new); |
4447 | new = NULL; | |
907860ed | 4448 | goto swap_buffers; |
2e72b634 KS |
4449 | } |
4450 | ||
2c488db2 | 4451 | new->size = size; |
2e72b634 KS |
4452 | |
4453 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4454 | new->current_threshold = -1; |
4455 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4456 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4457 | continue; |
4458 | ||
2c488db2 | 4459 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4460 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4461 | /* |
2c488db2 | 4462 | * new->current_threshold will not be used |
2e72b634 KS |
4463 | * until rcu_assign_pointer(), so it's safe to increment |
4464 | * it here. | |
4465 | */ | |
2c488db2 | 4466 | ++new->current_threshold; |
2e72b634 KS |
4467 | } |
4468 | j++; | |
4469 | } | |
4470 | ||
907860ed | 4471 | swap_buffers: |
2c488db2 KS |
4472 | /* Swap primary and spare array */ |
4473 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4474 | /* If all events are unregistered, free the spare array */ |
4475 | if (!new) { | |
4476 | kfree(thresholds->spare); | |
4477 | thresholds->spare = NULL; | |
4478 | } | |
4479 | ||
2c488db2 | 4480 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4481 | |
907860ed | 4482 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4483 | synchronize_rcu(); |
371528ca | 4484 | unlock: |
2e72b634 | 4485 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4486 | } |
c1e862c1 | 4487 | |
9490ff27 KH |
4488 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4489 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4490 | { | |
4491 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4492 | struct mem_cgroup_eventfd_list *event; | |
4493 | int type = MEMFILE_TYPE(cft->private); | |
4494 | ||
4495 | BUG_ON(type != _OOM_TYPE); | |
4496 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4497 | if (!event) | |
4498 | return -ENOMEM; | |
4499 | ||
1af8efe9 | 4500 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4501 | |
4502 | event->eventfd = eventfd; | |
4503 | list_add(&event->list, &memcg->oom_notify); | |
4504 | ||
4505 | /* already in OOM ? */ | |
79dfdacc | 4506 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4507 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4508 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4509 | |
4510 | return 0; | |
4511 | } | |
4512 | ||
907860ed | 4513 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4514 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4515 | { | |
c0ff4b85 | 4516 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 KH |
4517 | struct mem_cgroup_eventfd_list *ev, *tmp; |
4518 | int type = MEMFILE_TYPE(cft->private); | |
4519 | ||
4520 | BUG_ON(type != _OOM_TYPE); | |
4521 | ||
1af8efe9 | 4522 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4523 | |
c0ff4b85 | 4524 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4525 | if (ev->eventfd == eventfd) { |
4526 | list_del(&ev->list); | |
4527 | kfree(ev); | |
4528 | } | |
4529 | } | |
4530 | ||
1af8efe9 | 4531 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4532 | } |
4533 | ||
3c11ecf4 KH |
4534 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4535 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4536 | { | |
c0ff4b85 | 4537 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4538 | |
c0ff4b85 | 4539 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4540 | |
c0ff4b85 | 4541 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4542 | cb->fill(cb, "under_oom", 1); |
4543 | else | |
4544 | cb->fill(cb, "under_oom", 0); | |
4545 | return 0; | |
4546 | } | |
4547 | ||
3c11ecf4 KH |
4548 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4549 | struct cftype *cft, u64 val) | |
4550 | { | |
c0ff4b85 | 4551 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4552 | struct mem_cgroup *parent; |
4553 | ||
4554 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4555 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4556 | return -EINVAL; | |
4557 | ||
4558 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4559 | ||
4560 | cgroup_lock(); | |
4561 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4562 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4563 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4564 | cgroup_unlock(); |
4565 | return -EINVAL; | |
4566 | } | |
c0ff4b85 | 4567 | memcg->oom_kill_disable = val; |
4d845ebf | 4568 | if (!val) |
c0ff4b85 | 4569 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4570 | cgroup_unlock(); |
4571 | return 0; | |
4572 | } | |
4573 | ||
e5671dfa | 4574 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
cbe128e3 | 4575 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4576 | { |
1d62e436 | 4577 | return mem_cgroup_sockets_init(memcg, ss); |
e5671dfa GC |
4578 | }; |
4579 | ||
1d62e436 | 4580 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 | 4581 | { |
1d62e436 | 4582 | mem_cgroup_sockets_destroy(memcg); |
d1a4c0b3 | 4583 | } |
e5671dfa | 4584 | #else |
cbe128e3 | 4585 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4586 | { |
4587 | return 0; | |
4588 | } | |
d1a4c0b3 | 4589 | |
1d62e436 | 4590 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 GC |
4591 | { |
4592 | } | |
e5671dfa GC |
4593 | #endif |
4594 | ||
8cdea7c0 BS |
4595 | static struct cftype mem_cgroup_files[] = { |
4596 | { | |
0eea1030 | 4597 | .name = "usage_in_bytes", |
8c7c6e34 | 4598 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 4599 | .read = mem_cgroup_read, |
9490ff27 KH |
4600 | .register_event = mem_cgroup_usage_register_event, |
4601 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4602 | }, |
c84872e1 PE |
4603 | { |
4604 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4605 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4606 | .trigger = mem_cgroup_reset, |
af36f906 | 4607 | .read = mem_cgroup_read, |
c84872e1 | 4608 | }, |
8cdea7c0 | 4609 | { |
0eea1030 | 4610 | .name = "limit_in_bytes", |
8c7c6e34 | 4611 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4612 | .write_string = mem_cgroup_write, |
af36f906 | 4613 | .read = mem_cgroup_read, |
8cdea7c0 | 4614 | }, |
296c81d8 BS |
4615 | { |
4616 | .name = "soft_limit_in_bytes", | |
4617 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4618 | .write_string = mem_cgroup_write, | |
af36f906 | 4619 | .read = mem_cgroup_read, |
296c81d8 | 4620 | }, |
8cdea7c0 BS |
4621 | { |
4622 | .name = "failcnt", | |
8c7c6e34 | 4623 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4624 | .trigger = mem_cgroup_reset, |
af36f906 | 4625 | .read = mem_cgroup_read, |
8cdea7c0 | 4626 | }, |
d2ceb9b7 KH |
4627 | { |
4628 | .name = "stat", | |
78ccf5b5 | 4629 | .read_seq_string = mem_control_stat_show, |
d2ceb9b7 | 4630 | }, |
c1e862c1 KH |
4631 | { |
4632 | .name = "force_empty", | |
4633 | .trigger = mem_cgroup_force_empty_write, | |
4634 | }, | |
18f59ea7 BS |
4635 | { |
4636 | .name = "use_hierarchy", | |
4637 | .write_u64 = mem_cgroup_hierarchy_write, | |
4638 | .read_u64 = mem_cgroup_hierarchy_read, | |
4639 | }, | |
a7885eb8 KM |
4640 | { |
4641 | .name = "swappiness", | |
4642 | .read_u64 = mem_cgroup_swappiness_read, | |
4643 | .write_u64 = mem_cgroup_swappiness_write, | |
4644 | }, | |
7dc74be0 DN |
4645 | { |
4646 | .name = "move_charge_at_immigrate", | |
4647 | .read_u64 = mem_cgroup_move_charge_read, | |
4648 | .write_u64 = mem_cgroup_move_charge_write, | |
4649 | }, | |
9490ff27 KH |
4650 | { |
4651 | .name = "oom_control", | |
3c11ecf4 KH |
4652 | .read_map = mem_cgroup_oom_control_read, |
4653 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4654 | .register_event = mem_cgroup_oom_register_event, |
4655 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4656 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4657 | }, | |
406eb0c9 YH |
4658 | #ifdef CONFIG_NUMA |
4659 | { | |
4660 | .name = "numa_stat", | |
fada52ca | 4661 | .read_seq_string = mem_control_numa_stat_show, |
406eb0c9 YH |
4662 | }, |
4663 | #endif | |
8c7c6e34 | 4664 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
8c7c6e34 KH |
4665 | { |
4666 | .name = "memsw.usage_in_bytes", | |
4667 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
af36f906 | 4668 | .read = mem_cgroup_read, |
9490ff27 KH |
4669 | .register_event = mem_cgroup_usage_register_event, |
4670 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4671 | }, |
4672 | { | |
4673 | .name = "memsw.max_usage_in_bytes", | |
4674 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4675 | .trigger = mem_cgroup_reset, | |
af36f906 | 4676 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4677 | }, |
4678 | { | |
4679 | .name = "memsw.limit_in_bytes", | |
4680 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4681 | .write_string = mem_cgroup_write, | |
af36f906 | 4682 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4683 | }, |
4684 | { | |
4685 | .name = "memsw.failcnt", | |
4686 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4687 | .trigger = mem_cgroup_reset, | |
af36f906 | 4688 | .read = mem_cgroup_read, |
8c7c6e34 | 4689 | }, |
8c7c6e34 | 4690 | #endif |
6bc10349 | 4691 | { }, /* terminate */ |
af36f906 | 4692 | }; |
8c7c6e34 | 4693 | |
c0ff4b85 | 4694 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4695 | { |
4696 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4697 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4698 | int zone, tmp = node; |
1ecaab2b KH |
4699 | /* |
4700 | * This routine is called against possible nodes. | |
4701 | * But it's BUG to call kmalloc() against offline node. | |
4702 | * | |
4703 | * TODO: this routine can waste much memory for nodes which will | |
4704 | * never be onlined. It's better to use memory hotplug callback | |
4705 | * function. | |
4706 | */ | |
41e3355d KH |
4707 | if (!node_state(node, N_NORMAL_MEMORY)) |
4708 | tmp = -1; | |
17295c88 | 4709 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4710 | if (!pn) |
4711 | return 1; | |
1ecaab2b | 4712 | |
1ecaab2b KH |
4713 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4714 | mz = &pn->zoneinfo[zone]; | |
7f5e86c2 | 4715 | lruvec_init(&mz->lruvec, &NODE_DATA(node)->node_zones[zone]); |
f64c3f54 | 4716 | mz->usage_in_excess = 0; |
4e416953 | 4717 | mz->on_tree = false; |
d79154bb | 4718 | mz->memcg = memcg; |
1ecaab2b | 4719 | } |
0a619e58 | 4720 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4721 | return 0; |
4722 | } | |
4723 | ||
c0ff4b85 | 4724 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4725 | { |
c0ff4b85 | 4726 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4727 | } |
4728 | ||
33327948 KH |
4729 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4730 | { | |
d79154bb | 4731 | struct mem_cgroup *memcg; |
c62b1a3b | 4732 | int size = sizeof(struct mem_cgroup); |
33327948 | 4733 | |
c62b1a3b | 4734 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4735 | if (size < PAGE_SIZE) |
d79154bb | 4736 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 4737 | else |
d79154bb | 4738 | memcg = vzalloc(size); |
33327948 | 4739 | |
d79154bb | 4740 | if (!memcg) |
e7bbcdf3 DC |
4741 | return NULL; |
4742 | ||
d79154bb HD |
4743 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4744 | if (!memcg->stat) | |
d2e61b8d | 4745 | goto out_free; |
d79154bb HD |
4746 | spin_lock_init(&memcg->pcp_counter_lock); |
4747 | return memcg; | |
d2e61b8d DC |
4748 | |
4749 | out_free: | |
4750 | if (size < PAGE_SIZE) | |
d79154bb | 4751 | kfree(memcg); |
d2e61b8d | 4752 | else |
d79154bb | 4753 | vfree(memcg); |
d2e61b8d | 4754 | return NULL; |
33327948 KH |
4755 | } |
4756 | ||
59927fb9 HD |
4757 | /* |
4758 | * Helpers for freeing a vzalloc()ed mem_cgroup by RCU, | |
4759 | * but in process context. The work_freeing structure is overlaid | |
4760 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
4761 | */ | |
4762 | static void vfree_work(struct work_struct *work) | |
4763 | { | |
4764 | struct mem_cgroup *memcg; | |
4765 | ||
4766 | memcg = container_of(work, struct mem_cgroup, work_freeing); | |
4767 | vfree(memcg); | |
4768 | } | |
4769 | static void vfree_rcu(struct rcu_head *rcu_head) | |
4770 | { | |
4771 | struct mem_cgroup *memcg; | |
4772 | ||
4773 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); | |
4774 | INIT_WORK(&memcg->work_freeing, vfree_work); | |
4775 | schedule_work(&memcg->work_freeing); | |
4776 | } | |
4777 | ||
8c7c6e34 KH |
4778 | /* |
4779 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4780 | * (scanning all at force_empty is too costly...) | |
4781 | * | |
4782 | * Instead of clearing all references at force_empty, we remember | |
4783 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4784 | * it goes down to 0. | |
4785 | * | |
8c7c6e34 KH |
4786 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4787 | */ | |
4788 | ||
c0ff4b85 | 4789 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 4790 | { |
08e552c6 KH |
4791 | int node; |
4792 | ||
c0ff4b85 R |
4793 | mem_cgroup_remove_from_trees(memcg); |
4794 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 4795 | |
3ed28fa1 | 4796 | for_each_node(node) |
c0ff4b85 | 4797 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 4798 | |
c0ff4b85 | 4799 | free_percpu(memcg->stat); |
c62b1a3b | 4800 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) |
59927fb9 | 4801 | kfree_rcu(memcg, rcu_freeing); |
33327948 | 4802 | else |
59927fb9 | 4803 | call_rcu(&memcg->rcu_freeing, vfree_rcu); |
33327948 KH |
4804 | } |
4805 | ||
c0ff4b85 | 4806 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 4807 | { |
c0ff4b85 | 4808 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
4809 | } |
4810 | ||
c0ff4b85 | 4811 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 4812 | { |
c0ff4b85 R |
4813 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
4814 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4815 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
4816 | if (parent) |
4817 | mem_cgroup_put(parent); | |
4818 | } | |
8c7c6e34 KH |
4819 | } |
4820 | ||
c0ff4b85 | 4821 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 4822 | { |
c0ff4b85 | 4823 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
4824 | } |
4825 | ||
7bcc1bb1 DN |
4826 | /* |
4827 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4828 | */ | |
e1aab161 | 4829 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4830 | { |
c0ff4b85 | 4831 | if (!memcg->res.parent) |
7bcc1bb1 | 4832 | return NULL; |
c0ff4b85 | 4833 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 4834 | } |
e1aab161 | 4835 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4836 | |
c077719b KH |
4837 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4838 | static void __init enable_swap_cgroup(void) | |
4839 | { | |
f8d66542 | 4840 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4841 | do_swap_account = 1; |
4842 | } | |
4843 | #else | |
4844 | static void __init enable_swap_cgroup(void) | |
4845 | { | |
4846 | } | |
4847 | #endif | |
4848 | ||
f64c3f54 BS |
4849 | static int mem_cgroup_soft_limit_tree_init(void) |
4850 | { | |
4851 | struct mem_cgroup_tree_per_node *rtpn; | |
4852 | struct mem_cgroup_tree_per_zone *rtpz; | |
4853 | int tmp, node, zone; | |
4854 | ||
3ed28fa1 | 4855 | for_each_node(node) { |
f64c3f54 BS |
4856 | tmp = node; |
4857 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4858 | tmp = -1; | |
4859 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4860 | if (!rtpn) | |
c3cecc68 | 4861 | goto err_cleanup; |
f64c3f54 BS |
4862 | |
4863 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4864 | ||
4865 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4866 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4867 | rtpz->rb_root = RB_ROOT; | |
4868 | spin_lock_init(&rtpz->lock); | |
4869 | } | |
4870 | } | |
4871 | return 0; | |
c3cecc68 MH |
4872 | |
4873 | err_cleanup: | |
3ed28fa1 | 4874 | for_each_node(node) { |
c3cecc68 MH |
4875 | if (!soft_limit_tree.rb_tree_per_node[node]) |
4876 | break; | |
4877 | kfree(soft_limit_tree.rb_tree_per_node[node]); | |
4878 | soft_limit_tree.rb_tree_per_node[node] = NULL; | |
4879 | } | |
4880 | return 1; | |
4881 | ||
f64c3f54 BS |
4882 | } |
4883 | ||
0eb253e2 | 4884 | static struct cgroup_subsys_state * __ref |
761b3ef5 | 4885 | mem_cgroup_create(struct cgroup *cont) |
8cdea7c0 | 4886 | { |
c0ff4b85 | 4887 | struct mem_cgroup *memcg, *parent; |
04046e1a | 4888 | long error = -ENOMEM; |
6d12e2d8 | 4889 | int node; |
8cdea7c0 | 4890 | |
c0ff4b85 R |
4891 | memcg = mem_cgroup_alloc(); |
4892 | if (!memcg) | |
04046e1a | 4893 | return ERR_PTR(error); |
78fb7466 | 4894 | |
3ed28fa1 | 4895 | for_each_node(node) |
c0ff4b85 | 4896 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4897 | goto free_out; |
f64c3f54 | 4898 | |
c077719b | 4899 | /* root ? */ |
28dbc4b6 | 4900 | if (cont->parent == NULL) { |
cdec2e42 | 4901 | int cpu; |
c077719b | 4902 | enable_swap_cgroup(); |
28dbc4b6 | 4903 | parent = NULL; |
f64c3f54 BS |
4904 | if (mem_cgroup_soft_limit_tree_init()) |
4905 | goto free_out; | |
a41c58a6 | 4906 | root_mem_cgroup = memcg; |
cdec2e42 KH |
4907 | for_each_possible_cpu(cpu) { |
4908 | struct memcg_stock_pcp *stock = | |
4909 | &per_cpu(memcg_stock, cpu); | |
4910 | INIT_WORK(&stock->work, drain_local_stock); | |
4911 | } | |
711d3d2c | 4912 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4913 | } else { |
28dbc4b6 | 4914 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
4915 | memcg->use_hierarchy = parent->use_hierarchy; |
4916 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 4917 | } |
28dbc4b6 | 4918 | |
18f59ea7 | 4919 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
4920 | res_counter_init(&memcg->res, &parent->res); |
4921 | res_counter_init(&memcg->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4922 | /* |
4923 | * We increment refcnt of the parent to ensure that we can | |
4924 | * safely access it on res_counter_charge/uncharge. | |
4925 | * This refcnt will be decremented when freeing this | |
4926 | * mem_cgroup(see mem_cgroup_put). | |
4927 | */ | |
4928 | mem_cgroup_get(parent); | |
18f59ea7 | 4929 | } else { |
c0ff4b85 R |
4930 | res_counter_init(&memcg->res, NULL); |
4931 | res_counter_init(&memcg->memsw, NULL); | |
18f59ea7 | 4932 | } |
c0ff4b85 R |
4933 | memcg->last_scanned_node = MAX_NUMNODES; |
4934 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 4935 | |
a7885eb8 | 4936 | if (parent) |
c0ff4b85 R |
4937 | memcg->swappiness = mem_cgroup_swappiness(parent); |
4938 | atomic_set(&memcg->refcnt, 1); | |
4939 | memcg->move_charge_at_immigrate = 0; | |
4940 | mutex_init(&memcg->thresholds_lock); | |
312734c0 | 4941 | spin_lock_init(&memcg->move_lock); |
cbe128e3 GC |
4942 | |
4943 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
4944 | if (error) { | |
4945 | /* | |
4946 | * We call put now because our (and parent's) refcnts | |
4947 | * are already in place. mem_cgroup_put() will internally | |
4948 | * call __mem_cgroup_free, so return directly | |
4949 | */ | |
4950 | mem_cgroup_put(memcg); | |
4951 | return ERR_PTR(error); | |
4952 | } | |
c0ff4b85 | 4953 | return &memcg->css; |
6d12e2d8 | 4954 | free_out: |
c0ff4b85 | 4955 | __mem_cgroup_free(memcg); |
04046e1a | 4956 | return ERR_PTR(error); |
8cdea7c0 BS |
4957 | } |
4958 | ||
761b3ef5 | 4959 | static int mem_cgroup_pre_destroy(struct cgroup *cont) |
df878fb0 | 4960 | { |
c0ff4b85 | 4961 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 4962 | |
c0ff4b85 | 4963 | return mem_cgroup_force_empty(memcg, false); |
df878fb0 KH |
4964 | } |
4965 | ||
761b3ef5 | 4966 | static void mem_cgroup_destroy(struct cgroup *cont) |
8cdea7c0 | 4967 | { |
c0ff4b85 | 4968 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 4969 | |
1d62e436 | 4970 | kmem_cgroup_destroy(memcg); |
d1a4c0b3 | 4971 | |
c0ff4b85 | 4972 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
4973 | } |
4974 | ||
02491447 | 4975 | #ifdef CONFIG_MMU |
7dc74be0 | 4976 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
4977 | #define PRECHARGE_COUNT_AT_ONCE 256 |
4978 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 4979 | { |
854ffa8d DN |
4980 | int ret = 0; |
4981 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 4982 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 4983 | |
c0ff4b85 | 4984 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
4985 | mc.precharge += count; |
4986 | /* we don't need css_get for root */ | |
4987 | return ret; | |
4988 | } | |
4989 | /* try to charge at once */ | |
4990 | if (count > 1) { | |
4991 | struct res_counter *dummy; | |
4992 | /* | |
c0ff4b85 | 4993 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
4994 | * by cgroup_lock_live_cgroup() that it is not removed and we |
4995 | * are still under the same cgroup_mutex. So we can postpone | |
4996 | * css_get(). | |
4997 | */ | |
c0ff4b85 | 4998 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 4999 | goto one_by_one; |
c0ff4b85 | 5000 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5001 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5002 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5003 | goto one_by_one; |
5004 | } | |
5005 | mc.precharge += count; | |
854ffa8d DN |
5006 | return ret; |
5007 | } | |
5008 | one_by_one: | |
5009 | /* fall back to one by one charge */ | |
5010 | while (count--) { | |
5011 | if (signal_pending(current)) { | |
5012 | ret = -EINTR; | |
5013 | break; | |
5014 | } | |
5015 | if (!batch_count--) { | |
5016 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5017 | cond_resched(); | |
5018 | } | |
c0ff4b85 R |
5019 | ret = __mem_cgroup_try_charge(NULL, |
5020 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 5021 | if (ret) |
854ffa8d | 5022 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 5023 | return ret; |
854ffa8d DN |
5024 | mc.precharge++; |
5025 | } | |
4ffef5fe DN |
5026 | return ret; |
5027 | } | |
5028 | ||
5029 | /** | |
8d32ff84 | 5030 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5031 | * @vma: the vma the pte to be checked belongs |
5032 | * @addr: the address corresponding to the pte to be checked | |
5033 | * @ptent: the pte to be checked | |
02491447 | 5034 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5035 | * |
5036 | * Returns | |
5037 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5038 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5039 | * move charge. if @target is not NULL, the page is stored in target->page | |
5040 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5041 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5042 | * target for charge migration. if @target is not NULL, the entry is stored | |
5043 | * in target->ent. | |
4ffef5fe DN |
5044 | * |
5045 | * Called with pte lock held. | |
5046 | */ | |
4ffef5fe DN |
5047 | union mc_target { |
5048 | struct page *page; | |
02491447 | 5049 | swp_entry_t ent; |
4ffef5fe DN |
5050 | }; |
5051 | ||
4ffef5fe | 5052 | enum mc_target_type { |
8d32ff84 | 5053 | MC_TARGET_NONE = 0, |
4ffef5fe | 5054 | MC_TARGET_PAGE, |
02491447 | 5055 | MC_TARGET_SWAP, |
4ffef5fe DN |
5056 | }; |
5057 | ||
90254a65 DN |
5058 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5059 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5060 | { |
90254a65 | 5061 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5062 | |
90254a65 DN |
5063 | if (!page || !page_mapped(page)) |
5064 | return NULL; | |
5065 | if (PageAnon(page)) { | |
5066 | /* we don't move shared anon */ | |
4b91355e | 5067 | if (!move_anon()) |
90254a65 | 5068 | return NULL; |
87946a72 DN |
5069 | } else if (!move_file()) |
5070 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5071 | return NULL; |
5072 | if (!get_page_unless_zero(page)) | |
5073 | return NULL; | |
5074 | ||
5075 | return page; | |
5076 | } | |
5077 | ||
4b91355e | 5078 | #ifdef CONFIG_SWAP |
90254a65 DN |
5079 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5080 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5081 | { | |
90254a65 DN |
5082 | struct page *page = NULL; |
5083 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5084 | ||
5085 | if (!move_anon() || non_swap_entry(ent)) | |
5086 | return NULL; | |
4b91355e KH |
5087 | /* |
5088 | * Because lookup_swap_cache() updates some statistics counter, | |
5089 | * we call find_get_page() with swapper_space directly. | |
5090 | */ | |
5091 | page = find_get_page(&swapper_space, ent.val); | |
90254a65 DN |
5092 | if (do_swap_account) |
5093 | entry->val = ent.val; | |
5094 | ||
5095 | return page; | |
5096 | } | |
4b91355e KH |
5097 | #else |
5098 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5099 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5100 | { | |
5101 | return NULL; | |
5102 | } | |
5103 | #endif | |
90254a65 | 5104 | |
87946a72 DN |
5105 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5106 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5107 | { | |
5108 | struct page *page = NULL; | |
87946a72 DN |
5109 | struct address_space *mapping; |
5110 | pgoff_t pgoff; | |
5111 | ||
5112 | if (!vma->vm_file) /* anonymous vma */ | |
5113 | return NULL; | |
5114 | if (!move_file()) | |
5115 | return NULL; | |
5116 | ||
87946a72 DN |
5117 | mapping = vma->vm_file->f_mapping; |
5118 | if (pte_none(ptent)) | |
5119 | pgoff = linear_page_index(vma, addr); | |
5120 | else /* pte_file(ptent) is true */ | |
5121 | pgoff = pte_to_pgoff(ptent); | |
5122 | ||
5123 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5124 | page = find_get_page(mapping, pgoff); |
5125 | ||
5126 | #ifdef CONFIG_SWAP | |
5127 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5128 | if (radix_tree_exceptional_entry(page)) { | |
5129 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5130 | if (do_swap_account) |
aa3b1895 HD |
5131 | *entry = swap; |
5132 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5133 | } |
aa3b1895 | 5134 | #endif |
87946a72 DN |
5135 | return page; |
5136 | } | |
5137 | ||
8d32ff84 | 5138 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5139 | unsigned long addr, pte_t ptent, union mc_target *target) |
5140 | { | |
5141 | struct page *page = NULL; | |
5142 | struct page_cgroup *pc; | |
8d32ff84 | 5143 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5144 | swp_entry_t ent = { .val = 0 }; |
5145 | ||
5146 | if (pte_present(ptent)) | |
5147 | page = mc_handle_present_pte(vma, addr, ptent); | |
5148 | else if (is_swap_pte(ptent)) | |
5149 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5150 | else if (pte_none(ptent) || pte_file(ptent)) |
5151 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5152 | |
5153 | if (!page && !ent.val) | |
8d32ff84 | 5154 | return ret; |
02491447 DN |
5155 | if (page) { |
5156 | pc = lookup_page_cgroup(page); | |
5157 | /* | |
5158 | * Do only loose check w/o page_cgroup lock. | |
5159 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5160 | * the lock. | |
5161 | */ | |
5162 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5163 | ret = MC_TARGET_PAGE; | |
5164 | if (target) | |
5165 | target->page = page; | |
5166 | } | |
5167 | if (!ret || !target) | |
5168 | put_page(page); | |
5169 | } | |
90254a65 DN |
5170 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5171 | if (ent.val && !ret && | |
9fb4b7cc | 5172 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5173 | ret = MC_TARGET_SWAP; |
5174 | if (target) | |
5175 | target->ent = ent; | |
4ffef5fe | 5176 | } |
4ffef5fe DN |
5177 | return ret; |
5178 | } | |
5179 | ||
12724850 NH |
5180 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5181 | /* | |
5182 | * We don't consider swapping or file mapped pages because THP does not | |
5183 | * support them for now. | |
5184 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5185 | */ | |
5186 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5187 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5188 | { | |
5189 | struct page *page = NULL; | |
5190 | struct page_cgroup *pc; | |
5191 | enum mc_target_type ret = MC_TARGET_NONE; | |
5192 | ||
5193 | page = pmd_page(pmd); | |
5194 | VM_BUG_ON(!page || !PageHead(page)); | |
5195 | if (!move_anon()) | |
5196 | return ret; | |
5197 | pc = lookup_page_cgroup(page); | |
5198 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5199 | ret = MC_TARGET_PAGE; | |
5200 | if (target) { | |
5201 | get_page(page); | |
5202 | target->page = page; | |
5203 | } | |
5204 | } | |
5205 | return ret; | |
5206 | } | |
5207 | #else | |
5208 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5209 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5210 | { | |
5211 | return MC_TARGET_NONE; | |
5212 | } | |
5213 | #endif | |
5214 | ||
4ffef5fe DN |
5215 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5216 | unsigned long addr, unsigned long end, | |
5217 | struct mm_walk *walk) | |
5218 | { | |
5219 | struct vm_area_struct *vma = walk->private; | |
5220 | pte_t *pte; | |
5221 | spinlock_t *ptl; | |
5222 | ||
12724850 NH |
5223 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
5224 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
5225 | mc.precharge += HPAGE_PMD_NR; | |
5226 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5227 | return 0; |
12724850 | 5228 | } |
03319327 | 5229 | |
45f83cef AA |
5230 | if (pmd_trans_unstable(pmd)) |
5231 | return 0; | |
4ffef5fe DN |
5232 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5233 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5234 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5235 | mc.precharge++; /* increment precharge temporarily */ |
5236 | pte_unmap_unlock(pte - 1, ptl); | |
5237 | cond_resched(); | |
5238 | ||
7dc74be0 DN |
5239 | return 0; |
5240 | } | |
5241 | ||
4ffef5fe DN |
5242 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5243 | { | |
5244 | unsigned long precharge; | |
5245 | struct vm_area_struct *vma; | |
5246 | ||
dfe076b0 | 5247 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5248 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5249 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5250 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5251 | .mm = mm, | |
5252 | .private = vma, | |
5253 | }; | |
5254 | if (is_vm_hugetlb_page(vma)) | |
5255 | continue; | |
4ffef5fe DN |
5256 | walk_page_range(vma->vm_start, vma->vm_end, |
5257 | &mem_cgroup_count_precharge_walk); | |
5258 | } | |
dfe076b0 | 5259 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5260 | |
5261 | precharge = mc.precharge; | |
5262 | mc.precharge = 0; | |
5263 | ||
5264 | return precharge; | |
5265 | } | |
5266 | ||
4ffef5fe DN |
5267 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5268 | { | |
dfe076b0 DN |
5269 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5270 | ||
5271 | VM_BUG_ON(mc.moving_task); | |
5272 | mc.moving_task = current; | |
5273 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5274 | } |
5275 | ||
dfe076b0 DN |
5276 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5277 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5278 | { |
2bd9bb20 KH |
5279 | struct mem_cgroup *from = mc.from; |
5280 | struct mem_cgroup *to = mc.to; | |
5281 | ||
4ffef5fe | 5282 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5283 | if (mc.precharge) { |
5284 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5285 | mc.precharge = 0; | |
5286 | } | |
5287 | /* | |
5288 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5289 | * we must uncharge here. | |
5290 | */ | |
5291 | if (mc.moved_charge) { | |
5292 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5293 | mc.moved_charge = 0; | |
4ffef5fe | 5294 | } |
483c30b5 DN |
5295 | /* we must fixup refcnts and charges */ |
5296 | if (mc.moved_swap) { | |
483c30b5 DN |
5297 | /* uncharge swap account from the old cgroup */ |
5298 | if (!mem_cgroup_is_root(mc.from)) | |
5299 | res_counter_uncharge(&mc.from->memsw, | |
5300 | PAGE_SIZE * mc.moved_swap); | |
5301 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5302 | ||
5303 | if (!mem_cgroup_is_root(mc.to)) { | |
5304 | /* | |
5305 | * we charged both to->res and to->memsw, so we should | |
5306 | * uncharge to->res. | |
5307 | */ | |
5308 | res_counter_uncharge(&mc.to->res, | |
5309 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5310 | } |
5311 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5312 | mc.moved_swap = 0; |
5313 | } | |
dfe076b0 DN |
5314 | memcg_oom_recover(from); |
5315 | memcg_oom_recover(to); | |
5316 | wake_up_all(&mc.waitq); | |
5317 | } | |
5318 | ||
5319 | static void mem_cgroup_clear_mc(void) | |
5320 | { | |
5321 | struct mem_cgroup *from = mc.from; | |
5322 | ||
5323 | /* | |
5324 | * we must clear moving_task before waking up waiters at the end of | |
5325 | * task migration. | |
5326 | */ | |
5327 | mc.moving_task = NULL; | |
5328 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5329 | spin_lock(&mc.lock); |
4ffef5fe DN |
5330 | mc.from = NULL; |
5331 | mc.to = NULL; | |
2bd9bb20 | 5332 | spin_unlock(&mc.lock); |
32047e2a | 5333 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5334 | } |
5335 | ||
761b3ef5 LZ |
5336 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5337 | struct cgroup_taskset *tset) | |
7dc74be0 | 5338 | { |
2f7ee569 | 5339 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5340 | int ret = 0; |
c0ff4b85 | 5341 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5342 | |
c0ff4b85 | 5343 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5344 | struct mm_struct *mm; |
5345 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5346 | ||
c0ff4b85 | 5347 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5348 | |
5349 | mm = get_task_mm(p); | |
5350 | if (!mm) | |
5351 | return 0; | |
7dc74be0 | 5352 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5353 | if (mm->owner == p) { |
5354 | VM_BUG_ON(mc.from); | |
5355 | VM_BUG_ON(mc.to); | |
5356 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5357 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5358 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5359 | mem_cgroup_start_move(from); |
2bd9bb20 | 5360 | spin_lock(&mc.lock); |
4ffef5fe | 5361 | mc.from = from; |
c0ff4b85 | 5362 | mc.to = memcg; |
2bd9bb20 | 5363 | spin_unlock(&mc.lock); |
dfe076b0 | 5364 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5365 | |
5366 | ret = mem_cgroup_precharge_mc(mm); | |
5367 | if (ret) | |
5368 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5369 | } |
5370 | mmput(mm); | |
7dc74be0 DN |
5371 | } |
5372 | return ret; | |
5373 | } | |
5374 | ||
761b3ef5 LZ |
5375 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5376 | struct cgroup_taskset *tset) | |
7dc74be0 | 5377 | { |
4ffef5fe | 5378 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5379 | } |
5380 | ||
4ffef5fe DN |
5381 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5382 | unsigned long addr, unsigned long end, | |
5383 | struct mm_walk *walk) | |
7dc74be0 | 5384 | { |
4ffef5fe DN |
5385 | int ret = 0; |
5386 | struct vm_area_struct *vma = walk->private; | |
5387 | pte_t *pte; | |
5388 | spinlock_t *ptl; | |
12724850 NH |
5389 | enum mc_target_type target_type; |
5390 | union mc_target target; | |
5391 | struct page *page; | |
5392 | struct page_cgroup *pc; | |
4ffef5fe | 5393 | |
12724850 NH |
5394 | /* |
5395 | * We don't take compound_lock() here but no race with splitting thp | |
5396 | * happens because: | |
5397 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5398 | * under splitting, which means there's no concurrent thp split, | |
5399 | * - if another thread runs into split_huge_page() just after we | |
5400 | * entered this if-block, the thread must wait for page table lock | |
5401 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5402 | * part of thp split is not executed yet. | |
5403 | */ | |
5404 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 5405 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
5406 | spin_unlock(&vma->vm_mm->page_table_lock); |
5407 | return 0; | |
5408 | } | |
5409 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5410 | if (target_type == MC_TARGET_PAGE) { | |
5411 | page = target.page; | |
5412 | if (!isolate_lru_page(page)) { | |
5413 | pc = lookup_page_cgroup(page); | |
5414 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5415 | pc, mc.from, mc.to)) { |
12724850 NH |
5416 | mc.precharge -= HPAGE_PMD_NR; |
5417 | mc.moved_charge += HPAGE_PMD_NR; | |
5418 | } | |
5419 | putback_lru_page(page); | |
5420 | } | |
5421 | put_page(page); | |
5422 | } | |
5423 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5424 | return 0; |
12724850 NH |
5425 | } |
5426 | ||
45f83cef AA |
5427 | if (pmd_trans_unstable(pmd)) |
5428 | return 0; | |
4ffef5fe DN |
5429 | retry: |
5430 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5431 | for (; addr != end; addr += PAGE_SIZE) { | |
5432 | pte_t ptent = *(pte++); | |
02491447 | 5433 | swp_entry_t ent; |
4ffef5fe DN |
5434 | |
5435 | if (!mc.precharge) | |
5436 | break; | |
5437 | ||
8d32ff84 | 5438 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5439 | case MC_TARGET_PAGE: |
5440 | page = target.page; | |
5441 | if (isolate_lru_page(page)) | |
5442 | goto put; | |
5443 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5444 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5445 | mc.from, mc.to)) { |
4ffef5fe | 5446 | mc.precharge--; |
854ffa8d DN |
5447 | /* we uncharge from mc.from later. */ |
5448 | mc.moved_charge++; | |
4ffef5fe DN |
5449 | } |
5450 | putback_lru_page(page); | |
8d32ff84 | 5451 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5452 | put_page(page); |
5453 | break; | |
02491447 DN |
5454 | case MC_TARGET_SWAP: |
5455 | ent = target.ent; | |
e91cbb42 | 5456 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5457 | mc.precharge--; |
483c30b5 DN |
5458 | /* we fixup refcnts and charges later. */ |
5459 | mc.moved_swap++; | |
5460 | } | |
02491447 | 5461 | break; |
4ffef5fe DN |
5462 | default: |
5463 | break; | |
5464 | } | |
5465 | } | |
5466 | pte_unmap_unlock(pte - 1, ptl); | |
5467 | cond_resched(); | |
5468 | ||
5469 | if (addr != end) { | |
5470 | /* | |
5471 | * We have consumed all precharges we got in can_attach(). | |
5472 | * We try charge one by one, but don't do any additional | |
5473 | * charges to mc.to if we have failed in charge once in attach() | |
5474 | * phase. | |
5475 | */ | |
854ffa8d | 5476 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5477 | if (!ret) |
5478 | goto retry; | |
5479 | } | |
5480 | ||
5481 | return ret; | |
5482 | } | |
5483 | ||
5484 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5485 | { | |
5486 | struct vm_area_struct *vma; | |
5487 | ||
5488 | lru_add_drain_all(); | |
dfe076b0 DN |
5489 | retry: |
5490 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5491 | /* | |
5492 | * Someone who are holding the mmap_sem might be waiting in | |
5493 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5494 | * and retry. Because we cancel precharges, we might not be able | |
5495 | * to move enough charges, but moving charge is a best-effort | |
5496 | * feature anyway, so it wouldn't be a big problem. | |
5497 | */ | |
5498 | __mem_cgroup_clear_mc(); | |
5499 | cond_resched(); | |
5500 | goto retry; | |
5501 | } | |
4ffef5fe DN |
5502 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5503 | int ret; | |
5504 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5505 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5506 | .mm = mm, | |
5507 | .private = vma, | |
5508 | }; | |
5509 | if (is_vm_hugetlb_page(vma)) | |
5510 | continue; | |
4ffef5fe DN |
5511 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5512 | &mem_cgroup_move_charge_walk); | |
5513 | if (ret) | |
5514 | /* | |
5515 | * means we have consumed all precharges and failed in | |
5516 | * doing additional charge. Just abandon here. | |
5517 | */ | |
5518 | break; | |
5519 | } | |
dfe076b0 | 5520 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5521 | } |
5522 | ||
761b3ef5 LZ |
5523 | static void mem_cgroup_move_task(struct cgroup *cont, |
5524 | struct cgroup_taskset *tset) | |
67e465a7 | 5525 | { |
2f7ee569 | 5526 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5527 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5528 | |
dfe076b0 | 5529 | if (mm) { |
a433658c KM |
5530 | if (mc.to) |
5531 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5532 | mmput(mm); |
5533 | } | |
a433658c KM |
5534 | if (mc.to) |
5535 | mem_cgroup_clear_mc(); | |
67e465a7 | 5536 | } |
5cfb80a7 | 5537 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
5538 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5539 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5540 | { |
5541 | return 0; | |
5542 | } | |
761b3ef5 LZ |
5543 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5544 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5545 | { |
5546 | } | |
761b3ef5 LZ |
5547 | static void mem_cgroup_move_task(struct cgroup *cont, |
5548 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5549 | { |
5550 | } | |
5551 | #endif | |
67e465a7 | 5552 | |
8cdea7c0 BS |
5553 | struct cgroup_subsys mem_cgroup_subsys = { |
5554 | .name = "memory", | |
5555 | .subsys_id = mem_cgroup_subsys_id, | |
5556 | .create = mem_cgroup_create, | |
df878fb0 | 5557 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 | 5558 | .destroy = mem_cgroup_destroy, |
7dc74be0 DN |
5559 | .can_attach = mem_cgroup_can_attach, |
5560 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5561 | .attach = mem_cgroup_move_task, |
6bc10349 | 5562 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 5563 | .early_init = 0, |
04046e1a | 5564 | .use_id = 1, |
48ddbe19 | 5565 | .__DEPRECATED_clear_css_refs = true, |
8cdea7c0 | 5566 | }; |
c077719b KH |
5567 | |
5568 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5569 | static int __init enable_swap_account(char *s) |
5570 | { | |
5571 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5572 | if (!strcmp(s, "1")) |
a42c390c | 5573 | really_do_swap_account = 1; |
a2c8990a | 5574 | else if (!strcmp(s, "0")) |
a42c390c MH |
5575 | really_do_swap_account = 0; |
5576 | return 1; | |
5577 | } | |
a2c8990a | 5578 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5579 | |
c077719b | 5580 | #endif |