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 | |
c255a458 | 64 | #ifdef CONFIG_MEMCG_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*/ | |
c255a458 | 69 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
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 */ |
bff6bb83 | 90 | MEM_CGROUP_STAT_SWAP, /* # 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 | /* | |
3afe36b1 GC |
261 | * We also need some space for a worker in deferred freeing. |
262 | * By the time we call it, rcu_freeing is no longer in use. | |
59927fb9 HD |
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, | |
41326c17 | 381 | MEM_CGROUP_CHARGE_TYPE_ANON, |
4f98a2fe | 382 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
d13d1443 | 383 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 384 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
385 | NR_CHARGE_TYPE, |
386 | }; | |
387 | ||
8c7c6e34 | 388 | /* for encoding cft->private value on file */ |
65c64ce8 GC |
389 | #define _MEM (0) |
390 | #define _MEMSWAP (1) | |
391 | #define _OOM_TYPE (2) | |
a0db00fc KS |
392 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
393 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 394 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
395 | /* Used for OOM nofiier */ |
396 | #define OOM_CONTROL (0) | |
8c7c6e34 | 397 | |
75822b44 BS |
398 | /* |
399 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
400 | */ | |
401 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
402 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
403 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
404 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
405 | ||
c0ff4b85 R |
406 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
407 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 GC |
408 | |
409 | /* Writing them here to avoid exposing memcg's inner layout */ | |
c255a458 | 410 | #ifdef CONFIG_MEMCG_KMEM |
e1aab161 | 411 | #include <net/sock.h> |
d1a4c0b3 | 412 | #include <net/ip.h> |
e1aab161 GC |
413 | |
414 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); | |
415 | void sock_update_memcg(struct sock *sk) | |
416 | { | |
376be5ff | 417 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 418 | struct mem_cgroup *memcg; |
3f134619 | 419 | struct cg_proto *cg_proto; |
e1aab161 GC |
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); | |
3f134619 GC |
439 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
440 | if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) { | |
e1aab161 | 441 | mem_cgroup_get(memcg); |
3f134619 | 442 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
443 | } |
444 | rcu_read_unlock(); | |
445 | } | |
446 | } | |
447 | EXPORT_SYMBOL(sock_update_memcg); | |
448 | ||
449 | void sock_release_memcg(struct sock *sk) | |
450 | { | |
376be5ff | 451 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
452 | struct mem_cgroup *memcg; |
453 | WARN_ON(!sk->sk_cgrp->memcg); | |
454 | memcg = sk->sk_cgrp->memcg; | |
455 | mem_cgroup_put(memcg); | |
456 | } | |
457 | } | |
d1a4c0b3 | 458 | |
319d3b9c | 459 | #ifdef CONFIG_INET |
d1a4c0b3 GC |
460 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
461 | { | |
462 | if (!memcg || mem_cgroup_is_root(memcg)) | |
463 | return NULL; | |
464 | ||
465 | return &memcg->tcp_mem.cg_proto; | |
466 | } | |
467 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 468 | #endif /* CONFIG_INET */ |
c255a458 | 469 | #endif /* CONFIG_MEMCG_KMEM */ |
e1aab161 | 470 | |
c255a458 | 471 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
3f134619 GC |
472 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
473 | { | |
474 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
475 | return; | |
476 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
477 | } | |
478 | #else | |
479 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
480 | { | |
481 | } | |
482 | #endif | |
483 | ||
c0ff4b85 | 484 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 485 | |
f64c3f54 | 486 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 487 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 488 | { |
c0ff4b85 | 489 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
490 | } |
491 | ||
c0ff4b85 | 492 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 493 | { |
c0ff4b85 | 494 | return &memcg->css; |
d324236b WF |
495 | } |
496 | ||
f64c3f54 | 497 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 498 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 499 | { |
97a6c37b JW |
500 | int nid = page_to_nid(page); |
501 | int zid = page_zonenum(page); | |
f64c3f54 | 502 | |
c0ff4b85 | 503 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
504 | } |
505 | ||
506 | static struct mem_cgroup_tree_per_zone * | |
507 | soft_limit_tree_node_zone(int nid, int zid) | |
508 | { | |
509 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
510 | } | |
511 | ||
512 | static struct mem_cgroup_tree_per_zone * | |
513 | soft_limit_tree_from_page(struct page *page) | |
514 | { | |
515 | int nid = page_to_nid(page); | |
516 | int zid = page_zonenum(page); | |
517 | ||
518 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
519 | } | |
520 | ||
521 | static void | |
c0ff4b85 | 522 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 523 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
524 | struct mem_cgroup_tree_per_zone *mctz, |
525 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
526 | { |
527 | struct rb_node **p = &mctz->rb_root.rb_node; | |
528 | struct rb_node *parent = NULL; | |
529 | struct mem_cgroup_per_zone *mz_node; | |
530 | ||
531 | if (mz->on_tree) | |
532 | return; | |
533 | ||
ef8745c1 KH |
534 | mz->usage_in_excess = new_usage_in_excess; |
535 | if (!mz->usage_in_excess) | |
536 | return; | |
f64c3f54 BS |
537 | while (*p) { |
538 | parent = *p; | |
539 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
540 | tree_node); | |
541 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
542 | p = &(*p)->rb_left; | |
543 | /* | |
544 | * We can't avoid mem cgroups that are over their soft | |
545 | * limit by the same amount | |
546 | */ | |
547 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
548 | p = &(*p)->rb_right; | |
549 | } | |
550 | rb_link_node(&mz->tree_node, parent, p); | |
551 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
552 | mz->on_tree = true; | |
4e416953 BS |
553 | } |
554 | ||
555 | static void | |
c0ff4b85 | 556 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
557 | struct mem_cgroup_per_zone *mz, |
558 | struct mem_cgroup_tree_per_zone *mctz) | |
559 | { | |
560 | if (!mz->on_tree) | |
561 | return; | |
562 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
563 | mz->on_tree = false; | |
564 | } | |
565 | ||
f64c3f54 | 566 | static void |
c0ff4b85 | 567 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
568 | struct mem_cgroup_per_zone *mz, |
569 | struct mem_cgroup_tree_per_zone *mctz) | |
570 | { | |
571 | spin_lock(&mctz->lock); | |
c0ff4b85 | 572 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
573 | spin_unlock(&mctz->lock); |
574 | } | |
575 | ||
f64c3f54 | 576 | |
c0ff4b85 | 577 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 578 | { |
ef8745c1 | 579 | unsigned long long excess; |
f64c3f54 BS |
580 | struct mem_cgroup_per_zone *mz; |
581 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
582 | int nid = page_to_nid(page); |
583 | int zid = page_zonenum(page); | |
f64c3f54 BS |
584 | mctz = soft_limit_tree_from_page(page); |
585 | ||
586 | /* | |
4e649152 KH |
587 | * Necessary to update all ancestors when hierarchy is used. |
588 | * because their event counter is not touched. | |
f64c3f54 | 589 | */ |
c0ff4b85 R |
590 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
591 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
592 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
593 | /* |
594 | * We have to update the tree if mz is on RB-tree or | |
595 | * mem is over its softlimit. | |
596 | */ | |
ef8745c1 | 597 | if (excess || mz->on_tree) { |
4e649152 KH |
598 | spin_lock(&mctz->lock); |
599 | /* if on-tree, remove it */ | |
600 | if (mz->on_tree) | |
c0ff4b85 | 601 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 602 | /* |
ef8745c1 KH |
603 | * Insert again. mz->usage_in_excess will be updated. |
604 | * If excess is 0, no tree ops. | |
4e649152 | 605 | */ |
c0ff4b85 | 606 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
607 | spin_unlock(&mctz->lock); |
608 | } | |
f64c3f54 BS |
609 | } |
610 | } | |
611 | ||
c0ff4b85 | 612 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
613 | { |
614 | int node, zone; | |
615 | struct mem_cgroup_per_zone *mz; | |
616 | struct mem_cgroup_tree_per_zone *mctz; | |
617 | ||
3ed28fa1 | 618 | for_each_node(node) { |
f64c3f54 | 619 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 620 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 621 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 622 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
623 | } |
624 | } | |
625 | } | |
626 | ||
4e416953 BS |
627 | static struct mem_cgroup_per_zone * |
628 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
629 | { | |
630 | struct rb_node *rightmost = NULL; | |
26251eaf | 631 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
632 | |
633 | retry: | |
26251eaf | 634 | mz = NULL; |
4e416953 BS |
635 | rightmost = rb_last(&mctz->rb_root); |
636 | if (!rightmost) | |
637 | goto done; /* Nothing to reclaim from */ | |
638 | ||
639 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
640 | /* | |
641 | * Remove the node now but someone else can add it back, | |
642 | * we will to add it back at the end of reclaim to its correct | |
643 | * position in the tree. | |
644 | */ | |
d79154bb HD |
645 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
646 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
647 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
648 | goto retry; |
649 | done: | |
650 | return mz; | |
651 | } | |
652 | ||
653 | static struct mem_cgroup_per_zone * | |
654 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
655 | { | |
656 | struct mem_cgroup_per_zone *mz; | |
657 | ||
658 | spin_lock(&mctz->lock); | |
659 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
660 | spin_unlock(&mctz->lock); | |
661 | return mz; | |
662 | } | |
663 | ||
711d3d2c KH |
664 | /* |
665 | * Implementation Note: reading percpu statistics for memcg. | |
666 | * | |
667 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
668 | * synchronization to implement "quick" read. There are trade-off between | |
669 | * reading cost and precision of value. Then, we may have a chance to implement | |
670 | * a periodic synchronizion of counter in memcg's counter. | |
671 | * | |
672 | * But this _read() function is used for user interface now. The user accounts | |
673 | * memory usage by memory cgroup and he _always_ requires exact value because | |
674 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
675 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
676 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
677 | * | |
678 | * If there are kernel internal actions which can make use of some not-exact | |
679 | * value, and reading all cpu value can be performance bottleneck in some | |
680 | * common workload, threashold and synchonization as vmstat[] should be | |
681 | * implemented. | |
682 | */ | |
c0ff4b85 | 683 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 684 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 685 | { |
7a159cc9 | 686 | long val = 0; |
c62b1a3b | 687 | int cpu; |
c62b1a3b | 688 | |
711d3d2c KH |
689 | get_online_cpus(); |
690 | for_each_online_cpu(cpu) | |
c0ff4b85 | 691 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 692 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
693 | spin_lock(&memcg->pcp_counter_lock); |
694 | val += memcg->nocpu_base.count[idx]; | |
695 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
696 | #endif |
697 | put_online_cpus(); | |
c62b1a3b KH |
698 | return val; |
699 | } | |
700 | ||
c0ff4b85 | 701 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
702 | bool charge) |
703 | { | |
704 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 705 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
706 | } |
707 | ||
c0ff4b85 | 708 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
709 | enum mem_cgroup_events_index idx) |
710 | { | |
711 | unsigned long val = 0; | |
712 | int cpu; | |
713 | ||
714 | for_each_online_cpu(cpu) | |
c0ff4b85 | 715 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 716 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
717 | spin_lock(&memcg->pcp_counter_lock); |
718 | val += memcg->nocpu_base.events[idx]; | |
719 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
720 | #endif |
721 | return val; | |
722 | } | |
723 | ||
c0ff4b85 | 724 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 725 | bool anon, int nr_pages) |
d52aa412 | 726 | { |
c62b1a3b KH |
727 | preempt_disable(); |
728 | ||
b2402857 KH |
729 | /* |
730 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
731 | * counted as CACHE even if it's on ANON LRU. | |
732 | */ | |
733 | if (anon) | |
734 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 735 | nr_pages); |
d52aa412 | 736 | else |
b2402857 | 737 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 738 | nr_pages); |
55e462b0 | 739 | |
e401f176 KH |
740 | /* pagein of a big page is an event. So, ignore page size */ |
741 | if (nr_pages > 0) | |
c0ff4b85 | 742 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 743 | else { |
c0ff4b85 | 744 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
745 | nr_pages = -nr_pages; /* for event */ |
746 | } | |
e401f176 | 747 | |
13114716 | 748 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 749 | |
c62b1a3b | 750 | preempt_enable(); |
6d12e2d8 KH |
751 | } |
752 | ||
bb2a0de9 | 753 | unsigned long |
4d7dcca2 | 754 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
755 | { |
756 | struct mem_cgroup_per_zone *mz; | |
757 | ||
758 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
759 | return mz->lru_size[lru]; | |
760 | } | |
761 | ||
762 | static unsigned long | |
c0ff4b85 | 763 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 764 | unsigned int lru_mask) |
889976db YH |
765 | { |
766 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 767 | enum lru_list lru; |
bb2a0de9 KH |
768 | unsigned long ret = 0; |
769 | ||
c0ff4b85 | 770 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 771 | |
f156ab93 HD |
772 | for_each_lru(lru) { |
773 | if (BIT(lru) & lru_mask) | |
774 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
775 | } |
776 | return ret; | |
777 | } | |
778 | ||
779 | static unsigned long | |
c0ff4b85 | 780 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
781 | int nid, unsigned int lru_mask) |
782 | { | |
889976db YH |
783 | u64 total = 0; |
784 | int zid; | |
785 | ||
bb2a0de9 | 786 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
787 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
788 | nid, zid, lru_mask); | |
bb2a0de9 | 789 | |
889976db YH |
790 | return total; |
791 | } | |
bb2a0de9 | 792 | |
c0ff4b85 | 793 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 794 | unsigned int lru_mask) |
6d12e2d8 | 795 | { |
889976db | 796 | int nid; |
6d12e2d8 KH |
797 | u64 total = 0; |
798 | ||
bb2a0de9 | 799 | for_each_node_state(nid, N_HIGH_MEMORY) |
c0ff4b85 | 800 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 801 | return total; |
d52aa412 KH |
802 | } |
803 | ||
f53d7ce3 JW |
804 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
805 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
806 | { |
807 | unsigned long val, next; | |
808 | ||
13114716 | 809 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 810 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 811 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
812 | if ((long)next - (long)val < 0) { |
813 | switch (target) { | |
814 | case MEM_CGROUP_TARGET_THRESH: | |
815 | next = val + THRESHOLDS_EVENTS_TARGET; | |
816 | break; | |
817 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
818 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
819 | break; | |
820 | case MEM_CGROUP_TARGET_NUMAINFO: | |
821 | next = val + NUMAINFO_EVENTS_TARGET; | |
822 | break; | |
823 | default: | |
824 | break; | |
825 | } | |
826 | __this_cpu_write(memcg->stat->targets[target], next); | |
827 | return true; | |
7a159cc9 | 828 | } |
f53d7ce3 | 829 | return false; |
d2265e6f KH |
830 | } |
831 | ||
832 | /* | |
833 | * Check events in order. | |
834 | * | |
835 | */ | |
c0ff4b85 | 836 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 837 | { |
4799401f | 838 | preempt_disable(); |
d2265e6f | 839 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
840 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
841 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
842 | bool do_softlimit; |
843 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
844 | |
845 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
846 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
847 | #if MAX_NUMNODES > 1 | |
848 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
849 | MEM_CGROUP_TARGET_NUMAINFO); | |
850 | #endif | |
851 | preempt_enable(); | |
852 | ||
c0ff4b85 | 853 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 854 | if (unlikely(do_softlimit)) |
c0ff4b85 | 855 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 856 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 857 | if (unlikely(do_numainfo)) |
c0ff4b85 | 858 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 859 | #endif |
f53d7ce3 JW |
860 | } else |
861 | preempt_enable(); | |
d2265e6f KH |
862 | } |
863 | ||
d1a4c0b3 | 864 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
865 | { |
866 | return container_of(cgroup_subsys_state(cont, | |
867 | mem_cgroup_subsys_id), struct mem_cgroup, | |
868 | css); | |
869 | } | |
870 | ||
cf475ad2 | 871 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 872 | { |
31a78f23 BS |
873 | /* |
874 | * mm_update_next_owner() may clear mm->owner to NULL | |
875 | * if it races with swapoff, page migration, etc. | |
876 | * So this can be called with p == NULL. | |
877 | */ | |
878 | if (unlikely(!p)) | |
879 | return NULL; | |
880 | ||
78fb7466 PE |
881 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
882 | struct mem_cgroup, css); | |
883 | } | |
884 | ||
a433658c | 885 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 886 | { |
c0ff4b85 | 887 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
888 | |
889 | if (!mm) | |
890 | return NULL; | |
54595fe2 KH |
891 | /* |
892 | * Because we have no locks, mm->owner's may be being moved to other | |
893 | * cgroup. We use css_tryget() here even if this looks | |
894 | * pessimistic (rather than adding locks here). | |
895 | */ | |
896 | rcu_read_lock(); | |
897 | do { | |
c0ff4b85 R |
898 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
899 | if (unlikely(!memcg)) | |
54595fe2 | 900 | break; |
c0ff4b85 | 901 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 902 | rcu_read_unlock(); |
c0ff4b85 | 903 | return memcg; |
54595fe2 KH |
904 | } |
905 | ||
5660048c JW |
906 | /** |
907 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
908 | * @root: hierarchy root | |
909 | * @prev: previously returned memcg, NULL on first invocation | |
910 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
911 | * | |
912 | * Returns references to children of the hierarchy below @root, or | |
913 | * @root itself, or %NULL after a full round-trip. | |
914 | * | |
915 | * Caller must pass the return value in @prev on subsequent | |
916 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
917 | * to cancel a hierarchy walk before the round-trip is complete. | |
918 | * | |
919 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
920 | * divide up the memcgs in the hierarchy among all concurrent | |
921 | * reclaimers operating on the same zone and priority. | |
922 | */ | |
923 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
924 | struct mem_cgroup *prev, | |
925 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 926 | { |
9f3a0d09 JW |
927 | struct mem_cgroup *memcg = NULL; |
928 | int id = 0; | |
711d3d2c | 929 | |
5660048c JW |
930 | if (mem_cgroup_disabled()) |
931 | return NULL; | |
932 | ||
9f3a0d09 JW |
933 | if (!root) |
934 | root = root_mem_cgroup; | |
7d74b06f | 935 | |
9f3a0d09 JW |
936 | if (prev && !reclaim) |
937 | id = css_id(&prev->css); | |
14067bb3 | 938 | |
9f3a0d09 JW |
939 | if (prev && prev != root) |
940 | css_put(&prev->css); | |
14067bb3 | 941 | |
9f3a0d09 JW |
942 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
943 | if (prev) | |
944 | return NULL; | |
945 | return root; | |
946 | } | |
14067bb3 | 947 | |
9f3a0d09 | 948 | while (!memcg) { |
527a5ec9 | 949 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 950 | struct cgroup_subsys_state *css; |
711d3d2c | 951 | |
527a5ec9 JW |
952 | if (reclaim) { |
953 | int nid = zone_to_nid(reclaim->zone); | |
954 | int zid = zone_idx(reclaim->zone); | |
955 | struct mem_cgroup_per_zone *mz; | |
956 | ||
957 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
958 | iter = &mz->reclaim_iter[reclaim->priority]; | |
959 | if (prev && reclaim->generation != iter->generation) | |
960 | return NULL; | |
961 | id = iter->position; | |
962 | } | |
7d74b06f | 963 | |
9f3a0d09 JW |
964 | rcu_read_lock(); |
965 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
966 | if (css) { | |
967 | if (css == &root->css || css_tryget(css)) | |
968 | memcg = container_of(css, | |
969 | struct mem_cgroup, css); | |
970 | } else | |
971 | id = 0; | |
14067bb3 | 972 | rcu_read_unlock(); |
14067bb3 | 973 | |
527a5ec9 JW |
974 | if (reclaim) { |
975 | iter->position = id; | |
976 | if (!css) | |
977 | iter->generation++; | |
978 | else if (!prev && memcg) | |
979 | reclaim->generation = iter->generation; | |
980 | } | |
9f3a0d09 JW |
981 | |
982 | if (prev && !css) | |
983 | return NULL; | |
984 | } | |
985 | return memcg; | |
14067bb3 | 986 | } |
7d74b06f | 987 | |
5660048c JW |
988 | /** |
989 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
990 | * @root: hierarchy root | |
991 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
992 | */ | |
993 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
994 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
995 | { |
996 | if (!root) | |
997 | root = root_mem_cgroup; | |
998 | if (prev && prev != root) | |
999 | css_put(&prev->css); | |
1000 | } | |
7d74b06f | 1001 | |
9f3a0d09 JW |
1002 | /* |
1003 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1004 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1005 | * be used for reference counting. | |
1006 | */ | |
1007 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1008 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1009 | iter != NULL; \ |
527a5ec9 | 1010 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1011 | |
9f3a0d09 | 1012 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1013 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1014 | iter != NULL; \ |
527a5ec9 | 1015 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1016 | |
c0ff4b85 | 1017 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
4b3bde4c | 1018 | { |
c0ff4b85 | 1019 | return (memcg == root_mem_cgroup); |
4b3bde4c BS |
1020 | } |
1021 | ||
456f998e YH |
1022 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
1023 | { | |
c0ff4b85 | 1024 | struct mem_cgroup *memcg; |
456f998e YH |
1025 | |
1026 | if (!mm) | |
1027 | return; | |
1028 | ||
1029 | rcu_read_lock(); | |
c0ff4b85 R |
1030 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1031 | if (unlikely(!memcg)) | |
456f998e YH |
1032 | goto out; |
1033 | ||
1034 | switch (idx) { | |
456f998e | 1035 | case PGFAULT: |
0e574a93 JW |
1036 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1037 | break; | |
1038 | case PGMAJFAULT: | |
1039 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1040 | break; |
1041 | default: | |
1042 | BUG(); | |
1043 | } | |
1044 | out: | |
1045 | rcu_read_unlock(); | |
1046 | } | |
1047 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
1048 | ||
925b7673 JW |
1049 | /** |
1050 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1051 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1052 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1053 | * |
1054 | * Returns the lru list vector holding pages for the given @zone and | |
1055 | * @mem. This can be the global zone lruvec, if the memory controller | |
1056 | * is disabled. | |
1057 | */ | |
1058 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1059 | struct mem_cgroup *memcg) | |
1060 | { | |
1061 | struct mem_cgroup_per_zone *mz; | |
1062 | ||
1063 | if (mem_cgroup_disabled()) | |
1064 | return &zone->lruvec; | |
1065 | ||
1066 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
1067 | return &mz->lruvec; | |
1068 | } | |
1069 | ||
08e552c6 KH |
1070 | /* |
1071 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1072 | * Operations are called by routine of global LRU independently from memcg. | |
1073 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1074 | * | |
1075 | * Changes to pc->mem_cgroup happens when | |
1076 | * 1. charge | |
1077 | * 2. moving account | |
1078 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1079 | * It is added to LRU before charge. | |
1080 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1081 | * When moving account, the page is not on LRU. It's isolated. | |
1082 | */ | |
4f98a2fe | 1083 | |
925b7673 | 1084 | /** |
fa9add64 | 1085 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1086 | * @page: the page |
fa9add64 | 1087 | * @zone: zone of the page |
925b7673 | 1088 | */ |
fa9add64 | 1089 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1090 | { |
08e552c6 | 1091 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1092 | struct mem_cgroup *memcg; |
1093 | struct page_cgroup *pc; | |
6d12e2d8 | 1094 | |
f8d66542 | 1095 | if (mem_cgroup_disabled()) |
925b7673 JW |
1096 | return &zone->lruvec; |
1097 | ||
08e552c6 | 1098 | pc = lookup_page_cgroup(page); |
38c5d72f | 1099 | memcg = pc->mem_cgroup; |
7512102c HD |
1100 | |
1101 | /* | |
fa9add64 | 1102 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1103 | * an uncharged page off lru does nothing to secure |
1104 | * its former mem_cgroup from sudden removal. | |
1105 | * | |
1106 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1107 | * under page_cgroup lock: between them, they make all uses | |
1108 | * of pc->mem_cgroup safe. | |
1109 | */ | |
fa9add64 | 1110 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1111 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1112 | ||
925b7673 | 1113 | mz = page_cgroup_zoneinfo(memcg, page); |
925b7673 | 1114 | return &mz->lruvec; |
08e552c6 | 1115 | } |
b69408e8 | 1116 | |
925b7673 | 1117 | /** |
fa9add64 HD |
1118 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1119 | * @lruvec: mem_cgroup per zone lru vector | |
1120 | * @lru: index of lru list the page is sitting on | |
1121 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1122 | * |
fa9add64 HD |
1123 | * This function must be called when a page is added to or removed from an |
1124 | * lru list. | |
3f58a829 | 1125 | */ |
fa9add64 HD |
1126 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1127 | int nr_pages) | |
3f58a829 MK |
1128 | { |
1129 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1130 | unsigned long *lru_size; |
3f58a829 MK |
1131 | |
1132 | if (mem_cgroup_disabled()) | |
1133 | return; | |
1134 | ||
fa9add64 HD |
1135 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1136 | lru_size = mz->lru_size + lru; | |
1137 | *lru_size += nr_pages; | |
1138 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1139 | } |
544122e5 | 1140 | |
3e92041d | 1141 | /* |
c0ff4b85 | 1142 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1143 | * hierarchy subtree |
1144 | */ | |
c3ac9a8a JW |
1145 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1146 | struct mem_cgroup *memcg) | |
3e92041d | 1147 | { |
91c63734 JW |
1148 | if (root_memcg == memcg) |
1149 | return true; | |
3a981f48 | 1150 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1151 | return false; |
c3ac9a8a JW |
1152 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1153 | } | |
1154 | ||
1155 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1156 | struct mem_cgroup *memcg) | |
1157 | { | |
1158 | bool ret; | |
1159 | ||
91c63734 | 1160 | rcu_read_lock(); |
c3ac9a8a | 1161 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1162 | rcu_read_unlock(); |
1163 | return ret; | |
3e92041d MH |
1164 | } |
1165 | ||
c0ff4b85 | 1166 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1167 | { |
1168 | int ret; | |
0b7f569e | 1169 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1170 | struct task_struct *p; |
4c4a2214 | 1171 | |
158e0a2d | 1172 | p = find_lock_task_mm(task); |
de077d22 DR |
1173 | if (p) { |
1174 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1175 | task_unlock(p); | |
1176 | } else { | |
1177 | /* | |
1178 | * All threads may have already detached their mm's, but the oom | |
1179 | * killer still needs to detect if they have already been oom | |
1180 | * killed to prevent needlessly killing additional tasks. | |
1181 | */ | |
1182 | task_lock(task); | |
1183 | curr = mem_cgroup_from_task(task); | |
1184 | if (curr) | |
1185 | css_get(&curr->css); | |
1186 | task_unlock(task); | |
1187 | } | |
0b7f569e KH |
1188 | if (!curr) |
1189 | return 0; | |
d31f56db | 1190 | /* |
c0ff4b85 | 1191 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1192 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1193 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1194 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1195 | */ |
c0ff4b85 | 1196 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1197 | css_put(&curr->css); |
4c4a2214 DR |
1198 | return ret; |
1199 | } | |
1200 | ||
c56d5c7d | 1201 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1202 | { |
9b272977 | 1203 | unsigned long inactive_ratio; |
14797e23 | 1204 | unsigned long inactive; |
9b272977 | 1205 | unsigned long active; |
c772be93 | 1206 | unsigned long gb; |
14797e23 | 1207 | |
4d7dcca2 HD |
1208 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1209 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1210 | |
c772be93 KM |
1211 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1212 | if (gb) | |
1213 | inactive_ratio = int_sqrt(10 * gb); | |
1214 | else | |
1215 | inactive_ratio = 1; | |
1216 | ||
9b272977 | 1217 | return inactive * inactive_ratio < active; |
14797e23 KM |
1218 | } |
1219 | ||
c56d5c7d | 1220 | int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 RR |
1221 | { |
1222 | unsigned long active; | |
1223 | unsigned long inactive; | |
1224 | ||
4d7dcca2 HD |
1225 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE); |
1226 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 RR |
1227 | |
1228 | return (active > inactive); | |
1229 | } | |
1230 | ||
6d61ef40 BS |
1231 | #define mem_cgroup_from_res_counter(counter, member) \ |
1232 | container_of(counter, struct mem_cgroup, member) | |
1233 | ||
19942822 | 1234 | /** |
9d11ea9f | 1235 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1236 | * @memcg: the memory cgroup |
19942822 | 1237 | * |
9d11ea9f | 1238 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1239 | * pages. |
19942822 | 1240 | */ |
c0ff4b85 | 1241 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1242 | { |
9d11ea9f JW |
1243 | unsigned long long margin; |
1244 | ||
c0ff4b85 | 1245 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1246 | if (do_swap_account) |
c0ff4b85 | 1247 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1248 | return margin >> PAGE_SHIFT; |
19942822 JW |
1249 | } |
1250 | ||
1f4c025b | 1251 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1252 | { |
1253 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1254 | |
1255 | /* root ? */ | |
1256 | if (cgrp->parent == NULL) | |
1257 | return vm_swappiness; | |
1258 | ||
bf1ff263 | 1259 | return memcg->swappiness; |
a7885eb8 KM |
1260 | } |
1261 | ||
619d094b KH |
1262 | /* |
1263 | * memcg->moving_account is used for checking possibility that some thread is | |
1264 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1265 | * a memcg, other threads should check memcg->moving_account under | |
1266 | * rcu_read_lock(), like this: | |
1267 | * | |
1268 | * CPU-A CPU-B | |
1269 | * rcu_read_lock() | |
1270 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1271 | * take heavy locks. | |
1272 | * synchronize_rcu() update something. | |
1273 | * rcu_read_unlock() | |
1274 | * start move here. | |
1275 | */ | |
4331f7d3 KH |
1276 | |
1277 | /* for quick checking without looking up memcg */ | |
1278 | atomic_t memcg_moving __read_mostly; | |
1279 | ||
c0ff4b85 | 1280 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1281 | { |
4331f7d3 | 1282 | atomic_inc(&memcg_moving); |
619d094b | 1283 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1284 | synchronize_rcu(); |
1285 | } | |
1286 | ||
c0ff4b85 | 1287 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1288 | { |
619d094b KH |
1289 | /* |
1290 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1291 | * We check NULL in callee rather than caller. | |
1292 | */ | |
4331f7d3 KH |
1293 | if (memcg) { |
1294 | atomic_dec(&memcg_moving); | |
619d094b | 1295 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1296 | } |
32047e2a | 1297 | } |
619d094b | 1298 | |
32047e2a KH |
1299 | /* |
1300 | * 2 routines for checking "mem" is under move_account() or not. | |
1301 | * | |
13fd1dd9 AM |
1302 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1303 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1304 | * pc->mem_cgroup may be overwritten. |
1305 | * | |
1306 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1307 | * under hierarchy of moving cgroups. This is for | |
1308 | * waiting at hith-memory prressure caused by "move". | |
1309 | */ | |
1310 | ||
13fd1dd9 | 1311 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1312 | { |
1313 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1314 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1315 | } |
4b534334 | 1316 | |
c0ff4b85 | 1317 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1318 | { |
2bd9bb20 KH |
1319 | struct mem_cgroup *from; |
1320 | struct mem_cgroup *to; | |
4b534334 | 1321 | bool ret = false; |
2bd9bb20 KH |
1322 | /* |
1323 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1324 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1325 | */ | |
1326 | spin_lock(&mc.lock); | |
1327 | from = mc.from; | |
1328 | to = mc.to; | |
1329 | if (!from) | |
1330 | goto unlock; | |
3e92041d | 1331 | |
c0ff4b85 R |
1332 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1333 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1334 | unlock: |
1335 | spin_unlock(&mc.lock); | |
4b534334 KH |
1336 | return ret; |
1337 | } | |
1338 | ||
c0ff4b85 | 1339 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1340 | { |
1341 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1342 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1343 | DEFINE_WAIT(wait); |
1344 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1345 | /* moving charge context might have finished. */ | |
1346 | if (mc.moving_task) | |
1347 | schedule(); | |
1348 | finish_wait(&mc.waitq, &wait); | |
1349 | return true; | |
1350 | } | |
1351 | } | |
1352 | return false; | |
1353 | } | |
1354 | ||
312734c0 KH |
1355 | /* |
1356 | * Take this lock when | |
1357 | * - a code tries to modify page's memcg while it's USED. | |
1358 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1359 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1360 | */ |
1361 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1362 | unsigned long *flags) | |
1363 | { | |
1364 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1365 | } | |
1366 | ||
1367 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1368 | unsigned long *flags) | |
1369 | { | |
1370 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1371 | } | |
1372 | ||
e222432b | 1373 | /** |
6a6135b6 | 1374 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1375 | * @memcg: The memory cgroup that went over limit |
1376 | * @p: Task that is going to be killed | |
1377 | * | |
1378 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1379 | * enabled | |
1380 | */ | |
1381 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1382 | { | |
1383 | struct cgroup *task_cgrp; | |
1384 | struct cgroup *mem_cgrp; | |
1385 | /* | |
1386 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1387 | * on the assumption that OOM is serialized for memory controller. | |
1388 | * If this assumption is broken, revisit this code. | |
1389 | */ | |
1390 | static char memcg_name[PATH_MAX]; | |
1391 | int ret; | |
1392 | ||
d31f56db | 1393 | if (!memcg || !p) |
e222432b BS |
1394 | return; |
1395 | ||
e222432b BS |
1396 | rcu_read_lock(); |
1397 | ||
1398 | mem_cgrp = memcg->css.cgroup; | |
1399 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1400 | ||
1401 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1402 | if (ret < 0) { | |
1403 | /* | |
1404 | * Unfortunately, we are unable to convert to a useful name | |
1405 | * But we'll still print out the usage information | |
1406 | */ | |
1407 | rcu_read_unlock(); | |
1408 | goto done; | |
1409 | } | |
1410 | rcu_read_unlock(); | |
1411 | ||
1412 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1413 | ||
1414 | rcu_read_lock(); | |
1415 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1416 | if (ret < 0) { | |
1417 | rcu_read_unlock(); | |
1418 | goto done; | |
1419 | } | |
1420 | rcu_read_unlock(); | |
1421 | ||
1422 | /* | |
1423 | * Continues from above, so we don't need an KERN_ level | |
1424 | */ | |
1425 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1426 | done: | |
1427 | ||
1428 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1429 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1430 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1431 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1432 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1433 | "failcnt %llu\n", | |
1434 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1435 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1436 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1437 | } | |
1438 | ||
81d39c20 KH |
1439 | /* |
1440 | * This function returns the number of memcg under hierarchy tree. Returns | |
1441 | * 1(self count) if no children. | |
1442 | */ | |
c0ff4b85 | 1443 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1444 | { |
1445 | int num = 0; | |
7d74b06f KH |
1446 | struct mem_cgroup *iter; |
1447 | ||
c0ff4b85 | 1448 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1449 | num++; |
81d39c20 KH |
1450 | return num; |
1451 | } | |
1452 | ||
a63d83f4 DR |
1453 | /* |
1454 | * Return the memory (and swap, if configured) limit for a memcg. | |
1455 | */ | |
9cbb78bb | 1456 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1457 | { |
1458 | u64 limit; | |
1459 | u64 memsw; | |
1460 | ||
f3e8eb70 JW |
1461 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1462 | limit += total_swap_pages << PAGE_SHIFT; | |
1463 | ||
a63d83f4 DR |
1464 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1465 | /* | |
1466 | * If memsw is finite and limits the amount of swap space available | |
1467 | * to this memcg, return that limit. | |
1468 | */ | |
1469 | return min(limit, memsw); | |
1470 | } | |
1471 | ||
9cbb78bb DR |
1472 | void __mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1473 | int order) | |
1474 | { | |
1475 | struct mem_cgroup *iter; | |
1476 | unsigned long chosen_points = 0; | |
1477 | unsigned long totalpages; | |
1478 | unsigned int points = 0; | |
1479 | struct task_struct *chosen = NULL; | |
1480 | ||
1481 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; | |
1482 | for_each_mem_cgroup_tree(iter, memcg) { | |
1483 | struct cgroup *cgroup = iter->css.cgroup; | |
1484 | struct cgroup_iter it; | |
1485 | struct task_struct *task; | |
1486 | ||
1487 | cgroup_iter_start(cgroup, &it); | |
1488 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
1489 | switch (oom_scan_process_thread(task, totalpages, NULL, | |
1490 | false)) { | |
1491 | case OOM_SCAN_SELECT: | |
1492 | if (chosen) | |
1493 | put_task_struct(chosen); | |
1494 | chosen = task; | |
1495 | chosen_points = ULONG_MAX; | |
1496 | get_task_struct(chosen); | |
1497 | /* fall through */ | |
1498 | case OOM_SCAN_CONTINUE: | |
1499 | continue; | |
1500 | case OOM_SCAN_ABORT: | |
1501 | cgroup_iter_end(cgroup, &it); | |
1502 | mem_cgroup_iter_break(memcg, iter); | |
1503 | if (chosen) | |
1504 | put_task_struct(chosen); | |
1505 | return; | |
1506 | case OOM_SCAN_OK: | |
1507 | break; | |
1508 | }; | |
1509 | points = oom_badness(task, memcg, NULL, totalpages); | |
1510 | if (points > chosen_points) { | |
1511 | if (chosen) | |
1512 | put_task_struct(chosen); | |
1513 | chosen = task; | |
1514 | chosen_points = points; | |
1515 | get_task_struct(chosen); | |
1516 | } | |
1517 | } | |
1518 | cgroup_iter_end(cgroup, &it); | |
1519 | } | |
1520 | ||
1521 | if (!chosen) | |
1522 | return; | |
1523 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1524 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1525 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1526 | } |
1527 | ||
5660048c JW |
1528 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1529 | gfp_t gfp_mask, | |
1530 | unsigned long flags) | |
1531 | { | |
1532 | unsigned long total = 0; | |
1533 | bool noswap = false; | |
1534 | int loop; | |
1535 | ||
1536 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1537 | noswap = true; | |
1538 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1539 | noswap = true; | |
1540 | ||
1541 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1542 | if (loop) | |
1543 | drain_all_stock_async(memcg); | |
1544 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1545 | /* | |
1546 | * Allow limit shrinkers, which are triggered directly | |
1547 | * by userspace, to catch signals and stop reclaim | |
1548 | * after minimal progress, regardless of the margin. | |
1549 | */ | |
1550 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1551 | break; | |
1552 | if (mem_cgroup_margin(memcg)) | |
1553 | break; | |
1554 | /* | |
1555 | * If nothing was reclaimed after two attempts, there | |
1556 | * may be no reclaimable pages in this hierarchy. | |
1557 | */ | |
1558 | if (loop && !total) | |
1559 | break; | |
1560 | } | |
1561 | return total; | |
1562 | } | |
1563 | ||
4d0c066d KH |
1564 | /** |
1565 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1566 | * @memcg: the target memcg |
4d0c066d KH |
1567 | * @nid: the node ID to be checked. |
1568 | * @noswap : specify true here if the user wants flle only information. | |
1569 | * | |
1570 | * This function returns whether the specified memcg contains any | |
1571 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1572 | * pages in the node. | |
1573 | */ | |
c0ff4b85 | 1574 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1575 | int nid, bool noswap) |
1576 | { | |
c0ff4b85 | 1577 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1578 | return true; |
1579 | if (noswap || !total_swap_pages) | |
1580 | return false; | |
c0ff4b85 | 1581 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1582 | return true; |
1583 | return false; | |
1584 | ||
1585 | } | |
889976db YH |
1586 | #if MAX_NUMNODES > 1 |
1587 | ||
1588 | /* | |
1589 | * Always updating the nodemask is not very good - even if we have an empty | |
1590 | * list or the wrong list here, we can start from some node and traverse all | |
1591 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1592 | * | |
1593 | */ | |
c0ff4b85 | 1594 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1595 | { |
1596 | int nid; | |
453a9bf3 KH |
1597 | /* |
1598 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1599 | * pagein/pageout changes since the last update. | |
1600 | */ | |
c0ff4b85 | 1601 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1602 | return; |
c0ff4b85 | 1603 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1604 | return; |
1605 | ||
889976db | 1606 | /* make a nodemask where this memcg uses memory from */ |
c0ff4b85 | 1607 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
889976db YH |
1608 | |
1609 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1610 | ||
c0ff4b85 R |
1611 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1612 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1613 | } |
453a9bf3 | 1614 | |
c0ff4b85 R |
1615 | atomic_set(&memcg->numainfo_events, 0); |
1616 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1617 | } |
1618 | ||
1619 | /* | |
1620 | * Selecting a node where we start reclaim from. Because what we need is just | |
1621 | * reducing usage counter, start from anywhere is O,K. Considering | |
1622 | * memory reclaim from current node, there are pros. and cons. | |
1623 | * | |
1624 | * Freeing memory from current node means freeing memory from a node which | |
1625 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1626 | * hit limits, it will see a contention on a node. But freeing from remote | |
1627 | * node means more costs for memory reclaim because of memory latency. | |
1628 | * | |
1629 | * Now, we use round-robin. Better algorithm is welcomed. | |
1630 | */ | |
c0ff4b85 | 1631 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1632 | { |
1633 | int node; | |
1634 | ||
c0ff4b85 R |
1635 | mem_cgroup_may_update_nodemask(memcg); |
1636 | node = memcg->last_scanned_node; | |
889976db | 1637 | |
c0ff4b85 | 1638 | node = next_node(node, memcg->scan_nodes); |
889976db | 1639 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1640 | node = first_node(memcg->scan_nodes); |
889976db YH |
1641 | /* |
1642 | * We call this when we hit limit, not when pages are added to LRU. | |
1643 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1644 | * memcg is too small and all pages are not on LRU. In that case, | |
1645 | * we use curret node. | |
1646 | */ | |
1647 | if (unlikely(node == MAX_NUMNODES)) | |
1648 | node = numa_node_id(); | |
1649 | ||
c0ff4b85 | 1650 | memcg->last_scanned_node = node; |
889976db YH |
1651 | return node; |
1652 | } | |
1653 | ||
4d0c066d KH |
1654 | /* |
1655 | * Check all nodes whether it contains reclaimable pages or not. | |
1656 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1657 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1658 | * enough new information. We need to do double check. | |
1659 | */ | |
6bbda35c | 1660 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1661 | { |
1662 | int nid; | |
1663 | ||
1664 | /* | |
1665 | * quick check...making use of scan_node. | |
1666 | * We can skip unused nodes. | |
1667 | */ | |
c0ff4b85 R |
1668 | if (!nodes_empty(memcg->scan_nodes)) { |
1669 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1670 | nid < MAX_NUMNODES; |
c0ff4b85 | 1671 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1672 | |
c0ff4b85 | 1673 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1674 | return true; |
1675 | } | |
1676 | } | |
1677 | /* | |
1678 | * Check rest of nodes. | |
1679 | */ | |
1680 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
c0ff4b85 | 1681 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1682 | continue; |
c0ff4b85 | 1683 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1684 | return true; |
1685 | } | |
1686 | return false; | |
1687 | } | |
1688 | ||
889976db | 1689 | #else |
c0ff4b85 | 1690 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1691 | { |
1692 | return 0; | |
1693 | } | |
4d0c066d | 1694 | |
6bbda35c | 1695 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1696 | { |
c0ff4b85 | 1697 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1698 | } |
889976db YH |
1699 | #endif |
1700 | ||
5660048c JW |
1701 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1702 | struct zone *zone, | |
1703 | gfp_t gfp_mask, | |
1704 | unsigned long *total_scanned) | |
6d61ef40 | 1705 | { |
9f3a0d09 | 1706 | struct mem_cgroup *victim = NULL; |
5660048c | 1707 | int total = 0; |
04046e1a | 1708 | int loop = 0; |
9d11ea9f | 1709 | unsigned long excess; |
185efc0f | 1710 | unsigned long nr_scanned; |
527a5ec9 JW |
1711 | struct mem_cgroup_reclaim_cookie reclaim = { |
1712 | .zone = zone, | |
1713 | .priority = 0, | |
1714 | }; | |
9d11ea9f | 1715 | |
c0ff4b85 | 1716 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1717 | |
4e416953 | 1718 | while (1) { |
527a5ec9 | 1719 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1720 | if (!victim) { |
04046e1a | 1721 | loop++; |
4e416953 BS |
1722 | if (loop >= 2) { |
1723 | /* | |
1724 | * If we have not been able to reclaim | |
1725 | * anything, it might because there are | |
1726 | * no reclaimable pages under this hierarchy | |
1727 | */ | |
5660048c | 1728 | if (!total) |
4e416953 | 1729 | break; |
4e416953 | 1730 | /* |
25985edc | 1731 | * We want to do more targeted reclaim. |
4e416953 BS |
1732 | * excess >> 2 is not to excessive so as to |
1733 | * reclaim too much, nor too less that we keep | |
1734 | * coming back to reclaim from this cgroup | |
1735 | */ | |
1736 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1737 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1738 | break; |
4e416953 | 1739 | } |
9f3a0d09 | 1740 | continue; |
4e416953 | 1741 | } |
5660048c | 1742 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1743 | continue; |
5660048c JW |
1744 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1745 | zone, &nr_scanned); | |
1746 | *total_scanned += nr_scanned; | |
1747 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1748 | break; |
6d61ef40 | 1749 | } |
9f3a0d09 | 1750 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1751 | return total; |
6d61ef40 BS |
1752 | } |
1753 | ||
867578cb KH |
1754 | /* |
1755 | * Check OOM-Killer is already running under our hierarchy. | |
1756 | * If someone is running, return false. | |
1af8efe9 | 1757 | * Has to be called with memcg_oom_lock |
867578cb | 1758 | */ |
c0ff4b85 | 1759 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1760 | { |
79dfdacc | 1761 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1762 | |
9f3a0d09 | 1763 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1764 | if (iter->oom_lock) { |
79dfdacc MH |
1765 | /* |
1766 | * this subtree of our hierarchy is already locked | |
1767 | * so we cannot give a lock. | |
1768 | */ | |
79dfdacc | 1769 | failed = iter; |
9f3a0d09 JW |
1770 | mem_cgroup_iter_break(memcg, iter); |
1771 | break; | |
23751be0 JW |
1772 | } else |
1773 | iter->oom_lock = true; | |
7d74b06f | 1774 | } |
867578cb | 1775 | |
79dfdacc | 1776 | if (!failed) |
23751be0 | 1777 | return true; |
79dfdacc MH |
1778 | |
1779 | /* | |
1780 | * OK, we failed to lock the whole subtree so we have to clean up | |
1781 | * what we set up to the failing subtree | |
1782 | */ | |
9f3a0d09 | 1783 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1784 | if (iter == failed) { |
9f3a0d09 JW |
1785 | mem_cgroup_iter_break(memcg, iter); |
1786 | break; | |
79dfdacc MH |
1787 | } |
1788 | iter->oom_lock = false; | |
1789 | } | |
23751be0 | 1790 | return false; |
a636b327 | 1791 | } |
0b7f569e | 1792 | |
79dfdacc | 1793 | /* |
1af8efe9 | 1794 | * Has to be called with memcg_oom_lock |
79dfdacc | 1795 | */ |
c0ff4b85 | 1796 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1797 | { |
7d74b06f KH |
1798 | struct mem_cgroup *iter; |
1799 | ||
c0ff4b85 | 1800 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1801 | iter->oom_lock = false; |
1802 | return 0; | |
1803 | } | |
1804 | ||
c0ff4b85 | 1805 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1806 | { |
1807 | struct mem_cgroup *iter; | |
1808 | ||
c0ff4b85 | 1809 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1810 | atomic_inc(&iter->under_oom); |
1811 | } | |
1812 | ||
c0ff4b85 | 1813 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1814 | { |
1815 | struct mem_cgroup *iter; | |
1816 | ||
867578cb KH |
1817 | /* |
1818 | * When a new child is created while the hierarchy is under oom, | |
1819 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1820 | * atomic_add_unless() here. | |
1821 | */ | |
c0ff4b85 | 1822 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1823 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1824 | } |
1825 | ||
1af8efe9 | 1826 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1827 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1828 | ||
dc98df5a | 1829 | struct oom_wait_info { |
d79154bb | 1830 | struct mem_cgroup *memcg; |
dc98df5a KH |
1831 | wait_queue_t wait; |
1832 | }; | |
1833 | ||
1834 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1835 | unsigned mode, int sync, void *arg) | |
1836 | { | |
d79154bb HD |
1837 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1838 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1839 | struct oom_wait_info *oom_wait_info; |
1840 | ||
1841 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1842 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1843 | |
dc98df5a | 1844 | /* |
d79154bb | 1845 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1846 | * Then we can use css_is_ancestor without taking care of RCU. |
1847 | */ | |
c0ff4b85 R |
1848 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1849 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1850 | return 0; |
dc98df5a KH |
1851 | return autoremove_wake_function(wait, mode, sync, arg); |
1852 | } | |
1853 | ||
c0ff4b85 | 1854 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1855 | { |
c0ff4b85 R |
1856 | /* for filtering, pass "memcg" as argument. */ |
1857 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1858 | } |
1859 | ||
c0ff4b85 | 1860 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1861 | { |
c0ff4b85 R |
1862 | if (memcg && atomic_read(&memcg->under_oom)) |
1863 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1864 | } |
1865 | ||
867578cb KH |
1866 | /* |
1867 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1868 | */ | |
6bbda35c KS |
1869 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
1870 | int order) | |
0b7f569e | 1871 | { |
dc98df5a | 1872 | struct oom_wait_info owait; |
3c11ecf4 | 1873 | bool locked, need_to_kill; |
867578cb | 1874 | |
d79154bb | 1875 | owait.memcg = memcg; |
dc98df5a KH |
1876 | owait.wait.flags = 0; |
1877 | owait.wait.func = memcg_oom_wake_function; | |
1878 | owait.wait.private = current; | |
1879 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1880 | need_to_kill = true; |
c0ff4b85 | 1881 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1882 | |
c0ff4b85 | 1883 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1884 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1885 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1886 | /* |
1887 | * Even if signal_pending(), we can't quit charge() loop without | |
1888 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1889 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1890 | */ | |
3c11ecf4 | 1891 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1892 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1893 | need_to_kill = false; |
1894 | if (locked) | |
c0ff4b85 | 1895 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1896 | spin_unlock(&memcg_oom_lock); |
867578cb | 1897 | |
3c11ecf4 KH |
1898 | if (need_to_kill) { |
1899 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 1900 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 1901 | } else { |
867578cb | 1902 | schedule(); |
dc98df5a | 1903 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1904 | } |
1af8efe9 | 1905 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1906 | if (locked) |
c0ff4b85 R |
1907 | mem_cgroup_oom_unlock(memcg); |
1908 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1909 | spin_unlock(&memcg_oom_lock); |
867578cb | 1910 | |
c0ff4b85 | 1911 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1912 | |
867578cb KH |
1913 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1914 | return false; | |
1915 | /* Give chance to dying process */ | |
715a5ee8 | 1916 | schedule_timeout_uninterruptible(1); |
867578cb | 1917 | return true; |
0b7f569e KH |
1918 | } |
1919 | ||
d69b042f BS |
1920 | /* |
1921 | * Currently used to update mapped file statistics, but the routine can be | |
1922 | * generalized to update other statistics as well. | |
32047e2a KH |
1923 | * |
1924 | * Notes: Race condition | |
1925 | * | |
1926 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1927 | * it tends to be costly. But considering some conditions, we doesn't need | |
1928 | * to do so _always_. | |
1929 | * | |
1930 | * Considering "charge", lock_page_cgroup() is not required because all | |
1931 | * file-stat operations happen after a page is attached to radix-tree. There | |
1932 | * are no race with "charge". | |
1933 | * | |
1934 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1935 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1936 | * if there are race with "uncharge". Statistics itself is properly handled | |
1937 | * by flags. | |
1938 | * | |
1939 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
1940 | * small, we check mm->moving_account and detect there are possibility of race |
1941 | * If there is, we take a lock. | |
d69b042f | 1942 | */ |
26174efd | 1943 | |
89c06bd5 KH |
1944 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
1945 | bool *locked, unsigned long *flags) | |
1946 | { | |
1947 | struct mem_cgroup *memcg; | |
1948 | struct page_cgroup *pc; | |
1949 | ||
1950 | pc = lookup_page_cgroup(page); | |
1951 | again: | |
1952 | memcg = pc->mem_cgroup; | |
1953 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
1954 | return; | |
1955 | /* | |
1956 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 1957 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 1958 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 1959 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 1960 | */ |
13fd1dd9 | 1961 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
1962 | return; |
1963 | ||
1964 | move_lock_mem_cgroup(memcg, flags); | |
1965 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
1966 | move_unlock_mem_cgroup(memcg, flags); | |
1967 | goto again; | |
1968 | } | |
1969 | *locked = true; | |
1970 | } | |
1971 | ||
1972 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
1973 | { | |
1974 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1975 | ||
1976 | /* | |
1977 | * It's guaranteed that pc->mem_cgroup never changes while | |
1978 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 1979 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
1980 | */ |
1981 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
1982 | } | |
1983 | ||
2a7106f2 GT |
1984 | void mem_cgroup_update_page_stat(struct page *page, |
1985 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 1986 | { |
c0ff4b85 | 1987 | struct mem_cgroup *memcg; |
32047e2a | 1988 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 1989 | unsigned long uninitialized_var(flags); |
d69b042f | 1990 | |
cfa44946 | 1991 | if (mem_cgroup_disabled()) |
d69b042f | 1992 | return; |
89c06bd5 | 1993 | |
c0ff4b85 R |
1994 | memcg = pc->mem_cgroup; |
1995 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 1996 | return; |
26174efd | 1997 | |
26174efd | 1998 | switch (idx) { |
2a7106f2 | 1999 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2000 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2001 | break; |
2002 | default: | |
2003 | BUG(); | |
8725d541 | 2004 | } |
d69b042f | 2005 | |
c0ff4b85 | 2006 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2007 | } |
26174efd | 2008 | |
cdec2e42 KH |
2009 | /* |
2010 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2011 | * TODO: maybe necessary to use big numbers in big irons. | |
2012 | */ | |
7ec99d62 | 2013 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2014 | struct memcg_stock_pcp { |
2015 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2016 | unsigned int nr_pages; |
cdec2e42 | 2017 | struct work_struct work; |
26fe6168 | 2018 | unsigned long flags; |
a0db00fc | 2019 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2020 | }; |
2021 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2022 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
2023 | |
2024 | /* | |
11c9ea4e | 2025 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
2026 | * from local stock and true is returned. If the stock is 0 or charges from a |
2027 | * cgroup which is not current target, returns false. This stock will be | |
2028 | * refilled. | |
2029 | */ | |
c0ff4b85 | 2030 | static bool consume_stock(struct mem_cgroup *memcg) |
cdec2e42 KH |
2031 | { |
2032 | struct memcg_stock_pcp *stock; | |
2033 | bool ret = true; | |
2034 | ||
2035 | stock = &get_cpu_var(memcg_stock); | |
c0ff4b85 | 2036 | if (memcg == stock->cached && stock->nr_pages) |
11c9ea4e | 2037 | stock->nr_pages--; |
cdec2e42 KH |
2038 | else /* need to call res_counter_charge */ |
2039 | ret = false; | |
2040 | put_cpu_var(memcg_stock); | |
2041 | return ret; | |
2042 | } | |
2043 | ||
2044 | /* | |
2045 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2046 | */ | |
2047 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2048 | { | |
2049 | struct mem_cgroup *old = stock->cached; | |
2050 | ||
11c9ea4e JW |
2051 | if (stock->nr_pages) { |
2052 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2053 | ||
2054 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2055 | if (do_swap_account) |
11c9ea4e JW |
2056 | res_counter_uncharge(&old->memsw, bytes); |
2057 | stock->nr_pages = 0; | |
cdec2e42 KH |
2058 | } |
2059 | stock->cached = NULL; | |
cdec2e42 KH |
2060 | } |
2061 | ||
2062 | /* | |
2063 | * This must be called under preempt disabled or must be called by | |
2064 | * a thread which is pinned to local cpu. | |
2065 | */ | |
2066 | static void drain_local_stock(struct work_struct *dummy) | |
2067 | { | |
2068 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2069 | drain_stock(stock); | |
26fe6168 | 2070 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2071 | } |
2072 | ||
2073 | /* | |
2074 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2075 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2076 | */ |
c0ff4b85 | 2077 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2078 | { |
2079 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2080 | ||
c0ff4b85 | 2081 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2082 | drain_stock(stock); |
c0ff4b85 | 2083 | stock->cached = memcg; |
cdec2e42 | 2084 | } |
11c9ea4e | 2085 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2086 | put_cpu_var(memcg_stock); |
2087 | } | |
2088 | ||
2089 | /* | |
c0ff4b85 | 2090 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2091 | * of the hierarchy under it. sync flag says whether we should block |
2092 | * until the work is done. | |
cdec2e42 | 2093 | */ |
c0ff4b85 | 2094 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2095 | { |
26fe6168 | 2096 | int cpu, curcpu; |
d38144b7 | 2097 | |
cdec2e42 | 2098 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2099 | get_online_cpus(); |
5af12d0e | 2100 | curcpu = get_cpu(); |
cdec2e42 KH |
2101 | for_each_online_cpu(cpu) { |
2102 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2103 | struct mem_cgroup *memcg; |
26fe6168 | 2104 | |
c0ff4b85 R |
2105 | memcg = stock->cached; |
2106 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2107 | continue; |
c0ff4b85 | 2108 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2109 | continue; |
d1a05b69 MH |
2110 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2111 | if (cpu == curcpu) | |
2112 | drain_local_stock(&stock->work); | |
2113 | else | |
2114 | schedule_work_on(cpu, &stock->work); | |
2115 | } | |
cdec2e42 | 2116 | } |
5af12d0e | 2117 | put_cpu(); |
d38144b7 MH |
2118 | |
2119 | if (!sync) | |
2120 | goto out; | |
2121 | ||
2122 | for_each_online_cpu(cpu) { | |
2123 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2124 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2125 | flush_work(&stock->work); |
2126 | } | |
2127 | out: | |
cdec2e42 | 2128 | put_online_cpus(); |
d38144b7 MH |
2129 | } |
2130 | ||
2131 | /* | |
2132 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2133 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2134 | * expects some charges will be back to res_counter later but cannot wait for | |
2135 | * it. | |
2136 | */ | |
c0ff4b85 | 2137 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2138 | { |
9f50fad6 MH |
2139 | /* |
2140 | * If someone calls draining, avoid adding more kworker runs. | |
2141 | */ | |
2142 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2143 | return; | |
c0ff4b85 | 2144 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2145 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2146 | } |
2147 | ||
2148 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2149 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2150 | { |
2151 | /* called when force_empty is called */ | |
9f50fad6 | 2152 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2153 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2154 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2155 | } |
2156 | ||
711d3d2c KH |
2157 | /* |
2158 | * This function drains percpu counter value from DEAD cpu and | |
2159 | * move it to local cpu. Note that this function can be preempted. | |
2160 | */ | |
c0ff4b85 | 2161 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2162 | { |
2163 | int i; | |
2164 | ||
c0ff4b85 | 2165 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2166 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2167 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2168 | |
c0ff4b85 R |
2169 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2170 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2171 | } |
e9f8974f | 2172 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2173 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2174 | |
c0ff4b85 R |
2175 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2176 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2177 | } |
c0ff4b85 | 2178 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2179 | } |
2180 | ||
2181 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2182 | unsigned long action, |
2183 | void *hcpu) | |
2184 | { | |
2185 | int cpu = (unsigned long)hcpu; | |
2186 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2187 | struct mem_cgroup *iter; |
cdec2e42 | 2188 | |
619d094b | 2189 | if (action == CPU_ONLINE) |
1489ebad | 2190 | return NOTIFY_OK; |
1489ebad | 2191 | |
d833049b | 2192 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2193 | return NOTIFY_OK; |
711d3d2c | 2194 | |
9f3a0d09 | 2195 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2196 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2197 | ||
cdec2e42 KH |
2198 | stock = &per_cpu(memcg_stock, cpu); |
2199 | drain_stock(stock); | |
2200 | return NOTIFY_OK; | |
2201 | } | |
2202 | ||
4b534334 KH |
2203 | |
2204 | /* See __mem_cgroup_try_charge() for details */ | |
2205 | enum { | |
2206 | CHARGE_OK, /* success */ | |
2207 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2208 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2209 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2210 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2211 | }; | |
2212 | ||
c0ff4b85 | 2213 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
7ec99d62 | 2214 | unsigned int nr_pages, bool oom_check) |
4b534334 | 2215 | { |
7ec99d62 | 2216 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2217 | struct mem_cgroup *mem_over_limit; |
2218 | struct res_counter *fail_res; | |
2219 | unsigned long flags = 0; | |
2220 | int ret; | |
2221 | ||
c0ff4b85 | 2222 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2223 | |
2224 | if (likely(!ret)) { | |
2225 | if (!do_swap_account) | |
2226 | return CHARGE_OK; | |
c0ff4b85 | 2227 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2228 | if (likely(!ret)) |
2229 | return CHARGE_OK; | |
2230 | ||
c0ff4b85 | 2231 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2232 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2233 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2234 | } else | |
2235 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2236 | /* |
7ec99d62 JW |
2237 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2238 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2239 | * |
2240 | * Never reclaim on behalf of optional batching, retry with a | |
2241 | * single page instead. | |
2242 | */ | |
7ec99d62 | 2243 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2244 | return CHARGE_RETRY; |
2245 | ||
2246 | if (!(gfp_mask & __GFP_WAIT)) | |
2247 | return CHARGE_WOULDBLOCK; | |
2248 | ||
5660048c | 2249 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2250 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2251 | return CHARGE_RETRY; |
4b534334 | 2252 | /* |
19942822 JW |
2253 | * Even though the limit is exceeded at this point, reclaim |
2254 | * may have been able to free some pages. Retry the charge | |
2255 | * before killing the task. | |
2256 | * | |
2257 | * Only for regular pages, though: huge pages are rather | |
2258 | * unlikely to succeed so close to the limit, and we fall back | |
2259 | * to regular pages anyway in case of failure. | |
4b534334 | 2260 | */ |
7ec99d62 | 2261 | if (nr_pages == 1 && ret) |
4b534334 KH |
2262 | return CHARGE_RETRY; |
2263 | ||
2264 | /* | |
2265 | * At task move, charge accounts can be doubly counted. So, it's | |
2266 | * better to wait until the end of task_move if something is going on. | |
2267 | */ | |
2268 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2269 | return CHARGE_RETRY; | |
2270 | ||
2271 | /* If we don't need to call oom-killer at el, return immediately */ | |
2272 | if (!oom_check) | |
2273 | return CHARGE_NOMEM; | |
2274 | /* check OOM */ | |
e845e199 | 2275 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2276 | return CHARGE_OOM_DIE; |
2277 | ||
2278 | return CHARGE_RETRY; | |
2279 | } | |
2280 | ||
f817ed48 | 2281 | /* |
38c5d72f KH |
2282 | * __mem_cgroup_try_charge() does |
2283 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2284 | * 2. update res_counter | |
2285 | * 3. call memory reclaim if necessary. | |
2286 | * | |
2287 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2288 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2289 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2290 | * as possible without any hazards. 2: all pages should have a valid | |
2291 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2292 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2293 | * | |
2294 | * So __mem_cgroup_try_charge() will return | |
2295 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2296 | * -ENOMEM ... charge failure because of resource limits. | |
2297 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2298 | * | |
2299 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2300 | * the oom-killer can be invoked. | |
8a9f3ccd | 2301 | */ |
f817ed48 | 2302 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2303 | gfp_t gfp_mask, |
7ec99d62 | 2304 | unsigned int nr_pages, |
c0ff4b85 | 2305 | struct mem_cgroup **ptr, |
7ec99d62 | 2306 | bool oom) |
8a9f3ccd | 2307 | { |
7ec99d62 | 2308 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2309 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2310 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2311 | int ret; |
a636b327 | 2312 | |
867578cb KH |
2313 | /* |
2314 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2315 | * in system level. So, allow to go ahead dying process in addition to | |
2316 | * MEMDIE process. | |
2317 | */ | |
2318 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2319 | || fatal_signal_pending(current))) | |
2320 | goto bypass; | |
a636b327 | 2321 | |
8a9f3ccd | 2322 | /* |
3be91277 HD |
2323 | * We always charge the cgroup the mm_struct belongs to. |
2324 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2325 | * thread group leader migrates. It's possible that mm is not |
2326 | * set, if so charge the init_mm (happens for pagecache usage). | |
2327 | */ | |
c0ff4b85 | 2328 | if (!*ptr && !mm) |
38c5d72f | 2329 | *ptr = root_mem_cgroup; |
f75ca962 | 2330 | again: |
c0ff4b85 R |
2331 | if (*ptr) { /* css should be a valid one */ |
2332 | memcg = *ptr; | |
2333 | VM_BUG_ON(css_is_removed(&memcg->css)); | |
2334 | if (mem_cgroup_is_root(memcg)) | |
f75ca962 | 2335 | goto done; |
c0ff4b85 | 2336 | if (nr_pages == 1 && consume_stock(memcg)) |
f75ca962 | 2337 | goto done; |
c0ff4b85 | 2338 | css_get(&memcg->css); |
4b534334 | 2339 | } else { |
f75ca962 | 2340 | struct task_struct *p; |
54595fe2 | 2341 | |
f75ca962 KH |
2342 | rcu_read_lock(); |
2343 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2344 | /* |
ebb76ce1 | 2345 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2346 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2347 | * race with swapoff. Then, we have small risk of mis-accouning. |
2348 | * But such kind of mis-account by race always happens because | |
2349 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2350 | * small race, here. | |
2351 | * (*) swapoff at el will charge against mm-struct not against | |
2352 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2353 | */ |
c0ff4b85 | 2354 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2355 | if (!memcg) |
2356 | memcg = root_mem_cgroup; | |
2357 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2358 | rcu_read_unlock(); |
2359 | goto done; | |
2360 | } | |
c0ff4b85 | 2361 | if (nr_pages == 1 && consume_stock(memcg)) { |
f75ca962 KH |
2362 | /* |
2363 | * It seems dagerous to access memcg without css_get(). | |
2364 | * But considering how consume_stok works, it's not | |
2365 | * necessary. If consume_stock success, some charges | |
2366 | * from this memcg are cached on this cpu. So, we | |
2367 | * don't need to call css_get()/css_tryget() before | |
2368 | * calling consume_stock(). | |
2369 | */ | |
2370 | rcu_read_unlock(); | |
2371 | goto done; | |
2372 | } | |
2373 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2374 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2375 | rcu_read_unlock(); |
2376 | goto again; | |
2377 | } | |
2378 | rcu_read_unlock(); | |
2379 | } | |
8a9f3ccd | 2380 | |
4b534334 KH |
2381 | do { |
2382 | bool oom_check; | |
7a81b88c | 2383 | |
4b534334 | 2384 | /* If killed, bypass charge */ |
f75ca962 | 2385 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2386 | css_put(&memcg->css); |
4b534334 | 2387 | goto bypass; |
f75ca962 | 2388 | } |
6d61ef40 | 2389 | |
4b534334 KH |
2390 | oom_check = false; |
2391 | if (oom && !nr_oom_retries) { | |
2392 | oom_check = true; | |
2393 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2394 | } |
66e1707b | 2395 | |
c0ff4b85 | 2396 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
4b534334 KH |
2397 | switch (ret) { |
2398 | case CHARGE_OK: | |
2399 | break; | |
2400 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2401 | batch = nr_pages; |
c0ff4b85 R |
2402 | css_put(&memcg->css); |
2403 | memcg = NULL; | |
f75ca962 | 2404 | goto again; |
4b534334 | 2405 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2406 | css_put(&memcg->css); |
4b534334 KH |
2407 | goto nomem; |
2408 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2409 | if (!oom) { |
c0ff4b85 | 2410 | css_put(&memcg->css); |
867578cb | 2411 | goto nomem; |
f75ca962 | 2412 | } |
4b534334 KH |
2413 | /* If oom, we never return -ENOMEM */ |
2414 | nr_oom_retries--; | |
2415 | break; | |
2416 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2417 | css_put(&memcg->css); |
867578cb | 2418 | goto bypass; |
66e1707b | 2419 | } |
4b534334 KH |
2420 | } while (ret != CHARGE_OK); |
2421 | ||
7ec99d62 | 2422 | if (batch > nr_pages) |
c0ff4b85 R |
2423 | refill_stock(memcg, batch - nr_pages); |
2424 | css_put(&memcg->css); | |
0c3e73e8 | 2425 | done: |
c0ff4b85 | 2426 | *ptr = memcg; |
7a81b88c KH |
2427 | return 0; |
2428 | nomem: | |
c0ff4b85 | 2429 | *ptr = NULL; |
7a81b88c | 2430 | return -ENOMEM; |
867578cb | 2431 | bypass: |
38c5d72f KH |
2432 | *ptr = root_mem_cgroup; |
2433 | return -EINTR; | |
7a81b88c | 2434 | } |
8a9f3ccd | 2435 | |
a3032a2c DN |
2436 | /* |
2437 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2438 | * This function is for that and do uncharge, put css's refcnt. | |
2439 | * gotten by try_charge(). | |
2440 | */ | |
c0ff4b85 | 2441 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2442 | unsigned int nr_pages) |
a3032a2c | 2443 | { |
c0ff4b85 | 2444 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2445 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2446 | ||
c0ff4b85 | 2447 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2448 | if (do_swap_account) |
c0ff4b85 | 2449 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2450 | } |
854ffa8d DN |
2451 | } |
2452 | ||
d01dd17f KH |
2453 | /* |
2454 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2455 | * This is useful when moving usage to parent cgroup. | |
2456 | */ | |
2457 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2458 | unsigned int nr_pages) | |
2459 | { | |
2460 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2461 | ||
2462 | if (mem_cgroup_is_root(memcg)) | |
2463 | return; | |
2464 | ||
2465 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2466 | if (do_swap_account) | |
2467 | res_counter_uncharge_until(&memcg->memsw, | |
2468 | memcg->memsw.parent, bytes); | |
2469 | } | |
2470 | ||
a3b2d692 KH |
2471 | /* |
2472 | * A helper function to get mem_cgroup from ID. must be called under | |
2473 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2474 | * it's concern. (dropping refcnt from swap can be called against removed | |
2475 | * memcg.) | |
2476 | */ | |
2477 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2478 | { | |
2479 | struct cgroup_subsys_state *css; | |
2480 | ||
2481 | /* ID 0 is unused ID */ | |
2482 | if (!id) | |
2483 | return NULL; | |
2484 | css = css_lookup(&mem_cgroup_subsys, id); | |
2485 | if (!css) | |
2486 | return NULL; | |
2487 | return container_of(css, struct mem_cgroup, css); | |
2488 | } | |
2489 | ||
e42d9d5d | 2490 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2491 | { |
c0ff4b85 | 2492 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2493 | struct page_cgroup *pc; |
a3b2d692 | 2494 | unsigned short id; |
b5a84319 KH |
2495 | swp_entry_t ent; |
2496 | ||
3c776e64 DN |
2497 | VM_BUG_ON(!PageLocked(page)); |
2498 | ||
3c776e64 | 2499 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2500 | lock_page_cgroup(pc); |
a3b2d692 | 2501 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2502 | memcg = pc->mem_cgroup; |
2503 | if (memcg && !css_tryget(&memcg->css)) | |
2504 | memcg = NULL; | |
e42d9d5d | 2505 | } else if (PageSwapCache(page)) { |
3c776e64 | 2506 | ent.val = page_private(page); |
9fb4b7cc | 2507 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2508 | rcu_read_lock(); |
c0ff4b85 R |
2509 | memcg = mem_cgroup_lookup(id); |
2510 | if (memcg && !css_tryget(&memcg->css)) | |
2511 | memcg = NULL; | |
a3b2d692 | 2512 | rcu_read_unlock(); |
3c776e64 | 2513 | } |
c0bd3f63 | 2514 | unlock_page_cgroup(pc); |
c0ff4b85 | 2515 | return memcg; |
b5a84319 KH |
2516 | } |
2517 | ||
c0ff4b85 | 2518 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2519 | struct page *page, |
7ec99d62 | 2520 | unsigned int nr_pages, |
9ce70c02 HD |
2521 | enum charge_type ctype, |
2522 | bool lrucare) | |
7a81b88c | 2523 | { |
ce587e65 | 2524 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2525 | struct zone *uninitialized_var(zone); |
fa9add64 | 2526 | struct lruvec *lruvec; |
9ce70c02 | 2527 | bool was_on_lru = false; |
b2402857 | 2528 | bool anon; |
9ce70c02 | 2529 | |
ca3e0214 KH |
2530 | lock_page_cgroup(pc); |
2531 | if (unlikely(PageCgroupUsed(pc))) { | |
2532 | unlock_page_cgroup(pc); | |
c0ff4b85 | 2533 | __mem_cgroup_cancel_charge(memcg, nr_pages); |
ca3e0214 KH |
2534 | return; |
2535 | } | |
2536 | /* | |
2537 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2538 | * accessed by any other context at this point. | |
2539 | */ | |
9ce70c02 HD |
2540 | |
2541 | /* | |
2542 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2543 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2544 | */ | |
2545 | if (lrucare) { | |
2546 | zone = page_zone(page); | |
2547 | spin_lock_irq(&zone->lru_lock); | |
2548 | if (PageLRU(page)) { | |
fa9add64 | 2549 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2550 | ClearPageLRU(page); |
fa9add64 | 2551 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2552 | was_on_lru = true; |
2553 | } | |
2554 | } | |
2555 | ||
c0ff4b85 | 2556 | pc->mem_cgroup = memcg; |
261fb61a KH |
2557 | /* |
2558 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2559 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2560 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2561 | * before USED bit, we need memory barrier here. | |
2562 | * See mem_cgroup_add_lru_list(), etc. | |
2563 | */ | |
08e552c6 | 2564 | smp_wmb(); |
b2402857 | 2565 | SetPageCgroupUsed(pc); |
3be91277 | 2566 | |
9ce70c02 HD |
2567 | if (lrucare) { |
2568 | if (was_on_lru) { | |
fa9add64 | 2569 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2570 | VM_BUG_ON(PageLRU(page)); |
2571 | SetPageLRU(page); | |
fa9add64 | 2572 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2573 | } |
2574 | spin_unlock_irq(&zone->lru_lock); | |
2575 | } | |
2576 | ||
41326c17 | 2577 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2578 | anon = true; |
2579 | else | |
2580 | anon = false; | |
2581 | ||
2582 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2583 | unlock_page_cgroup(pc); |
9ce70c02 | 2584 | |
430e4863 KH |
2585 | /* |
2586 | * "charge_statistics" updated event counter. Then, check it. | |
2587 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2588 | * if they exceeds softlimit. | |
2589 | */ | |
c0ff4b85 | 2590 | memcg_check_events(memcg, page); |
7a81b88c | 2591 | } |
66e1707b | 2592 | |
ca3e0214 KH |
2593 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2594 | ||
a0db00fc | 2595 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
2596 | /* |
2597 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2598 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2599 | * charge/uncharge will be never happen and move_account() is done under | |
2600 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2601 | */ |
e94c8a9c | 2602 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
2603 | { |
2604 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
2605 | struct page_cgroup *pc; |
2606 | int i; | |
ca3e0214 | 2607 | |
3d37c4a9 KH |
2608 | if (mem_cgroup_disabled()) |
2609 | return; | |
e94c8a9c KH |
2610 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
2611 | pc = head_pc + i; | |
2612 | pc->mem_cgroup = head_pc->mem_cgroup; | |
2613 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
2614 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2615 | } | |
ca3e0214 | 2616 | } |
12d27107 | 2617 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2618 | |
f817ed48 | 2619 | /** |
de3638d9 | 2620 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2621 | * @page: the page |
7ec99d62 | 2622 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2623 | * @pc: page_cgroup of the page. |
2624 | * @from: mem_cgroup which the page is moved from. | |
2625 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2626 | * | |
2627 | * The caller must confirm following. | |
08e552c6 | 2628 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2629 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2630 | * |
2f3479b1 KH |
2631 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2632 | * from old cgroup. | |
f817ed48 | 2633 | */ |
7ec99d62 JW |
2634 | static int mem_cgroup_move_account(struct page *page, |
2635 | unsigned int nr_pages, | |
2636 | struct page_cgroup *pc, | |
2637 | struct mem_cgroup *from, | |
2f3479b1 | 2638 | struct mem_cgroup *to) |
f817ed48 | 2639 | { |
de3638d9 JW |
2640 | unsigned long flags; |
2641 | int ret; | |
b2402857 | 2642 | bool anon = PageAnon(page); |
987eba66 | 2643 | |
f817ed48 | 2644 | VM_BUG_ON(from == to); |
5564e88b | 2645 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2646 | /* |
2647 | * The page is isolated from LRU. So, collapse function | |
2648 | * will not handle this page. But page splitting can happen. | |
2649 | * Do this check under compound_page_lock(). The caller should | |
2650 | * hold it. | |
2651 | */ | |
2652 | ret = -EBUSY; | |
7ec99d62 | 2653 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2654 | goto out; |
2655 | ||
2656 | lock_page_cgroup(pc); | |
2657 | ||
2658 | ret = -EINVAL; | |
2659 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2660 | goto unlock; | |
2661 | ||
312734c0 | 2662 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 2663 | |
2ff76f11 | 2664 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
2665 | /* Update mapped_file data for mem_cgroup */ |
2666 | preempt_disable(); | |
2667 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2668 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2669 | preempt_enable(); | |
d69b042f | 2670 | } |
b2402857 | 2671 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
d69b042f | 2672 | |
854ffa8d | 2673 | /* caller should have done css_get */ |
08e552c6 | 2674 | pc->mem_cgroup = to; |
b2402857 | 2675 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
88703267 KH |
2676 | /* |
2677 | * We charges against "to" which may not have any tasks. Then, "to" | |
2678 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2679 | * this function is just force_empty() and move charge, so it's |
25985edc | 2680 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2681 | * status here. |
88703267 | 2682 | */ |
312734c0 | 2683 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
2684 | ret = 0; |
2685 | unlock: | |
57f9fd7d | 2686 | unlock_page_cgroup(pc); |
d2265e6f KH |
2687 | /* |
2688 | * check events | |
2689 | */ | |
5564e88b JW |
2690 | memcg_check_events(to, page); |
2691 | memcg_check_events(from, page); | |
de3638d9 | 2692 | out: |
f817ed48 KH |
2693 | return ret; |
2694 | } | |
2695 | ||
2696 | /* | |
2697 | * move charges to its parent. | |
2698 | */ | |
2699 | ||
5564e88b JW |
2700 | static int mem_cgroup_move_parent(struct page *page, |
2701 | struct page_cgroup *pc, | |
6068bf01 | 2702 | struct mem_cgroup *child) |
f817ed48 | 2703 | { |
f817ed48 | 2704 | struct mem_cgroup *parent; |
7ec99d62 | 2705 | unsigned int nr_pages; |
4be4489f | 2706 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2707 | int ret; |
2708 | ||
2709 | /* Is ROOT ? */ | |
cc926f78 | 2710 | if (mem_cgroup_is_root(child)) |
f817ed48 KH |
2711 | return -EINVAL; |
2712 | ||
57f9fd7d DN |
2713 | ret = -EBUSY; |
2714 | if (!get_page_unless_zero(page)) | |
2715 | goto out; | |
2716 | if (isolate_lru_page(page)) | |
2717 | goto put; | |
52dbb905 | 2718 | |
7ec99d62 | 2719 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2720 | |
cc926f78 KH |
2721 | parent = parent_mem_cgroup(child); |
2722 | /* | |
2723 | * If no parent, move charges to root cgroup. | |
2724 | */ | |
2725 | if (!parent) | |
2726 | parent = root_mem_cgroup; | |
f817ed48 | 2727 | |
7ec99d62 | 2728 | if (nr_pages > 1) |
987eba66 KH |
2729 | flags = compound_lock_irqsave(page); |
2730 | ||
cc926f78 | 2731 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 2732 | pc, child, parent); |
cc926f78 KH |
2733 | if (!ret) |
2734 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 2735 | |
7ec99d62 | 2736 | if (nr_pages > 1) |
987eba66 | 2737 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 2738 | putback_lru_page(page); |
57f9fd7d | 2739 | put: |
40d58138 | 2740 | put_page(page); |
57f9fd7d | 2741 | out: |
f817ed48 KH |
2742 | return ret; |
2743 | } | |
2744 | ||
7a81b88c KH |
2745 | /* |
2746 | * Charge the memory controller for page usage. | |
2747 | * Return | |
2748 | * 0 if the charge was successful | |
2749 | * < 0 if the cgroup is over its limit | |
2750 | */ | |
2751 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2752 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2753 | { |
c0ff4b85 | 2754 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2755 | unsigned int nr_pages = 1; |
8493ae43 | 2756 | bool oom = true; |
7a81b88c | 2757 | int ret; |
ec168510 | 2758 | |
37c2ac78 | 2759 | if (PageTransHuge(page)) { |
7ec99d62 | 2760 | nr_pages <<= compound_order(page); |
37c2ac78 | 2761 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2762 | /* |
2763 | * Never OOM-kill a process for a huge page. The | |
2764 | * fault handler will fall back to regular pages. | |
2765 | */ | |
2766 | oom = false; | |
37c2ac78 | 2767 | } |
7a81b88c | 2768 | |
c0ff4b85 | 2769 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 2770 | if (ret == -ENOMEM) |
7a81b88c | 2771 | return ret; |
ce587e65 | 2772 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 2773 | return 0; |
8a9f3ccd BS |
2774 | } |
2775 | ||
7a81b88c KH |
2776 | int mem_cgroup_newpage_charge(struct page *page, |
2777 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2778 | { |
f8d66542 | 2779 | if (mem_cgroup_disabled()) |
cede86ac | 2780 | return 0; |
7a0524cf JW |
2781 | VM_BUG_ON(page_mapped(page)); |
2782 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
2783 | VM_BUG_ON(!mm); | |
217bc319 | 2784 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 2785 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
2786 | } |
2787 | ||
83aae4c7 DN |
2788 | static void |
2789 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2790 | enum charge_type ctype); | |
2791 | ||
e1a1cd59 BS |
2792 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2793 | gfp_t gfp_mask) | |
8697d331 | 2794 | { |
c0ff4b85 | 2795 | struct mem_cgroup *memcg = NULL; |
dc67d504 | 2796 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
b5a84319 KH |
2797 | int ret; |
2798 | ||
f8d66542 | 2799 | if (mem_cgroup_disabled()) |
cede86ac | 2800 | return 0; |
52d4b9ac KH |
2801 | if (PageCompound(page)) |
2802 | return 0; | |
accf163e | 2803 | |
73045c47 | 2804 | if (unlikely(!mm)) |
8697d331 | 2805 | mm = &init_mm; |
dc67d504 KH |
2806 | if (!page_is_file_cache(page)) |
2807 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
accf163e | 2808 | |
38c5d72f | 2809 | if (!PageSwapCache(page)) |
dc67d504 | 2810 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); |
38c5d72f | 2811 | else { /* page is swapcache/shmem */ |
c0ff4b85 | 2812 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg); |
83aae4c7 | 2813 | if (!ret) |
dc67d504 KH |
2814 | __mem_cgroup_commit_charge_swapin(page, memcg, type); |
2815 | } | |
b5a84319 | 2816 | return ret; |
e8589cc1 KH |
2817 | } |
2818 | ||
54595fe2 KH |
2819 | /* |
2820 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2821 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2822 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2823 | * "commit()" or removed by "cancel()" |
2824 | */ | |
8c7c6e34 KH |
2825 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2826 | struct page *page, | |
72835c86 | 2827 | gfp_t mask, struct mem_cgroup **memcgp) |
8c7c6e34 | 2828 | { |
c0ff4b85 | 2829 | struct mem_cgroup *memcg; |
54595fe2 | 2830 | int ret; |
8c7c6e34 | 2831 | |
72835c86 | 2832 | *memcgp = NULL; |
56039efa | 2833 | |
f8d66542 | 2834 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2835 | return 0; |
2836 | ||
2837 | if (!do_swap_account) | |
2838 | goto charge_cur_mm; | |
8c7c6e34 KH |
2839 | /* |
2840 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2841 | * the pte, and even removed page from swap cache: in those cases |
2842 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2843 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2844 | */ |
2845 | if (!PageSwapCache(page)) | |
407f9c8b | 2846 | goto charge_cur_mm; |
c0ff4b85 R |
2847 | memcg = try_get_mem_cgroup_from_page(page); |
2848 | if (!memcg) | |
54595fe2 | 2849 | goto charge_cur_mm; |
72835c86 JW |
2850 | *memcgp = memcg; |
2851 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 2852 | css_put(&memcg->css); |
38c5d72f KH |
2853 | if (ret == -EINTR) |
2854 | ret = 0; | |
54595fe2 | 2855 | return ret; |
8c7c6e34 KH |
2856 | charge_cur_mm: |
2857 | if (unlikely(!mm)) | |
2858 | mm = &init_mm; | |
38c5d72f KH |
2859 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
2860 | if (ret == -EINTR) | |
2861 | ret = 0; | |
2862 | return ret; | |
8c7c6e34 KH |
2863 | } |
2864 | ||
83aae4c7 | 2865 | static void |
72835c86 | 2866 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 2867 | enum charge_type ctype) |
7a81b88c | 2868 | { |
f8d66542 | 2869 | if (mem_cgroup_disabled()) |
7a81b88c | 2870 | return; |
72835c86 | 2871 | if (!memcg) |
7a81b88c | 2872 | return; |
72835c86 | 2873 | cgroup_exclude_rmdir(&memcg->css); |
5a6475a4 | 2874 | |
ce587e65 | 2875 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
2876 | /* |
2877 | * Now swap is on-memory. This means this page may be | |
2878 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2879 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2880 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2881 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2882 | */ |
03f3c433 | 2883 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2884 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 2885 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 2886 | } |
88703267 KH |
2887 | /* |
2888 | * At swapin, we may charge account against cgroup which has no tasks. | |
2889 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2890 | * In that case, we need to call pre_destroy() again. check it here. | |
2891 | */ | |
72835c86 | 2892 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
7a81b88c KH |
2893 | } |
2894 | ||
72835c86 JW |
2895 | void mem_cgroup_commit_charge_swapin(struct page *page, |
2896 | struct mem_cgroup *memcg) | |
83aae4c7 | 2897 | { |
72835c86 | 2898 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 2899 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
2900 | } |
2901 | ||
c0ff4b85 | 2902 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
7a81b88c | 2903 | { |
f8d66542 | 2904 | if (mem_cgroup_disabled()) |
7a81b88c | 2905 | return; |
c0ff4b85 | 2906 | if (!memcg) |
7a81b88c | 2907 | return; |
c0ff4b85 | 2908 | __mem_cgroup_cancel_charge(memcg, 1); |
7a81b88c KH |
2909 | } |
2910 | ||
c0ff4b85 | 2911 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2912 | unsigned int nr_pages, |
2913 | const enum charge_type ctype) | |
569b846d KH |
2914 | { |
2915 | struct memcg_batch_info *batch = NULL; | |
2916 | bool uncharge_memsw = true; | |
7ec99d62 | 2917 | |
569b846d KH |
2918 | /* If swapout, usage of swap doesn't decrease */ |
2919 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2920 | uncharge_memsw = false; | |
569b846d KH |
2921 | |
2922 | batch = ¤t->memcg_batch; | |
2923 | /* | |
2924 | * In usual, we do css_get() when we remember memcg pointer. | |
2925 | * But in this case, we keep res->usage until end of a series of | |
2926 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2927 | */ | |
2928 | if (!batch->memcg) | |
c0ff4b85 | 2929 | batch->memcg = memcg; |
3c11ecf4 KH |
2930 | /* |
2931 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2932 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2933 | * the same cgroup and we have chance to coalesce uncharges. |
2934 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2935 | * because we want to do uncharge as soon as possible. | |
2936 | */ | |
2937 | ||
2938 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2939 | goto direct_uncharge; | |
2940 | ||
7ec99d62 | 2941 | if (nr_pages > 1) |
ec168510 AA |
2942 | goto direct_uncharge; |
2943 | ||
569b846d KH |
2944 | /* |
2945 | * In typical case, batch->memcg == mem. This means we can | |
2946 | * merge a series of uncharges to an uncharge of res_counter. | |
2947 | * If not, we uncharge res_counter ony by one. | |
2948 | */ | |
c0ff4b85 | 2949 | if (batch->memcg != memcg) |
569b846d KH |
2950 | goto direct_uncharge; |
2951 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2952 | batch->nr_pages++; |
569b846d | 2953 | if (uncharge_memsw) |
7ffd4ca7 | 2954 | batch->memsw_nr_pages++; |
569b846d KH |
2955 | return; |
2956 | direct_uncharge: | |
c0ff4b85 | 2957 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 2958 | if (uncharge_memsw) |
c0ff4b85 R |
2959 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
2960 | if (unlikely(batch->memcg != memcg)) | |
2961 | memcg_oom_recover(memcg); | |
569b846d | 2962 | } |
7a81b88c | 2963 | |
8a9f3ccd | 2964 | /* |
69029cd5 | 2965 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2966 | */ |
8c7c6e34 | 2967 | static struct mem_cgroup * |
69029cd5 | 2968 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2969 | { |
c0ff4b85 | 2970 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
2971 | unsigned int nr_pages = 1; |
2972 | struct page_cgroup *pc; | |
b2402857 | 2973 | bool anon; |
8a9f3ccd | 2974 | |
f8d66542 | 2975 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2976 | return NULL; |
4077960e | 2977 | |
d13d1443 | 2978 | if (PageSwapCache(page)) |
8c7c6e34 | 2979 | return NULL; |
d13d1443 | 2980 | |
37c2ac78 | 2981 | if (PageTransHuge(page)) { |
7ec99d62 | 2982 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
2983 | VM_BUG_ON(!PageTransHuge(page)); |
2984 | } | |
8697d331 | 2985 | /* |
3c541e14 | 2986 | * Check if our page_cgroup is valid |
8697d331 | 2987 | */ |
52d4b9ac | 2988 | pc = lookup_page_cgroup(page); |
cfa44946 | 2989 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 2990 | return NULL; |
b9c565d5 | 2991 | |
52d4b9ac | 2992 | lock_page_cgroup(pc); |
d13d1443 | 2993 | |
c0ff4b85 | 2994 | memcg = pc->mem_cgroup; |
8c7c6e34 | 2995 | |
d13d1443 KH |
2996 | if (!PageCgroupUsed(pc)) |
2997 | goto unlock_out; | |
2998 | ||
b2402857 KH |
2999 | anon = PageAnon(page); |
3000 | ||
d13d1443 | 3001 | switch (ctype) { |
41326c17 | 3002 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
3003 | /* |
3004 | * Generally PageAnon tells if it's the anon statistics to be | |
3005 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
3006 | * used before page reached the stage of being marked PageAnon. | |
3007 | */ | |
b2402857 KH |
3008 | anon = true; |
3009 | /* fallthrough */ | |
8a9478ca | 3010 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 3011 | /* See mem_cgroup_prepare_migration() */ |
3012 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
3013 | goto unlock_out; |
3014 | break; | |
3015 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
3016 | if (!PageAnon(page)) { /* Shared memory */ | |
3017 | if (page->mapping && !page_is_file_cache(page)) | |
3018 | goto unlock_out; | |
3019 | } else if (page_mapped(page)) /* Anon */ | |
3020 | goto unlock_out; | |
3021 | break; | |
3022 | default: | |
3023 | break; | |
52d4b9ac | 3024 | } |
d13d1443 | 3025 | |
b2402857 | 3026 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 3027 | |
52d4b9ac | 3028 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3029 | /* |
3030 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3031 | * freed from LRU. This is safe because uncharged page is expected not | |
3032 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3033 | * special functions. | |
3034 | */ | |
b9c565d5 | 3035 | |
52d4b9ac | 3036 | unlock_page_cgroup(pc); |
f75ca962 | 3037 | /* |
c0ff4b85 | 3038 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
3039 | * will never be freed. |
3040 | */ | |
c0ff4b85 | 3041 | memcg_check_events(memcg, page); |
f75ca962 | 3042 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
3043 | mem_cgroup_swap_statistics(memcg, true); |
3044 | mem_cgroup_get(memcg); | |
f75ca962 | 3045 | } |
c0ff4b85 R |
3046 | if (!mem_cgroup_is_root(memcg)) |
3047 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); | |
6d12e2d8 | 3048 | |
c0ff4b85 | 3049 | return memcg; |
d13d1443 KH |
3050 | |
3051 | unlock_out: | |
3052 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3053 | return NULL; |
3c541e14 BS |
3054 | } |
3055 | ||
69029cd5 KH |
3056 | void mem_cgroup_uncharge_page(struct page *page) |
3057 | { | |
52d4b9ac KH |
3058 | /* early check. */ |
3059 | if (page_mapped(page)) | |
3060 | return; | |
40f23a21 | 3061 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
41326c17 | 3062 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON); |
69029cd5 KH |
3063 | } |
3064 | ||
3065 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3066 | { | |
3067 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3068 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
3069 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
3070 | } | |
3071 | ||
569b846d KH |
3072 | /* |
3073 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3074 | * In that cases, pages are freed continuously and we can expect pages | |
3075 | * are in the same memcg. All these calls itself limits the number of | |
3076 | * pages freed at once, then uncharge_start/end() is called properly. | |
3077 | * This may be called prural(2) times in a context, | |
3078 | */ | |
3079 | ||
3080 | void mem_cgroup_uncharge_start(void) | |
3081 | { | |
3082 | current->memcg_batch.do_batch++; | |
3083 | /* We can do nest. */ | |
3084 | if (current->memcg_batch.do_batch == 1) { | |
3085 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3086 | current->memcg_batch.nr_pages = 0; |
3087 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3088 | } |
3089 | } | |
3090 | ||
3091 | void mem_cgroup_uncharge_end(void) | |
3092 | { | |
3093 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3094 | ||
3095 | if (!batch->do_batch) | |
3096 | return; | |
3097 | ||
3098 | batch->do_batch--; | |
3099 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3100 | return; | |
3101 | ||
3102 | if (!batch->memcg) | |
3103 | return; | |
3104 | /* | |
3105 | * This "batch->memcg" is valid without any css_get/put etc... | |
3106 | * bacause we hide charges behind us. | |
3107 | */ | |
7ffd4ca7 JW |
3108 | if (batch->nr_pages) |
3109 | res_counter_uncharge(&batch->memcg->res, | |
3110 | batch->nr_pages * PAGE_SIZE); | |
3111 | if (batch->memsw_nr_pages) | |
3112 | res_counter_uncharge(&batch->memcg->memsw, | |
3113 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3114 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3115 | /* forget this pointer (for sanity check) */ |
3116 | batch->memcg = NULL; | |
3117 | } | |
3118 | ||
e767e056 | 3119 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3120 | /* |
e767e056 | 3121 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3122 | * memcg information is recorded to swap_cgroup of "ent" |
3123 | */ | |
8a9478ca KH |
3124 | void |
3125 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3126 | { |
3127 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3128 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3129 | ||
3130 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3131 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3132 | ||
3133 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3134 | |
f75ca962 KH |
3135 | /* |
3136 | * record memcg information, if swapout && memcg != NULL, | |
3137 | * mem_cgroup_get() was called in uncharge(). | |
3138 | */ | |
3139 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3140 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3141 | } |
e767e056 | 3142 | #endif |
8c7c6e34 | 3143 | |
c255a458 | 3144 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
3145 | /* |
3146 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3147 | * uncharge "memsw" account. | |
3148 | */ | |
3149 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3150 | { |
8c7c6e34 | 3151 | struct mem_cgroup *memcg; |
a3b2d692 | 3152 | unsigned short id; |
8c7c6e34 KH |
3153 | |
3154 | if (!do_swap_account) | |
3155 | return; | |
3156 | ||
a3b2d692 KH |
3157 | id = swap_cgroup_record(ent, 0); |
3158 | rcu_read_lock(); | |
3159 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3160 | if (memcg) { |
a3b2d692 KH |
3161 | /* |
3162 | * We uncharge this because swap is freed. | |
3163 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3164 | */ | |
0c3e73e8 | 3165 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3166 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3167 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3168 | mem_cgroup_put(memcg); |
3169 | } | |
a3b2d692 | 3170 | rcu_read_unlock(); |
d13d1443 | 3171 | } |
02491447 DN |
3172 | |
3173 | /** | |
3174 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3175 | * @entry: swap entry to be moved | |
3176 | * @from: mem_cgroup which the entry is moved from | |
3177 | * @to: mem_cgroup which the entry is moved to | |
3178 | * | |
3179 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3180 | * as the mem_cgroup's id of @from. | |
3181 | * | |
3182 | * Returns 0 on success, -EINVAL on failure. | |
3183 | * | |
3184 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3185 | * both res and memsw, and called css_get(). | |
3186 | */ | |
3187 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3188 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3189 | { |
3190 | unsigned short old_id, new_id; | |
3191 | ||
3192 | old_id = css_id(&from->css); | |
3193 | new_id = css_id(&to->css); | |
3194 | ||
3195 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3196 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3197 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3198 | /* |
483c30b5 DN |
3199 | * This function is only called from task migration context now. |
3200 | * It postpones res_counter and refcount handling till the end | |
3201 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3202 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3203 | * because if the process that has been moved to @to does | |
3204 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3205 | */ |
02491447 | 3206 | mem_cgroup_get(to); |
02491447 DN |
3207 | return 0; |
3208 | } | |
3209 | return -EINVAL; | |
3210 | } | |
3211 | #else | |
3212 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3213 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3214 | { |
3215 | return -EINVAL; | |
3216 | } | |
8c7c6e34 | 3217 | #endif |
d13d1443 | 3218 | |
ae41be37 | 3219 | /* |
01b1ae63 KH |
3220 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3221 | * page belongs to. | |
ae41be37 | 3222 | */ |
ac39cf8c | 3223 | int mem_cgroup_prepare_migration(struct page *page, |
72835c86 | 3224 | struct page *newpage, struct mem_cgroup **memcgp, gfp_t gfp_mask) |
ae41be37 | 3225 | { |
c0ff4b85 | 3226 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3227 | struct page_cgroup *pc; |
ac39cf8c | 3228 | enum charge_type ctype; |
e8589cc1 | 3229 | int ret = 0; |
8869b8f6 | 3230 | |
72835c86 | 3231 | *memcgp = NULL; |
56039efa | 3232 | |
ec168510 | 3233 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3234 | if (mem_cgroup_disabled()) |
4077960e BS |
3235 | return 0; |
3236 | ||
52d4b9ac KH |
3237 | pc = lookup_page_cgroup(page); |
3238 | lock_page_cgroup(pc); | |
3239 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3240 | memcg = pc->mem_cgroup; |
3241 | css_get(&memcg->css); | |
ac39cf8c | 3242 | /* |
3243 | * At migrating an anonymous page, its mapcount goes down | |
3244 | * to 0 and uncharge() will be called. But, even if it's fully | |
3245 | * unmapped, migration may fail and this page has to be | |
3246 | * charged again. We set MIGRATION flag here and delay uncharge | |
3247 | * until end_migration() is called | |
3248 | * | |
3249 | * Corner Case Thinking | |
3250 | * A) | |
3251 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3252 | * while migration was ongoing. | |
3253 | * If unmap finds the old page, uncharge() of it will be delayed | |
3254 | * until end_migration(). If unmap finds a new page, it's | |
3255 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3256 | * finds swap_migration_entry, the new page will not be mapped | |
3257 | * and end_migration() will find it(mapcount==0). | |
3258 | * | |
3259 | * B) | |
3260 | * When the old page was mapped but migraion fails, the kernel | |
3261 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3262 | * if mapcount goes down to 0. We can do remap successfully | |
3263 | * without charging it again. | |
3264 | * | |
3265 | * C) | |
3266 | * The "old" page is under lock_page() until the end of | |
3267 | * migration, so, the old page itself will not be swapped-out. | |
3268 | * If the new page is swapped out before end_migraton, our | |
3269 | * hook to usual swap-out path will catch the event. | |
3270 | */ | |
3271 | if (PageAnon(page)) | |
3272 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3273 | } |
52d4b9ac | 3274 | unlock_page_cgroup(pc); |
ac39cf8c | 3275 | /* |
3276 | * If the page is not charged at this point, | |
3277 | * we return here. | |
3278 | */ | |
c0ff4b85 | 3279 | if (!memcg) |
ac39cf8c | 3280 | return 0; |
01b1ae63 | 3281 | |
72835c86 JW |
3282 | *memcgp = memcg; |
3283 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, memcgp, false); | |
c0ff4b85 | 3284 | css_put(&memcg->css);/* drop extra refcnt */ |
38c5d72f | 3285 | if (ret) { |
ac39cf8c | 3286 | if (PageAnon(page)) { |
3287 | lock_page_cgroup(pc); | |
3288 | ClearPageCgroupMigration(pc); | |
3289 | unlock_page_cgroup(pc); | |
3290 | /* | |
3291 | * The old page may be fully unmapped while we kept it. | |
3292 | */ | |
3293 | mem_cgroup_uncharge_page(page); | |
3294 | } | |
38c5d72f | 3295 | /* we'll need to revisit this error code (we have -EINTR) */ |
ac39cf8c | 3296 | return -ENOMEM; |
e8589cc1 | 3297 | } |
ac39cf8c | 3298 | /* |
3299 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3300 | * is called before end_migration, we can catch all events on this new | |
3301 | * page. In the case new page is migrated but not remapped, new page's | |
3302 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3303 | */ | |
ac39cf8c | 3304 | if (PageAnon(page)) |
41326c17 | 3305 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 3306 | else if (page_is_file_cache(page)) |
3307 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3308 | else | |
3309 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
ce587e65 | 3310 | __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false); |
e8589cc1 | 3311 | return ret; |
ae41be37 | 3312 | } |
8869b8f6 | 3313 | |
69029cd5 | 3314 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3315 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3316 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3317 | { |
ac39cf8c | 3318 | struct page *used, *unused; |
01b1ae63 | 3319 | struct page_cgroup *pc; |
b2402857 | 3320 | bool anon; |
01b1ae63 | 3321 | |
c0ff4b85 | 3322 | if (!memcg) |
01b1ae63 | 3323 | return; |
ac39cf8c | 3324 | /* blocks rmdir() */ |
c0ff4b85 | 3325 | cgroup_exclude_rmdir(&memcg->css); |
50de1dd9 | 3326 | if (!migration_ok) { |
ac39cf8c | 3327 | used = oldpage; |
3328 | unused = newpage; | |
01b1ae63 | 3329 | } else { |
ac39cf8c | 3330 | used = newpage; |
01b1ae63 KH |
3331 | unused = oldpage; |
3332 | } | |
69029cd5 | 3333 | /* |
ac39cf8c | 3334 | * We disallowed uncharge of pages under migration because mapcount |
3335 | * of the page goes down to zero, temporarly. | |
3336 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3337 | */ |
ac39cf8c | 3338 | pc = lookup_page_cgroup(oldpage); |
3339 | lock_page_cgroup(pc); | |
3340 | ClearPageCgroupMigration(pc); | |
3341 | unlock_page_cgroup(pc); | |
b2402857 KH |
3342 | anon = PageAnon(used); |
3343 | __mem_cgroup_uncharge_common(unused, | |
41326c17 | 3344 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON |
b2402857 | 3345 | : MEM_CGROUP_CHARGE_TYPE_CACHE); |
ac39cf8c | 3346 | |
01b1ae63 | 3347 | /* |
ac39cf8c | 3348 | * If a page is a file cache, radix-tree replacement is very atomic |
3349 | * and we can skip this check. When it was an Anon page, its mapcount | |
3350 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3351 | * uncharged yet. There are several case but page->mapcount check | |
3352 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3353 | * check. (see prepare_charge() also) | |
69029cd5 | 3354 | */ |
b2402857 | 3355 | if (anon) |
ac39cf8c | 3356 | mem_cgroup_uncharge_page(used); |
88703267 | 3357 | /* |
ac39cf8c | 3358 | * At migration, we may charge account against cgroup which has no |
3359 | * tasks. | |
88703267 KH |
3360 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3361 | * In that case, we need to call pre_destroy() again. check it here. | |
3362 | */ | |
c0ff4b85 | 3363 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
ae41be37 | 3364 | } |
78fb7466 | 3365 | |
ab936cbc KH |
3366 | /* |
3367 | * At replace page cache, newpage is not under any memcg but it's on | |
3368 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
3369 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
3370 | */ | |
3371 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
3372 | struct page *newpage) | |
3373 | { | |
bde05d1c | 3374 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 3375 | struct page_cgroup *pc; |
ab936cbc | 3376 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
3377 | |
3378 | if (mem_cgroup_disabled()) | |
3379 | return; | |
3380 | ||
3381 | pc = lookup_page_cgroup(oldpage); | |
3382 | /* fix accounting on old pages */ | |
3383 | lock_page_cgroup(pc); | |
bde05d1c HD |
3384 | if (PageCgroupUsed(pc)) { |
3385 | memcg = pc->mem_cgroup; | |
3386 | mem_cgroup_charge_statistics(memcg, false, -1); | |
3387 | ClearPageCgroupUsed(pc); | |
3388 | } | |
ab936cbc KH |
3389 | unlock_page_cgroup(pc); |
3390 | ||
bde05d1c HD |
3391 | /* |
3392 | * When called from shmem_replace_page(), in some cases the | |
3393 | * oldpage has already been charged, and in some cases not. | |
3394 | */ | |
3395 | if (!memcg) | |
3396 | return; | |
3397 | ||
ab936cbc KH |
3398 | if (PageSwapBacked(oldpage)) |
3399 | type = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
3400 | ||
ab936cbc KH |
3401 | /* |
3402 | * Even if newpage->mapping was NULL before starting replacement, | |
3403 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
3404 | * LRU while we overwrite pc->mem_cgroup. | |
3405 | */ | |
ce587e65 | 3406 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
3407 | } |
3408 | ||
f212ad7c DN |
3409 | #ifdef CONFIG_DEBUG_VM |
3410 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3411 | { | |
3412 | struct page_cgroup *pc; | |
3413 | ||
3414 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3415 | /* |
3416 | * Can be NULL while feeding pages into the page allocator for | |
3417 | * the first time, i.e. during boot or memory hotplug; | |
3418 | * or when mem_cgroup_disabled(). | |
3419 | */ | |
f212ad7c DN |
3420 | if (likely(pc) && PageCgroupUsed(pc)) |
3421 | return pc; | |
3422 | return NULL; | |
3423 | } | |
3424 | ||
3425 | bool mem_cgroup_bad_page_check(struct page *page) | |
3426 | { | |
3427 | if (mem_cgroup_disabled()) | |
3428 | return false; | |
3429 | ||
3430 | return lookup_page_cgroup_used(page) != NULL; | |
3431 | } | |
3432 | ||
3433 | void mem_cgroup_print_bad_page(struct page *page) | |
3434 | { | |
3435 | struct page_cgroup *pc; | |
3436 | ||
3437 | pc = lookup_page_cgroup_used(page); | |
3438 | if (pc) { | |
90b3feae | 3439 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
f212ad7c | 3440 | pc, pc->flags, pc->mem_cgroup); |
f212ad7c DN |
3441 | } |
3442 | } | |
3443 | #endif | |
3444 | ||
8c7c6e34 KH |
3445 | static DEFINE_MUTEX(set_limit_mutex); |
3446 | ||
d38d2a75 | 3447 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3448 | unsigned long long val) |
628f4235 | 3449 | { |
81d39c20 | 3450 | int retry_count; |
3c11ecf4 | 3451 | u64 memswlimit, memlimit; |
628f4235 | 3452 | int ret = 0; |
81d39c20 KH |
3453 | int children = mem_cgroup_count_children(memcg); |
3454 | u64 curusage, oldusage; | |
3c11ecf4 | 3455 | int enlarge; |
81d39c20 KH |
3456 | |
3457 | /* | |
3458 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3459 | * is depends on callers. We set our retry-count to be function | |
3460 | * of # of children which we should visit in this loop. | |
3461 | */ | |
3462 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3463 | ||
3464 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3465 | |
3c11ecf4 | 3466 | enlarge = 0; |
8c7c6e34 | 3467 | while (retry_count) { |
628f4235 KH |
3468 | if (signal_pending(current)) { |
3469 | ret = -EINTR; | |
3470 | break; | |
3471 | } | |
8c7c6e34 KH |
3472 | /* |
3473 | * Rather than hide all in some function, I do this in | |
3474 | * open coded manner. You see what this really does. | |
aaad153e | 3475 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3476 | */ |
3477 | mutex_lock(&set_limit_mutex); | |
3478 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3479 | if (memswlimit < val) { | |
3480 | ret = -EINVAL; | |
3481 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3482 | break; |
3483 | } | |
3c11ecf4 KH |
3484 | |
3485 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3486 | if (memlimit < val) | |
3487 | enlarge = 1; | |
3488 | ||
8c7c6e34 | 3489 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3490 | if (!ret) { |
3491 | if (memswlimit == val) | |
3492 | memcg->memsw_is_minimum = true; | |
3493 | else | |
3494 | memcg->memsw_is_minimum = false; | |
3495 | } | |
8c7c6e34 KH |
3496 | mutex_unlock(&set_limit_mutex); |
3497 | ||
3498 | if (!ret) | |
3499 | break; | |
3500 | ||
5660048c JW |
3501 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3502 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3503 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3504 | /* Usage is reduced ? */ | |
3505 | if (curusage >= oldusage) | |
3506 | retry_count--; | |
3507 | else | |
3508 | oldusage = curusage; | |
8c7c6e34 | 3509 | } |
3c11ecf4 KH |
3510 | if (!ret && enlarge) |
3511 | memcg_oom_recover(memcg); | |
14797e23 | 3512 | |
8c7c6e34 KH |
3513 | return ret; |
3514 | } | |
3515 | ||
338c8431 LZ |
3516 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3517 | unsigned long long val) | |
8c7c6e34 | 3518 | { |
81d39c20 | 3519 | int retry_count; |
3c11ecf4 | 3520 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3521 | int children = mem_cgroup_count_children(memcg); |
3522 | int ret = -EBUSY; | |
3c11ecf4 | 3523 | int enlarge = 0; |
8c7c6e34 | 3524 | |
81d39c20 KH |
3525 | /* see mem_cgroup_resize_res_limit */ |
3526 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3527 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3528 | while (retry_count) { |
3529 | if (signal_pending(current)) { | |
3530 | ret = -EINTR; | |
3531 | break; | |
3532 | } | |
3533 | /* | |
3534 | * Rather than hide all in some function, I do this in | |
3535 | * open coded manner. You see what this really does. | |
aaad153e | 3536 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3537 | */ |
3538 | mutex_lock(&set_limit_mutex); | |
3539 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3540 | if (memlimit > val) { | |
3541 | ret = -EINVAL; | |
3542 | mutex_unlock(&set_limit_mutex); | |
3543 | break; | |
3544 | } | |
3c11ecf4 KH |
3545 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3546 | if (memswlimit < val) | |
3547 | enlarge = 1; | |
8c7c6e34 | 3548 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3549 | if (!ret) { |
3550 | if (memlimit == val) | |
3551 | memcg->memsw_is_minimum = true; | |
3552 | else | |
3553 | memcg->memsw_is_minimum = false; | |
3554 | } | |
8c7c6e34 KH |
3555 | mutex_unlock(&set_limit_mutex); |
3556 | ||
3557 | if (!ret) | |
3558 | break; | |
3559 | ||
5660048c JW |
3560 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3561 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3562 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3563 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3564 | /* Usage is reduced ? */ |
8c7c6e34 | 3565 | if (curusage >= oldusage) |
628f4235 | 3566 | retry_count--; |
81d39c20 KH |
3567 | else |
3568 | oldusage = curusage; | |
628f4235 | 3569 | } |
3c11ecf4 KH |
3570 | if (!ret && enlarge) |
3571 | memcg_oom_recover(memcg); | |
628f4235 KH |
3572 | return ret; |
3573 | } | |
3574 | ||
4e416953 | 3575 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3576 | gfp_t gfp_mask, |
3577 | unsigned long *total_scanned) | |
4e416953 BS |
3578 | { |
3579 | unsigned long nr_reclaimed = 0; | |
3580 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3581 | unsigned long reclaimed; | |
3582 | int loop = 0; | |
3583 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3584 | unsigned long long excess; |
0ae5e89c | 3585 | unsigned long nr_scanned; |
4e416953 BS |
3586 | |
3587 | if (order > 0) | |
3588 | return 0; | |
3589 | ||
00918b6a | 3590 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3591 | /* |
3592 | * This loop can run a while, specially if mem_cgroup's continuously | |
3593 | * keep exceeding their soft limit and putting the system under | |
3594 | * pressure | |
3595 | */ | |
3596 | do { | |
3597 | if (next_mz) | |
3598 | mz = next_mz; | |
3599 | else | |
3600 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3601 | if (!mz) | |
3602 | break; | |
3603 | ||
0ae5e89c | 3604 | nr_scanned = 0; |
d79154bb | 3605 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 3606 | gfp_mask, &nr_scanned); |
4e416953 | 3607 | nr_reclaimed += reclaimed; |
0ae5e89c | 3608 | *total_scanned += nr_scanned; |
4e416953 BS |
3609 | spin_lock(&mctz->lock); |
3610 | ||
3611 | /* | |
3612 | * If we failed to reclaim anything from this memory cgroup | |
3613 | * it is time to move on to the next cgroup | |
3614 | */ | |
3615 | next_mz = NULL; | |
3616 | if (!reclaimed) { | |
3617 | do { | |
3618 | /* | |
3619 | * Loop until we find yet another one. | |
3620 | * | |
3621 | * By the time we get the soft_limit lock | |
3622 | * again, someone might have aded the | |
3623 | * group back on the RB tree. Iterate to | |
3624 | * make sure we get a different mem. | |
3625 | * mem_cgroup_largest_soft_limit_node returns | |
3626 | * NULL if no other cgroup is present on | |
3627 | * the tree | |
3628 | */ | |
3629 | next_mz = | |
3630 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3631 | if (next_mz == mz) |
d79154bb | 3632 | css_put(&next_mz->memcg->css); |
39cc98f1 | 3633 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3634 | break; |
3635 | } while (1); | |
3636 | } | |
d79154bb HD |
3637 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
3638 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
3639 | /* |
3640 | * One school of thought says that we should not add | |
3641 | * back the node to the tree if reclaim returns 0. | |
3642 | * But our reclaim could return 0, simply because due | |
3643 | * to priority we are exposing a smaller subset of | |
3644 | * memory to reclaim from. Consider this as a longer | |
3645 | * term TODO. | |
3646 | */ | |
ef8745c1 | 3647 | /* If excess == 0, no tree ops */ |
d79154bb | 3648 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 3649 | spin_unlock(&mctz->lock); |
d79154bb | 3650 | css_put(&mz->memcg->css); |
4e416953 BS |
3651 | loop++; |
3652 | /* | |
3653 | * Could not reclaim anything and there are no more | |
3654 | * mem cgroups to try or we seem to be looping without | |
3655 | * reclaiming anything. | |
3656 | */ | |
3657 | if (!nr_reclaimed && | |
3658 | (next_mz == NULL || | |
3659 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3660 | break; | |
3661 | } while (!nr_reclaimed); | |
3662 | if (next_mz) | |
d79154bb | 3663 | css_put(&next_mz->memcg->css); |
4e416953 BS |
3664 | return nr_reclaimed; |
3665 | } | |
3666 | ||
cc847582 | 3667 | /* |
3c935d18 KH |
3668 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
3669 | * reclaim the pages page themselves - it just removes the page_cgroups. | |
3670 | * Returns true if some page_cgroups were not freed, indicating that the caller | |
3671 | * must retry this operation. | |
cc847582 | 3672 | */ |
3c935d18 | 3673 | static bool mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3674 | int node, int zid, enum lru_list lru) |
cc847582 | 3675 | { |
08e552c6 | 3676 | struct mem_cgroup_per_zone *mz; |
08e552c6 | 3677 | unsigned long flags, loop; |
072c56c1 | 3678 | struct list_head *list; |
925b7673 JW |
3679 | struct page *busy; |
3680 | struct zone *zone; | |
072c56c1 | 3681 | |
08e552c6 | 3682 | zone = &NODE_DATA(node)->node_zones[zid]; |
c0ff4b85 | 3683 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
6290df54 | 3684 | list = &mz->lruvec.lists[lru]; |
cc847582 | 3685 | |
1eb49272 | 3686 | loop = mz->lru_size[lru]; |
f817ed48 KH |
3687 | /* give some margin against EBUSY etc...*/ |
3688 | loop += 256; | |
3689 | busy = NULL; | |
3690 | while (loop--) { | |
925b7673 | 3691 | struct page_cgroup *pc; |
5564e88b JW |
3692 | struct page *page; |
3693 | ||
08e552c6 | 3694 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3695 | if (list_empty(list)) { |
08e552c6 | 3696 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3697 | break; |
f817ed48 | 3698 | } |
925b7673 JW |
3699 | page = list_entry(list->prev, struct page, lru); |
3700 | if (busy == page) { | |
3701 | list_move(&page->lru, list); | |
648bcc77 | 3702 | busy = NULL; |
08e552c6 | 3703 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3704 | continue; |
3705 | } | |
08e552c6 | 3706 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3707 | |
925b7673 | 3708 | pc = lookup_page_cgroup(page); |
5564e88b | 3709 | |
3c935d18 | 3710 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 3711 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 3712 | busy = page; |
f817ed48 KH |
3713 | cond_resched(); |
3714 | } else | |
3715 | busy = NULL; | |
cc847582 | 3716 | } |
3c935d18 | 3717 | return !list_empty(list); |
cc847582 KH |
3718 | } |
3719 | ||
3720 | /* | |
3721 | * make mem_cgroup's charge to be 0 if there is no task. | |
3722 | * This enables deleting this mem_cgroup. | |
3723 | */ | |
c0ff4b85 | 3724 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all) |
cc847582 | 3725 | { |
f817ed48 KH |
3726 | int ret; |
3727 | int node, zid, shrink; | |
3728 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c0ff4b85 | 3729 | struct cgroup *cgrp = memcg->css.cgroup; |
8869b8f6 | 3730 | |
c0ff4b85 | 3731 | css_get(&memcg->css); |
f817ed48 KH |
3732 | |
3733 | shrink = 0; | |
c1e862c1 KH |
3734 | /* should free all ? */ |
3735 | if (free_all) | |
3736 | goto try_to_free; | |
f817ed48 | 3737 | move_account: |
fce66477 | 3738 | do { |
f817ed48 | 3739 | ret = -EBUSY; |
c1e862c1 KH |
3740 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3741 | goto out; | |
52d4b9ac KH |
3742 | /* This is for making all *used* pages to be on LRU. */ |
3743 | lru_add_drain_all(); | |
c0ff4b85 | 3744 | drain_all_stock_sync(memcg); |
f817ed48 | 3745 | ret = 0; |
c0ff4b85 | 3746 | mem_cgroup_start_move(memcg); |
299b4eaa | 3747 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3748 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3749 | enum lru_list lru; |
3750 | for_each_lru(lru) { | |
c0ff4b85 | 3751 | ret = mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3752 | node, zid, lru); |
f817ed48 KH |
3753 | if (ret) |
3754 | break; | |
3755 | } | |
1ecaab2b | 3756 | } |
f817ed48 KH |
3757 | if (ret) |
3758 | break; | |
3759 | } | |
c0ff4b85 R |
3760 | mem_cgroup_end_move(memcg); |
3761 | memcg_oom_recover(memcg); | |
52d4b9ac | 3762 | cond_resched(); |
fce66477 | 3763 | /* "ret" should also be checked to ensure all lists are empty. */ |
569530fb | 3764 | } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret); |
cc847582 | 3765 | out: |
c0ff4b85 | 3766 | css_put(&memcg->css); |
cc847582 | 3767 | return ret; |
f817ed48 KH |
3768 | |
3769 | try_to_free: | |
c1e862c1 KH |
3770 | /* returns EBUSY if there is a task or if we come here twice. */ |
3771 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3772 | ret = -EBUSY; |
3773 | goto out; | |
3774 | } | |
c1e862c1 KH |
3775 | /* we call try-to-free pages for make this cgroup empty */ |
3776 | lru_add_drain_all(); | |
f817ed48 KH |
3777 | /* try to free all pages in this cgroup */ |
3778 | shrink = 1; | |
569530fb | 3779 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 3780 | int progress; |
c1e862c1 KH |
3781 | |
3782 | if (signal_pending(current)) { | |
3783 | ret = -EINTR; | |
3784 | goto out; | |
3785 | } | |
c0ff4b85 | 3786 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3787 | false); |
c1e862c1 | 3788 | if (!progress) { |
f817ed48 | 3789 | nr_retries--; |
c1e862c1 | 3790 | /* maybe some writeback is necessary */ |
8aa7e847 | 3791 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3792 | } |
f817ed48 KH |
3793 | |
3794 | } | |
08e552c6 | 3795 | lru_add_drain(); |
f817ed48 | 3796 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3797 | goto move_account; |
cc847582 KH |
3798 | } |
3799 | ||
6bbda35c | 3800 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 KH |
3801 | { |
3802 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3803 | } | |
3804 | ||
3805 | ||
18f59ea7 BS |
3806 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3807 | { | |
3808 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3809 | } | |
3810 | ||
3811 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3812 | u64 val) | |
3813 | { | |
3814 | int retval = 0; | |
c0ff4b85 | 3815 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3816 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3817 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3818 | |
3819 | if (parent) | |
c0ff4b85 | 3820 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3821 | |
3822 | cgroup_lock(); | |
567fb435 GC |
3823 | |
3824 | if (memcg->use_hierarchy == val) | |
3825 | goto out; | |
3826 | ||
18f59ea7 | 3827 | /* |
af901ca1 | 3828 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3829 | * in the child subtrees. If it is unset, then the change can |
3830 | * occur, provided the current cgroup has no children. | |
3831 | * | |
3832 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3833 | * set if there are no children. | |
3834 | */ | |
c0ff4b85 | 3835 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3836 | (val == 1 || val == 0)) { |
3837 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3838 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3839 | else |
3840 | retval = -EBUSY; | |
3841 | } else | |
3842 | retval = -EINVAL; | |
567fb435 GC |
3843 | |
3844 | out: | |
18f59ea7 BS |
3845 | cgroup_unlock(); |
3846 | ||
3847 | return retval; | |
3848 | } | |
3849 | ||
0c3e73e8 | 3850 | |
c0ff4b85 | 3851 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3852 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3853 | { |
7d74b06f | 3854 | struct mem_cgroup *iter; |
7a159cc9 | 3855 | long val = 0; |
0c3e73e8 | 3856 | |
7a159cc9 | 3857 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3858 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3859 | val += mem_cgroup_read_stat(iter, idx); |
3860 | ||
3861 | if (val < 0) /* race ? */ | |
3862 | val = 0; | |
3863 | return val; | |
0c3e73e8 BS |
3864 | } |
3865 | ||
c0ff4b85 | 3866 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3867 | { |
7d74b06f | 3868 | u64 val; |
104f3928 | 3869 | |
c0ff4b85 | 3870 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 3871 | if (!swap) |
65c64ce8 | 3872 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3873 | else |
65c64ce8 | 3874 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
3875 | } |
3876 | ||
c0ff4b85 R |
3877 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3878 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3879 | |
7d74b06f | 3880 | if (swap) |
bff6bb83 | 3881 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
3882 | |
3883 | return val << PAGE_SHIFT; | |
3884 | } | |
3885 | ||
af36f906 TH |
3886 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
3887 | struct file *file, char __user *buf, | |
3888 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 3889 | { |
c0ff4b85 | 3890 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 3891 | char str[64]; |
104f3928 | 3892 | u64 val; |
af36f906 | 3893 | int type, name, len; |
8c7c6e34 KH |
3894 | |
3895 | type = MEMFILE_TYPE(cft->private); | |
3896 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3897 | |
3898 | if (!do_swap_account && type == _MEMSWAP) | |
3899 | return -EOPNOTSUPP; | |
3900 | ||
8c7c6e34 KH |
3901 | switch (type) { |
3902 | case _MEM: | |
104f3928 | 3903 | if (name == RES_USAGE) |
c0ff4b85 | 3904 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3905 | else |
c0ff4b85 | 3906 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3907 | break; |
3908 | case _MEMSWAP: | |
104f3928 | 3909 | if (name == RES_USAGE) |
c0ff4b85 | 3910 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 3911 | else |
c0ff4b85 | 3912 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 KH |
3913 | break; |
3914 | default: | |
3915 | BUG(); | |
8c7c6e34 | 3916 | } |
af36f906 TH |
3917 | |
3918 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
3919 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 3920 | } |
628f4235 KH |
3921 | /* |
3922 | * The user of this function is... | |
3923 | * RES_LIMIT. | |
3924 | */ | |
856c13aa PM |
3925 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3926 | const char *buffer) | |
8cdea7c0 | 3927 | { |
628f4235 | 3928 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3929 | int type, name; |
628f4235 KH |
3930 | unsigned long long val; |
3931 | int ret; | |
3932 | ||
8c7c6e34 KH |
3933 | type = MEMFILE_TYPE(cft->private); |
3934 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3935 | |
3936 | if (!do_swap_account && type == _MEMSWAP) | |
3937 | return -EOPNOTSUPP; | |
3938 | ||
8c7c6e34 | 3939 | switch (name) { |
628f4235 | 3940 | case RES_LIMIT: |
4b3bde4c BS |
3941 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3942 | ret = -EINVAL; | |
3943 | break; | |
3944 | } | |
628f4235 KH |
3945 | /* This function does all necessary parse...reuse it */ |
3946 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3947 | if (ret) |
3948 | break; | |
3949 | if (type == _MEM) | |
628f4235 | 3950 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3951 | else |
3952 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3953 | break; |
296c81d8 BS |
3954 | case RES_SOFT_LIMIT: |
3955 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3956 | if (ret) | |
3957 | break; | |
3958 | /* | |
3959 | * For memsw, soft limits are hard to implement in terms | |
3960 | * of semantics, for now, we support soft limits for | |
3961 | * control without swap | |
3962 | */ | |
3963 | if (type == _MEM) | |
3964 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3965 | else | |
3966 | ret = -EINVAL; | |
3967 | break; | |
628f4235 KH |
3968 | default: |
3969 | ret = -EINVAL; /* should be BUG() ? */ | |
3970 | break; | |
3971 | } | |
3972 | return ret; | |
8cdea7c0 BS |
3973 | } |
3974 | ||
fee7b548 KH |
3975 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3976 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3977 | { | |
3978 | struct cgroup *cgroup; | |
3979 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3980 | ||
3981 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3982 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3983 | cgroup = memcg->css.cgroup; | |
3984 | if (!memcg->use_hierarchy) | |
3985 | goto out; | |
3986 | ||
3987 | while (cgroup->parent) { | |
3988 | cgroup = cgroup->parent; | |
3989 | memcg = mem_cgroup_from_cont(cgroup); | |
3990 | if (!memcg->use_hierarchy) | |
3991 | break; | |
3992 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3993 | min_limit = min(min_limit, tmp); | |
3994 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3995 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3996 | } | |
3997 | out: | |
3998 | *mem_limit = min_limit; | |
3999 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
4000 | } |
4001 | ||
29f2a4da | 4002 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 4003 | { |
af36f906 | 4004 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 4005 | int type, name; |
c84872e1 | 4006 | |
8c7c6e34 KH |
4007 | type = MEMFILE_TYPE(event); |
4008 | name = MEMFILE_ATTR(event); | |
af36f906 TH |
4009 | |
4010 | if (!do_swap_account && type == _MEMSWAP) | |
4011 | return -EOPNOTSUPP; | |
4012 | ||
8c7c6e34 | 4013 | switch (name) { |
29f2a4da | 4014 | case RES_MAX_USAGE: |
8c7c6e34 | 4015 | if (type == _MEM) |
c0ff4b85 | 4016 | res_counter_reset_max(&memcg->res); |
8c7c6e34 | 4017 | else |
c0ff4b85 | 4018 | res_counter_reset_max(&memcg->memsw); |
29f2a4da PE |
4019 | break; |
4020 | case RES_FAILCNT: | |
8c7c6e34 | 4021 | if (type == _MEM) |
c0ff4b85 | 4022 | res_counter_reset_failcnt(&memcg->res); |
8c7c6e34 | 4023 | else |
c0ff4b85 | 4024 | res_counter_reset_failcnt(&memcg->memsw); |
29f2a4da PE |
4025 | break; |
4026 | } | |
f64c3f54 | 4027 | |
85cc59db | 4028 | return 0; |
c84872e1 PE |
4029 | } |
4030 | ||
7dc74be0 DN |
4031 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
4032 | struct cftype *cft) | |
4033 | { | |
4034 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
4035 | } | |
4036 | ||
02491447 | 4037 | #ifdef CONFIG_MMU |
7dc74be0 DN |
4038 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
4039 | struct cftype *cft, u64 val) | |
4040 | { | |
c0ff4b85 | 4041 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
4042 | |
4043 | if (val >= (1 << NR_MOVE_TYPE)) | |
4044 | return -EINVAL; | |
4045 | /* | |
4046 | * We check this value several times in both in can_attach() and | |
4047 | * attach(), so we need cgroup lock to prevent this value from being | |
4048 | * inconsistent. | |
4049 | */ | |
4050 | cgroup_lock(); | |
c0ff4b85 | 4051 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4052 | cgroup_unlock(); |
4053 | ||
4054 | return 0; | |
4055 | } | |
02491447 DN |
4056 | #else |
4057 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4058 | struct cftype *cft, u64 val) | |
4059 | { | |
4060 | return -ENOSYS; | |
4061 | } | |
4062 | #endif | |
7dc74be0 | 4063 | |
406eb0c9 | 4064 | #ifdef CONFIG_NUMA |
ab215884 | 4065 | static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft, |
fada52ca | 4066 | struct seq_file *m) |
406eb0c9 YH |
4067 | { |
4068 | int nid; | |
4069 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4070 | unsigned long node_nr; | |
d79154bb | 4071 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 4072 | |
d79154bb | 4073 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 YH |
4074 | seq_printf(m, "total=%lu", total_nr); |
4075 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4076 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
4077 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4078 | } | |
4079 | seq_putc(m, '\n'); | |
4080 | ||
d79154bb | 4081 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 YH |
4082 | seq_printf(m, "file=%lu", file_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_FILE); |
406eb0c9 YH |
4086 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4087 | } | |
4088 | seq_putc(m, '\n'); | |
4089 | ||
d79154bb | 4090 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 YH |
4091 | seq_printf(m, "anon=%lu", anon_nr); |
4092 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4093 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4094 | LRU_ALL_ANON); |
406eb0c9 YH |
4095 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4096 | } | |
4097 | seq_putc(m, '\n'); | |
4098 | ||
d79154bb | 4099 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4100 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4101 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4102 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4103 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4104 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4105 | } | |
4106 | seq_putc(m, '\n'); | |
4107 | return 0; | |
4108 | } | |
4109 | #endif /* CONFIG_NUMA */ | |
4110 | ||
af7c4b0e JW |
4111 | static const char * const mem_cgroup_lru_names[] = { |
4112 | "inactive_anon", | |
4113 | "active_anon", | |
4114 | "inactive_file", | |
4115 | "active_file", | |
4116 | "unevictable", | |
4117 | }; | |
4118 | ||
4119 | static inline void mem_cgroup_lru_names_not_uptodate(void) | |
4120 | { | |
4121 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4122 | } | |
4123 | ||
ab215884 | 4124 | static int memcg_stat_show(struct cgroup *cont, struct cftype *cft, |
78ccf5b5 | 4125 | struct seq_file *m) |
d2ceb9b7 | 4126 | { |
d79154bb | 4127 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af7c4b0e JW |
4128 | struct mem_cgroup *mi; |
4129 | unsigned int i; | |
406eb0c9 | 4130 | |
af7c4b0e | 4131 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4132 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4133 | continue; |
af7c4b0e JW |
4134 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4135 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4136 | } |
7b854121 | 4137 | |
af7c4b0e JW |
4138 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4139 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4140 | mem_cgroup_read_events(memcg, i)); | |
4141 | ||
4142 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4143 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4144 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4145 | ||
14067bb3 | 4146 | /* Hierarchical information */ |
fee7b548 KH |
4147 | { |
4148 | unsigned long long limit, memsw_limit; | |
d79154bb | 4149 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 4150 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 4151 | if (do_swap_account) |
78ccf5b5 JW |
4152 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4153 | memsw_limit); | |
fee7b548 | 4154 | } |
7f016ee8 | 4155 | |
af7c4b0e JW |
4156 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4157 | long long val = 0; | |
4158 | ||
bff6bb83 | 4159 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4160 | continue; |
af7c4b0e JW |
4161 | for_each_mem_cgroup_tree(mi, memcg) |
4162 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4163 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4164 | } | |
4165 | ||
4166 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4167 | unsigned long long val = 0; | |
4168 | ||
4169 | for_each_mem_cgroup_tree(mi, memcg) | |
4170 | val += mem_cgroup_read_events(mi, i); | |
4171 | seq_printf(m, "total_%s %llu\n", | |
4172 | mem_cgroup_events_names[i], val); | |
4173 | } | |
4174 | ||
4175 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4176 | unsigned long long val = 0; | |
4177 | ||
4178 | for_each_mem_cgroup_tree(mi, memcg) | |
4179 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4180 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4181 | } |
14067bb3 | 4182 | |
7f016ee8 | 4183 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4184 | { |
4185 | int nid, zid; | |
4186 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4187 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4188 | unsigned long recent_rotated[2] = {0, 0}; |
4189 | unsigned long recent_scanned[2] = {0, 0}; | |
4190 | ||
4191 | for_each_online_node(nid) | |
4192 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 4193 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 4194 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4195 | |
89abfab1 HD |
4196 | recent_rotated[0] += rstat->recent_rotated[0]; |
4197 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4198 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4199 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4200 | } |
78ccf5b5 JW |
4201 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4202 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4203 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4204 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4205 | } |
4206 | #endif | |
4207 | ||
d2ceb9b7 KH |
4208 | return 0; |
4209 | } | |
4210 | ||
a7885eb8 KM |
4211 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4212 | { | |
4213 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4214 | ||
1f4c025b | 4215 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4216 | } |
4217 | ||
4218 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4219 | u64 val) | |
4220 | { | |
4221 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4222 | struct mem_cgroup *parent; | |
068b38c1 | 4223 | |
a7885eb8 KM |
4224 | if (val > 100) |
4225 | return -EINVAL; | |
4226 | ||
4227 | if (cgrp->parent == NULL) | |
4228 | return -EINVAL; | |
4229 | ||
4230 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4231 | |
4232 | cgroup_lock(); | |
4233 | ||
a7885eb8 KM |
4234 | /* If under hierarchy, only empty-root can set this value */ |
4235 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4236 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4237 | cgroup_unlock(); | |
a7885eb8 | 4238 | return -EINVAL; |
068b38c1 | 4239 | } |
a7885eb8 | 4240 | |
a7885eb8 | 4241 | memcg->swappiness = val; |
a7885eb8 | 4242 | |
068b38c1 LZ |
4243 | cgroup_unlock(); |
4244 | ||
a7885eb8 KM |
4245 | return 0; |
4246 | } | |
4247 | ||
2e72b634 KS |
4248 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4249 | { | |
4250 | struct mem_cgroup_threshold_ary *t; | |
4251 | u64 usage; | |
4252 | int i; | |
4253 | ||
4254 | rcu_read_lock(); | |
4255 | if (!swap) | |
2c488db2 | 4256 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4257 | else |
2c488db2 | 4258 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4259 | |
4260 | if (!t) | |
4261 | goto unlock; | |
4262 | ||
4263 | usage = mem_cgroup_usage(memcg, swap); | |
4264 | ||
4265 | /* | |
748dad36 | 4266 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4267 | * If it's not true, a threshold was crossed after last |
4268 | * call of __mem_cgroup_threshold(). | |
4269 | */ | |
5407a562 | 4270 | i = t->current_threshold; |
2e72b634 KS |
4271 | |
4272 | /* | |
4273 | * Iterate backward over array of thresholds starting from | |
4274 | * current_threshold and check if a threshold is crossed. | |
4275 | * If none of thresholds below usage is crossed, we read | |
4276 | * only one element of the array here. | |
4277 | */ | |
4278 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4279 | eventfd_signal(t->entries[i].eventfd, 1); | |
4280 | ||
4281 | /* i = current_threshold + 1 */ | |
4282 | i++; | |
4283 | ||
4284 | /* | |
4285 | * Iterate forward over array of thresholds starting from | |
4286 | * current_threshold+1 and check if a threshold is crossed. | |
4287 | * If none of thresholds above usage is crossed, we read | |
4288 | * only one element of the array here. | |
4289 | */ | |
4290 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4291 | eventfd_signal(t->entries[i].eventfd, 1); | |
4292 | ||
4293 | /* Update current_threshold */ | |
5407a562 | 4294 | t->current_threshold = i - 1; |
2e72b634 KS |
4295 | unlock: |
4296 | rcu_read_unlock(); | |
4297 | } | |
4298 | ||
4299 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4300 | { | |
ad4ca5f4 KS |
4301 | while (memcg) { |
4302 | __mem_cgroup_threshold(memcg, false); | |
4303 | if (do_swap_account) | |
4304 | __mem_cgroup_threshold(memcg, true); | |
4305 | ||
4306 | memcg = parent_mem_cgroup(memcg); | |
4307 | } | |
2e72b634 KS |
4308 | } |
4309 | ||
4310 | static int compare_thresholds(const void *a, const void *b) | |
4311 | { | |
4312 | const struct mem_cgroup_threshold *_a = a; | |
4313 | const struct mem_cgroup_threshold *_b = b; | |
4314 | ||
4315 | return _a->threshold - _b->threshold; | |
4316 | } | |
4317 | ||
c0ff4b85 | 4318 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4319 | { |
4320 | struct mem_cgroup_eventfd_list *ev; | |
4321 | ||
c0ff4b85 | 4322 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4323 | eventfd_signal(ev->eventfd, 1); |
4324 | return 0; | |
4325 | } | |
4326 | ||
c0ff4b85 | 4327 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4328 | { |
7d74b06f KH |
4329 | struct mem_cgroup *iter; |
4330 | ||
c0ff4b85 | 4331 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4332 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4333 | } |
4334 | ||
4335 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4336 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4337 | { |
4338 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4339 | struct mem_cgroup_thresholds *thresholds; |
4340 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4341 | int type = MEMFILE_TYPE(cft->private); |
4342 | u64 threshold, usage; | |
2c488db2 | 4343 | int i, size, ret; |
2e72b634 KS |
4344 | |
4345 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4346 | if (ret) | |
4347 | return ret; | |
4348 | ||
4349 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4350 | |
2e72b634 | 4351 | if (type == _MEM) |
2c488db2 | 4352 | thresholds = &memcg->thresholds; |
2e72b634 | 4353 | else if (type == _MEMSWAP) |
2c488db2 | 4354 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4355 | else |
4356 | BUG(); | |
4357 | ||
4358 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4359 | ||
4360 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4361 | if (thresholds->primary) |
2e72b634 KS |
4362 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4363 | ||
2c488db2 | 4364 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4365 | |
4366 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4367 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4368 | GFP_KERNEL); |
2c488db2 | 4369 | if (!new) { |
2e72b634 KS |
4370 | ret = -ENOMEM; |
4371 | goto unlock; | |
4372 | } | |
2c488db2 | 4373 | new->size = size; |
2e72b634 KS |
4374 | |
4375 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4376 | if (thresholds->primary) { |
4377 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4378 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4379 | } |
4380 | ||
2e72b634 | 4381 | /* Add new threshold */ |
2c488db2 KS |
4382 | new->entries[size - 1].eventfd = eventfd; |
4383 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4384 | |
4385 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4386 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4387 | compare_thresholds, NULL); |
4388 | ||
4389 | /* Find current threshold */ | |
2c488db2 | 4390 | new->current_threshold = -1; |
2e72b634 | 4391 | for (i = 0; i < size; i++) { |
748dad36 | 4392 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4393 | /* |
2c488db2 KS |
4394 | * new->current_threshold will not be used until |
4395 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4396 | * it here. |
4397 | */ | |
2c488db2 | 4398 | ++new->current_threshold; |
748dad36 SZ |
4399 | } else |
4400 | break; | |
2e72b634 KS |
4401 | } |
4402 | ||
2c488db2 KS |
4403 | /* Free old spare buffer and save old primary buffer as spare */ |
4404 | kfree(thresholds->spare); | |
4405 | thresholds->spare = thresholds->primary; | |
4406 | ||
4407 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4408 | |
907860ed | 4409 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4410 | synchronize_rcu(); |
4411 | ||
2e72b634 KS |
4412 | unlock: |
4413 | mutex_unlock(&memcg->thresholds_lock); | |
4414 | ||
4415 | return ret; | |
4416 | } | |
4417 | ||
907860ed | 4418 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4419 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4420 | { |
4421 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4422 | struct mem_cgroup_thresholds *thresholds; |
4423 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4424 | int type = MEMFILE_TYPE(cft->private); |
4425 | u64 usage; | |
2c488db2 | 4426 | int i, j, size; |
2e72b634 KS |
4427 | |
4428 | mutex_lock(&memcg->thresholds_lock); | |
4429 | if (type == _MEM) | |
2c488db2 | 4430 | thresholds = &memcg->thresholds; |
2e72b634 | 4431 | else if (type == _MEMSWAP) |
2c488db2 | 4432 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4433 | else |
4434 | BUG(); | |
4435 | ||
371528ca AV |
4436 | if (!thresholds->primary) |
4437 | goto unlock; | |
4438 | ||
2e72b634 KS |
4439 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
4440 | ||
4441 | /* Check if a threshold crossed before removing */ | |
4442 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4443 | ||
4444 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4445 | size = 0; |
4446 | for (i = 0; i < thresholds->primary->size; i++) { | |
4447 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4448 | size++; |
4449 | } | |
4450 | ||
2c488db2 | 4451 | new = thresholds->spare; |
907860ed | 4452 | |
2e72b634 KS |
4453 | /* Set thresholds array to NULL if we don't have thresholds */ |
4454 | if (!size) { | |
2c488db2 KS |
4455 | kfree(new); |
4456 | new = NULL; | |
907860ed | 4457 | goto swap_buffers; |
2e72b634 KS |
4458 | } |
4459 | ||
2c488db2 | 4460 | new->size = size; |
2e72b634 KS |
4461 | |
4462 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4463 | new->current_threshold = -1; |
4464 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4465 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4466 | continue; |
4467 | ||
2c488db2 | 4468 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4469 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4470 | /* |
2c488db2 | 4471 | * new->current_threshold will not be used |
2e72b634 KS |
4472 | * until rcu_assign_pointer(), so it's safe to increment |
4473 | * it here. | |
4474 | */ | |
2c488db2 | 4475 | ++new->current_threshold; |
2e72b634 KS |
4476 | } |
4477 | j++; | |
4478 | } | |
4479 | ||
907860ed | 4480 | swap_buffers: |
2c488db2 KS |
4481 | /* Swap primary and spare array */ |
4482 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4483 | /* If all events are unregistered, free the spare array */ |
4484 | if (!new) { | |
4485 | kfree(thresholds->spare); | |
4486 | thresholds->spare = NULL; | |
4487 | } | |
4488 | ||
2c488db2 | 4489 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4490 | |
907860ed | 4491 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4492 | synchronize_rcu(); |
371528ca | 4493 | unlock: |
2e72b634 | 4494 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4495 | } |
c1e862c1 | 4496 | |
9490ff27 KH |
4497 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4498 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4499 | { | |
4500 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4501 | struct mem_cgroup_eventfd_list *event; | |
4502 | int type = MEMFILE_TYPE(cft->private); | |
4503 | ||
4504 | BUG_ON(type != _OOM_TYPE); | |
4505 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4506 | if (!event) | |
4507 | return -ENOMEM; | |
4508 | ||
1af8efe9 | 4509 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4510 | |
4511 | event->eventfd = eventfd; | |
4512 | list_add(&event->list, &memcg->oom_notify); | |
4513 | ||
4514 | /* already in OOM ? */ | |
79dfdacc | 4515 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4516 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4517 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4518 | |
4519 | return 0; | |
4520 | } | |
4521 | ||
907860ed | 4522 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4523 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4524 | { | |
c0ff4b85 | 4525 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 KH |
4526 | struct mem_cgroup_eventfd_list *ev, *tmp; |
4527 | int type = MEMFILE_TYPE(cft->private); | |
4528 | ||
4529 | BUG_ON(type != _OOM_TYPE); | |
4530 | ||
1af8efe9 | 4531 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4532 | |
c0ff4b85 | 4533 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4534 | if (ev->eventfd == eventfd) { |
4535 | list_del(&ev->list); | |
4536 | kfree(ev); | |
4537 | } | |
4538 | } | |
4539 | ||
1af8efe9 | 4540 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4541 | } |
4542 | ||
3c11ecf4 KH |
4543 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4544 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4545 | { | |
c0ff4b85 | 4546 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4547 | |
c0ff4b85 | 4548 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4549 | |
c0ff4b85 | 4550 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4551 | cb->fill(cb, "under_oom", 1); |
4552 | else | |
4553 | cb->fill(cb, "under_oom", 0); | |
4554 | return 0; | |
4555 | } | |
4556 | ||
3c11ecf4 KH |
4557 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4558 | struct cftype *cft, u64 val) | |
4559 | { | |
c0ff4b85 | 4560 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4561 | struct mem_cgroup *parent; |
4562 | ||
4563 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4564 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4565 | return -EINVAL; | |
4566 | ||
4567 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4568 | ||
4569 | cgroup_lock(); | |
4570 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4571 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4572 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4573 | cgroup_unlock(); |
4574 | return -EINVAL; | |
4575 | } | |
c0ff4b85 | 4576 | memcg->oom_kill_disable = val; |
4d845ebf | 4577 | if (!val) |
c0ff4b85 | 4578 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4579 | cgroup_unlock(); |
4580 | return 0; | |
4581 | } | |
4582 | ||
c255a458 | 4583 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4584 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4585 | { |
1d62e436 | 4586 | return mem_cgroup_sockets_init(memcg, ss); |
e5671dfa GC |
4587 | }; |
4588 | ||
1d62e436 | 4589 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 | 4590 | { |
1d62e436 | 4591 | mem_cgroup_sockets_destroy(memcg); |
d1a4c0b3 | 4592 | } |
e5671dfa | 4593 | #else |
cbe128e3 | 4594 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4595 | { |
4596 | return 0; | |
4597 | } | |
d1a4c0b3 | 4598 | |
1d62e436 | 4599 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 GC |
4600 | { |
4601 | } | |
e5671dfa GC |
4602 | #endif |
4603 | ||
8cdea7c0 BS |
4604 | static struct cftype mem_cgroup_files[] = { |
4605 | { | |
0eea1030 | 4606 | .name = "usage_in_bytes", |
8c7c6e34 | 4607 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 4608 | .read = mem_cgroup_read, |
9490ff27 KH |
4609 | .register_event = mem_cgroup_usage_register_event, |
4610 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4611 | }, |
c84872e1 PE |
4612 | { |
4613 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4614 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4615 | .trigger = mem_cgroup_reset, |
af36f906 | 4616 | .read = mem_cgroup_read, |
c84872e1 | 4617 | }, |
8cdea7c0 | 4618 | { |
0eea1030 | 4619 | .name = "limit_in_bytes", |
8c7c6e34 | 4620 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4621 | .write_string = mem_cgroup_write, |
af36f906 | 4622 | .read = mem_cgroup_read, |
8cdea7c0 | 4623 | }, |
296c81d8 BS |
4624 | { |
4625 | .name = "soft_limit_in_bytes", | |
4626 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4627 | .write_string = mem_cgroup_write, | |
af36f906 | 4628 | .read = mem_cgroup_read, |
296c81d8 | 4629 | }, |
8cdea7c0 BS |
4630 | { |
4631 | .name = "failcnt", | |
8c7c6e34 | 4632 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4633 | .trigger = mem_cgroup_reset, |
af36f906 | 4634 | .read = mem_cgroup_read, |
8cdea7c0 | 4635 | }, |
d2ceb9b7 KH |
4636 | { |
4637 | .name = "stat", | |
ab215884 | 4638 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 4639 | }, |
c1e862c1 KH |
4640 | { |
4641 | .name = "force_empty", | |
4642 | .trigger = mem_cgroup_force_empty_write, | |
4643 | }, | |
18f59ea7 BS |
4644 | { |
4645 | .name = "use_hierarchy", | |
4646 | .write_u64 = mem_cgroup_hierarchy_write, | |
4647 | .read_u64 = mem_cgroup_hierarchy_read, | |
4648 | }, | |
a7885eb8 KM |
4649 | { |
4650 | .name = "swappiness", | |
4651 | .read_u64 = mem_cgroup_swappiness_read, | |
4652 | .write_u64 = mem_cgroup_swappiness_write, | |
4653 | }, | |
7dc74be0 DN |
4654 | { |
4655 | .name = "move_charge_at_immigrate", | |
4656 | .read_u64 = mem_cgroup_move_charge_read, | |
4657 | .write_u64 = mem_cgroup_move_charge_write, | |
4658 | }, | |
9490ff27 KH |
4659 | { |
4660 | .name = "oom_control", | |
3c11ecf4 KH |
4661 | .read_map = mem_cgroup_oom_control_read, |
4662 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4663 | .register_event = mem_cgroup_oom_register_event, |
4664 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4665 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4666 | }, | |
406eb0c9 YH |
4667 | #ifdef CONFIG_NUMA |
4668 | { | |
4669 | .name = "numa_stat", | |
ab215884 | 4670 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
4671 | }, |
4672 | #endif | |
c255a458 | 4673 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4674 | { |
4675 | .name = "memsw.usage_in_bytes", | |
4676 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
af36f906 | 4677 | .read = mem_cgroup_read, |
9490ff27 KH |
4678 | .register_event = mem_cgroup_usage_register_event, |
4679 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4680 | }, |
4681 | { | |
4682 | .name = "memsw.max_usage_in_bytes", | |
4683 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4684 | .trigger = mem_cgroup_reset, | |
af36f906 | 4685 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4686 | }, |
4687 | { | |
4688 | .name = "memsw.limit_in_bytes", | |
4689 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4690 | .write_string = mem_cgroup_write, | |
af36f906 | 4691 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4692 | }, |
4693 | { | |
4694 | .name = "memsw.failcnt", | |
4695 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4696 | .trigger = mem_cgroup_reset, | |
af36f906 | 4697 | .read = mem_cgroup_read, |
8c7c6e34 | 4698 | }, |
8c7c6e34 | 4699 | #endif |
6bc10349 | 4700 | { }, /* terminate */ |
af36f906 | 4701 | }; |
8c7c6e34 | 4702 | |
c0ff4b85 | 4703 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4704 | { |
4705 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4706 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4707 | int zone, tmp = node; |
1ecaab2b KH |
4708 | /* |
4709 | * This routine is called against possible nodes. | |
4710 | * But it's BUG to call kmalloc() against offline node. | |
4711 | * | |
4712 | * TODO: this routine can waste much memory for nodes which will | |
4713 | * never be onlined. It's better to use memory hotplug callback | |
4714 | * function. | |
4715 | */ | |
41e3355d KH |
4716 | if (!node_state(node, N_NORMAL_MEMORY)) |
4717 | tmp = -1; | |
17295c88 | 4718 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4719 | if (!pn) |
4720 | return 1; | |
1ecaab2b | 4721 | |
1ecaab2b KH |
4722 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4723 | mz = &pn->zoneinfo[zone]; | |
7f5e86c2 | 4724 | lruvec_init(&mz->lruvec, &NODE_DATA(node)->node_zones[zone]); |
f64c3f54 | 4725 | mz->usage_in_excess = 0; |
4e416953 | 4726 | mz->on_tree = false; |
d79154bb | 4727 | mz->memcg = memcg; |
1ecaab2b | 4728 | } |
0a619e58 | 4729 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4730 | return 0; |
4731 | } | |
4732 | ||
c0ff4b85 | 4733 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4734 | { |
c0ff4b85 | 4735 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4736 | } |
4737 | ||
33327948 KH |
4738 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4739 | { | |
d79154bb | 4740 | struct mem_cgroup *memcg; |
c62b1a3b | 4741 | int size = sizeof(struct mem_cgroup); |
33327948 | 4742 | |
c62b1a3b | 4743 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4744 | if (size < PAGE_SIZE) |
d79154bb | 4745 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 4746 | else |
d79154bb | 4747 | memcg = vzalloc(size); |
33327948 | 4748 | |
d79154bb | 4749 | if (!memcg) |
e7bbcdf3 DC |
4750 | return NULL; |
4751 | ||
d79154bb HD |
4752 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4753 | if (!memcg->stat) | |
d2e61b8d | 4754 | goto out_free; |
d79154bb HD |
4755 | spin_lock_init(&memcg->pcp_counter_lock); |
4756 | return memcg; | |
d2e61b8d DC |
4757 | |
4758 | out_free: | |
4759 | if (size < PAGE_SIZE) | |
d79154bb | 4760 | kfree(memcg); |
d2e61b8d | 4761 | else |
d79154bb | 4762 | vfree(memcg); |
d2e61b8d | 4763 | return NULL; |
33327948 KH |
4764 | } |
4765 | ||
59927fb9 | 4766 | /* |
3afe36b1 | 4767 | * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU, |
59927fb9 HD |
4768 | * but in process context. The work_freeing structure is overlaid |
4769 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
4770 | */ | |
3afe36b1 | 4771 | static void free_work(struct work_struct *work) |
59927fb9 HD |
4772 | { |
4773 | struct mem_cgroup *memcg; | |
3afe36b1 | 4774 | int size = sizeof(struct mem_cgroup); |
59927fb9 HD |
4775 | |
4776 | memcg = container_of(work, struct mem_cgroup, work_freeing); | |
3f134619 GC |
4777 | /* |
4778 | * We need to make sure that (at least for now), the jump label | |
4779 | * destruction code runs outside of the cgroup lock. This is because | |
4780 | * get_online_cpus(), which is called from the static_branch update, | |
4781 | * can't be called inside the cgroup_lock. cpusets are the ones | |
4782 | * enforcing this dependency, so if they ever change, we might as well. | |
4783 | * | |
4784 | * schedule_work() will guarantee this happens. Be careful if you need | |
4785 | * to move this code around, and make sure it is outside | |
4786 | * the cgroup_lock. | |
4787 | */ | |
4788 | disarm_sock_keys(memcg); | |
3afe36b1 GC |
4789 | if (size < PAGE_SIZE) |
4790 | kfree(memcg); | |
4791 | else | |
4792 | vfree(memcg); | |
59927fb9 | 4793 | } |
3afe36b1 GC |
4794 | |
4795 | static void free_rcu(struct rcu_head *rcu_head) | |
59927fb9 HD |
4796 | { |
4797 | struct mem_cgroup *memcg; | |
4798 | ||
4799 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); | |
3afe36b1 | 4800 | INIT_WORK(&memcg->work_freeing, free_work); |
59927fb9 HD |
4801 | schedule_work(&memcg->work_freeing); |
4802 | } | |
4803 | ||
8c7c6e34 KH |
4804 | /* |
4805 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4806 | * (scanning all at force_empty is too costly...) | |
4807 | * | |
4808 | * Instead of clearing all references at force_empty, we remember | |
4809 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4810 | * it goes down to 0. | |
4811 | * | |
8c7c6e34 KH |
4812 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4813 | */ | |
4814 | ||
c0ff4b85 | 4815 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 4816 | { |
08e552c6 KH |
4817 | int node; |
4818 | ||
c0ff4b85 R |
4819 | mem_cgroup_remove_from_trees(memcg); |
4820 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 4821 | |
3ed28fa1 | 4822 | for_each_node(node) |
c0ff4b85 | 4823 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 4824 | |
c0ff4b85 | 4825 | free_percpu(memcg->stat); |
3afe36b1 | 4826 | call_rcu(&memcg->rcu_freeing, free_rcu); |
33327948 KH |
4827 | } |
4828 | ||
c0ff4b85 | 4829 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 4830 | { |
c0ff4b85 | 4831 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
4832 | } |
4833 | ||
c0ff4b85 | 4834 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 4835 | { |
c0ff4b85 R |
4836 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
4837 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4838 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
4839 | if (parent) |
4840 | mem_cgroup_put(parent); | |
4841 | } | |
8c7c6e34 KH |
4842 | } |
4843 | ||
c0ff4b85 | 4844 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 4845 | { |
c0ff4b85 | 4846 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
4847 | } |
4848 | ||
7bcc1bb1 DN |
4849 | /* |
4850 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4851 | */ | |
e1aab161 | 4852 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4853 | { |
c0ff4b85 | 4854 | if (!memcg->res.parent) |
7bcc1bb1 | 4855 | return NULL; |
c0ff4b85 | 4856 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 4857 | } |
e1aab161 | 4858 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4859 | |
c255a458 | 4860 | #ifdef CONFIG_MEMCG_SWAP |
c077719b KH |
4861 | static void __init enable_swap_cgroup(void) |
4862 | { | |
f8d66542 | 4863 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4864 | do_swap_account = 1; |
4865 | } | |
4866 | #else | |
4867 | static void __init enable_swap_cgroup(void) | |
4868 | { | |
4869 | } | |
4870 | #endif | |
4871 | ||
f64c3f54 BS |
4872 | static int mem_cgroup_soft_limit_tree_init(void) |
4873 | { | |
4874 | struct mem_cgroup_tree_per_node *rtpn; | |
4875 | struct mem_cgroup_tree_per_zone *rtpz; | |
4876 | int tmp, node, zone; | |
4877 | ||
3ed28fa1 | 4878 | for_each_node(node) { |
f64c3f54 BS |
4879 | tmp = node; |
4880 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4881 | tmp = -1; | |
4882 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4883 | if (!rtpn) | |
c3cecc68 | 4884 | goto err_cleanup; |
f64c3f54 BS |
4885 | |
4886 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4887 | ||
4888 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4889 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4890 | rtpz->rb_root = RB_ROOT; | |
4891 | spin_lock_init(&rtpz->lock); | |
4892 | } | |
4893 | } | |
4894 | return 0; | |
c3cecc68 MH |
4895 | |
4896 | err_cleanup: | |
3ed28fa1 | 4897 | for_each_node(node) { |
c3cecc68 MH |
4898 | if (!soft_limit_tree.rb_tree_per_node[node]) |
4899 | break; | |
4900 | kfree(soft_limit_tree.rb_tree_per_node[node]); | |
4901 | soft_limit_tree.rb_tree_per_node[node] = NULL; | |
4902 | } | |
4903 | return 1; | |
4904 | ||
f64c3f54 BS |
4905 | } |
4906 | ||
0eb253e2 | 4907 | static struct cgroup_subsys_state * __ref |
761b3ef5 | 4908 | mem_cgroup_create(struct cgroup *cont) |
8cdea7c0 | 4909 | { |
c0ff4b85 | 4910 | struct mem_cgroup *memcg, *parent; |
04046e1a | 4911 | long error = -ENOMEM; |
6d12e2d8 | 4912 | int node; |
8cdea7c0 | 4913 | |
c0ff4b85 R |
4914 | memcg = mem_cgroup_alloc(); |
4915 | if (!memcg) | |
04046e1a | 4916 | return ERR_PTR(error); |
78fb7466 | 4917 | |
3ed28fa1 | 4918 | for_each_node(node) |
c0ff4b85 | 4919 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4920 | goto free_out; |
f64c3f54 | 4921 | |
c077719b | 4922 | /* root ? */ |
28dbc4b6 | 4923 | if (cont->parent == NULL) { |
cdec2e42 | 4924 | int cpu; |
c077719b | 4925 | enable_swap_cgroup(); |
28dbc4b6 | 4926 | parent = NULL; |
f64c3f54 BS |
4927 | if (mem_cgroup_soft_limit_tree_init()) |
4928 | goto free_out; | |
a41c58a6 | 4929 | root_mem_cgroup = memcg; |
cdec2e42 KH |
4930 | for_each_possible_cpu(cpu) { |
4931 | struct memcg_stock_pcp *stock = | |
4932 | &per_cpu(memcg_stock, cpu); | |
4933 | INIT_WORK(&stock->work, drain_local_stock); | |
4934 | } | |
711d3d2c | 4935 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4936 | } else { |
28dbc4b6 | 4937 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
4938 | memcg->use_hierarchy = parent->use_hierarchy; |
4939 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 4940 | } |
28dbc4b6 | 4941 | |
18f59ea7 | 4942 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
4943 | res_counter_init(&memcg->res, &parent->res); |
4944 | res_counter_init(&memcg->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4945 | /* |
4946 | * We increment refcnt of the parent to ensure that we can | |
4947 | * safely access it on res_counter_charge/uncharge. | |
4948 | * This refcnt will be decremented when freeing this | |
4949 | * mem_cgroup(see mem_cgroup_put). | |
4950 | */ | |
4951 | mem_cgroup_get(parent); | |
18f59ea7 | 4952 | } else { |
c0ff4b85 R |
4953 | res_counter_init(&memcg->res, NULL); |
4954 | res_counter_init(&memcg->memsw, NULL); | |
18f59ea7 | 4955 | } |
c0ff4b85 R |
4956 | memcg->last_scanned_node = MAX_NUMNODES; |
4957 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 4958 | |
a7885eb8 | 4959 | if (parent) |
c0ff4b85 R |
4960 | memcg->swappiness = mem_cgroup_swappiness(parent); |
4961 | atomic_set(&memcg->refcnt, 1); | |
4962 | memcg->move_charge_at_immigrate = 0; | |
4963 | mutex_init(&memcg->thresholds_lock); | |
312734c0 | 4964 | spin_lock_init(&memcg->move_lock); |
cbe128e3 GC |
4965 | |
4966 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
4967 | if (error) { | |
4968 | /* | |
4969 | * We call put now because our (and parent's) refcnts | |
4970 | * are already in place. mem_cgroup_put() will internally | |
4971 | * call __mem_cgroup_free, so return directly | |
4972 | */ | |
4973 | mem_cgroup_put(memcg); | |
4974 | return ERR_PTR(error); | |
4975 | } | |
c0ff4b85 | 4976 | return &memcg->css; |
6d12e2d8 | 4977 | free_out: |
c0ff4b85 | 4978 | __mem_cgroup_free(memcg); |
04046e1a | 4979 | return ERR_PTR(error); |
8cdea7c0 BS |
4980 | } |
4981 | ||
761b3ef5 | 4982 | static int mem_cgroup_pre_destroy(struct cgroup *cont) |
df878fb0 | 4983 | { |
c0ff4b85 | 4984 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 4985 | |
c0ff4b85 | 4986 | return mem_cgroup_force_empty(memcg, false); |
df878fb0 KH |
4987 | } |
4988 | ||
761b3ef5 | 4989 | static void mem_cgroup_destroy(struct cgroup *cont) |
8cdea7c0 | 4990 | { |
c0ff4b85 | 4991 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 4992 | |
1d62e436 | 4993 | kmem_cgroup_destroy(memcg); |
d1a4c0b3 | 4994 | |
c0ff4b85 | 4995 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
4996 | } |
4997 | ||
02491447 | 4998 | #ifdef CONFIG_MMU |
7dc74be0 | 4999 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5000 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5001 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5002 | { |
854ffa8d DN |
5003 | int ret = 0; |
5004 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5005 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5006 | |
c0ff4b85 | 5007 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5008 | mc.precharge += count; |
5009 | /* we don't need css_get for root */ | |
5010 | return ret; | |
5011 | } | |
5012 | /* try to charge at once */ | |
5013 | if (count > 1) { | |
5014 | struct res_counter *dummy; | |
5015 | /* | |
c0ff4b85 | 5016 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5017 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5018 | * are still under the same cgroup_mutex. So we can postpone | |
5019 | * css_get(). | |
5020 | */ | |
c0ff4b85 | 5021 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5022 | goto one_by_one; |
c0ff4b85 | 5023 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5024 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5025 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5026 | goto one_by_one; |
5027 | } | |
5028 | mc.precharge += count; | |
854ffa8d DN |
5029 | return ret; |
5030 | } | |
5031 | one_by_one: | |
5032 | /* fall back to one by one charge */ | |
5033 | while (count--) { | |
5034 | if (signal_pending(current)) { | |
5035 | ret = -EINTR; | |
5036 | break; | |
5037 | } | |
5038 | if (!batch_count--) { | |
5039 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5040 | cond_resched(); | |
5041 | } | |
c0ff4b85 R |
5042 | ret = __mem_cgroup_try_charge(NULL, |
5043 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 5044 | if (ret) |
854ffa8d | 5045 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 5046 | return ret; |
854ffa8d DN |
5047 | mc.precharge++; |
5048 | } | |
4ffef5fe DN |
5049 | return ret; |
5050 | } | |
5051 | ||
5052 | /** | |
8d32ff84 | 5053 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5054 | * @vma: the vma the pte to be checked belongs |
5055 | * @addr: the address corresponding to the pte to be checked | |
5056 | * @ptent: the pte to be checked | |
02491447 | 5057 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5058 | * |
5059 | * Returns | |
5060 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5061 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5062 | * move charge. if @target is not NULL, the page is stored in target->page | |
5063 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5064 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5065 | * target for charge migration. if @target is not NULL, the entry is stored | |
5066 | * in target->ent. | |
4ffef5fe DN |
5067 | * |
5068 | * Called with pte lock held. | |
5069 | */ | |
4ffef5fe DN |
5070 | union mc_target { |
5071 | struct page *page; | |
02491447 | 5072 | swp_entry_t ent; |
4ffef5fe DN |
5073 | }; |
5074 | ||
4ffef5fe | 5075 | enum mc_target_type { |
8d32ff84 | 5076 | MC_TARGET_NONE = 0, |
4ffef5fe | 5077 | MC_TARGET_PAGE, |
02491447 | 5078 | MC_TARGET_SWAP, |
4ffef5fe DN |
5079 | }; |
5080 | ||
90254a65 DN |
5081 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5082 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5083 | { |
90254a65 | 5084 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5085 | |
90254a65 DN |
5086 | if (!page || !page_mapped(page)) |
5087 | return NULL; | |
5088 | if (PageAnon(page)) { | |
5089 | /* we don't move shared anon */ | |
4b91355e | 5090 | if (!move_anon()) |
90254a65 | 5091 | return NULL; |
87946a72 DN |
5092 | } else if (!move_file()) |
5093 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5094 | return NULL; |
5095 | if (!get_page_unless_zero(page)) | |
5096 | return NULL; | |
5097 | ||
5098 | return page; | |
5099 | } | |
5100 | ||
4b91355e | 5101 | #ifdef CONFIG_SWAP |
90254a65 DN |
5102 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5103 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5104 | { | |
90254a65 DN |
5105 | struct page *page = NULL; |
5106 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5107 | ||
5108 | if (!move_anon() || non_swap_entry(ent)) | |
5109 | return NULL; | |
4b91355e KH |
5110 | /* |
5111 | * Because lookup_swap_cache() updates some statistics counter, | |
5112 | * we call find_get_page() with swapper_space directly. | |
5113 | */ | |
5114 | page = find_get_page(&swapper_space, ent.val); | |
90254a65 DN |
5115 | if (do_swap_account) |
5116 | entry->val = ent.val; | |
5117 | ||
5118 | return page; | |
5119 | } | |
4b91355e KH |
5120 | #else |
5121 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5122 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5123 | { | |
5124 | return NULL; | |
5125 | } | |
5126 | #endif | |
90254a65 | 5127 | |
87946a72 DN |
5128 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5129 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5130 | { | |
5131 | struct page *page = NULL; | |
87946a72 DN |
5132 | struct address_space *mapping; |
5133 | pgoff_t pgoff; | |
5134 | ||
5135 | if (!vma->vm_file) /* anonymous vma */ | |
5136 | return NULL; | |
5137 | if (!move_file()) | |
5138 | return NULL; | |
5139 | ||
87946a72 DN |
5140 | mapping = vma->vm_file->f_mapping; |
5141 | if (pte_none(ptent)) | |
5142 | pgoff = linear_page_index(vma, addr); | |
5143 | else /* pte_file(ptent) is true */ | |
5144 | pgoff = pte_to_pgoff(ptent); | |
5145 | ||
5146 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5147 | page = find_get_page(mapping, pgoff); |
5148 | ||
5149 | #ifdef CONFIG_SWAP | |
5150 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5151 | if (radix_tree_exceptional_entry(page)) { | |
5152 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5153 | if (do_swap_account) |
aa3b1895 HD |
5154 | *entry = swap; |
5155 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5156 | } |
aa3b1895 | 5157 | #endif |
87946a72 DN |
5158 | return page; |
5159 | } | |
5160 | ||
8d32ff84 | 5161 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5162 | unsigned long addr, pte_t ptent, union mc_target *target) |
5163 | { | |
5164 | struct page *page = NULL; | |
5165 | struct page_cgroup *pc; | |
8d32ff84 | 5166 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5167 | swp_entry_t ent = { .val = 0 }; |
5168 | ||
5169 | if (pte_present(ptent)) | |
5170 | page = mc_handle_present_pte(vma, addr, ptent); | |
5171 | else if (is_swap_pte(ptent)) | |
5172 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5173 | else if (pte_none(ptent) || pte_file(ptent)) |
5174 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5175 | |
5176 | if (!page && !ent.val) | |
8d32ff84 | 5177 | return ret; |
02491447 DN |
5178 | if (page) { |
5179 | pc = lookup_page_cgroup(page); | |
5180 | /* | |
5181 | * Do only loose check w/o page_cgroup lock. | |
5182 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5183 | * the lock. | |
5184 | */ | |
5185 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5186 | ret = MC_TARGET_PAGE; | |
5187 | if (target) | |
5188 | target->page = page; | |
5189 | } | |
5190 | if (!ret || !target) | |
5191 | put_page(page); | |
5192 | } | |
90254a65 DN |
5193 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5194 | if (ent.val && !ret && | |
9fb4b7cc | 5195 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5196 | ret = MC_TARGET_SWAP; |
5197 | if (target) | |
5198 | target->ent = ent; | |
4ffef5fe | 5199 | } |
4ffef5fe DN |
5200 | return ret; |
5201 | } | |
5202 | ||
12724850 NH |
5203 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5204 | /* | |
5205 | * We don't consider swapping or file mapped pages because THP does not | |
5206 | * support them for now. | |
5207 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5208 | */ | |
5209 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5210 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5211 | { | |
5212 | struct page *page = NULL; | |
5213 | struct page_cgroup *pc; | |
5214 | enum mc_target_type ret = MC_TARGET_NONE; | |
5215 | ||
5216 | page = pmd_page(pmd); | |
5217 | VM_BUG_ON(!page || !PageHead(page)); | |
5218 | if (!move_anon()) | |
5219 | return ret; | |
5220 | pc = lookup_page_cgroup(page); | |
5221 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5222 | ret = MC_TARGET_PAGE; | |
5223 | if (target) { | |
5224 | get_page(page); | |
5225 | target->page = page; | |
5226 | } | |
5227 | } | |
5228 | return ret; | |
5229 | } | |
5230 | #else | |
5231 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5232 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5233 | { | |
5234 | return MC_TARGET_NONE; | |
5235 | } | |
5236 | #endif | |
5237 | ||
4ffef5fe DN |
5238 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5239 | unsigned long addr, unsigned long end, | |
5240 | struct mm_walk *walk) | |
5241 | { | |
5242 | struct vm_area_struct *vma = walk->private; | |
5243 | pte_t *pte; | |
5244 | spinlock_t *ptl; | |
5245 | ||
12724850 NH |
5246 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
5247 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
5248 | mc.precharge += HPAGE_PMD_NR; | |
5249 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5250 | return 0; |
12724850 | 5251 | } |
03319327 | 5252 | |
45f83cef AA |
5253 | if (pmd_trans_unstable(pmd)) |
5254 | return 0; | |
4ffef5fe DN |
5255 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5256 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5257 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5258 | mc.precharge++; /* increment precharge temporarily */ |
5259 | pte_unmap_unlock(pte - 1, ptl); | |
5260 | cond_resched(); | |
5261 | ||
7dc74be0 DN |
5262 | return 0; |
5263 | } | |
5264 | ||
4ffef5fe DN |
5265 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5266 | { | |
5267 | unsigned long precharge; | |
5268 | struct vm_area_struct *vma; | |
5269 | ||
dfe076b0 | 5270 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5271 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5272 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5273 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5274 | .mm = mm, | |
5275 | .private = vma, | |
5276 | }; | |
5277 | if (is_vm_hugetlb_page(vma)) | |
5278 | continue; | |
4ffef5fe DN |
5279 | walk_page_range(vma->vm_start, vma->vm_end, |
5280 | &mem_cgroup_count_precharge_walk); | |
5281 | } | |
dfe076b0 | 5282 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5283 | |
5284 | precharge = mc.precharge; | |
5285 | mc.precharge = 0; | |
5286 | ||
5287 | return precharge; | |
5288 | } | |
5289 | ||
4ffef5fe DN |
5290 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5291 | { | |
dfe076b0 DN |
5292 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5293 | ||
5294 | VM_BUG_ON(mc.moving_task); | |
5295 | mc.moving_task = current; | |
5296 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5297 | } |
5298 | ||
dfe076b0 DN |
5299 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5300 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5301 | { |
2bd9bb20 KH |
5302 | struct mem_cgroup *from = mc.from; |
5303 | struct mem_cgroup *to = mc.to; | |
5304 | ||
4ffef5fe | 5305 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5306 | if (mc.precharge) { |
5307 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5308 | mc.precharge = 0; | |
5309 | } | |
5310 | /* | |
5311 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5312 | * we must uncharge here. | |
5313 | */ | |
5314 | if (mc.moved_charge) { | |
5315 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5316 | mc.moved_charge = 0; | |
4ffef5fe | 5317 | } |
483c30b5 DN |
5318 | /* we must fixup refcnts and charges */ |
5319 | if (mc.moved_swap) { | |
483c30b5 DN |
5320 | /* uncharge swap account from the old cgroup */ |
5321 | if (!mem_cgroup_is_root(mc.from)) | |
5322 | res_counter_uncharge(&mc.from->memsw, | |
5323 | PAGE_SIZE * mc.moved_swap); | |
5324 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5325 | ||
5326 | if (!mem_cgroup_is_root(mc.to)) { | |
5327 | /* | |
5328 | * we charged both to->res and to->memsw, so we should | |
5329 | * uncharge to->res. | |
5330 | */ | |
5331 | res_counter_uncharge(&mc.to->res, | |
5332 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5333 | } |
5334 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5335 | mc.moved_swap = 0; |
5336 | } | |
dfe076b0 DN |
5337 | memcg_oom_recover(from); |
5338 | memcg_oom_recover(to); | |
5339 | wake_up_all(&mc.waitq); | |
5340 | } | |
5341 | ||
5342 | static void mem_cgroup_clear_mc(void) | |
5343 | { | |
5344 | struct mem_cgroup *from = mc.from; | |
5345 | ||
5346 | /* | |
5347 | * we must clear moving_task before waking up waiters at the end of | |
5348 | * task migration. | |
5349 | */ | |
5350 | mc.moving_task = NULL; | |
5351 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5352 | spin_lock(&mc.lock); |
4ffef5fe DN |
5353 | mc.from = NULL; |
5354 | mc.to = NULL; | |
2bd9bb20 | 5355 | spin_unlock(&mc.lock); |
32047e2a | 5356 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5357 | } |
5358 | ||
761b3ef5 LZ |
5359 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5360 | struct cgroup_taskset *tset) | |
7dc74be0 | 5361 | { |
2f7ee569 | 5362 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5363 | int ret = 0; |
c0ff4b85 | 5364 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5365 | |
c0ff4b85 | 5366 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5367 | struct mm_struct *mm; |
5368 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5369 | ||
c0ff4b85 | 5370 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5371 | |
5372 | mm = get_task_mm(p); | |
5373 | if (!mm) | |
5374 | return 0; | |
7dc74be0 | 5375 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5376 | if (mm->owner == p) { |
5377 | VM_BUG_ON(mc.from); | |
5378 | VM_BUG_ON(mc.to); | |
5379 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5380 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5381 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5382 | mem_cgroup_start_move(from); |
2bd9bb20 | 5383 | spin_lock(&mc.lock); |
4ffef5fe | 5384 | mc.from = from; |
c0ff4b85 | 5385 | mc.to = memcg; |
2bd9bb20 | 5386 | spin_unlock(&mc.lock); |
dfe076b0 | 5387 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5388 | |
5389 | ret = mem_cgroup_precharge_mc(mm); | |
5390 | if (ret) | |
5391 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5392 | } |
5393 | mmput(mm); | |
7dc74be0 DN |
5394 | } |
5395 | return ret; | |
5396 | } | |
5397 | ||
761b3ef5 LZ |
5398 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5399 | struct cgroup_taskset *tset) | |
7dc74be0 | 5400 | { |
4ffef5fe | 5401 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5402 | } |
5403 | ||
4ffef5fe DN |
5404 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5405 | unsigned long addr, unsigned long end, | |
5406 | struct mm_walk *walk) | |
7dc74be0 | 5407 | { |
4ffef5fe DN |
5408 | int ret = 0; |
5409 | struct vm_area_struct *vma = walk->private; | |
5410 | pte_t *pte; | |
5411 | spinlock_t *ptl; | |
12724850 NH |
5412 | enum mc_target_type target_type; |
5413 | union mc_target target; | |
5414 | struct page *page; | |
5415 | struct page_cgroup *pc; | |
4ffef5fe | 5416 | |
12724850 NH |
5417 | /* |
5418 | * We don't take compound_lock() here but no race with splitting thp | |
5419 | * happens because: | |
5420 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5421 | * under splitting, which means there's no concurrent thp split, | |
5422 | * - if another thread runs into split_huge_page() just after we | |
5423 | * entered this if-block, the thread must wait for page table lock | |
5424 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5425 | * part of thp split is not executed yet. | |
5426 | */ | |
5427 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 5428 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
5429 | spin_unlock(&vma->vm_mm->page_table_lock); |
5430 | return 0; | |
5431 | } | |
5432 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5433 | if (target_type == MC_TARGET_PAGE) { | |
5434 | page = target.page; | |
5435 | if (!isolate_lru_page(page)) { | |
5436 | pc = lookup_page_cgroup(page); | |
5437 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5438 | pc, mc.from, mc.to)) { |
12724850 NH |
5439 | mc.precharge -= HPAGE_PMD_NR; |
5440 | mc.moved_charge += HPAGE_PMD_NR; | |
5441 | } | |
5442 | putback_lru_page(page); | |
5443 | } | |
5444 | put_page(page); | |
5445 | } | |
5446 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5447 | return 0; |
12724850 NH |
5448 | } |
5449 | ||
45f83cef AA |
5450 | if (pmd_trans_unstable(pmd)) |
5451 | return 0; | |
4ffef5fe DN |
5452 | retry: |
5453 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5454 | for (; addr != end; addr += PAGE_SIZE) { | |
5455 | pte_t ptent = *(pte++); | |
02491447 | 5456 | swp_entry_t ent; |
4ffef5fe DN |
5457 | |
5458 | if (!mc.precharge) | |
5459 | break; | |
5460 | ||
8d32ff84 | 5461 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5462 | case MC_TARGET_PAGE: |
5463 | page = target.page; | |
5464 | if (isolate_lru_page(page)) | |
5465 | goto put; | |
5466 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5467 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5468 | mc.from, mc.to)) { |
4ffef5fe | 5469 | mc.precharge--; |
854ffa8d DN |
5470 | /* we uncharge from mc.from later. */ |
5471 | mc.moved_charge++; | |
4ffef5fe DN |
5472 | } |
5473 | putback_lru_page(page); | |
8d32ff84 | 5474 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5475 | put_page(page); |
5476 | break; | |
02491447 DN |
5477 | case MC_TARGET_SWAP: |
5478 | ent = target.ent; | |
e91cbb42 | 5479 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5480 | mc.precharge--; |
483c30b5 DN |
5481 | /* we fixup refcnts and charges later. */ |
5482 | mc.moved_swap++; | |
5483 | } | |
02491447 | 5484 | break; |
4ffef5fe DN |
5485 | default: |
5486 | break; | |
5487 | } | |
5488 | } | |
5489 | pte_unmap_unlock(pte - 1, ptl); | |
5490 | cond_resched(); | |
5491 | ||
5492 | if (addr != end) { | |
5493 | /* | |
5494 | * We have consumed all precharges we got in can_attach(). | |
5495 | * We try charge one by one, but don't do any additional | |
5496 | * charges to mc.to if we have failed in charge once in attach() | |
5497 | * phase. | |
5498 | */ | |
854ffa8d | 5499 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5500 | if (!ret) |
5501 | goto retry; | |
5502 | } | |
5503 | ||
5504 | return ret; | |
5505 | } | |
5506 | ||
5507 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5508 | { | |
5509 | struct vm_area_struct *vma; | |
5510 | ||
5511 | lru_add_drain_all(); | |
dfe076b0 DN |
5512 | retry: |
5513 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5514 | /* | |
5515 | * Someone who are holding the mmap_sem might be waiting in | |
5516 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5517 | * and retry. Because we cancel precharges, we might not be able | |
5518 | * to move enough charges, but moving charge is a best-effort | |
5519 | * feature anyway, so it wouldn't be a big problem. | |
5520 | */ | |
5521 | __mem_cgroup_clear_mc(); | |
5522 | cond_resched(); | |
5523 | goto retry; | |
5524 | } | |
4ffef5fe DN |
5525 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5526 | int ret; | |
5527 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5528 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5529 | .mm = mm, | |
5530 | .private = vma, | |
5531 | }; | |
5532 | if (is_vm_hugetlb_page(vma)) | |
5533 | continue; | |
4ffef5fe DN |
5534 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5535 | &mem_cgroup_move_charge_walk); | |
5536 | if (ret) | |
5537 | /* | |
5538 | * means we have consumed all precharges and failed in | |
5539 | * doing additional charge. Just abandon here. | |
5540 | */ | |
5541 | break; | |
5542 | } | |
dfe076b0 | 5543 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5544 | } |
5545 | ||
761b3ef5 LZ |
5546 | static void mem_cgroup_move_task(struct cgroup *cont, |
5547 | struct cgroup_taskset *tset) | |
67e465a7 | 5548 | { |
2f7ee569 | 5549 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5550 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5551 | |
dfe076b0 | 5552 | if (mm) { |
a433658c KM |
5553 | if (mc.to) |
5554 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5555 | mmput(mm); |
5556 | } | |
a433658c KM |
5557 | if (mc.to) |
5558 | mem_cgroup_clear_mc(); | |
67e465a7 | 5559 | } |
5cfb80a7 | 5560 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
5561 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5562 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5563 | { |
5564 | return 0; | |
5565 | } | |
761b3ef5 LZ |
5566 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5567 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5568 | { |
5569 | } | |
761b3ef5 LZ |
5570 | static void mem_cgroup_move_task(struct cgroup *cont, |
5571 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5572 | { |
5573 | } | |
5574 | #endif | |
67e465a7 | 5575 | |
8cdea7c0 BS |
5576 | struct cgroup_subsys mem_cgroup_subsys = { |
5577 | .name = "memory", | |
5578 | .subsys_id = mem_cgroup_subsys_id, | |
5579 | .create = mem_cgroup_create, | |
df878fb0 | 5580 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 | 5581 | .destroy = mem_cgroup_destroy, |
7dc74be0 DN |
5582 | .can_attach = mem_cgroup_can_attach, |
5583 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5584 | .attach = mem_cgroup_move_task, |
6bc10349 | 5585 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 5586 | .early_init = 0, |
04046e1a | 5587 | .use_id = 1, |
48ddbe19 | 5588 | .__DEPRECATED_clear_css_refs = true, |
8cdea7c0 | 5589 | }; |
c077719b | 5590 | |
c255a458 | 5591 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5592 | static int __init enable_swap_account(char *s) |
5593 | { | |
5594 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5595 | if (!strcmp(s, "1")) |
a42c390c | 5596 | really_do_swap_account = 1; |
a2c8990a | 5597 | else if (!strcmp(s, "0")) |
a42c390c MH |
5598 | really_do_swap_account = 0; |
5599 | return 1; | |
5600 | } | |
a2c8990a | 5601 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5602 | |
c077719b | 5603 | #endif |