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, |
d13d1443 | 382 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 383 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
384 | NR_CHARGE_TYPE, |
385 | }; | |
386 | ||
8c7c6e34 | 387 | /* for encoding cft->private value on file */ |
65c64ce8 GC |
388 | #define _MEM (0) |
389 | #define _MEMSWAP (1) | |
390 | #define _OOM_TYPE (2) | |
a0db00fc KS |
391 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
392 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 393 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
394 | /* Used for OOM nofiier */ |
395 | #define OOM_CONTROL (0) | |
8c7c6e34 | 396 | |
75822b44 BS |
397 | /* |
398 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
399 | */ | |
400 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
401 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
402 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
403 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
404 | ||
c0ff4b85 R |
405 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
406 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 GC |
407 | |
408 | /* Writing them here to avoid exposing memcg's inner layout */ | |
c255a458 | 409 | #ifdef CONFIG_MEMCG_KMEM |
e1aab161 | 410 | #include <net/sock.h> |
d1a4c0b3 | 411 | #include <net/ip.h> |
e1aab161 GC |
412 | |
413 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); | |
414 | void sock_update_memcg(struct sock *sk) | |
415 | { | |
376be5ff | 416 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 417 | struct mem_cgroup *memcg; |
3f134619 | 418 | struct cg_proto *cg_proto; |
e1aab161 GC |
419 | |
420 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
421 | ||
f3f511e1 GC |
422 | /* Socket cloning can throw us here with sk_cgrp already |
423 | * filled. It won't however, necessarily happen from | |
424 | * process context. So the test for root memcg given | |
425 | * the current task's memcg won't help us in this case. | |
426 | * | |
427 | * Respecting the original socket's memcg is a better | |
428 | * decision in this case. | |
429 | */ | |
430 | if (sk->sk_cgrp) { | |
431 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
432 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
433 | return; | |
434 | } | |
435 | ||
e1aab161 GC |
436 | rcu_read_lock(); |
437 | memcg = mem_cgroup_from_task(current); | |
3f134619 GC |
438 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
439 | if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) { | |
e1aab161 | 440 | mem_cgroup_get(memcg); |
3f134619 | 441 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
442 | } |
443 | rcu_read_unlock(); | |
444 | } | |
445 | } | |
446 | EXPORT_SYMBOL(sock_update_memcg); | |
447 | ||
448 | void sock_release_memcg(struct sock *sk) | |
449 | { | |
376be5ff | 450 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
451 | struct mem_cgroup *memcg; |
452 | WARN_ON(!sk->sk_cgrp->memcg); | |
453 | memcg = sk->sk_cgrp->memcg; | |
454 | mem_cgroup_put(memcg); | |
455 | } | |
456 | } | |
d1a4c0b3 | 457 | |
319d3b9c | 458 | #ifdef CONFIG_INET |
d1a4c0b3 GC |
459 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
460 | { | |
461 | if (!memcg || mem_cgroup_is_root(memcg)) | |
462 | return NULL; | |
463 | ||
464 | return &memcg->tcp_mem.cg_proto; | |
465 | } | |
466 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 467 | #endif /* CONFIG_INET */ |
c255a458 | 468 | #endif /* CONFIG_MEMCG_KMEM */ |
e1aab161 | 469 | |
c255a458 | 470 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
3f134619 GC |
471 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
472 | { | |
473 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
474 | return; | |
475 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
476 | } | |
477 | #else | |
478 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
479 | { | |
480 | } | |
481 | #endif | |
482 | ||
c0ff4b85 | 483 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 484 | |
f64c3f54 | 485 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 486 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 487 | { |
c0ff4b85 | 488 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
489 | } |
490 | ||
c0ff4b85 | 491 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 492 | { |
c0ff4b85 | 493 | return &memcg->css; |
d324236b WF |
494 | } |
495 | ||
f64c3f54 | 496 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 497 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 498 | { |
97a6c37b JW |
499 | int nid = page_to_nid(page); |
500 | int zid = page_zonenum(page); | |
f64c3f54 | 501 | |
c0ff4b85 | 502 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
503 | } |
504 | ||
505 | static struct mem_cgroup_tree_per_zone * | |
506 | soft_limit_tree_node_zone(int nid, int zid) | |
507 | { | |
508 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
509 | } | |
510 | ||
511 | static struct mem_cgroup_tree_per_zone * | |
512 | soft_limit_tree_from_page(struct page *page) | |
513 | { | |
514 | int nid = page_to_nid(page); | |
515 | int zid = page_zonenum(page); | |
516 | ||
517 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
518 | } | |
519 | ||
520 | static void | |
c0ff4b85 | 521 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 522 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
523 | struct mem_cgroup_tree_per_zone *mctz, |
524 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
525 | { |
526 | struct rb_node **p = &mctz->rb_root.rb_node; | |
527 | struct rb_node *parent = NULL; | |
528 | struct mem_cgroup_per_zone *mz_node; | |
529 | ||
530 | if (mz->on_tree) | |
531 | return; | |
532 | ||
ef8745c1 KH |
533 | mz->usage_in_excess = new_usage_in_excess; |
534 | if (!mz->usage_in_excess) | |
535 | return; | |
f64c3f54 BS |
536 | while (*p) { |
537 | parent = *p; | |
538 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
539 | tree_node); | |
540 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
541 | p = &(*p)->rb_left; | |
542 | /* | |
543 | * We can't avoid mem cgroups that are over their soft | |
544 | * limit by the same amount | |
545 | */ | |
546 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
547 | p = &(*p)->rb_right; | |
548 | } | |
549 | rb_link_node(&mz->tree_node, parent, p); | |
550 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
551 | mz->on_tree = true; | |
4e416953 BS |
552 | } |
553 | ||
554 | static void | |
c0ff4b85 | 555 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
556 | struct mem_cgroup_per_zone *mz, |
557 | struct mem_cgroup_tree_per_zone *mctz) | |
558 | { | |
559 | if (!mz->on_tree) | |
560 | return; | |
561 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
562 | mz->on_tree = false; | |
563 | } | |
564 | ||
f64c3f54 | 565 | static void |
c0ff4b85 | 566 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
567 | struct mem_cgroup_per_zone *mz, |
568 | struct mem_cgroup_tree_per_zone *mctz) | |
569 | { | |
570 | spin_lock(&mctz->lock); | |
c0ff4b85 | 571 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
572 | spin_unlock(&mctz->lock); |
573 | } | |
574 | ||
f64c3f54 | 575 | |
c0ff4b85 | 576 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 577 | { |
ef8745c1 | 578 | unsigned long long excess; |
f64c3f54 BS |
579 | struct mem_cgroup_per_zone *mz; |
580 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
581 | int nid = page_to_nid(page); |
582 | int zid = page_zonenum(page); | |
f64c3f54 BS |
583 | mctz = soft_limit_tree_from_page(page); |
584 | ||
585 | /* | |
4e649152 KH |
586 | * Necessary to update all ancestors when hierarchy is used. |
587 | * because their event counter is not touched. | |
f64c3f54 | 588 | */ |
c0ff4b85 R |
589 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
590 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
591 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
592 | /* |
593 | * We have to update the tree if mz is on RB-tree or | |
594 | * mem is over its softlimit. | |
595 | */ | |
ef8745c1 | 596 | if (excess || mz->on_tree) { |
4e649152 KH |
597 | spin_lock(&mctz->lock); |
598 | /* if on-tree, remove it */ | |
599 | if (mz->on_tree) | |
c0ff4b85 | 600 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 601 | /* |
ef8745c1 KH |
602 | * Insert again. mz->usage_in_excess will be updated. |
603 | * If excess is 0, no tree ops. | |
4e649152 | 604 | */ |
c0ff4b85 | 605 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
606 | spin_unlock(&mctz->lock); |
607 | } | |
f64c3f54 BS |
608 | } |
609 | } | |
610 | ||
c0ff4b85 | 611 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
612 | { |
613 | int node, zone; | |
614 | struct mem_cgroup_per_zone *mz; | |
615 | struct mem_cgroup_tree_per_zone *mctz; | |
616 | ||
3ed28fa1 | 617 | for_each_node(node) { |
f64c3f54 | 618 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 619 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 620 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 621 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
622 | } |
623 | } | |
624 | } | |
625 | ||
4e416953 BS |
626 | static struct mem_cgroup_per_zone * |
627 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
628 | { | |
629 | struct rb_node *rightmost = NULL; | |
26251eaf | 630 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
631 | |
632 | retry: | |
26251eaf | 633 | mz = NULL; |
4e416953 BS |
634 | rightmost = rb_last(&mctz->rb_root); |
635 | if (!rightmost) | |
636 | goto done; /* Nothing to reclaim from */ | |
637 | ||
638 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
639 | /* | |
640 | * Remove the node now but someone else can add it back, | |
641 | * we will to add it back at the end of reclaim to its correct | |
642 | * position in the tree. | |
643 | */ | |
d79154bb HD |
644 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
645 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
646 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
647 | goto retry; |
648 | done: | |
649 | return mz; | |
650 | } | |
651 | ||
652 | static struct mem_cgroup_per_zone * | |
653 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
654 | { | |
655 | struct mem_cgroup_per_zone *mz; | |
656 | ||
657 | spin_lock(&mctz->lock); | |
658 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
659 | spin_unlock(&mctz->lock); | |
660 | return mz; | |
661 | } | |
662 | ||
711d3d2c KH |
663 | /* |
664 | * Implementation Note: reading percpu statistics for memcg. | |
665 | * | |
666 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
667 | * synchronization to implement "quick" read. There are trade-off between | |
668 | * reading cost and precision of value. Then, we may have a chance to implement | |
669 | * a periodic synchronizion of counter in memcg's counter. | |
670 | * | |
671 | * But this _read() function is used for user interface now. The user accounts | |
672 | * memory usage by memory cgroup and he _always_ requires exact value because | |
673 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
674 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
675 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
676 | * | |
677 | * If there are kernel internal actions which can make use of some not-exact | |
678 | * value, and reading all cpu value can be performance bottleneck in some | |
679 | * common workload, threashold and synchonization as vmstat[] should be | |
680 | * implemented. | |
681 | */ | |
c0ff4b85 | 682 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 683 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 684 | { |
7a159cc9 | 685 | long val = 0; |
c62b1a3b | 686 | int cpu; |
c62b1a3b | 687 | |
711d3d2c KH |
688 | get_online_cpus(); |
689 | for_each_online_cpu(cpu) | |
c0ff4b85 | 690 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 691 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
692 | spin_lock(&memcg->pcp_counter_lock); |
693 | val += memcg->nocpu_base.count[idx]; | |
694 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
695 | #endif |
696 | put_online_cpus(); | |
c62b1a3b KH |
697 | return val; |
698 | } | |
699 | ||
c0ff4b85 | 700 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
701 | bool charge) |
702 | { | |
703 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 704 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
705 | } |
706 | ||
c0ff4b85 | 707 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
708 | enum mem_cgroup_events_index idx) |
709 | { | |
710 | unsigned long val = 0; | |
711 | int cpu; | |
712 | ||
713 | for_each_online_cpu(cpu) | |
c0ff4b85 | 714 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 715 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
716 | spin_lock(&memcg->pcp_counter_lock); |
717 | val += memcg->nocpu_base.events[idx]; | |
718 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
719 | #endif |
720 | return val; | |
721 | } | |
722 | ||
c0ff4b85 | 723 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 724 | bool anon, int nr_pages) |
d52aa412 | 725 | { |
c62b1a3b KH |
726 | preempt_disable(); |
727 | ||
b2402857 KH |
728 | /* |
729 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
730 | * counted as CACHE even if it's on ANON LRU. | |
731 | */ | |
732 | if (anon) | |
733 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 734 | nr_pages); |
d52aa412 | 735 | else |
b2402857 | 736 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 737 | nr_pages); |
55e462b0 | 738 | |
e401f176 KH |
739 | /* pagein of a big page is an event. So, ignore page size */ |
740 | if (nr_pages > 0) | |
c0ff4b85 | 741 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 742 | else { |
c0ff4b85 | 743 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
744 | nr_pages = -nr_pages; /* for event */ |
745 | } | |
e401f176 | 746 | |
13114716 | 747 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 748 | |
c62b1a3b | 749 | preempt_enable(); |
6d12e2d8 KH |
750 | } |
751 | ||
bb2a0de9 | 752 | unsigned long |
4d7dcca2 | 753 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
754 | { |
755 | struct mem_cgroup_per_zone *mz; | |
756 | ||
757 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
758 | return mz->lru_size[lru]; | |
759 | } | |
760 | ||
761 | static unsigned long | |
c0ff4b85 | 762 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 763 | unsigned int lru_mask) |
889976db YH |
764 | { |
765 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 766 | enum lru_list lru; |
bb2a0de9 KH |
767 | unsigned long ret = 0; |
768 | ||
c0ff4b85 | 769 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 770 | |
f156ab93 HD |
771 | for_each_lru(lru) { |
772 | if (BIT(lru) & lru_mask) | |
773 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
774 | } |
775 | return ret; | |
776 | } | |
777 | ||
778 | static unsigned long | |
c0ff4b85 | 779 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
780 | int nid, unsigned int lru_mask) |
781 | { | |
889976db YH |
782 | u64 total = 0; |
783 | int zid; | |
784 | ||
bb2a0de9 | 785 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
786 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
787 | nid, zid, lru_mask); | |
bb2a0de9 | 788 | |
889976db YH |
789 | return total; |
790 | } | |
bb2a0de9 | 791 | |
c0ff4b85 | 792 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 793 | unsigned int lru_mask) |
6d12e2d8 | 794 | { |
889976db | 795 | int nid; |
6d12e2d8 KH |
796 | u64 total = 0; |
797 | ||
bb2a0de9 | 798 | for_each_node_state(nid, N_HIGH_MEMORY) |
c0ff4b85 | 799 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 800 | return total; |
d52aa412 KH |
801 | } |
802 | ||
f53d7ce3 JW |
803 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
804 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
805 | { |
806 | unsigned long val, next; | |
807 | ||
13114716 | 808 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 809 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 810 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
811 | if ((long)next - (long)val < 0) { |
812 | switch (target) { | |
813 | case MEM_CGROUP_TARGET_THRESH: | |
814 | next = val + THRESHOLDS_EVENTS_TARGET; | |
815 | break; | |
816 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
817 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
818 | break; | |
819 | case MEM_CGROUP_TARGET_NUMAINFO: | |
820 | next = val + NUMAINFO_EVENTS_TARGET; | |
821 | break; | |
822 | default: | |
823 | break; | |
824 | } | |
825 | __this_cpu_write(memcg->stat->targets[target], next); | |
826 | return true; | |
7a159cc9 | 827 | } |
f53d7ce3 | 828 | return false; |
d2265e6f KH |
829 | } |
830 | ||
831 | /* | |
832 | * Check events in order. | |
833 | * | |
834 | */ | |
c0ff4b85 | 835 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 836 | { |
4799401f | 837 | preempt_disable(); |
d2265e6f | 838 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
839 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
840 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
841 | bool do_softlimit; |
842 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
843 | |
844 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
845 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
846 | #if MAX_NUMNODES > 1 | |
847 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
848 | MEM_CGROUP_TARGET_NUMAINFO); | |
849 | #endif | |
850 | preempt_enable(); | |
851 | ||
c0ff4b85 | 852 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 853 | if (unlikely(do_softlimit)) |
c0ff4b85 | 854 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 855 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 856 | if (unlikely(do_numainfo)) |
c0ff4b85 | 857 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 858 | #endif |
f53d7ce3 JW |
859 | } else |
860 | preempt_enable(); | |
d2265e6f KH |
861 | } |
862 | ||
d1a4c0b3 | 863 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
864 | { |
865 | return container_of(cgroup_subsys_state(cont, | |
866 | mem_cgroup_subsys_id), struct mem_cgroup, | |
867 | css); | |
868 | } | |
869 | ||
cf475ad2 | 870 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 871 | { |
31a78f23 BS |
872 | /* |
873 | * mm_update_next_owner() may clear mm->owner to NULL | |
874 | * if it races with swapoff, page migration, etc. | |
875 | * So this can be called with p == NULL. | |
876 | */ | |
877 | if (unlikely(!p)) | |
878 | return NULL; | |
879 | ||
78fb7466 PE |
880 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
881 | struct mem_cgroup, css); | |
882 | } | |
883 | ||
a433658c | 884 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 885 | { |
c0ff4b85 | 886 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
887 | |
888 | if (!mm) | |
889 | return NULL; | |
54595fe2 KH |
890 | /* |
891 | * Because we have no locks, mm->owner's may be being moved to other | |
892 | * cgroup. We use css_tryget() here even if this looks | |
893 | * pessimistic (rather than adding locks here). | |
894 | */ | |
895 | rcu_read_lock(); | |
896 | do { | |
c0ff4b85 R |
897 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
898 | if (unlikely(!memcg)) | |
54595fe2 | 899 | break; |
c0ff4b85 | 900 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 901 | rcu_read_unlock(); |
c0ff4b85 | 902 | return memcg; |
54595fe2 KH |
903 | } |
904 | ||
5660048c JW |
905 | /** |
906 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
907 | * @root: hierarchy root | |
908 | * @prev: previously returned memcg, NULL on first invocation | |
909 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
910 | * | |
911 | * Returns references to children of the hierarchy below @root, or | |
912 | * @root itself, or %NULL after a full round-trip. | |
913 | * | |
914 | * Caller must pass the return value in @prev on subsequent | |
915 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
916 | * to cancel a hierarchy walk before the round-trip is complete. | |
917 | * | |
918 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
919 | * divide up the memcgs in the hierarchy among all concurrent | |
920 | * reclaimers operating on the same zone and priority. | |
921 | */ | |
922 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
923 | struct mem_cgroup *prev, | |
924 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 925 | { |
9f3a0d09 JW |
926 | struct mem_cgroup *memcg = NULL; |
927 | int id = 0; | |
711d3d2c | 928 | |
5660048c JW |
929 | if (mem_cgroup_disabled()) |
930 | return NULL; | |
931 | ||
9f3a0d09 JW |
932 | if (!root) |
933 | root = root_mem_cgroup; | |
7d74b06f | 934 | |
9f3a0d09 JW |
935 | if (prev && !reclaim) |
936 | id = css_id(&prev->css); | |
14067bb3 | 937 | |
9f3a0d09 JW |
938 | if (prev && prev != root) |
939 | css_put(&prev->css); | |
14067bb3 | 940 | |
9f3a0d09 JW |
941 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
942 | if (prev) | |
943 | return NULL; | |
944 | return root; | |
945 | } | |
14067bb3 | 946 | |
9f3a0d09 | 947 | while (!memcg) { |
527a5ec9 | 948 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 949 | struct cgroup_subsys_state *css; |
711d3d2c | 950 | |
527a5ec9 JW |
951 | if (reclaim) { |
952 | int nid = zone_to_nid(reclaim->zone); | |
953 | int zid = zone_idx(reclaim->zone); | |
954 | struct mem_cgroup_per_zone *mz; | |
955 | ||
956 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
957 | iter = &mz->reclaim_iter[reclaim->priority]; | |
958 | if (prev && reclaim->generation != iter->generation) | |
959 | return NULL; | |
960 | id = iter->position; | |
961 | } | |
7d74b06f | 962 | |
9f3a0d09 JW |
963 | rcu_read_lock(); |
964 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
965 | if (css) { | |
966 | if (css == &root->css || css_tryget(css)) | |
967 | memcg = container_of(css, | |
968 | struct mem_cgroup, css); | |
969 | } else | |
970 | id = 0; | |
14067bb3 | 971 | rcu_read_unlock(); |
14067bb3 | 972 | |
527a5ec9 JW |
973 | if (reclaim) { |
974 | iter->position = id; | |
975 | if (!css) | |
976 | iter->generation++; | |
977 | else if (!prev && memcg) | |
978 | reclaim->generation = iter->generation; | |
979 | } | |
9f3a0d09 JW |
980 | |
981 | if (prev && !css) | |
982 | return NULL; | |
983 | } | |
984 | return memcg; | |
14067bb3 | 985 | } |
7d74b06f | 986 | |
5660048c JW |
987 | /** |
988 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
989 | * @root: hierarchy root | |
990 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
991 | */ | |
992 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
993 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
994 | { |
995 | if (!root) | |
996 | root = root_mem_cgroup; | |
997 | if (prev && prev != root) | |
998 | css_put(&prev->css); | |
999 | } | |
7d74b06f | 1000 | |
9f3a0d09 JW |
1001 | /* |
1002 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1003 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1004 | * be used for reference counting. | |
1005 | */ | |
1006 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1007 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1008 | iter != NULL; \ |
527a5ec9 | 1009 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1010 | |
9f3a0d09 | 1011 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1012 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1013 | iter != NULL; \ |
527a5ec9 | 1014 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1015 | |
c0ff4b85 | 1016 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
4b3bde4c | 1017 | { |
c0ff4b85 | 1018 | return (memcg == root_mem_cgroup); |
4b3bde4c BS |
1019 | } |
1020 | ||
456f998e YH |
1021 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
1022 | { | |
c0ff4b85 | 1023 | struct mem_cgroup *memcg; |
456f998e YH |
1024 | |
1025 | if (!mm) | |
1026 | return; | |
1027 | ||
1028 | rcu_read_lock(); | |
c0ff4b85 R |
1029 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1030 | if (unlikely(!memcg)) | |
456f998e YH |
1031 | goto out; |
1032 | ||
1033 | switch (idx) { | |
456f998e | 1034 | case PGFAULT: |
0e574a93 JW |
1035 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1036 | break; | |
1037 | case PGMAJFAULT: | |
1038 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1039 | break; |
1040 | default: | |
1041 | BUG(); | |
1042 | } | |
1043 | out: | |
1044 | rcu_read_unlock(); | |
1045 | } | |
1046 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
1047 | ||
925b7673 JW |
1048 | /** |
1049 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1050 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1051 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1052 | * |
1053 | * Returns the lru list vector holding pages for the given @zone and | |
1054 | * @mem. This can be the global zone lruvec, if the memory controller | |
1055 | * is disabled. | |
1056 | */ | |
1057 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1058 | struct mem_cgroup *memcg) | |
1059 | { | |
1060 | struct mem_cgroup_per_zone *mz; | |
1061 | ||
1062 | if (mem_cgroup_disabled()) | |
1063 | return &zone->lruvec; | |
1064 | ||
1065 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
1066 | return &mz->lruvec; | |
1067 | } | |
1068 | ||
08e552c6 KH |
1069 | /* |
1070 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1071 | * Operations are called by routine of global LRU independently from memcg. | |
1072 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1073 | * | |
1074 | * Changes to pc->mem_cgroup happens when | |
1075 | * 1. charge | |
1076 | * 2. moving account | |
1077 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1078 | * It is added to LRU before charge. | |
1079 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1080 | * When moving account, the page is not on LRU. It's isolated. | |
1081 | */ | |
4f98a2fe | 1082 | |
925b7673 | 1083 | /** |
fa9add64 | 1084 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1085 | * @page: the page |
fa9add64 | 1086 | * @zone: zone of the page |
925b7673 | 1087 | */ |
fa9add64 | 1088 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1089 | { |
08e552c6 | 1090 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1091 | struct mem_cgroup *memcg; |
1092 | struct page_cgroup *pc; | |
6d12e2d8 | 1093 | |
f8d66542 | 1094 | if (mem_cgroup_disabled()) |
925b7673 JW |
1095 | return &zone->lruvec; |
1096 | ||
08e552c6 | 1097 | pc = lookup_page_cgroup(page); |
38c5d72f | 1098 | memcg = pc->mem_cgroup; |
7512102c HD |
1099 | |
1100 | /* | |
fa9add64 | 1101 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1102 | * an uncharged page off lru does nothing to secure |
1103 | * its former mem_cgroup from sudden removal. | |
1104 | * | |
1105 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1106 | * under page_cgroup lock: between them, they make all uses | |
1107 | * of pc->mem_cgroup safe. | |
1108 | */ | |
fa9add64 | 1109 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1110 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1111 | ||
925b7673 | 1112 | mz = page_cgroup_zoneinfo(memcg, page); |
925b7673 | 1113 | return &mz->lruvec; |
08e552c6 | 1114 | } |
b69408e8 | 1115 | |
925b7673 | 1116 | /** |
fa9add64 HD |
1117 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1118 | * @lruvec: mem_cgroup per zone lru vector | |
1119 | * @lru: index of lru list the page is sitting on | |
1120 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1121 | * |
fa9add64 HD |
1122 | * This function must be called when a page is added to or removed from an |
1123 | * lru list. | |
3f58a829 | 1124 | */ |
fa9add64 HD |
1125 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1126 | int nr_pages) | |
3f58a829 MK |
1127 | { |
1128 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1129 | unsigned long *lru_size; |
3f58a829 MK |
1130 | |
1131 | if (mem_cgroup_disabled()) | |
1132 | return; | |
1133 | ||
fa9add64 HD |
1134 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1135 | lru_size = mz->lru_size + lru; | |
1136 | *lru_size += nr_pages; | |
1137 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1138 | } |
544122e5 | 1139 | |
3e92041d | 1140 | /* |
c0ff4b85 | 1141 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1142 | * hierarchy subtree |
1143 | */ | |
c3ac9a8a JW |
1144 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1145 | struct mem_cgroup *memcg) | |
3e92041d | 1146 | { |
91c63734 JW |
1147 | if (root_memcg == memcg) |
1148 | return true; | |
3a981f48 | 1149 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1150 | return false; |
c3ac9a8a JW |
1151 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1152 | } | |
1153 | ||
1154 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1155 | struct mem_cgroup *memcg) | |
1156 | { | |
1157 | bool ret; | |
1158 | ||
91c63734 | 1159 | rcu_read_lock(); |
c3ac9a8a | 1160 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1161 | rcu_read_unlock(); |
1162 | return ret; | |
3e92041d MH |
1163 | } |
1164 | ||
c0ff4b85 | 1165 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1166 | { |
1167 | int ret; | |
0b7f569e | 1168 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1169 | struct task_struct *p; |
4c4a2214 | 1170 | |
158e0a2d | 1171 | p = find_lock_task_mm(task); |
de077d22 DR |
1172 | if (p) { |
1173 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1174 | task_unlock(p); | |
1175 | } else { | |
1176 | /* | |
1177 | * All threads may have already detached their mm's, but the oom | |
1178 | * killer still needs to detect if they have already been oom | |
1179 | * killed to prevent needlessly killing additional tasks. | |
1180 | */ | |
1181 | task_lock(task); | |
1182 | curr = mem_cgroup_from_task(task); | |
1183 | if (curr) | |
1184 | css_get(&curr->css); | |
1185 | task_unlock(task); | |
1186 | } | |
0b7f569e KH |
1187 | if (!curr) |
1188 | return 0; | |
d31f56db | 1189 | /* |
c0ff4b85 | 1190 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1191 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1192 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1193 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1194 | */ |
c0ff4b85 | 1195 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1196 | css_put(&curr->css); |
4c4a2214 DR |
1197 | return ret; |
1198 | } | |
1199 | ||
c56d5c7d | 1200 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1201 | { |
9b272977 | 1202 | unsigned long inactive_ratio; |
14797e23 | 1203 | unsigned long inactive; |
9b272977 | 1204 | unsigned long active; |
c772be93 | 1205 | unsigned long gb; |
14797e23 | 1206 | |
4d7dcca2 HD |
1207 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1208 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1209 | |
c772be93 KM |
1210 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1211 | if (gb) | |
1212 | inactive_ratio = int_sqrt(10 * gb); | |
1213 | else | |
1214 | inactive_ratio = 1; | |
1215 | ||
9b272977 | 1216 | return inactive * inactive_ratio < active; |
14797e23 KM |
1217 | } |
1218 | ||
c56d5c7d | 1219 | int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 RR |
1220 | { |
1221 | unsigned long active; | |
1222 | unsigned long inactive; | |
1223 | ||
4d7dcca2 HD |
1224 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE); |
1225 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 RR |
1226 | |
1227 | return (active > inactive); | |
1228 | } | |
1229 | ||
6d61ef40 BS |
1230 | #define mem_cgroup_from_res_counter(counter, member) \ |
1231 | container_of(counter, struct mem_cgroup, member) | |
1232 | ||
19942822 | 1233 | /** |
9d11ea9f | 1234 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1235 | * @memcg: the memory cgroup |
19942822 | 1236 | * |
9d11ea9f | 1237 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1238 | * pages. |
19942822 | 1239 | */ |
c0ff4b85 | 1240 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1241 | { |
9d11ea9f JW |
1242 | unsigned long long margin; |
1243 | ||
c0ff4b85 | 1244 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1245 | if (do_swap_account) |
c0ff4b85 | 1246 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1247 | return margin >> PAGE_SHIFT; |
19942822 JW |
1248 | } |
1249 | ||
1f4c025b | 1250 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1251 | { |
1252 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1253 | |
1254 | /* root ? */ | |
1255 | if (cgrp->parent == NULL) | |
1256 | return vm_swappiness; | |
1257 | ||
bf1ff263 | 1258 | return memcg->swappiness; |
a7885eb8 KM |
1259 | } |
1260 | ||
619d094b KH |
1261 | /* |
1262 | * memcg->moving_account is used for checking possibility that some thread is | |
1263 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1264 | * a memcg, other threads should check memcg->moving_account under | |
1265 | * rcu_read_lock(), like this: | |
1266 | * | |
1267 | * CPU-A CPU-B | |
1268 | * rcu_read_lock() | |
1269 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1270 | * take heavy locks. | |
1271 | * synchronize_rcu() update something. | |
1272 | * rcu_read_unlock() | |
1273 | * start move here. | |
1274 | */ | |
4331f7d3 KH |
1275 | |
1276 | /* for quick checking without looking up memcg */ | |
1277 | atomic_t memcg_moving __read_mostly; | |
1278 | ||
c0ff4b85 | 1279 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1280 | { |
4331f7d3 | 1281 | atomic_inc(&memcg_moving); |
619d094b | 1282 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1283 | synchronize_rcu(); |
1284 | } | |
1285 | ||
c0ff4b85 | 1286 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1287 | { |
619d094b KH |
1288 | /* |
1289 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1290 | * We check NULL in callee rather than caller. | |
1291 | */ | |
4331f7d3 KH |
1292 | if (memcg) { |
1293 | atomic_dec(&memcg_moving); | |
619d094b | 1294 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1295 | } |
32047e2a | 1296 | } |
619d094b | 1297 | |
32047e2a KH |
1298 | /* |
1299 | * 2 routines for checking "mem" is under move_account() or not. | |
1300 | * | |
13fd1dd9 AM |
1301 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1302 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1303 | * pc->mem_cgroup may be overwritten. |
1304 | * | |
1305 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1306 | * under hierarchy of moving cgroups. This is for | |
1307 | * waiting at hith-memory prressure caused by "move". | |
1308 | */ | |
1309 | ||
13fd1dd9 | 1310 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1311 | { |
1312 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1313 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1314 | } |
4b534334 | 1315 | |
c0ff4b85 | 1316 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1317 | { |
2bd9bb20 KH |
1318 | struct mem_cgroup *from; |
1319 | struct mem_cgroup *to; | |
4b534334 | 1320 | bool ret = false; |
2bd9bb20 KH |
1321 | /* |
1322 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1323 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1324 | */ | |
1325 | spin_lock(&mc.lock); | |
1326 | from = mc.from; | |
1327 | to = mc.to; | |
1328 | if (!from) | |
1329 | goto unlock; | |
3e92041d | 1330 | |
c0ff4b85 R |
1331 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1332 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1333 | unlock: |
1334 | spin_unlock(&mc.lock); | |
4b534334 KH |
1335 | return ret; |
1336 | } | |
1337 | ||
c0ff4b85 | 1338 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1339 | { |
1340 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1341 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1342 | DEFINE_WAIT(wait); |
1343 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1344 | /* moving charge context might have finished. */ | |
1345 | if (mc.moving_task) | |
1346 | schedule(); | |
1347 | finish_wait(&mc.waitq, &wait); | |
1348 | return true; | |
1349 | } | |
1350 | } | |
1351 | return false; | |
1352 | } | |
1353 | ||
312734c0 KH |
1354 | /* |
1355 | * Take this lock when | |
1356 | * - a code tries to modify page's memcg while it's USED. | |
1357 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1358 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1359 | */ |
1360 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1361 | unsigned long *flags) | |
1362 | { | |
1363 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1364 | } | |
1365 | ||
1366 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1367 | unsigned long *flags) | |
1368 | { | |
1369 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1370 | } | |
1371 | ||
e222432b | 1372 | /** |
6a6135b6 | 1373 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1374 | * @memcg: The memory cgroup that went over limit |
1375 | * @p: Task that is going to be killed | |
1376 | * | |
1377 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1378 | * enabled | |
1379 | */ | |
1380 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1381 | { | |
1382 | struct cgroup *task_cgrp; | |
1383 | struct cgroup *mem_cgrp; | |
1384 | /* | |
1385 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1386 | * on the assumption that OOM is serialized for memory controller. | |
1387 | * If this assumption is broken, revisit this code. | |
1388 | */ | |
1389 | static char memcg_name[PATH_MAX]; | |
1390 | int ret; | |
1391 | ||
d31f56db | 1392 | if (!memcg || !p) |
e222432b BS |
1393 | return; |
1394 | ||
e222432b BS |
1395 | rcu_read_lock(); |
1396 | ||
1397 | mem_cgrp = memcg->css.cgroup; | |
1398 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1399 | ||
1400 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1401 | if (ret < 0) { | |
1402 | /* | |
1403 | * Unfortunately, we are unable to convert to a useful name | |
1404 | * But we'll still print out the usage information | |
1405 | */ | |
1406 | rcu_read_unlock(); | |
1407 | goto done; | |
1408 | } | |
1409 | rcu_read_unlock(); | |
1410 | ||
1411 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1412 | ||
1413 | rcu_read_lock(); | |
1414 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1415 | if (ret < 0) { | |
1416 | rcu_read_unlock(); | |
1417 | goto done; | |
1418 | } | |
1419 | rcu_read_unlock(); | |
1420 | ||
1421 | /* | |
1422 | * Continues from above, so we don't need an KERN_ level | |
1423 | */ | |
1424 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1425 | done: | |
1426 | ||
1427 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1428 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1429 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1430 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1431 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1432 | "failcnt %llu\n", | |
1433 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1434 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1435 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1436 | } | |
1437 | ||
81d39c20 KH |
1438 | /* |
1439 | * This function returns the number of memcg under hierarchy tree. Returns | |
1440 | * 1(self count) if no children. | |
1441 | */ | |
c0ff4b85 | 1442 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1443 | { |
1444 | int num = 0; | |
7d74b06f KH |
1445 | struct mem_cgroup *iter; |
1446 | ||
c0ff4b85 | 1447 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1448 | num++; |
81d39c20 KH |
1449 | return num; |
1450 | } | |
1451 | ||
a63d83f4 DR |
1452 | /* |
1453 | * Return the memory (and swap, if configured) limit for a memcg. | |
1454 | */ | |
9cbb78bb | 1455 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1456 | { |
1457 | u64 limit; | |
1458 | u64 memsw; | |
1459 | ||
f3e8eb70 JW |
1460 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1461 | limit += total_swap_pages << PAGE_SHIFT; | |
1462 | ||
a63d83f4 DR |
1463 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1464 | /* | |
1465 | * If memsw is finite and limits the amount of swap space available | |
1466 | * to this memcg, return that limit. | |
1467 | */ | |
1468 | return min(limit, memsw); | |
1469 | } | |
1470 | ||
876aafbf DR |
1471 | void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1472 | int order) | |
9cbb78bb DR |
1473 | { |
1474 | struct mem_cgroup *iter; | |
1475 | unsigned long chosen_points = 0; | |
1476 | unsigned long totalpages; | |
1477 | unsigned int points = 0; | |
1478 | struct task_struct *chosen = NULL; | |
1479 | ||
876aafbf DR |
1480 | /* |
1481 | * If current has a pending SIGKILL, then automatically select it. The | |
1482 | * goal is to allow it to allocate so that it may quickly exit and free | |
1483 | * its memory. | |
1484 | */ | |
1485 | if (fatal_signal_pending(current)) { | |
1486 | set_thread_flag(TIF_MEMDIE); | |
1487 | return; | |
1488 | } | |
1489 | ||
1490 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1491 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1492 | for_each_mem_cgroup_tree(iter, memcg) { | |
1493 | struct cgroup *cgroup = iter->css.cgroup; | |
1494 | struct cgroup_iter it; | |
1495 | struct task_struct *task; | |
1496 | ||
1497 | cgroup_iter_start(cgroup, &it); | |
1498 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
1499 | switch (oom_scan_process_thread(task, totalpages, NULL, | |
1500 | false)) { | |
1501 | case OOM_SCAN_SELECT: | |
1502 | if (chosen) | |
1503 | put_task_struct(chosen); | |
1504 | chosen = task; | |
1505 | chosen_points = ULONG_MAX; | |
1506 | get_task_struct(chosen); | |
1507 | /* fall through */ | |
1508 | case OOM_SCAN_CONTINUE: | |
1509 | continue; | |
1510 | case OOM_SCAN_ABORT: | |
1511 | cgroup_iter_end(cgroup, &it); | |
1512 | mem_cgroup_iter_break(memcg, iter); | |
1513 | if (chosen) | |
1514 | put_task_struct(chosen); | |
1515 | return; | |
1516 | case OOM_SCAN_OK: | |
1517 | break; | |
1518 | }; | |
1519 | points = oom_badness(task, memcg, NULL, totalpages); | |
1520 | if (points > chosen_points) { | |
1521 | if (chosen) | |
1522 | put_task_struct(chosen); | |
1523 | chosen = task; | |
1524 | chosen_points = points; | |
1525 | get_task_struct(chosen); | |
1526 | } | |
1527 | } | |
1528 | cgroup_iter_end(cgroup, &it); | |
1529 | } | |
1530 | ||
1531 | if (!chosen) | |
1532 | return; | |
1533 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1534 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1535 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1536 | } |
1537 | ||
5660048c JW |
1538 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1539 | gfp_t gfp_mask, | |
1540 | unsigned long flags) | |
1541 | { | |
1542 | unsigned long total = 0; | |
1543 | bool noswap = false; | |
1544 | int loop; | |
1545 | ||
1546 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1547 | noswap = true; | |
1548 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1549 | noswap = true; | |
1550 | ||
1551 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1552 | if (loop) | |
1553 | drain_all_stock_async(memcg); | |
1554 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1555 | /* | |
1556 | * Allow limit shrinkers, which are triggered directly | |
1557 | * by userspace, to catch signals and stop reclaim | |
1558 | * after minimal progress, regardless of the margin. | |
1559 | */ | |
1560 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1561 | break; | |
1562 | if (mem_cgroup_margin(memcg)) | |
1563 | break; | |
1564 | /* | |
1565 | * If nothing was reclaimed after two attempts, there | |
1566 | * may be no reclaimable pages in this hierarchy. | |
1567 | */ | |
1568 | if (loop && !total) | |
1569 | break; | |
1570 | } | |
1571 | return total; | |
1572 | } | |
1573 | ||
4d0c066d KH |
1574 | /** |
1575 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1576 | * @memcg: the target memcg |
4d0c066d KH |
1577 | * @nid: the node ID to be checked. |
1578 | * @noswap : specify true here if the user wants flle only information. | |
1579 | * | |
1580 | * This function returns whether the specified memcg contains any | |
1581 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1582 | * pages in the node. | |
1583 | */ | |
c0ff4b85 | 1584 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1585 | int nid, bool noswap) |
1586 | { | |
c0ff4b85 | 1587 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1588 | return true; |
1589 | if (noswap || !total_swap_pages) | |
1590 | return false; | |
c0ff4b85 | 1591 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1592 | return true; |
1593 | return false; | |
1594 | ||
1595 | } | |
889976db YH |
1596 | #if MAX_NUMNODES > 1 |
1597 | ||
1598 | /* | |
1599 | * Always updating the nodemask is not very good - even if we have an empty | |
1600 | * list or the wrong list here, we can start from some node and traverse all | |
1601 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1602 | * | |
1603 | */ | |
c0ff4b85 | 1604 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1605 | { |
1606 | int nid; | |
453a9bf3 KH |
1607 | /* |
1608 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1609 | * pagein/pageout changes since the last update. | |
1610 | */ | |
c0ff4b85 | 1611 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1612 | return; |
c0ff4b85 | 1613 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1614 | return; |
1615 | ||
889976db | 1616 | /* make a nodemask where this memcg uses memory from */ |
c0ff4b85 | 1617 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
889976db YH |
1618 | |
1619 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1620 | ||
c0ff4b85 R |
1621 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1622 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1623 | } |
453a9bf3 | 1624 | |
c0ff4b85 R |
1625 | atomic_set(&memcg->numainfo_events, 0); |
1626 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1627 | } |
1628 | ||
1629 | /* | |
1630 | * Selecting a node where we start reclaim from. Because what we need is just | |
1631 | * reducing usage counter, start from anywhere is O,K. Considering | |
1632 | * memory reclaim from current node, there are pros. and cons. | |
1633 | * | |
1634 | * Freeing memory from current node means freeing memory from a node which | |
1635 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1636 | * hit limits, it will see a contention on a node. But freeing from remote | |
1637 | * node means more costs for memory reclaim because of memory latency. | |
1638 | * | |
1639 | * Now, we use round-robin. Better algorithm is welcomed. | |
1640 | */ | |
c0ff4b85 | 1641 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1642 | { |
1643 | int node; | |
1644 | ||
c0ff4b85 R |
1645 | mem_cgroup_may_update_nodemask(memcg); |
1646 | node = memcg->last_scanned_node; | |
889976db | 1647 | |
c0ff4b85 | 1648 | node = next_node(node, memcg->scan_nodes); |
889976db | 1649 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1650 | node = first_node(memcg->scan_nodes); |
889976db YH |
1651 | /* |
1652 | * We call this when we hit limit, not when pages are added to LRU. | |
1653 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1654 | * memcg is too small and all pages are not on LRU. In that case, | |
1655 | * we use curret node. | |
1656 | */ | |
1657 | if (unlikely(node == MAX_NUMNODES)) | |
1658 | node = numa_node_id(); | |
1659 | ||
c0ff4b85 | 1660 | memcg->last_scanned_node = node; |
889976db YH |
1661 | return node; |
1662 | } | |
1663 | ||
4d0c066d KH |
1664 | /* |
1665 | * Check all nodes whether it contains reclaimable pages or not. | |
1666 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1667 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1668 | * enough new information. We need to do double check. | |
1669 | */ | |
6bbda35c | 1670 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1671 | { |
1672 | int nid; | |
1673 | ||
1674 | /* | |
1675 | * quick check...making use of scan_node. | |
1676 | * We can skip unused nodes. | |
1677 | */ | |
c0ff4b85 R |
1678 | if (!nodes_empty(memcg->scan_nodes)) { |
1679 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1680 | nid < MAX_NUMNODES; |
c0ff4b85 | 1681 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1682 | |
c0ff4b85 | 1683 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1684 | return true; |
1685 | } | |
1686 | } | |
1687 | /* | |
1688 | * Check rest of nodes. | |
1689 | */ | |
1690 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
c0ff4b85 | 1691 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1692 | continue; |
c0ff4b85 | 1693 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1694 | return true; |
1695 | } | |
1696 | return false; | |
1697 | } | |
1698 | ||
889976db | 1699 | #else |
c0ff4b85 | 1700 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1701 | { |
1702 | return 0; | |
1703 | } | |
4d0c066d | 1704 | |
6bbda35c | 1705 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1706 | { |
c0ff4b85 | 1707 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1708 | } |
889976db YH |
1709 | #endif |
1710 | ||
5660048c JW |
1711 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1712 | struct zone *zone, | |
1713 | gfp_t gfp_mask, | |
1714 | unsigned long *total_scanned) | |
6d61ef40 | 1715 | { |
9f3a0d09 | 1716 | struct mem_cgroup *victim = NULL; |
5660048c | 1717 | int total = 0; |
04046e1a | 1718 | int loop = 0; |
9d11ea9f | 1719 | unsigned long excess; |
185efc0f | 1720 | unsigned long nr_scanned; |
527a5ec9 JW |
1721 | struct mem_cgroup_reclaim_cookie reclaim = { |
1722 | .zone = zone, | |
1723 | .priority = 0, | |
1724 | }; | |
9d11ea9f | 1725 | |
c0ff4b85 | 1726 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1727 | |
4e416953 | 1728 | while (1) { |
527a5ec9 | 1729 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1730 | if (!victim) { |
04046e1a | 1731 | loop++; |
4e416953 BS |
1732 | if (loop >= 2) { |
1733 | /* | |
1734 | * If we have not been able to reclaim | |
1735 | * anything, it might because there are | |
1736 | * no reclaimable pages under this hierarchy | |
1737 | */ | |
5660048c | 1738 | if (!total) |
4e416953 | 1739 | break; |
4e416953 | 1740 | /* |
25985edc | 1741 | * We want to do more targeted reclaim. |
4e416953 BS |
1742 | * excess >> 2 is not to excessive so as to |
1743 | * reclaim too much, nor too less that we keep | |
1744 | * coming back to reclaim from this cgroup | |
1745 | */ | |
1746 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1747 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1748 | break; |
4e416953 | 1749 | } |
9f3a0d09 | 1750 | continue; |
4e416953 | 1751 | } |
5660048c | 1752 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1753 | continue; |
5660048c JW |
1754 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1755 | zone, &nr_scanned); | |
1756 | *total_scanned += nr_scanned; | |
1757 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1758 | break; |
6d61ef40 | 1759 | } |
9f3a0d09 | 1760 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1761 | return total; |
6d61ef40 BS |
1762 | } |
1763 | ||
867578cb KH |
1764 | /* |
1765 | * Check OOM-Killer is already running under our hierarchy. | |
1766 | * If someone is running, return false. | |
1af8efe9 | 1767 | * Has to be called with memcg_oom_lock |
867578cb | 1768 | */ |
c0ff4b85 | 1769 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1770 | { |
79dfdacc | 1771 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1772 | |
9f3a0d09 | 1773 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1774 | if (iter->oom_lock) { |
79dfdacc MH |
1775 | /* |
1776 | * this subtree of our hierarchy is already locked | |
1777 | * so we cannot give a lock. | |
1778 | */ | |
79dfdacc | 1779 | failed = iter; |
9f3a0d09 JW |
1780 | mem_cgroup_iter_break(memcg, iter); |
1781 | break; | |
23751be0 JW |
1782 | } else |
1783 | iter->oom_lock = true; | |
7d74b06f | 1784 | } |
867578cb | 1785 | |
79dfdacc | 1786 | if (!failed) |
23751be0 | 1787 | return true; |
79dfdacc MH |
1788 | |
1789 | /* | |
1790 | * OK, we failed to lock the whole subtree so we have to clean up | |
1791 | * what we set up to the failing subtree | |
1792 | */ | |
9f3a0d09 | 1793 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1794 | if (iter == failed) { |
9f3a0d09 JW |
1795 | mem_cgroup_iter_break(memcg, iter); |
1796 | break; | |
79dfdacc MH |
1797 | } |
1798 | iter->oom_lock = false; | |
1799 | } | |
23751be0 | 1800 | return false; |
a636b327 | 1801 | } |
0b7f569e | 1802 | |
79dfdacc | 1803 | /* |
1af8efe9 | 1804 | * Has to be called with memcg_oom_lock |
79dfdacc | 1805 | */ |
c0ff4b85 | 1806 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1807 | { |
7d74b06f KH |
1808 | struct mem_cgroup *iter; |
1809 | ||
c0ff4b85 | 1810 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1811 | iter->oom_lock = false; |
1812 | return 0; | |
1813 | } | |
1814 | ||
c0ff4b85 | 1815 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1816 | { |
1817 | struct mem_cgroup *iter; | |
1818 | ||
c0ff4b85 | 1819 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1820 | atomic_inc(&iter->under_oom); |
1821 | } | |
1822 | ||
c0ff4b85 | 1823 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1824 | { |
1825 | struct mem_cgroup *iter; | |
1826 | ||
867578cb KH |
1827 | /* |
1828 | * When a new child is created while the hierarchy is under oom, | |
1829 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1830 | * atomic_add_unless() here. | |
1831 | */ | |
c0ff4b85 | 1832 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1833 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1834 | } |
1835 | ||
1af8efe9 | 1836 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1837 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1838 | ||
dc98df5a | 1839 | struct oom_wait_info { |
d79154bb | 1840 | struct mem_cgroup *memcg; |
dc98df5a KH |
1841 | wait_queue_t wait; |
1842 | }; | |
1843 | ||
1844 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1845 | unsigned mode, int sync, void *arg) | |
1846 | { | |
d79154bb HD |
1847 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1848 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1849 | struct oom_wait_info *oom_wait_info; |
1850 | ||
1851 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1852 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1853 | |
dc98df5a | 1854 | /* |
d79154bb | 1855 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1856 | * Then we can use css_is_ancestor without taking care of RCU. |
1857 | */ | |
c0ff4b85 R |
1858 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1859 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1860 | return 0; |
dc98df5a KH |
1861 | return autoremove_wake_function(wait, mode, sync, arg); |
1862 | } | |
1863 | ||
c0ff4b85 | 1864 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1865 | { |
c0ff4b85 R |
1866 | /* for filtering, pass "memcg" as argument. */ |
1867 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1868 | } |
1869 | ||
c0ff4b85 | 1870 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1871 | { |
c0ff4b85 R |
1872 | if (memcg && atomic_read(&memcg->under_oom)) |
1873 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1874 | } |
1875 | ||
867578cb KH |
1876 | /* |
1877 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1878 | */ | |
6bbda35c KS |
1879 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
1880 | int order) | |
0b7f569e | 1881 | { |
dc98df5a | 1882 | struct oom_wait_info owait; |
3c11ecf4 | 1883 | bool locked, need_to_kill; |
867578cb | 1884 | |
d79154bb | 1885 | owait.memcg = memcg; |
dc98df5a KH |
1886 | owait.wait.flags = 0; |
1887 | owait.wait.func = memcg_oom_wake_function; | |
1888 | owait.wait.private = current; | |
1889 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1890 | need_to_kill = true; |
c0ff4b85 | 1891 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1892 | |
c0ff4b85 | 1893 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1894 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1895 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1896 | /* |
1897 | * Even if signal_pending(), we can't quit charge() loop without | |
1898 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1899 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1900 | */ | |
3c11ecf4 | 1901 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1902 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1903 | need_to_kill = false; |
1904 | if (locked) | |
c0ff4b85 | 1905 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1906 | spin_unlock(&memcg_oom_lock); |
867578cb | 1907 | |
3c11ecf4 KH |
1908 | if (need_to_kill) { |
1909 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 1910 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 1911 | } else { |
867578cb | 1912 | schedule(); |
dc98df5a | 1913 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1914 | } |
1af8efe9 | 1915 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1916 | if (locked) |
c0ff4b85 R |
1917 | mem_cgroup_oom_unlock(memcg); |
1918 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1919 | spin_unlock(&memcg_oom_lock); |
867578cb | 1920 | |
c0ff4b85 | 1921 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1922 | |
867578cb KH |
1923 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1924 | return false; | |
1925 | /* Give chance to dying process */ | |
715a5ee8 | 1926 | schedule_timeout_uninterruptible(1); |
867578cb | 1927 | return true; |
0b7f569e KH |
1928 | } |
1929 | ||
d69b042f BS |
1930 | /* |
1931 | * Currently used to update mapped file statistics, but the routine can be | |
1932 | * generalized to update other statistics as well. | |
32047e2a KH |
1933 | * |
1934 | * Notes: Race condition | |
1935 | * | |
1936 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1937 | * it tends to be costly. But considering some conditions, we doesn't need | |
1938 | * to do so _always_. | |
1939 | * | |
1940 | * Considering "charge", lock_page_cgroup() is not required because all | |
1941 | * file-stat operations happen after a page is attached to radix-tree. There | |
1942 | * are no race with "charge". | |
1943 | * | |
1944 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1945 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1946 | * if there are race with "uncharge". Statistics itself is properly handled | |
1947 | * by flags. | |
1948 | * | |
1949 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
1950 | * small, we check mm->moving_account and detect there are possibility of race |
1951 | * If there is, we take a lock. | |
d69b042f | 1952 | */ |
26174efd | 1953 | |
89c06bd5 KH |
1954 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
1955 | bool *locked, unsigned long *flags) | |
1956 | { | |
1957 | struct mem_cgroup *memcg; | |
1958 | struct page_cgroup *pc; | |
1959 | ||
1960 | pc = lookup_page_cgroup(page); | |
1961 | again: | |
1962 | memcg = pc->mem_cgroup; | |
1963 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
1964 | return; | |
1965 | /* | |
1966 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 1967 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 1968 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 1969 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 1970 | */ |
13fd1dd9 | 1971 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
1972 | return; |
1973 | ||
1974 | move_lock_mem_cgroup(memcg, flags); | |
1975 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
1976 | move_unlock_mem_cgroup(memcg, flags); | |
1977 | goto again; | |
1978 | } | |
1979 | *locked = true; | |
1980 | } | |
1981 | ||
1982 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
1983 | { | |
1984 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1985 | ||
1986 | /* | |
1987 | * It's guaranteed that pc->mem_cgroup never changes while | |
1988 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 1989 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
1990 | */ |
1991 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
1992 | } | |
1993 | ||
2a7106f2 GT |
1994 | void mem_cgroup_update_page_stat(struct page *page, |
1995 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 1996 | { |
c0ff4b85 | 1997 | struct mem_cgroup *memcg; |
32047e2a | 1998 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 1999 | unsigned long uninitialized_var(flags); |
d69b042f | 2000 | |
cfa44946 | 2001 | if (mem_cgroup_disabled()) |
d69b042f | 2002 | return; |
89c06bd5 | 2003 | |
c0ff4b85 R |
2004 | memcg = pc->mem_cgroup; |
2005 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2006 | return; |
26174efd | 2007 | |
26174efd | 2008 | switch (idx) { |
2a7106f2 | 2009 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2010 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2011 | break; |
2012 | default: | |
2013 | BUG(); | |
8725d541 | 2014 | } |
d69b042f | 2015 | |
c0ff4b85 | 2016 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2017 | } |
26174efd | 2018 | |
cdec2e42 KH |
2019 | /* |
2020 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2021 | * TODO: maybe necessary to use big numbers in big irons. | |
2022 | */ | |
7ec99d62 | 2023 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2024 | struct memcg_stock_pcp { |
2025 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2026 | unsigned int nr_pages; |
cdec2e42 | 2027 | struct work_struct work; |
26fe6168 | 2028 | unsigned long flags; |
a0db00fc | 2029 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2030 | }; |
2031 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2032 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
2033 | |
2034 | /* | |
11c9ea4e | 2035 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
2036 | * from local stock and true is returned. If the stock is 0 or charges from a |
2037 | * cgroup which is not current target, returns false. This stock will be | |
2038 | * refilled. | |
2039 | */ | |
c0ff4b85 | 2040 | static bool consume_stock(struct mem_cgroup *memcg) |
cdec2e42 KH |
2041 | { |
2042 | struct memcg_stock_pcp *stock; | |
2043 | bool ret = true; | |
2044 | ||
2045 | stock = &get_cpu_var(memcg_stock); | |
c0ff4b85 | 2046 | if (memcg == stock->cached && stock->nr_pages) |
11c9ea4e | 2047 | stock->nr_pages--; |
cdec2e42 KH |
2048 | else /* need to call res_counter_charge */ |
2049 | ret = false; | |
2050 | put_cpu_var(memcg_stock); | |
2051 | return ret; | |
2052 | } | |
2053 | ||
2054 | /* | |
2055 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2056 | */ | |
2057 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2058 | { | |
2059 | struct mem_cgroup *old = stock->cached; | |
2060 | ||
11c9ea4e JW |
2061 | if (stock->nr_pages) { |
2062 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2063 | ||
2064 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2065 | if (do_swap_account) |
11c9ea4e JW |
2066 | res_counter_uncharge(&old->memsw, bytes); |
2067 | stock->nr_pages = 0; | |
cdec2e42 KH |
2068 | } |
2069 | stock->cached = NULL; | |
cdec2e42 KH |
2070 | } |
2071 | ||
2072 | /* | |
2073 | * This must be called under preempt disabled or must be called by | |
2074 | * a thread which is pinned to local cpu. | |
2075 | */ | |
2076 | static void drain_local_stock(struct work_struct *dummy) | |
2077 | { | |
2078 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2079 | drain_stock(stock); | |
26fe6168 | 2080 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2081 | } |
2082 | ||
2083 | /* | |
2084 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2085 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2086 | */ |
c0ff4b85 | 2087 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2088 | { |
2089 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2090 | ||
c0ff4b85 | 2091 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2092 | drain_stock(stock); |
c0ff4b85 | 2093 | stock->cached = memcg; |
cdec2e42 | 2094 | } |
11c9ea4e | 2095 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2096 | put_cpu_var(memcg_stock); |
2097 | } | |
2098 | ||
2099 | /* | |
c0ff4b85 | 2100 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2101 | * of the hierarchy under it. sync flag says whether we should block |
2102 | * until the work is done. | |
cdec2e42 | 2103 | */ |
c0ff4b85 | 2104 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2105 | { |
26fe6168 | 2106 | int cpu, curcpu; |
d38144b7 | 2107 | |
cdec2e42 | 2108 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2109 | get_online_cpus(); |
5af12d0e | 2110 | curcpu = get_cpu(); |
cdec2e42 KH |
2111 | for_each_online_cpu(cpu) { |
2112 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2113 | struct mem_cgroup *memcg; |
26fe6168 | 2114 | |
c0ff4b85 R |
2115 | memcg = stock->cached; |
2116 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2117 | continue; |
c0ff4b85 | 2118 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2119 | continue; |
d1a05b69 MH |
2120 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2121 | if (cpu == curcpu) | |
2122 | drain_local_stock(&stock->work); | |
2123 | else | |
2124 | schedule_work_on(cpu, &stock->work); | |
2125 | } | |
cdec2e42 | 2126 | } |
5af12d0e | 2127 | put_cpu(); |
d38144b7 MH |
2128 | |
2129 | if (!sync) | |
2130 | goto out; | |
2131 | ||
2132 | for_each_online_cpu(cpu) { | |
2133 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2134 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2135 | flush_work(&stock->work); |
2136 | } | |
2137 | out: | |
cdec2e42 | 2138 | put_online_cpus(); |
d38144b7 MH |
2139 | } |
2140 | ||
2141 | /* | |
2142 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2143 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2144 | * expects some charges will be back to res_counter later but cannot wait for | |
2145 | * it. | |
2146 | */ | |
c0ff4b85 | 2147 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2148 | { |
9f50fad6 MH |
2149 | /* |
2150 | * If someone calls draining, avoid adding more kworker runs. | |
2151 | */ | |
2152 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2153 | return; | |
c0ff4b85 | 2154 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2155 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2156 | } |
2157 | ||
2158 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2159 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2160 | { |
2161 | /* called when force_empty is called */ | |
9f50fad6 | 2162 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2163 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2164 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2165 | } |
2166 | ||
711d3d2c KH |
2167 | /* |
2168 | * This function drains percpu counter value from DEAD cpu and | |
2169 | * move it to local cpu. Note that this function can be preempted. | |
2170 | */ | |
c0ff4b85 | 2171 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2172 | { |
2173 | int i; | |
2174 | ||
c0ff4b85 | 2175 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2176 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2177 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2178 | |
c0ff4b85 R |
2179 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2180 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2181 | } |
e9f8974f | 2182 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2183 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2184 | |
c0ff4b85 R |
2185 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2186 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2187 | } |
c0ff4b85 | 2188 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2189 | } |
2190 | ||
2191 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2192 | unsigned long action, |
2193 | void *hcpu) | |
2194 | { | |
2195 | int cpu = (unsigned long)hcpu; | |
2196 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2197 | struct mem_cgroup *iter; |
cdec2e42 | 2198 | |
619d094b | 2199 | if (action == CPU_ONLINE) |
1489ebad | 2200 | return NOTIFY_OK; |
1489ebad | 2201 | |
d833049b | 2202 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2203 | return NOTIFY_OK; |
711d3d2c | 2204 | |
9f3a0d09 | 2205 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2206 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2207 | ||
cdec2e42 KH |
2208 | stock = &per_cpu(memcg_stock, cpu); |
2209 | drain_stock(stock); | |
2210 | return NOTIFY_OK; | |
2211 | } | |
2212 | ||
4b534334 KH |
2213 | |
2214 | /* See __mem_cgroup_try_charge() for details */ | |
2215 | enum { | |
2216 | CHARGE_OK, /* success */ | |
2217 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2218 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2219 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2220 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2221 | }; | |
2222 | ||
c0ff4b85 | 2223 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
7ec99d62 | 2224 | unsigned int nr_pages, bool oom_check) |
4b534334 | 2225 | { |
7ec99d62 | 2226 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2227 | struct mem_cgroup *mem_over_limit; |
2228 | struct res_counter *fail_res; | |
2229 | unsigned long flags = 0; | |
2230 | int ret; | |
2231 | ||
c0ff4b85 | 2232 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2233 | |
2234 | if (likely(!ret)) { | |
2235 | if (!do_swap_account) | |
2236 | return CHARGE_OK; | |
c0ff4b85 | 2237 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2238 | if (likely(!ret)) |
2239 | return CHARGE_OK; | |
2240 | ||
c0ff4b85 | 2241 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2242 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2243 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2244 | } else | |
2245 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2246 | /* |
7ec99d62 JW |
2247 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2248 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2249 | * |
2250 | * Never reclaim on behalf of optional batching, retry with a | |
2251 | * single page instead. | |
2252 | */ | |
7ec99d62 | 2253 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2254 | return CHARGE_RETRY; |
2255 | ||
2256 | if (!(gfp_mask & __GFP_WAIT)) | |
2257 | return CHARGE_WOULDBLOCK; | |
2258 | ||
5660048c | 2259 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2260 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2261 | return CHARGE_RETRY; |
4b534334 | 2262 | /* |
19942822 JW |
2263 | * Even though the limit is exceeded at this point, reclaim |
2264 | * may have been able to free some pages. Retry the charge | |
2265 | * before killing the task. | |
2266 | * | |
2267 | * Only for regular pages, though: huge pages are rather | |
2268 | * unlikely to succeed so close to the limit, and we fall back | |
2269 | * to regular pages anyway in case of failure. | |
4b534334 | 2270 | */ |
7ec99d62 | 2271 | if (nr_pages == 1 && ret) |
4b534334 KH |
2272 | return CHARGE_RETRY; |
2273 | ||
2274 | /* | |
2275 | * At task move, charge accounts can be doubly counted. So, it's | |
2276 | * better to wait until the end of task_move if something is going on. | |
2277 | */ | |
2278 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2279 | return CHARGE_RETRY; | |
2280 | ||
2281 | /* If we don't need to call oom-killer at el, return immediately */ | |
2282 | if (!oom_check) | |
2283 | return CHARGE_NOMEM; | |
2284 | /* check OOM */ | |
e845e199 | 2285 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2286 | return CHARGE_OOM_DIE; |
2287 | ||
2288 | return CHARGE_RETRY; | |
2289 | } | |
2290 | ||
f817ed48 | 2291 | /* |
38c5d72f KH |
2292 | * __mem_cgroup_try_charge() does |
2293 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2294 | * 2. update res_counter | |
2295 | * 3. call memory reclaim if necessary. | |
2296 | * | |
2297 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2298 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2299 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2300 | * as possible without any hazards. 2: all pages should have a valid | |
2301 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2302 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2303 | * | |
2304 | * So __mem_cgroup_try_charge() will return | |
2305 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2306 | * -ENOMEM ... charge failure because of resource limits. | |
2307 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2308 | * | |
2309 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2310 | * the oom-killer can be invoked. | |
8a9f3ccd | 2311 | */ |
f817ed48 | 2312 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2313 | gfp_t gfp_mask, |
7ec99d62 | 2314 | unsigned int nr_pages, |
c0ff4b85 | 2315 | struct mem_cgroup **ptr, |
7ec99d62 | 2316 | bool oom) |
8a9f3ccd | 2317 | { |
7ec99d62 | 2318 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2319 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2320 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2321 | int ret; |
a636b327 | 2322 | |
867578cb KH |
2323 | /* |
2324 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2325 | * in system level. So, allow to go ahead dying process in addition to | |
2326 | * MEMDIE process. | |
2327 | */ | |
2328 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2329 | || fatal_signal_pending(current))) | |
2330 | goto bypass; | |
a636b327 | 2331 | |
8a9f3ccd | 2332 | /* |
3be91277 HD |
2333 | * We always charge the cgroup the mm_struct belongs to. |
2334 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2335 | * thread group leader migrates. It's possible that mm is not |
2336 | * set, if so charge the init_mm (happens for pagecache usage). | |
2337 | */ | |
c0ff4b85 | 2338 | if (!*ptr && !mm) |
38c5d72f | 2339 | *ptr = root_mem_cgroup; |
f75ca962 | 2340 | again: |
c0ff4b85 R |
2341 | if (*ptr) { /* css should be a valid one */ |
2342 | memcg = *ptr; | |
2343 | VM_BUG_ON(css_is_removed(&memcg->css)); | |
2344 | if (mem_cgroup_is_root(memcg)) | |
f75ca962 | 2345 | goto done; |
c0ff4b85 | 2346 | if (nr_pages == 1 && consume_stock(memcg)) |
f75ca962 | 2347 | goto done; |
c0ff4b85 | 2348 | css_get(&memcg->css); |
4b534334 | 2349 | } else { |
f75ca962 | 2350 | struct task_struct *p; |
54595fe2 | 2351 | |
f75ca962 KH |
2352 | rcu_read_lock(); |
2353 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2354 | /* |
ebb76ce1 | 2355 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2356 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2357 | * race with swapoff. Then, we have small risk of mis-accouning. |
2358 | * But such kind of mis-account by race always happens because | |
2359 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2360 | * small race, here. | |
2361 | * (*) swapoff at el will charge against mm-struct not against | |
2362 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2363 | */ |
c0ff4b85 | 2364 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2365 | if (!memcg) |
2366 | memcg = root_mem_cgroup; | |
2367 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2368 | rcu_read_unlock(); |
2369 | goto done; | |
2370 | } | |
c0ff4b85 | 2371 | if (nr_pages == 1 && consume_stock(memcg)) { |
f75ca962 KH |
2372 | /* |
2373 | * It seems dagerous to access memcg without css_get(). | |
2374 | * But considering how consume_stok works, it's not | |
2375 | * necessary. If consume_stock success, some charges | |
2376 | * from this memcg are cached on this cpu. So, we | |
2377 | * don't need to call css_get()/css_tryget() before | |
2378 | * calling consume_stock(). | |
2379 | */ | |
2380 | rcu_read_unlock(); | |
2381 | goto done; | |
2382 | } | |
2383 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2384 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2385 | rcu_read_unlock(); |
2386 | goto again; | |
2387 | } | |
2388 | rcu_read_unlock(); | |
2389 | } | |
8a9f3ccd | 2390 | |
4b534334 KH |
2391 | do { |
2392 | bool oom_check; | |
7a81b88c | 2393 | |
4b534334 | 2394 | /* If killed, bypass charge */ |
f75ca962 | 2395 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2396 | css_put(&memcg->css); |
4b534334 | 2397 | goto bypass; |
f75ca962 | 2398 | } |
6d61ef40 | 2399 | |
4b534334 KH |
2400 | oom_check = false; |
2401 | if (oom && !nr_oom_retries) { | |
2402 | oom_check = true; | |
2403 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2404 | } |
66e1707b | 2405 | |
c0ff4b85 | 2406 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
4b534334 KH |
2407 | switch (ret) { |
2408 | case CHARGE_OK: | |
2409 | break; | |
2410 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2411 | batch = nr_pages; |
c0ff4b85 R |
2412 | css_put(&memcg->css); |
2413 | memcg = NULL; | |
f75ca962 | 2414 | goto again; |
4b534334 | 2415 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2416 | css_put(&memcg->css); |
4b534334 KH |
2417 | goto nomem; |
2418 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2419 | if (!oom) { |
c0ff4b85 | 2420 | css_put(&memcg->css); |
867578cb | 2421 | goto nomem; |
f75ca962 | 2422 | } |
4b534334 KH |
2423 | /* If oom, we never return -ENOMEM */ |
2424 | nr_oom_retries--; | |
2425 | break; | |
2426 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2427 | css_put(&memcg->css); |
867578cb | 2428 | goto bypass; |
66e1707b | 2429 | } |
4b534334 KH |
2430 | } while (ret != CHARGE_OK); |
2431 | ||
7ec99d62 | 2432 | if (batch > nr_pages) |
c0ff4b85 R |
2433 | refill_stock(memcg, batch - nr_pages); |
2434 | css_put(&memcg->css); | |
0c3e73e8 | 2435 | done: |
c0ff4b85 | 2436 | *ptr = memcg; |
7a81b88c KH |
2437 | return 0; |
2438 | nomem: | |
c0ff4b85 | 2439 | *ptr = NULL; |
7a81b88c | 2440 | return -ENOMEM; |
867578cb | 2441 | bypass: |
38c5d72f KH |
2442 | *ptr = root_mem_cgroup; |
2443 | return -EINTR; | |
7a81b88c | 2444 | } |
8a9f3ccd | 2445 | |
a3032a2c DN |
2446 | /* |
2447 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2448 | * This function is for that and do uncharge, put css's refcnt. | |
2449 | * gotten by try_charge(). | |
2450 | */ | |
c0ff4b85 | 2451 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2452 | unsigned int nr_pages) |
a3032a2c | 2453 | { |
c0ff4b85 | 2454 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2455 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2456 | ||
c0ff4b85 | 2457 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2458 | if (do_swap_account) |
c0ff4b85 | 2459 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2460 | } |
854ffa8d DN |
2461 | } |
2462 | ||
d01dd17f KH |
2463 | /* |
2464 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2465 | * This is useful when moving usage to parent cgroup. | |
2466 | */ | |
2467 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2468 | unsigned int nr_pages) | |
2469 | { | |
2470 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2471 | ||
2472 | if (mem_cgroup_is_root(memcg)) | |
2473 | return; | |
2474 | ||
2475 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2476 | if (do_swap_account) | |
2477 | res_counter_uncharge_until(&memcg->memsw, | |
2478 | memcg->memsw.parent, bytes); | |
2479 | } | |
2480 | ||
a3b2d692 KH |
2481 | /* |
2482 | * A helper function to get mem_cgroup from ID. must be called under | |
2483 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2484 | * it's concern. (dropping refcnt from swap can be called against removed | |
2485 | * memcg.) | |
2486 | */ | |
2487 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2488 | { | |
2489 | struct cgroup_subsys_state *css; | |
2490 | ||
2491 | /* ID 0 is unused ID */ | |
2492 | if (!id) | |
2493 | return NULL; | |
2494 | css = css_lookup(&mem_cgroup_subsys, id); | |
2495 | if (!css) | |
2496 | return NULL; | |
2497 | return container_of(css, struct mem_cgroup, css); | |
2498 | } | |
2499 | ||
e42d9d5d | 2500 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2501 | { |
c0ff4b85 | 2502 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2503 | struct page_cgroup *pc; |
a3b2d692 | 2504 | unsigned short id; |
b5a84319 KH |
2505 | swp_entry_t ent; |
2506 | ||
3c776e64 DN |
2507 | VM_BUG_ON(!PageLocked(page)); |
2508 | ||
3c776e64 | 2509 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2510 | lock_page_cgroup(pc); |
a3b2d692 | 2511 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2512 | memcg = pc->mem_cgroup; |
2513 | if (memcg && !css_tryget(&memcg->css)) | |
2514 | memcg = NULL; | |
e42d9d5d | 2515 | } else if (PageSwapCache(page)) { |
3c776e64 | 2516 | ent.val = page_private(page); |
9fb4b7cc | 2517 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2518 | rcu_read_lock(); |
c0ff4b85 R |
2519 | memcg = mem_cgroup_lookup(id); |
2520 | if (memcg && !css_tryget(&memcg->css)) | |
2521 | memcg = NULL; | |
a3b2d692 | 2522 | rcu_read_unlock(); |
3c776e64 | 2523 | } |
c0bd3f63 | 2524 | unlock_page_cgroup(pc); |
c0ff4b85 | 2525 | return memcg; |
b5a84319 KH |
2526 | } |
2527 | ||
c0ff4b85 | 2528 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2529 | struct page *page, |
7ec99d62 | 2530 | unsigned int nr_pages, |
9ce70c02 HD |
2531 | enum charge_type ctype, |
2532 | bool lrucare) | |
7a81b88c | 2533 | { |
ce587e65 | 2534 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2535 | struct zone *uninitialized_var(zone); |
fa9add64 | 2536 | struct lruvec *lruvec; |
9ce70c02 | 2537 | bool was_on_lru = false; |
b2402857 | 2538 | bool anon; |
9ce70c02 | 2539 | |
ca3e0214 KH |
2540 | lock_page_cgroup(pc); |
2541 | if (unlikely(PageCgroupUsed(pc))) { | |
2542 | unlock_page_cgroup(pc); | |
c0ff4b85 | 2543 | __mem_cgroup_cancel_charge(memcg, nr_pages); |
ca3e0214 KH |
2544 | return; |
2545 | } | |
2546 | /* | |
2547 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2548 | * accessed by any other context at this point. | |
2549 | */ | |
9ce70c02 HD |
2550 | |
2551 | /* | |
2552 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2553 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2554 | */ | |
2555 | if (lrucare) { | |
2556 | zone = page_zone(page); | |
2557 | spin_lock_irq(&zone->lru_lock); | |
2558 | if (PageLRU(page)) { | |
fa9add64 | 2559 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2560 | ClearPageLRU(page); |
fa9add64 | 2561 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2562 | was_on_lru = true; |
2563 | } | |
2564 | } | |
2565 | ||
c0ff4b85 | 2566 | pc->mem_cgroup = memcg; |
261fb61a KH |
2567 | /* |
2568 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2569 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2570 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2571 | * before USED bit, we need memory barrier here. | |
2572 | * See mem_cgroup_add_lru_list(), etc. | |
2573 | */ | |
08e552c6 | 2574 | smp_wmb(); |
b2402857 | 2575 | SetPageCgroupUsed(pc); |
3be91277 | 2576 | |
9ce70c02 HD |
2577 | if (lrucare) { |
2578 | if (was_on_lru) { | |
fa9add64 | 2579 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2580 | VM_BUG_ON(PageLRU(page)); |
2581 | SetPageLRU(page); | |
fa9add64 | 2582 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2583 | } |
2584 | spin_unlock_irq(&zone->lru_lock); | |
2585 | } | |
2586 | ||
41326c17 | 2587 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2588 | anon = true; |
2589 | else | |
2590 | anon = false; | |
2591 | ||
2592 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2593 | unlock_page_cgroup(pc); |
9ce70c02 | 2594 | |
430e4863 KH |
2595 | /* |
2596 | * "charge_statistics" updated event counter. Then, check it. | |
2597 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2598 | * if they exceeds softlimit. | |
2599 | */ | |
c0ff4b85 | 2600 | memcg_check_events(memcg, page); |
7a81b88c | 2601 | } |
66e1707b | 2602 | |
ca3e0214 KH |
2603 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2604 | ||
a0db00fc | 2605 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
2606 | /* |
2607 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2608 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2609 | * charge/uncharge will be never happen and move_account() is done under | |
2610 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2611 | */ |
e94c8a9c | 2612 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
2613 | { |
2614 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
2615 | struct page_cgroup *pc; |
2616 | int i; | |
ca3e0214 | 2617 | |
3d37c4a9 KH |
2618 | if (mem_cgroup_disabled()) |
2619 | return; | |
e94c8a9c KH |
2620 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
2621 | pc = head_pc + i; | |
2622 | pc->mem_cgroup = head_pc->mem_cgroup; | |
2623 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
2624 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2625 | } | |
ca3e0214 | 2626 | } |
12d27107 | 2627 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2628 | |
f817ed48 | 2629 | /** |
de3638d9 | 2630 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2631 | * @page: the page |
7ec99d62 | 2632 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2633 | * @pc: page_cgroup of the page. |
2634 | * @from: mem_cgroup which the page is moved from. | |
2635 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2636 | * | |
2637 | * The caller must confirm following. | |
08e552c6 | 2638 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2639 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2640 | * |
2f3479b1 KH |
2641 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2642 | * from old cgroup. | |
f817ed48 | 2643 | */ |
7ec99d62 JW |
2644 | static int mem_cgroup_move_account(struct page *page, |
2645 | unsigned int nr_pages, | |
2646 | struct page_cgroup *pc, | |
2647 | struct mem_cgroup *from, | |
2f3479b1 | 2648 | struct mem_cgroup *to) |
f817ed48 | 2649 | { |
de3638d9 JW |
2650 | unsigned long flags; |
2651 | int ret; | |
b2402857 | 2652 | bool anon = PageAnon(page); |
987eba66 | 2653 | |
f817ed48 | 2654 | VM_BUG_ON(from == to); |
5564e88b | 2655 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2656 | /* |
2657 | * The page is isolated from LRU. So, collapse function | |
2658 | * will not handle this page. But page splitting can happen. | |
2659 | * Do this check under compound_page_lock(). The caller should | |
2660 | * hold it. | |
2661 | */ | |
2662 | ret = -EBUSY; | |
7ec99d62 | 2663 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2664 | goto out; |
2665 | ||
2666 | lock_page_cgroup(pc); | |
2667 | ||
2668 | ret = -EINVAL; | |
2669 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2670 | goto unlock; | |
2671 | ||
312734c0 | 2672 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 2673 | |
2ff76f11 | 2674 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
2675 | /* Update mapped_file data for mem_cgroup */ |
2676 | preempt_disable(); | |
2677 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2678 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2679 | preempt_enable(); | |
d69b042f | 2680 | } |
b2402857 | 2681 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
d69b042f | 2682 | |
854ffa8d | 2683 | /* caller should have done css_get */ |
08e552c6 | 2684 | pc->mem_cgroup = to; |
b2402857 | 2685 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
88703267 KH |
2686 | /* |
2687 | * We charges against "to" which may not have any tasks. Then, "to" | |
2688 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2689 | * this function is just force_empty() and move charge, so it's |
25985edc | 2690 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2691 | * status here. |
88703267 | 2692 | */ |
312734c0 | 2693 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
2694 | ret = 0; |
2695 | unlock: | |
57f9fd7d | 2696 | unlock_page_cgroup(pc); |
d2265e6f KH |
2697 | /* |
2698 | * check events | |
2699 | */ | |
5564e88b JW |
2700 | memcg_check_events(to, page); |
2701 | memcg_check_events(from, page); | |
de3638d9 | 2702 | out: |
f817ed48 KH |
2703 | return ret; |
2704 | } | |
2705 | ||
2706 | /* | |
2707 | * move charges to its parent. | |
2708 | */ | |
2709 | ||
5564e88b JW |
2710 | static int mem_cgroup_move_parent(struct page *page, |
2711 | struct page_cgroup *pc, | |
6068bf01 | 2712 | struct mem_cgroup *child) |
f817ed48 | 2713 | { |
f817ed48 | 2714 | struct mem_cgroup *parent; |
7ec99d62 | 2715 | unsigned int nr_pages; |
4be4489f | 2716 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2717 | int ret; |
2718 | ||
2719 | /* Is ROOT ? */ | |
cc926f78 | 2720 | if (mem_cgroup_is_root(child)) |
f817ed48 KH |
2721 | return -EINVAL; |
2722 | ||
57f9fd7d DN |
2723 | ret = -EBUSY; |
2724 | if (!get_page_unless_zero(page)) | |
2725 | goto out; | |
2726 | if (isolate_lru_page(page)) | |
2727 | goto put; | |
52dbb905 | 2728 | |
7ec99d62 | 2729 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2730 | |
cc926f78 KH |
2731 | parent = parent_mem_cgroup(child); |
2732 | /* | |
2733 | * If no parent, move charges to root cgroup. | |
2734 | */ | |
2735 | if (!parent) | |
2736 | parent = root_mem_cgroup; | |
f817ed48 | 2737 | |
7ec99d62 | 2738 | if (nr_pages > 1) |
987eba66 KH |
2739 | flags = compound_lock_irqsave(page); |
2740 | ||
cc926f78 | 2741 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 2742 | pc, child, parent); |
cc926f78 KH |
2743 | if (!ret) |
2744 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 2745 | |
7ec99d62 | 2746 | if (nr_pages > 1) |
987eba66 | 2747 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 2748 | putback_lru_page(page); |
57f9fd7d | 2749 | put: |
40d58138 | 2750 | put_page(page); |
57f9fd7d | 2751 | out: |
f817ed48 KH |
2752 | return ret; |
2753 | } | |
2754 | ||
7a81b88c KH |
2755 | /* |
2756 | * Charge the memory controller for page usage. | |
2757 | * Return | |
2758 | * 0 if the charge was successful | |
2759 | * < 0 if the cgroup is over its limit | |
2760 | */ | |
2761 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2762 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2763 | { |
c0ff4b85 | 2764 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2765 | unsigned int nr_pages = 1; |
8493ae43 | 2766 | bool oom = true; |
7a81b88c | 2767 | int ret; |
ec168510 | 2768 | |
37c2ac78 | 2769 | if (PageTransHuge(page)) { |
7ec99d62 | 2770 | nr_pages <<= compound_order(page); |
37c2ac78 | 2771 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2772 | /* |
2773 | * Never OOM-kill a process for a huge page. The | |
2774 | * fault handler will fall back to regular pages. | |
2775 | */ | |
2776 | oom = false; | |
37c2ac78 | 2777 | } |
7a81b88c | 2778 | |
c0ff4b85 | 2779 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 2780 | if (ret == -ENOMEM) |
7a81b88c | 2781 | return ret; |
ce587e65 | 2782 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 2783 | return 0; |
8a9f3ccd BS |
2784 | } |
2785 | ||
7a81b88c KH |
2786 | int mem_cgroup_newpage_charge(struct page *page, |
2787 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2788 | { |
f8d66542 | 2789 | if (mem_cgroup_disabled()) |
cede86ac | 2790 | return 0; |
7a0524cf JW |
2791 | VM_BUG_ON(page_mapped(page)); |
2792 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
2793 | VM_BUG_ON(!mm); | |
217bc319 | 2794 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 2795 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
2796 | } |
2797 | ||
54595fe2 KH |
2798 | /* |
2799 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2800 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2801 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2802 | * "commit()" or removed by "cancel()" |
2803 | */ | |
8c7c6e34 KH |
2804 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2805 | struct page *page, | |
72835c86 | 2806 | gfp_t mask, struct mem_cgroup **memcgp) |
8c7c6e34 | 2807 | { |
c0ff4b85 | 2808 | struct mem_cgroup *memcg; |
54595fe2 | 2809 | int ret; |
8c7c6e34 | 2810 | |
72835c86 | 2811 | *memcgp = NULL; |
56039efa | 2812 | |
f8d66542 | 2813 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2814 | return 0; |
2815 | ||
2816 | if (!do_swap_account) | |
2817 | goto charge_cur_mm; | |
8c7c6e34 KH |
2818 | /* |
2819 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2820 | * the pte, and even removed page from swap cache: in those cases |
2821 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2822 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2823 | */ |
2824 | if (!PageSwapCache(page)) | |
407f9c8b | 2825 | goto charge_cur_mm; |
c0ff4b85 R |
2826 | memcg = try_get_mem_cgroup_from_page(page); |
2827 | if (!memcg) | |
54595fe2 | 2828 | goto charge_cur_mm; |
72835c86 JW |
2829 | *memcgp = memcg; |
2830 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 2831 | css_put(&memcg->css); |
38c5d72f KH |
2832 | if (ret == -EINTR) |
2833 | ret = 0; | |
54595fe2 | 2834 | return ret; |
8c7c6e34 KH |
2835 | charge_cur_mm: |
2836 | if (unlikely(!mm)) | |
2837 | mm = &init_mm; | |
38c5d72f KH |
2838 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
2839 | if (ret == -EINTR) | |
2840 | ret = 0; | |
2841 | return ret; | |
8c7c6e34 KH |
2842 | } |
2843 | ||
827a03d2 JW |
2844 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
2845 | { | |
2846 | if (mem_cgroup_disabled()) | |
2847 | return; | |
2848 | if (!memcg) | |
2849 | return; | |
2850 | __mem_cgroup_cancel_charge(memcg, 1); | |
2851 | } | |
2852 | ||
83aae4c7 | 2853 | static void |
72835c86 | 2854 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 2855 | enum charge_type ctype) |
7a81b88c | 2856 | { |
f8d66542 | 2857 | if (mem_cgroup_disabled()) |
7a81b88c | 2858 | return; |
72835c86 | 2859 | if (!memcg) |
7a81b88c | 2860 | return; |
72835c86 | 2861 | cgroup_exclude_rmdir(&memcg->css); |
5a6475a4 | 2862 | |
ce587e65 | 2863 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
2864 | /* |
2865 | * Now swap is on-memory. This means this page may be | |
2866 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2867 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2868 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2869 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2870 | */ |
03f3c433 | 2871 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2872 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 2873 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 2874 | } |
88703267 KH |
2875 | /* |
2876 | * At swapin, we may charge account against cgroup which has no tasks. | |
2877 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2878 | * In that case, we need to call pre_destroy() again. check it here. | |
2879 | */ | |
72835c86 | 2880 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
7a81b88c KH |
2881 | } |
2882 | ||
72835c86 JW |
2883 | void mem_cgroup_commit_charge_swapin(struct page *page, |
2884 | struct mem_cgroup *memcg) | |
83aae4c7 | 2885 | { |
72835c86 | 2886 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 2887 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
2888 | } |
2889 | ||
827a03d2 JW |
2890 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2891 | gfp_t gfp_mask) | |
7a81b88c | 2892 | { |
827a03d2 JW |
2893 | struct mem_cgroup *memcg = NULL; |
2894 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2895 | int ret; | |
2896 | ||
f8d66542 | 2897 | if (mem_cgroup_disabled()) |
827a03d2 JW |
2898 | return 0; |
2899 | if (PageCompound(page)) | |
2900 | return 0; | |
2901 | ||
2902 | if (unlikely(!mm)) | |
2903 | mm = &init_mm; | |
827a03d2 JW |
2904 | |
2905 | if (!PageSwapCache(page)) | |
2906 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
2907 | else { /* page is swapcache/shmem */ | |
2908 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg); | |
2909 | if (!ret) | |
2910 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
2911 | } | |
2912 | return ret; | |
7a81b88c KH |
2913 | } |
2914 | ||
c0ff4b85 | 2915 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2916 | unsigned int nr_pages, |
2917 | const enum charge_type ctype) | |
569b846d KH |
2918 | { |
2919 | struct memcg_batch_info *batch = NULL; | |
2920 | bool uncharge_memsw = true; | |
7ec99d62 | 2921 | |
569b846d KH |
2922 | /* If swapout, usage of swap doesn't decrease */ |
2923 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2924 | uncharge_memsw = false; | |
569b846d KH |
2925 | |
2926 | batch = ¤t->memcg_batch; | |
2927 | /* | |
2928 | * In usual, we do css_get() when we remember memcg pointer. | |
2929 | * But in this case, we keep res->usage until end of a series of | |
2930 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2931 | */ | |
2932 | if (!batch->memcg) | |
c0ff4b85 | 2933 | batch->memcg = memcg; |
3c11ecf4 KH |
2934 | /* |
2935 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2936 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2937 | * the same cgroup and we have chance to coalesce uncharges. |
2938 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2939 | * because we want to do uncharge as soon as possible. | |
2940 | */ | |
2941 | ||
2942 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2943 | goto direct_uncharge; | |
2944 | ||
7ec99d62 | 2945 | if (nr_pages > 1) |
ec168510 AA |
2946 | goto direct_uncharge; |
2947 | ||
569b846d KH |
2948 | /* |
2949 | * In typical case, batch->memcg == mem. This means we can | |
2950 | * merge a series of uncharges to an uncharge of res_counter. | |
2951 | * If not, we uncharge res_counter ony by one. | |
2952 | */ | |
c0ff4b85 | 2953 | if (batch->memcg != memcg) |
569b846d KH |
2954 | goto direct_uncharge; |
2955 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2956 | batch->nr_pages++; |
569b846d | 2957 | if (uncharge_memsw) |
7ffd4ca7 | 2958 | batch->memsw_nr_pages++; |
569b846d KH |
2959 | return; |
2960 | direct_uncharge: | |
c0ff4b85 | 2961 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 2962 | if (uncharge_memsw) |
c0ff4b85 R |
2963 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
2964 | if (unlikely(batch->memcg != memcg)) | |
2965 | memcg_oom_recover(memcg); | |
569b846d | 2966 | } |
7a81b88c | 2967 | |
8a9f3ccd | 2968 | /* |
69029cd5 | 2969 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2970 | */ |
8c7c6e34 | 2971 | static struct mem_cgroup * |
0030f535 JW |
2972 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
2973 | bool end_migration) | |
8a9f3ccd | 2974 | { |
c0ff4b85 | 2975 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
2976 | unsigned int nr_pages = 1; |
2977 | struct page_cgroup *pc; | |
b2402857 | 2978 | bool anon; |
8a9f3ccd | 2979 | |
f8d66542 | 2980 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2981 | return NULL; |
4077960e | 2982 | |
0c59b89c | 2983 | VM_BUG_ON(PageSwapCache(page)); |
d13d1443 | 2984 | |
37c2ac78 | 2985 | if (PageTransHuge(page)) { |
7ec99d62 | 2986 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
2987 | VM_BUG_ON(!PageTransHuge(page)); |
2988 | } | |
8697d331 | 2989 | /* |
3c541e14 | 2990 | * Check if our page_cgroup is valid |
8697d331 | 2991 | */ |
52d4b9ac | 2992 | pc = lookup_page_cgroup(page); |
cfa44946 | 2993 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 2994 | return NULL; |
b9c565d5 | 2995 | |
52d4b9ac | 2996 | lock_page_cgroup(pc); |
d13d1443 | 2997 | |
c0ff4b85 | 2998 | memcg = pc->mem_cgroup; |
8c7c6e34 | 2999 | |
d13d1443 KH |
3000 | if (!PageCgroupUsed(pc)) |
3001 | goto unlock_out; | |
3002 | ||
b2402857 KH |
3003 | anon = PageAnon(page); |
3004 | ||
d13d1443 | 3005 | switch (ctype) { |
41326c17 | 3006 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
3007 | /* |
3008 | * Generally PageAnon tells if it's the anon statistics to be | |
3009 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
3010 | * used before page reached the stage of being marked PageAnon. | |
3011 | */ | |
b2402857 KH |
3012 | anon = true; |
3013 | /* fallthrough */ | |
8a9478ca | 3014 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 3015 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
3016 | if (page_mapped(page)) |
3017 | goto unlock_out; | |
3018 | /* | |
3019 | * Pages under migration may not be uncharged. But | |
3020 | * end_migration() /must/ be the one uncharging the | |
3021 | * unused post-migration page and so it has to call | |
3022 | * here with the migration bit still set. See the | |
3023 | * res_counter handling below. | |
3024 | */ | |
3025 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
3026 | goto unlock_out; |
3027 | break; | |
3028 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
3029 | if (!PageAnon(page)) { /* Shared memory */ | |
3030 | if (page->mapping && !page_is_file_cache(page)) | |
3031 | goto unlock_out; | |
3032 | } else if (page_mapped(page)) /* Anon */ | |
3033 | goto unlock_out; | |
3034 | break; | |
3035 | default: | |
3036 | break; | |
52d4b9ac | 3037 | } |
d13d1443 | 3038 | |
b2402857 | 3039 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 3040 | |
52d4b9ac | 3041 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3042 | /* |
3043 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3044 | * freed from LRU. This is safe because uncharged page is expected not | |
3045 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3046 | * special functions. | |
3047 | */ | |
b9c565d5 | 3048 | |
52d4b9ac | 3049 | unlock_page_cgroup(pc); |
f75ca962 | 3050 | /* |
c0ff4b85 | 3051 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
3052 | * will never be freed. |
3053 | */ | |
c0ff4b85 | 3054 | memcg_check_events(memcg, page); |
f75ca962 | 3055 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
3056 | mem_cgroup_swap_statistics(memcg, true); |
3057 | mem_cgroup_get(memcg); | |
f75ca962 | 3058 | } |
0030f535 JW |
3059 | /* |
3060 | * Migration does not charge the res_counter for the | |
3061 | * replacement page, so leave it alone when phasing out the | |
3062 | * page that is unused after the migration. | |
3063 | */ | |
3064 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 3065 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 3066 | |
c0ff4b85 | 3067 | return memcg; |
d13d1443 KH |
3068 | |
3069 | unlock_out: | |
3070 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3071 | return NULL; |
3c541e14 BS |
3072 | } |
3073 | ||
69029cd5 KH |
3074 | void mem_cgroup_uncharge_page(struct page *page) |
3075 | { | |
52d4b9ac KH |
3076 | /* early check. */ |
3077 | if (page_mapped(page)) | |
3078 | return; | |
40f23a21 | 3079 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
0c59b89c JW |
3080 | if (PageSwapCache(page)) |
3081 | return; | |
0030f535 | 3082 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
3083 | } |
3084 | ||
3085 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3086 | { | |
3087 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3088 | VM_BUG_ON(page->mapping); |
0030f535 | 3089 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
3090 | } |
3091 | ||
569b846d KH |
3092 | /* |
3093 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3094 | * In that cases, pages are freed continuously and we can expect pages | |
3095 | * are in the same memcg. All these calls itself limits the number of | |
3096 | * pages freed at once, then uncharge_start/end() is called properly. | |
3097 | * This may be called prural(2) times in a context, | |
3098 | */ | |
3099 | ||
3100 | void mem_cgroup_uncharge_start(void) | |
3101 | { | |
3102 | current->memcg_batch.do_batch++; | |
3103 | /* We can do nest. */ | |
3104 | if (current->memcg_batch.do_batch == 1) { | |
3105 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3106 | current->memcg_batch.nr_pages = 0; |
3107 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3108 | } |
3109 | } | |
3110 | ||
3111 | void mem_cgroup_uncharge_end(void) | |
3112 | { | |
3113 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3114 | ||
3115 | if (!batch->do_batch) | |
3116 | return; | |
3117 | ||
3118 | batch->do_batch--; | |
3119 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3120 | return; | |
3121 | ||
3122 | if (!batch->memcg) | |
3123 | return; | |
3124 | /* | |
3125 | * This "batch->memcg" is valid without any css_get/put etc... | |
3126 | * bacause we hide charges behind us. | |
3127 | */ | |
7ffd4ca7 JW |
3128 | if (batch->nr_pages) |
3129 | res_counter_uncharge(&batch->memcg->res, | |
3130 | batch->nr_pages * PAGE_SIZE); | |
3131 | if (batch->memsw_nr_pages) | |
3132 | res_counter_uncharge(&batch->memcg->memsw, | |
3133 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3134 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3135 | /* forget this pointer (for sanity check) */ |
3136 | batch->memcg = NULL; | |
3137 | } | |
3138 | ||
e767e056 | 3139 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3140 | /* |
e767e056 | 3141 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3142 | * memcg information is recorded to swap_cgroup of "ent" |
3143 | */ | |
8a9478ca KH |
3144 | void |
3145 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3146 | { |
3147 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3148 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3149 | ||
3150 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3151 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3152 | ||
0030f535 | 3153 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 3154 | |
f75ca962 KH |
3155 | /* |
3156 | * record memcg information, if swapout && memcg != NULL, | |
3157 | * mem_cgroup_get() was called in uncharge(). | |
3158 | */ | |
3159 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3160 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3161 | } |
e767e056 | 3162 | #endif |
8c7c6e34 | 3163 | |
c255a458 | 3164 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
3165 | /* |
3166 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3167 | * uncharge "memsw" account. | |
3168 | */ | |
3169 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3170 | { |
8c7c6e34 | 3171 | struct mem_cgroup *memcg; |
a3b2d692 | 3172 | unsigned short id; |
8c7c6e34 KH |
3173 | |
3174 | if (!do_swap_account) | |
3175 | return; | |
3176 | ||
a3b2d692 KH |
3177 | id = swap_cgroup_record(ent, 0); |
3178 | rcu_read_lock(); | |
3179 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3180 | if (memcg) { |
a3b2d692 KH |
3181 | /* |
3182 | * We uncharge this because swap is freed. | |
3183 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3184 | */ | |
0c3e73e8 | 3185 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3186 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3187 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3188 | mem_cgroup_put(memcg); |
3189 | } | |
a3b2d692 | 3190 | rcu_read_unlock(); |
d13d1443 | 3191 | } |
02491447 DN |
3192 | |
3193 | /** | |
3194 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3195 | * @entry: swap entry to be moved | |
3196 | * @from: mem_cgroup which the entry is moved from | |
3197 | * @to: mem_cgroup which the entry is moved to | |
3198 | * | |
3199 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3200 | * as the mem_cgroup's id of @from. | |
3201 | * | |
3202 | * Returns 0 on success, -EINVAL on failure. | |
3203 | * | |
3204 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3205 | * both res and memsw, and called css_get(). | |
3206 | */ | |
3207 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3208 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3209 | { |
3210 | unsigned short old_id, new_id; | |
3211 | ||
3212 | old_id = css_id(&from->css); | |
3213 | new_id = css_id(&to->css); | |
3214 | ||
3215 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3216 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3217 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3218 | /* |
483c30b5 DN |
3219 | * This function is only called from task migration context now. |
3220 | * It postpones res_counter and refcount handling till the end | |
3221 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3222 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3223 | * because if the process that has been moved to @to does | |
3224 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3225 | */ |
02491447 | 3226 | mem_cgroup_get(to); |
02491447 DN |
3227 | return 0; |
3228 | } | |
3229 | return -EINVAL; | |
3230 | } | |
3231 | #else | |
3232 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3233 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3234 | { |
3235 | return -EINVAL; | |
3236 | } | |
8c7c6e34 | 3237 | #endif |
d13d1443 | 3238 | |
ae41be37 | 3239 | /* |
01b1ae63 KH |
3240 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3241 | * page belongs to. | |
ae41be37 | 3242 | */ |
0030f535 JW |
3243 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
3244 | struct mem_cgroup **memcgp) | |
ae41be37 | 3245 | { |
c0ff4b85 | 3246 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3247 | struct page_cgroup *pc; |
ac39cf8c | 3248 | enum charge_type ctype; |
8869b8f6 | 3249 | |
72835c86 | 3250 | *memcgp = NULL; |
56039efa | 3251 | |
ec168510 | 3252 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3253 | if (mem_cgroup_disabled()) |
0030f535 | 3254 | return; |
4077960e | 3255 | |
52d4b9ac KH |
3256 | pc = lookup_page_cgroup(page); |
3257 | lock_page_cgroup(pc); | |
3258 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3259 | memcg = pc->mem_cgroup; |
3260 | css_get(&memcg->css); | |
ac39cf8c | 3261 | /* |
3262 | * At migrating an anonymous page, its mapcount goes down | |
3263 | * to 0 and uncharge() will be called. But, even if it's fully | |
3264 | * unmapped, migration may fail and this page has to be | |
3265 | * charged again. We set MIGRATION flag here and delay uncharge | |
3266 | * until end_migration() is called | |
3267 | * | |
3268 | * Corner Case Thinking | |
3269 | * A) | |
3270 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3271 | * while migration was ongoing. | |
3272 | * If unmap finds the old page, uncharge() of it will be delayed | |
3273 | * until end_migration(). If unmap finds a new page, it's | |
3274 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3275 | * finds swap_migration_entry, the new page will not be mapped | |
3276 | * and end_migration() will find it(mapcount==0). | |
3277 | * | |
3278 | * B) | |
3279 | * When the old page was mapped but migraion fails, the kernel | |
3280 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3281 | * if mapcount goes down to 0. We can do remap successfully | |
3282 | * without charging it again. | |
3283 | * | |
3284 | * C) | |
3285 | * The "old" page is under lock_page() until the end of | |
3286 | * migration, so, the old page itself will not be swapped-out. | |
3287 | * If the new page is swapped out before end_migraton, our | |
3288 | * hook to usual swap-out path will catch the event. | |
3289 | */ | |
3290 | if (PageAnon(page)) | |
3291 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3292 | } |
52d4b9ac | 3293 | unlock_page_cgroup(pc); |
ac39cf8c | 3294 | /* |
3295 | * If the page is not charged at this point, | |
3296 | * we return here. | |
3297 | */ | |
c0ff4b85 | 3298 | if (!memcg) |
0030f535 | 3299 | return; |
01b1ae63 | 3300 | |
72835c86 | 3301 | *memcgp = memcg; |
ac39cf8c | 3302 | /* |
3303 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3304 | * is called before end_migration, we can catch all events on this new | |
3305 | * page. In the case new page is migrated but not remapped, new page's | |
3306 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3307 | */ | |
ac39cf8c | 3308 | if (PageAnon(page)) |
41326c17 | 3309 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 3310 | else |
62ba7442 | 3311 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
3312 | /* |
3313 | * The page is committed to the memcg, but it's not actually | |
3314 | * charged to the res_counter since we plan on replacing the | |
3315 | * old one and only one page is going to be left afterwards. | |
3316 | */ | |
ce587e65 | 3317 | __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false); |
ae41be37 | 3318 | } |
8869b8f6 | 3319 | |
69029cd5 | 3320 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3321 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3322 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3323 | { |
ac39cf8c | 3324 | struct page *used, *unused; |
01b1ae63 | 3325 | struct page_cgroup *pc; |
b2402857 | 3326 | bool anon; |
01b1ae63 | 3327 | |
c0ff4b85 | 3328 | if (!memcg) |
01b1ae63 | 3329 | return; |
ac39cf8c | 3330 | /* blocks rmdir() */ |
c0ff4b85 | 3331 | cgroup_exclude_rmdir(&memcg->css); |
50de1dd9 | 3332 | if (!migration_ok) { |
ac39cf8c | 3333 | used = oldpage; |
3334 | unused = newpage; | |
01b1ae63 | 3335 | } else { |
ac39cf8c | 3336 | used = newpage; |
01b1ae63 KH |
3337 | unused = oldpage; |
3338 | } | |
0030f535 | 3339 | anon = PageAnon(used); |
7d188958 JW |
3340 | __mem_cgroup_uncharge_common(unused, |
3341 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
3342 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
3343 | true); | |
0030f535 | 3344 | css_put(&memcg->css); |
69029cd5 | 3345 | /* |
ac39cf8c | 3346 | * We disallowed uncharge of pages under migration because mapcount |
3347 | * of the page goes down to zero, temporarly. | |
3348 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3349 | */ |
ac39cf8c | 3350 | pc = lookup_page_cgroup(oldpage); |
3351 | lock_page_cgroup(pc); | |
3352 | ClearPageCgroupMigration(pc); | |
3353 | unlock_page_cgroup(pc); | |
ac39cf8c | 3354 | |
01b1ae63 | 3355 | /* |
ac39cf8c | 3356 | * If a page is a file cache, radix-tree replacement is very atomic |
3357 | * and we can skip this check. When it was an Anon page, its mapcount | |
3358 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3359 | * uncharged yet. There are several case but page->mapcount check | |
3360 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3361 | * check. (see prepare_charge() also) | |
69029cd5 | 3362 | */ |
b2402857 | 3363 | if (anon) |
ac39cf8c | 3364 | mem_cgroup_uncharge_page(used); |
88703267 | 3365 | /* |
ac39cf8c | 3366 | * At migration, we may charge account against cgroup which has no |
3367 | * tasks. | |
88703267 KH |
3368 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3369 | * In that case, we need to call pre_destroy() again. check it here. | |
3370 | */ | |
c0ff4b85 | 3371 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
ae41be37 | 3372 | } |
78fb7466 | 3373 | |
ab936cbc KH |
3374 | /* |
3375 | * At replace page cache, newpage is not under any memcg but it's on | |
3376 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
3377 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
3378 | */ | |
3379 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
3380 | struct page *newpage) | |
3381 | { | |
bde05d1c | 3382 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 3383 | struct page_cgroup *pc; |
ab936cbc | 3384 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
3385 | |
3386 | if (mem_cgroup_disabled()) | |
3387 | return; | |
3388 | ||
3389 | pc = lookup_page_cgroup(oldpage); | |
3390 | /* fix accounting on old pages */ | |
3391 | lock_page_cgroup(pc); | |
bde05d1c HD |
3392 | if (PageCgroupUsed(pc)) { |
3393 | memcg = pc->mem_cgroup; | |
3394 | mem_cgroup_charge_statistics(memcg, false, -1); | |
3395 | ClearPageCgroupUsed(pc); | |
3396 | } | |
ab936cbc KH |
3397 | unlock_page_cgroup(pc); |
3398 | ||
bde05d1c HD |
3399 | /* |
3400 | * When called from shmem_replace_page(), in some cases the | |
3401 | * oldpage has already been charged, and in some cases not. | |
3402 | */ | |
3403 | if (!memcg) | |
3404 | return; | |
ab936cbc KH |
3405 | /* |
3406 | * Even if newpage->mapping was NULL before starting replacement, | |
3407 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
3408 | * LRU while we overwrite pc->mem_cgroup. | |
3409 | */ | |
ce587e65 | 3410 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
3411 | } |
3412 | ||
f212ad7c DN |
3413 | #ifdef CONFIG_DEBUG_VM |
3414 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3415 | { | |
3416 | struct page_cgroup *pc; | |
3417 | ||
3418 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3419 | /* |
3420 | * Can be NULL while feeding pages into the page allocator for | |
3421 | * the first time, i.e. during boot or memory hotplug; | |
3422 | * or when mem_cgroup_disabled(). | |
3423 | */ | |
f212ad7c DN |
3424 | if (likely(pc) && PageCgroupUsed(pc)) |
3425 | return pc; | |
3426 | return NULL; | |
3427 | } | |
3428 | ||
3429 | bool mem_cgroup_bad_page_check(struct page *page) | |
3430 | { | |
3431 | if (mem_cgroup_disabled()) | |
3432 | return false; | |
3433 | ||
3434 | return lookup_page_cgroup_used(page) != NULL; | |
3435 | } | |
3436 | ||
3437 | void mem_cgroup_print_bad_page(struct page *page) | |
3438 | { | |
3439 | struct page_cgroup *pc; | |
3440 | ||
3441 | pc = lookup_page_cgroup_used(page); | |
3442 | if (pc) { | |
90b3feae | 3443 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
f212ad7c | 3444 | pc, pc->flags, pc->mem_cgroup); |
f212ad7c DN |
3445 | } |
3446 | } | |
3447 | #endif | |
3448 | ||
8c7c6e34 KH |
3449 | static DEFINE_MUTEX(set_limit_mutex); |
3450 | ||
d38d2a75 | 3451 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3452 | unsigned long long val) |
628f4235 | 3453 | { |
81d39c20 | 3454 | int retry_count; |
3c11ecf4 | 3455 | u64 memswlimit, memlimit; |
628f4235 | 3456 | int ret = 0; |
81d39c20 KH |
3457 | int children = mem_cgroup_count_children(memcg); |
3458 | u64 curusage, oldusage; | |
3c11ecf4 | 3459 | int enlarge; |
81d39c20 KH |
3460 | |
3461 | /* | |
3462 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3463 | * is depends on callers. We set our retry-count to be function | |
3464 | * of # of children which we should visit in this loop. | |
3465 | */ | |
3466 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3467 | ||
3468 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3469 | |
3c11ecf4 | 3470 | enlarge = 0; |
8c7c6e34 | 3471 | while (retry_count) { |
628f4235 KH |
3472 | if (signal_pending(current)) { |
3473 | ret = -EINTR; | |
3474 | break; | |
3475 | } | |
8c7c6e34 KH |
3476 | /* |
3477 | * Rather than hide all in some function, I do this in | |
3478 | * open coded manner. You see what this really does. | |
aaad153e | 3479 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3480 | */ |
3481 | mutex_lock(&set_limit_mutex); | |
3482 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3483 | if (memswlimit < val) { | |
3484 | ret = -EINVAL; | |
3485 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3486 | break; |
3487 | } | |
3c11ecf4 KH |
3488 | |
3489 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3490 | if (memlimit < val) | |
3491 | enlarge = 1; | |
3492 | ||
8c7c6e34 | 3493 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3494 | if (!ret) { |
3495 | if (memswlimit == val) | |
3496 | memcg->memsw_is_minimum = true; | |
3497 | else | |
3498 | memcg->memsw_is_minimum = false; | |
3499 | } | |
8c7c6e34 KH |
3500 | mutex_unlock(&set_limit_mutex); |
3501 | ||
3502 | if (!ret) | |
3503 | break; | |
3504 | ||
5660048c JW |
3505 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3506 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3507 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3508 | /* Usage is reduced ? */ | |
3509 | if (curusage >= oldusage) | |
3510 | retry_count--; | |
3511 | else | |
3512 | oldusage = curusage; | |
8c7c6e34 | 3513 | } |
3c11ecf4 KH |
3514 | if (!ret && enlarge) |
3515 | memcg_oom_recover(memcg); | |
14797e23 | 3516 | |
8c7c6e34 KH |
3517 | return ret; |
3518 | } | |
3519 | ||
338c8431 LZ |
3520 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3521 | unsigned long long val) | |
8c7c6e34 | 3522 | { |
81d39c20 | 3523 | int retry_count; |
3c11ecf4 | 3524 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3525 | int children = mem_cgroup_count_children(memcg); |
3526 | int ret = -EBUSY; | |
3c11ecf4 | 3527 | int enlarge = 0; |
8c7c6e34 | 3528 | |
81d39c20 KH |
3529 | /* see mem_cgroup_resize_res_limit */ |
3530 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3531 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3532 | while (retry_count) { |
3533 | if (signal_pending(current)) { | |
3534 | ret = -EINTR; | |
3535 | break; | |
3536 | } | |
3537 | /* | |
3538 | * Rather than hide all in some function, I do this in | |
3539 | * open coded manner. You see what this really does. | |
aaad153e | 3540 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3541 | */ |
3542 | mutex_lock(&set_limit_mutex); | |
3543 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3544 | if (memlimit > val) { | |
3545 | ret = -EINVAL; | |
3546 | mutex_unlock(&set_limit_mutex); | |
3547 | break; | |
3548 | } | |
3c11ecf4 KH |
3549 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3550 | if (memswlimit < val) | |
3551 | enlarge = 1; | |
8c7c6e34 | 3552 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3553 | if (!ret) { |
3554 | if (memlimit == val) | |
3555 | memcg->memsw_is_minimum = true; | |
3556 | else | |
3557 | memcg->memsw_is_minimum = false; | |
3558 | } | |
8c7c6e34 KH |
3559 | mutex_unlock(&set_limit_mutex); |
3560 | ||
3561 | if (!ret) | |
3562 | break; | |
3563 | ||
5660048c JW |
3564 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3565 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3566 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3567 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3568 | /* Usage is reduced ? */ |
8c7c6e34 | 3569 | if (curusage >= oldusage) |
628f4235 | 3570 | retry_count--; |
81d39c20 KH |
3571 | else |
3572 | oldusage = curusage; | |
628f4235 | 3573 | } |
3c11ecf4 KH |
3574 | if (!ret && enlarge) |
3575 | memcg_oom_recover(memcg); | |
628f4235 KH |
3576 | return ret; |
3577 | } | |
3578 | ||
4e416953 | 3579 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3580 | gfp_t gfp_mask, |
3581 | unsigned long *total_scanned) | |
4e416953 BS |
3582 | { |
3583 | unsigned long nr_reclaimed = 0; | |
3584 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3585 | unsigned long reclaimed; | |
3586 | int loop = 0; | |
3587 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3588 | unsigned long long excess; |
0ae5e89c | 3589 | unsigned long nr_scanned; |
4e416953 BS |
3590 | |
3591 | if (order > 0) | |
3592 | return 0; | |
3593 | ||
00918b6a | 3594 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3595 | /* |
3596 | * This loop can run a while, specially if mem_cgroup's continuously | |
3597 | * keep exceeding their soft limit and putting the system under | |
3598 | * pressure | |
3599 | */ | |
3600 | do { | |
3601 | if (next_mz) | |
3602 | mz = next_mz; | |
3603 | else | |
3604 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3605 | if (!mz) | |
3606 | break; | |
3607 | ||
0ae5e89c | 3608 | nr_scanned = 0; |
d79154bb | 3609 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 3610 | gfp_mask, &nr_scanned); |
4e416953 | 3611 | nr_reclaimed += reclaimed; |
0ae5e89c | 3612 | *total_scanned += nr_scanned; |
4e416953 BS |
3613 | spin_lock(&mctz->lock); |
3614 | ||
3615 | /* | |
3616 | * If we failed to reclaim anything from this memory cgroup | |
3617 | * it is time to move on to the next cgroup | |
3618 | */ | |
3619 | next_mz = NULL; | |
3620 | if (!reclaimed) { | |
3621 | do { | |
3622 | /* | |
3623 | * Loop until we find yet another one. | |
3624 | * | |
3625 | * By the time we get the soft_limit lock | |
3626 | * again, someone might have aded the | |
3627 | * group back on the RB tree. Iterate to | |
3628 | * make sure we get a different mem. | |
3629 | * mem_cgroup_largest_soft_limit_node returns | |
3630 | * NULL if no other cgroup is present on | |
3631 | * the tree | |
3632 | */ | |
3633 | next_mz = | |
3634 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3635 | if (next_mz == mz) |
d79154bb | 3636 | css_put(&next_mz->memcg->css); |
39cc98f1 | 3637 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3638 | break; |
3639 | } while (1); | |
3640 | } | |
d79154bb HD |
3641 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
3642 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
3643 | /* |
3644 | * One school of thought says that we should not add | |
3645 | * back the node to the tree if reclaim returns 0. | |
3646 | * But our reclaim could return 0, simply because due | |
3647 | * to priority we are exposing a smaller subset of | |
3648 | * memory to reclaim from. Consider this as a longer | |
3649 | * term TODO. | |
3650 | */ | |
ef8745c1 | 3651 | /* If excess == 0, no tree ops */ |
d79154bb | 3652 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 3653 | spin_unlock(&mctz->lock); |
d79154bb | 3654 | css_put(&mz->memcg->css); |
4e416953 BS |
3655 | loop++; |
3656 | /* | |
3657 | * Could not reclaim anything and there are no more | |
3658 | * mem cgroups to try or we seem to be looping without | |
3659 | * reclaiming anything. | |
3660 | */ | |
3661 | if (!nr_reclaimed && | |
3662 | (next_mz == NULL || | |
3663 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3664 | break; | |
3665 | } while (!nr_reclaimed); | |
3666 | if (next_mz) | |
d79154bb | 3667 | css_put(&next_mz->memcg->css); |
4e416953 BS |
3668 | return nr_reclaimed; |
3669 | } | |
3670 | ||
cc847582 | 3671 | /* |
3c935d18 KH |
3672 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
3673 | * reclaim the pages page themselves - it just removes the page_cgroups. | |
3674 | * Returns true if some page_cgroups were not freed, indicating that the caller | |
3675 | * must retry this operation. | |
cc847582 | 3676 | */ |
3c935d18 | 3677 | static bool mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3678 | int node, int zid, enum lru_list lru) |
cc847582 | 3679 | { |
08e552c6 | 3680 | struct mem_cgroup_per_zone *mz; |
08e552c6 | 3681 | unsigned long flags, loop; |
072c56c1 | 3682 | struct list_head *list; |
925b7673 JW |
3683 | struct page *busy; |
3684 | struct zone *zone; | |
072c56c1 | 3685 | |
08e552c6 | 3686 | zone = &NODE_DATA(node)->node_zones[zid]; |
c0ff4b85 | 3687 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
6290df54 | 3688 | list = &mz->lruvec.lists[lru]; |
cc847582 | 3689 | |
1eb49272 | 3690 | loop = mz->lru_size[lru]; |
f817ed48 KH |
3691 | /* give some margin against EBUSY etc...*/ |
3692 | loop += 256; | |
3693 | busy = NULL; | |
3694 | while (loop--) { | |
925b7673 | 3695 | struct page_cgroup *pc; |
5564e88b JW |
3696 | struct page *page; |
3697 | ||
08e552c6 | 3698 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3699 | if (list_empty(list)) { |
08e552c6 | 3700 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3701 | break; |
f817ed48 | 3702 | } |
925b7673 JW |
3703 | page = list_entry(list->prev, struct page, lru); |
3704 | if (busy == page) { | |
3705 | list_move(&page->lru, list); | |
648bcc77 | 3706 | busy = NULL; |
08e552c6 | 3707 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3708 | continue; |
3709 | } | |
08e552c6 | 3710 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3711 | |
925b7673 | 3712 | pc = lookup_page_cgroup(page); |
5564e88b | 3713 | |
3c935d18 | 3714 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 3715 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 3716 | busy = page; |
f817ed48 KH |
3717 | cond_resched(); |
3718 | } else | |
3719 | busy = NULL; | |
cc847582 | 3720 | } |
3c935d18 | 3721 | return !list_empty(list); |
cc847582 KH |
3722 | } |
3723 | ||
3724 | /* | |
3725 | * make mem_cgroup's charge to be 0 if there is no task. | |
3726 | * This enables deleting this mem_cgroup. | |
3727 | */ | |
c0ff4b85 | 3728 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg, bool free_all) |
cc847582 | 3729 | { |
f817ed48 KH |
3730 | int ret; |
3731 | int node, zid, shrink; | |
3732 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c0ff4b85 | 3733 | struct cgroup *cgrp = memcg->css.cgroup; |
8869b8f6 | 3734 | |
c0ff4b85 | 3735 | css_get(&memcg->css); |
f817ed48 KH |
3736 | |
3737 | shrink = 0; | |
c1e862c1 KH |
3738 | /* should free all ? */ |
3739 | if (free_all) | |
3740 | goto try_to_free; | |
f817ed48 | 3741 | move_account: |
fce66477 | 3742 | do { |
f817ed48 | 3743 | ret = -EBUSY; |
c1e862c1 KH |
3744 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3745 | goto out; | |
52d4b9ac KH |
3746 | /* This is for making all *used* pages to be on LRU. */ |
3747 | lru_add_drain_all(); | |
c0ff4b85 | 3748 | drain_all_stock_sync(memcg); |
f817ed48 | 3749 | ret = 0; |
c0ff4b85 | 3750 | mem_cgroup_start_move(memcg); |
299b4eaa | 3751 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3752 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3753 | enum lru_list lru; |
3754 | for_each_lru(lru) { | |
c0ff4b85 | 3755 | ret = mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3756 | node, zid, lru); |
f817ed48 KH |
3757 | if (ret) |
3758 | break; | |
3759 | } | |
1ecaab2b | 3760 | } |
f817ed48 KH |
3761 | if (ret) |
3762 | break; | |
3763 | } | |
c0ff4b85 R |
3764 | mem_cgroup_end_move(memcg); |
3765 | memcg_oom_recover(memcg); | |
52d4b9ac | 3766 | cond_resched(); |
fce66477 | 3767 | /* "ret" should also be checked to ensure all lists are empty. */ |
569530fb | 3768 | } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0 || ret); |
cc847582 | 3769 | out: |
c0ff4b85 | 3770 | css_put(&memcg->css); |
cc847582 | 3771 | return ret; |
f817ed48 KH |
3772 | |
3773 | try_to_free: | |
c1e862c1 KH |
3774 | /* returns EBUSY if there is a task or if we come here twice. */ |
3775 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3776 | ret = -EBUSY; |
3777 | goto out; | |
3778 | } | |
c1e862c1 KH |
3779 | /* we call try-to-free pages for make this cgroup empty */ |
3780 | lru_add_drain_all(); | |
f817ed48 KH |
3781 | /* try to free all pages in this cgroup */ |
3782 | shrink = 1; | |
569530fb | 3783 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 3784 | int progress; |
c1e862c1 KH |
3785 | |
3786 | if (signal_pending(current)) { | |
3787 | ret = -EINTR; | |
3788 | goto out; | |
3789 | } | |
c0ff4b85 | 3790 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3791 | false); |
c1e862c1 | 3792 | if (!progress) { |
f817ed48 | 3793 | nr_retries--; |
c1e862c1 | 3794 | /* maybe some writeback is necessary */ |
8aa7e847 | 3795 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3796 | } |
f817ed48 KH |
3797 | |
3798 | } | |
08e552c6 | 3799 | lru_add_drain(); |
f817ed48 | 3800 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3801 | goto move_account; |
cc847582 KH |
3802 | } |
3803 | ||
6bbda35c | 3804 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 KH |
3805 | { |
3806 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3807 | } | |
3808 | ||
3809 | ||
18f59ea7 BS |
3810 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3811 | { | |
3812 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3813 | } | |
3814 | ||
3815 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3816 | u64 val) | |
3817 | { | |
3818 | int retval = 0; | |
c0ff4b85 | 3819 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3820 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3821 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3822 | |
3823 | if (parent) | |
c0ff4b85 | 3824 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3825 | |
3826 | cgroup_lock(); | |
567fb435 GC |
3827 | |
3828 | if (memcg->use_hierarchy == val) | |
3829 | goto out; | |
3830 | ||
18f59ea7 | 3831 | /* |
af901ca1 | 3832 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3833 | * in the child subtrees. If it is unset, then the change can |
3834 | * occur, provided the current cgroup has no children. | |
3835 | * | |
3836 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3837 | * set if there are no children. | |
3838 | */ | |
c0ff4b85 | 3839 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3840 | (val == 1 || val == 0)) { |
3841 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3842 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3843 | else |
3844 | retval = -EBUSY; | |
3845 | } else | |
3846 | retval = -EINVAL; | |
567fb435 GC |
3847 | |
3848 | out: | |
18f59ea7 BS |
3849 | cgroup_unlock(); |
3850 | ||
3851 | return retval; | |
3852 | } | |
3853 | ||
0c3e73e8 | 3854 | |
c0ff4b85 | 3855 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3856 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3857 | { |
7d74b06f | 3858 | struct mem_cgroup *iter; |
7a159cc9 | 3859 | long val = 0; |
0c3e73e8 | 3860 | |
7a159cc9 | 3861 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3862 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3863 | val += mem_cgroup_read_stat(iter, idx); |
3864 | ||
3865 | if (val < 0) /* race ? */ | |
3866 | val = 0; | |
3867 | return val; | |
0c3e73e8 BS |
3868 | } |
3869 | ||
c0ff4b85 | 3870 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3871 | { |
7d74b06f | 3872 | u64 val; |
104f3928 | 3873 | |
c0ff4b85 | 3874 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 3875 | if (!swap) |
65c64ce8 | 3876 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3877 | else |
65c64ce8 | 3878 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
3879 | } |
3880 | ||
c0ff4b85 R |
3881 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3882 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3883 | |
7d74b06f | 3884 | if (swap) |
bff6bb83 | 3885 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
3886 | |
3887 | return val << PAGE_SHIFT; | |
3888 | } | |
3889 | ||
af36f906 TH |
3890 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
3891 | struct file *file, char __user *buf, | |
3892 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 3893 | { |
c0ff4b85 | 3894 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 3895 | char str[64]; |
104f3928 | 3896 | u64 val; |
af36f906 | 3897 | int type, name, len; |
8c7c6e34 KH |
3898 | |
3899 | type = MEMFILE_TYPE(cft->private); | |
3900 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3901 | |
3902 | if (!do_swap_account && type == _MEMSWAP) | |
3903 | return -EOPNOTSUPP; | |
3904 | ||
8c7c6e34 KH |
3905 | switch (type) { |
3906 | case _MEM: | |
104f3928 | 3907 | if (name == RES_USAGE) |
c0ff4b85 | 3908 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3909 | else |
c0ff4b85 | 3910 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3911 | break; |
3912 | case _MEMSWAP: | |
104f3928 | 3913 | if (name == RES_USAGE) |
c0ff4b85 | 3914 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 3915 | else |
c0ff4b85 | 3916 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 KH |
3917 | break; |
3918 | default: | |
3919 | BUG(); | |
8c7c6e34 | 3920 | } |
af36f906 TH |
3921 | |
3922 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
3923 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 3924 | } |
628f4235 KH |
3925 | /* |
3926 | * The user of this function is... | |
3927 | * RES_LIMIT. | |
3928 | */ | |
856c13aa PM |
3929 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3930 | const char *buffer) | |
8cdea7c0 | 3931 | { |
628f4235 | 3932 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3933 | int type, name; |
628f4235 KH |
3934 | unsigned long long val; |
3935 | int ret; | |
3936 | ||
8c7c6e34 KH |
3937 | type = MEMFILE_TYPE(cft->private); |
3938 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3939 | |
3940 | if (!do_swap_account && type == _MEMSWAP) | |
3941 | return -EOPNOTSUPP; | |
3942 | ||
8c7c6e34 | 3943 | switch (name) { |
628f4235 | 3944 | case RES_LIMIT: |
4b3bde4c BS |
3945 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3946 | ret = -EINVAL; | |
3947 | break; | |
3948 | } | |
628f4235 KH |
3949 | /* This function does all necessary parse...reuse it */ |
3950 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3951 | if (ret) |
3952 | break; | |
3953 | if (type == _MEM) | |
628f4235 | 3954 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3955 | else |
3956 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3957 | break; |
296c81d8 BS |
3958 | case RES_SOFT_LIMIT: |
3959 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3960 | if (ret) | |
3961 | break; | |
3962 | /* | |
3963 | * For memsw, soft limits are hard to implement in terms | |
3964 | * of semantics, for now, we support soft limits for | |
3965 | * control without swap | |
3966 | */ | |
3967 | if (type == _MEM) | |
3968 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3969 | else | |
3970 | ret = -EINVAL; | |
3971 | break; | |
628f4235 KH |
3972 | default: |
3973 | ret = -EINVAL; /* should be BUG() ? */ | |
3974 | break; | |
3975 | } | |
3976 | return ret; | |
8cdea7c0 BS |
3977 | } |
3978 | ||
fee7b548 KH |
3979 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3980 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3981 | { | |
3982 | struct cgroup *cgroup; | |
3983 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3984 | ||
3985 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3986 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3987 | cgroup = memcg->css.cgroup; | |
3988 | if (!memcg->use_hierarchy) | |
3989 | goto out; | |
3990 | ||
3991 | while (cgroup->parent) { | |
3992 | cgroup = cgroup->parent; | |
3993 | memcg = mem_cgroup_from_cont(cgroup); | |
3994 | if (!memcg->use_hierarchy) | |
3995 | break; | |
3996 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3997 | min_limit = min(min_limit, tmp); | |
3998 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3999 | min_memsw_limit = min(min_memsw_limit, tmp); | |
4000 | } | |
4001 | out: | |
4002 | *mem_limit = min_limit; | |
4003 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
4004 | } |
4005 | ||
29f2a4da | 4006 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 4007 | { |
af36f906 | 4008 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 4009 | int type, name; |
c84872e1 | 4010 | |
8c7c6e34 KH |
4011 | type = MEMFILE_TYPE(event); |
4012 | name = MEMFILE_ATTR(event); | |
af36f906 TH |
4013 | |
4014 | if (!do_swap_account && type == _MEMSWAP) | |
4015 | return -EOPNOTSUPP; | |
4016 | ||
8c7c6e34 | 4017 | switch (name) { |
29f2a4da | 4018 | case RES_MAX_USAGE: |
8c7c6e34 | 4019 | if (type == _MEM) |
c0ff4b85 | 4020 | res_counter_reset_max(&memcg->res); |
8c7c6e34 | 4021 | else |
c0ff4b85 | 4022 | res_counter_reset_max(&memcg->memsw); |
29f2a4da PE |
4023 | break; |
4024 | case RES_FAILCNT: | |
8c7c6e34 | 4025 | if (type == _MEM) |
c0ff4b85 | 4026 | res_counter_reset_failcnt(&memcg->res); |
8c7c6e34 | 4027 | else |
c0ff4b85 | 4028 | res_counter_reset_failcnt(&memcg->memsw); |
29f2a4da PE |
4029 | break; |
4030 | } | |
f64c3f54 | 4031 | |
85cc59db | 4032 | return 0; |
c84872e1 PE |
4033 | } |
4034 | ||
7dc74be0 DN |
4035 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
4036 | struct cftype *cft) | |
4037 | { | |
4038 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
4039 | } | |
4040 | ||
02491447 | 4041 | #ifdef CONFIG_MMU |
7dc74be0 DN |
4042 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
4043 | struct cftype *cft, u64 val) | |
4044 | { | |
c0ff4b85 | 4045 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
4046 | |
4047 | if (val >= (1 << NR_MOVE_TYPE)) | |
4048 | return -EINVAL; | |
4049 | /* | |
4050 | * We check this value several times in both in can_attach() and | |
4051 | * attach(), so we need cgroup lock to prevent this value from being | |
4052 | * inconsistent. | |
4053 | */ | |
4054 | cgroup_lock(); | |
c0ff4b85 | 4055 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4056 | cgroup_unlock(); |
4057 | ||
4058 | return 0; | |
4059 | } | |
02491447 DN |
4060 | #else |
4061 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4062 | struct cftype *cft, u64 val) | |
4063 | { | |
4064 | return -ENOSYS; | |
4065 | } | |
4066 | #endif | |
7dc74be0 | 4067 | |
406eb0c9 | 4068 | #ifdef CONFIG_NUMA |
ab215884 | 4069 | static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft, |
fada52ca | 4070 | struct seq_file *m) |
406eb0c9 YH |
4071 | { |
4072 | int nid; | |
4073 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4074 | unsigned long node_nr; | |
d79154bb | 4075 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 4076 | |
d79154bb | 4077 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 YH |
4078 | seq_printf(m, "total=%lu", total_nr); |
4079 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4080 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
4081 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4082 | } | |
4083 | seq_putc(m, '\n'); | |
4084 | ||
d79154bb | 4085 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 YH |
4086 | seq_printf(m, "file=%lu", file_nr); |
4087 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4088 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4089 | LRU_ALL_FILE); |
406eb0c9 YH |
4090 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4091 | } | |
4092 | seq_putc(m, '\n'); | |
4093 | ||
d79154bb | 4094 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 YH |
4095 | seq_printf(m, "anon=%lu", anon_nr); |
4096 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4097 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4098 | LRU_ALL_ANON); |
406eb0c9 YH |
4099 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4100 | } | |
4101 | seq_putc(m, '\n'); | |
4102 | ||
d79154bb | 4103 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4104 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4105 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4106 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4107 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4108 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4109 | } | |
4110 | seq_putc(m, '\n'); | |
4111 | return 0; | |
4112 | } | |
4113 | #endif /* CONFIG_NUMA */ | |
4114 | ||
af7c4b0e JW |
4115 | static const char * const mem_cgroup_lru_names[] = { |
4116 | "inactive_anon", | |
4117 | "active_anon", | |
4118 | "inactive_file", | |
4119 | "active_file", | |
4120 | "unevictable", | |
4121 | }; | |
4122 | ||
4123 | static inline void mem_cgroup_lru_names_not_uptodate(void) | |
4124 | { | |
4125 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4126 | } | |
4127 | ||
ab215884 | 4128 | static int memcg_stat_show(struct cgroup *cont, struct cftype *cft, |
78ccf5b5 | 4129 | struct seq_file *m) |
d2ceb9b7 | 4130 | { |
d79154bb | 4131 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af7c4b0e JW |
4132 | struct mem_cgroup *mi; |
4133 | unsigned int i; | |
406eb0c9 | 4134 | |
af7c4b0e | 4135 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4136 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4137 | continue; |
af7c4b0e JW |
4138 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4139 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4140 | } |
7b854121 | 4141 | |
af7c4b0e JW |
4142 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4143 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4144 | mem_cgroup_read_events(memcg, i)); | |
4145 | ||
4146 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4147 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4148 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4149 | ||
14067bb3 | 4150 | /* Hierarchical information */ |
fee7b548 KH |
4151 | { |
4152 | unsigned long long limit, memsw_limit; | |
d79154bb | 4153 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 4154 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 4155 | if (do_swap_account) |
78ccf5b5 JW |
4156 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4157 | memsw_limit); | |
fee7b548 | 4158 | } |
7f016ee8 | 4159 | |
af7c4b0e JW |
4160 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4161 | long long val = 0; | |
4162 | ||
bff6bb83 | 4163 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4164 | continue; |
af7c4b0e JW |
4165 | for_each_mem_cgroup_tree(mi, memcg) |
4166 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4167 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4168 | } | |
4169 | ||
4170 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4171 | unsigned long long val = 0; | |
4172 | ||
4173 | for_each_mem_cgroup_tree(mi, memcg) | |
4174 | val += mem_cgroup_read_events(mi, i); | |
4175 | seq_printf(m, "total_%s %llu\n", | |
4176 | mem_cgroup_events_names[i], val); | |
4177 | } | |
4178 | ||
4179 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4180 | unsigned long long val = 0; | |
4181 | ||
4182 | for_each_mem_cgroup_tree(mi, memcg) | |
4183 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4184 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4185 | } |
14067bb3 | 4186 | |
7f016ee8 | 4187 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4188 | { |
4189 | int nid, zid; | |
4190 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4191 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4192 | unsigned long recent_rotated[2] = {0, 0}; |
4193 | unsigned long recent_scanned[2] = {0, 0}; | |
4194 | ||
4195 | for_each_online_node(nid) | |
4196 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 4197 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 4198 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4199 | |
89abfab1 HD |
4200 | recent_rotated[0] += rstat->recent_rotated[0]; |
4201 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4202 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4203 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4204 | } |
78ccf5b5 JW |
4205 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4206 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4207 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4208 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4209 | } |
4210 | #endif | |
4211 | ||
d2ceb9b7 KH |
4212 | return 0; |
4213 | } | |
4214 | ||
a7885eb8 KM |
4215 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4216 | { | |
4217 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4218 | ||
1f4c025b | 4219 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4220 | } |
4221 | ||
4222 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4223 | u64 val) | |
4224 | { | |
4225 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4226 | struct mem_cgroup *parent; | |
068b38c1 | 4227 | |
a7885eb8 KM |
4228 | if (val > 100) |
4229 | return -EINVAL; | |
4230 | ||
4231 | if (cgrp->parent == NULL) | |
4232 | return -EINVAL; | |
4233 | ||
4234 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4235 | |
4236 | cgroup_lock(); | |
4237 | ||
a7885eb8 KM |
4238 | /* If under hierarchy, only empty-root can set this value */ |
4239 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4240 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4241 | cgroup_unlock(); | |
a7885eb8 | 4242 | return -EINVAL; |
068b38c1 | 4243 | } |
a7885eb8 | 4244 | |
a7885eb8 | 4245 | memcg->swappiness = val; |
a7885eb8 | 4246 | |
068b38c1 LZ |
4247 | cgroup_unlock(); |
4248 | ||
a7885eb8 KM |
4249 | return 0; |
4250 | } | |
4251 | ||
2e72b634 KS |
4252 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4253 | { | |
4254 | struct mem_cgroup_threshold_ary *t; | |
4255 | u64 usage; | |
4256 | int i; | |
4257 | ||
4258 | rcu_read_lock(); | |
4259 | if (!swap) | |
2c488db2 | 4260 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4261 | else |
2c488db2 | 4262 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4263 | |
4264 | if (!t) | |
4265 | goto unlock; | |
4266 | ||
4267 | usage = mem_cgroup_usage(memcg, swap); | |
4268 | ||
4269 | /* | |
748dad36 | 4270 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4271 | * If it's not true, a threshold was crossed after last |
4272 | * call of __mem_cgroup_threshold(). | |
4273 | */ | |
5407a562 | 4274 | i = t->current_threshold; |
2e72b634 KS |
4275 | |
4276 | /* | |
4277 | * Iterate backward over array of thresholds starting from | |
4278 | * current_threshold and check if a threshold is crossed. | |
4279 | * If none of thresholds below usage is crossed, we read | |
4280 | * only one element of the array here. | |
4281 | */ | |
4282 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4283 | eventfd_signal(t->entries[i].eventfd, 1); | |
4284 | ||
4285 | /* i = current_threshold + 1 */ | |
4286 | i++; | |
4287 | ||
4288 | /* | |
4289 | * Iterate forward over array of thresholds starting from | |
4290 | * current_threshold+1 and check if a threshold is crossed. | |
4291 | * If none of thresholds above usage is crossed, we read | |
4292 | * only one element of the array here. | |
4293 | */ | |
4294 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4295 | eventfd_signal(t->entries[i].eventfd, 1); | |
4296 | ||
4297 | /* Update current_threshold */ | |
5407a562 | 4298 | t->current_threshold = i - 1; |
2e72b634 KS |
4299 | unlock: |
4300 | rcu_read_unlock(); | |
4301 | } | |
4302 | ||
4303 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4304 | { | |
ad4ca5f4 KS |
4305 | while (memcg) { |
4306 | __mem_cgroup_threshold(memcg, false); | |
4307 | if (do_swap_account) | |
4308 | __mem_cgroup_threshold(memcg, true); | |
4309 | ||
4310 | memcg = parent_mem_cgroup(memcg); | |
4311 | } | |
2e72b634 KS |
4312 | } |
4313 | ||
4314 | static int compare_thresholds(const void *a, const void *b) | |
4315 | { | |
4316 | const struct mem_cgroup_threshold *_a = a; | |
4317 | const struct mem_cgroup_threshold *_b = b; | |
4318 | ||
4319 | return _a->threshold - _b->threshold; | |
4320 | } | |
4321 | ||
c0ff4b85 | 4322 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4323 | { |
4324 | struct mem_cgroup_eventfd_list *ev; | |
4325 | ||
c0ff4b85 | 4326 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4327 | eventfd_signal(ev->eventfd, 1); |
4328 | return 0; | |
4329 | } | |
4330 | ||
c0ff4b85 | 4331 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4332 | { |
7d74b06f KH |
4333 | struct mem_cgroup *iter; |
4334 | ||
c0ff4b85 | 4335 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4336 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4337 | } |
4338 | ||
4339 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4340 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4341 | { |
4342 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4343 | struct mem_cgroup_thresholds *thresholds; |
4344 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4345 | int type = MEMFILE_TYPE(cft->private); |
4346 | u64 threshold, usage; | |
2c488db2 | 4347 | int i, size, ret; |
2e72b634 KS |
4348 | |
4349 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4350 | if (ret) | |
4351 | return ret; | |
4352 | ||
4353 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4354 | |
2e72b634 | 4355 | if (type == _MEM) |
2c488db2 | 4356 | thresholds = &memcg->thresholds; |
2e72b634 | 4357 | else if (type == _MEMSWAP) |
2c488db2 | 4358 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4359 | else |
4360 | BUG(); | |
4361 | ||
4362 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4363 | ||
4364 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4365 | if (thresholds->primary) |
2e72b634 KS |
4366 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4367 | ||
2c488db2 | 4368 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4369 | |
4370 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4371 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4372 | GFP_KERNEL); |
2c488db2 | 4373 | if (!new) { |
2e72b634 KS |
4374 | ret = -ENOMEM; |
4375 | goto unlock; | |
4376 | } | |
2c488db2 | 4377 | new->size = size; |
2e72b634 KS |
4378 | |
4379 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4380 | if (thresholds->primary) { |
4381 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4382 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4383 | } |
4384 | ||
2e72b634 | 4385 | /* Add new threshold */ |
2c488db2 KS |
4386 | new->entries[size - 1].eventfd = eventfd; |
4387 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4388 | |
4389 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4390 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4391 | compare_thresholds, NULL); |
4392 | ||
4393 | /* Find current threshold */ | |
2c488db2 | 4394 | new->current_threshold = -1; |
2e72b634 | 4395 | for (i = 0; i < size; i++) { |
748dad36 | 4396 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4397 | /* |
2c488db2 KS |
4398 | * new->current_threshold will not be used until |
4399 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4400 | * it here. |
4401 | */ | |
2c488db2 | 4402 | ++new->current_threshold; |
748dad36 SZ |
4403 | } else |
4404 | break; | |
2e72b634 KS |
4405 | } |
4406 | ||
2c488db2 KS |
4407 | /* Free old spare buffer and save old primary buffer as spare */ |
4408 | kfree(thresholds->spare); | |
4409 | thresholds->spare = thresholds->primary; | |
4410 | ||
4411 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4412 | |
907860ed | 4413 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4414 | synchronize_rcu(); |
4415 | ||
2e72b634 KS |
4416 | unlock: |
4417 | mutex_unlock(&memcg->thresholds_lock); | |
4418 | ||
4419 | return ret; | |
4420 | } | |
4421 | ||
907860ed | 4422 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4423 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4424 | { |
4425 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4426 | struct mem_cgroup_thresholds *thresholds; |
4427 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4428 | int type = MEMFILE_TYPE(cft->private); |
4429 | u64 usage; | |
2c488db2 | 4430 | int i, j, size; |
2e72b634 KS |
4431 | |
4432 | mutex_lock(&memcg->thresholds_lock); | |
4433 | if (type == _MEM) | |
2c488db2 | 4434 | thresholds = &memcg->thresholds; |
2e72b634 | 4435 | else if (type == _MEMSWAP) |
2c488db2 | 4436 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4437 | else |
4438 | BUG(); | |
4439 | ||
371528ca AV |
4440 | if (!thresholds->primary) |
4441 | goto unlock; | |
4442 | ||
2e72b634 KS |
4443 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
4444 | ||
4445 | /* Check if a threshold crossed before removing */ | |
4446 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4447 | ||
4448 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4449 | size = 0; |
4450 | for (i = 0; i < thresholds->primary->size; i++) { | |
4451 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4452 | size++; |
4453 | } | |
4454 | ||
2c488db2 | 4455 | new = thresholds->spare; |
907860ed | 4456 | |
2e72b634 KS |
4457 | /* Set thresholds array to NULL if we don't have thresholds */ |
4458 | if (!size) { | |
2c488db2 KS |
4459 | kfree(new); |
4460 | new = NULL; | |
907860ed | 4461 | goto swap_buffers; |
2e72b634 KS |
4462 | } |
4463 | ||
2c488db2 | 4464 | new->size = size; |
2e72b634 KS |
4465 | |
4466 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4467 | new->current_threshold = -1; |
4468 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4469 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4470 | continue; |
4471 | ||
2c488db2 | 4472 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4473 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4474 | /* |
2c488db2 | 4475 | * new->current_threshold will not be used |
2e72b634 KS |
4476 | * until rcu_assign_pointer(), so it's safe to increment |
4477 | * it here. | |
4478 | */ | |
2c488db2 | 4479 | ++new->current_threshold; |
2e72b634 KS |
4480 | } |
4481 | j++; | |
4482 | } | |
4483 | ||
907860ed | 4484 | swap_buffers: |
2c488db2 KS |
4485 | /* Swap primary and spare array */ |
4486 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4487 | /* If all events are unregistered, free the spare array */ |
4488 | if (!new) { | |
4489 | kfree(thresholds->spare); | |
4490 | thresholds->spare = NULL; | |
4491 | } | |
4492 | ||
2c488db2 | 4493 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4494 | |
907860ed | 4495 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4496 | synchronize_rcu(); |
371528ca | 4497 | unlock: |
2e72b634 | 4498 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4499 | } |
c1e862c1 | 4500 | |
9490ff27 KH |
4501 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4502 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4503 | { | |
4504 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4505 | struct mem_cgroup_eventfd_list *event; | |
4506 | int type = MEMFILE_TYPE(cft->private); | |
4507 | ||
4508 | BUG_ON(type != _OOM_TYPE); | |
4509 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4510 | if (!event) | |
4511 | return -ENOMEM; | |
4512 | ||
1af8efe9 | 4513 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4514 | |
4515 | event->eventfd = eventfd; | |
4516 | list_add(&event->list, &memcg->oom_notify); | |
4517 | ||
4518 | /* already in OOM ? */ | |
79dfdacc | 4519 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4520 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4521 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4522 | |
4523 | return 0; | |
4524 | } | |
4525 | ||
907860ed | 4526 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4527 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4528 | { | |
c0ff4b85 | 4529 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 KH |
4530 | struct mem_cgroup_eventfd_list *ev, *tmp; |
4531 | int type = MEMFILE_TYPE(cft->private); | |
4532 | ||
4533 | BUG_ON(type != _OOM_TYPE); | |
4534 | ||
1af8efe9 | 4535 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4536 | |
c0ff4b85 | 4537 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4538 | if (ev->eventfd == eventfd) { |
4539 | list_del(&ev->list); | |
4540 | kfree(ev); | |
4541 | } | |
4542 | } | |
4543 | ||
1af8efe9 | 4544 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4545 | } |
4546 | ||
3c11ecf4 KH |
4547 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4548 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4549 | { | |
c0ff4b85 | 4550 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4551 | |
c0ff4b85 | 4552 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4553 | |
c0ff4b85 | 4554 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4555 | cb->fill(cb, "under_oom", 1); |
4556 | else | |
4557 | cb->fill(cb, "under_oom", 0); | |
4558 | return 0; | |
4559 | } | |
4560 | ||
3c11ecf4 KH |
4561 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4562 | struct cftype *cft, u64 val) | |
4563 | { | |
c0ff4b85 | 4564 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4565 | struct mem_cgroup *parent; |
4566 | ||
4567 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4568 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4569 | return -EINVAL; | |
4570 | ||
4571 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4572 | ||
4573 | cgroup_lock(); | |
4574 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4575 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4576 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4577 | cgroup_unlock(); |
4578 | return -EINVAL; | |
4579 | } | |
c0ff4b85 | 4580 | memcg->oom_kill_disable = val; |
4d845ebf | 4581 | if (!val) |
c0ff4b85 | 4582 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4583 | cgroup_unlock(); |
4584 | return 0; | |
4585 | } | |
4586 | ||
c255a458 | 4587 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4588 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4589 | { |
1d62e436 | 4590 | return mem_cgroup_sockets_init(memcg, ss); |
e5671dfa GC |
4591 | }; |
4592 | ||
1d62e436 | 4593 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 | 4594 | { |
1d62e436 | 4595 | mem_cgroup_sockets_destroy(memcg); |
d1a4c0b3 | 4596 | } |
e5671dfa | 4597 | #else |
cbe128e3 | 4598 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4599 | { |
4600 | return 0; | |
4601 | } | |
d1a4c0b3 | 4602 | |
1d62e436 | 4603 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 GC |
4604 | { |
4605 | } | |
e5671dfa GC |
4606 | #endif |
4607 | ||
8cdea7c0 BS |
4608 | static struct cftype mem_cgroup_files[] = { |
4609 | { | |
0eea1030 | 4610 | .name = "usage_in_bytes", |
8c7c6e34 | 4611 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 4612 | .read = mem_cgroup_read, |
9490ff27 KH |
4613 | .register_event = mem_cgroup_usage_register_event, |
4614 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4615 | }, |
c84872e1 PE |
4616 | { |
4617 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4618 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4619 | .trigger = mem_cgroup_reset, |
af36f906 | 4620 | .read = mem_cgroup_read, |
c84872e1 | 4621 | }, |
8cdea7c0 | 4622 | { |
0eea1030 | 4623 | .name = "limit_in_bytes", |
8c7c6e34 | 4624 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4625 | .write_string = mem_cgroup_write, |
af36f906 | 4626 | .read = mem_cgroup_read, |
8cdea7c0 | 4627 | }, |
296c81d8 BS |
4628 | { |
4629 | .name = "soft_limit_in_bytes", | |
4630 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4631 | .write_string = mem_cgroup_write, | |
af36f906 | 4632 | .read = mem_cgroup_read, |
296c81d8 | 4633 | }, |
8cdea7c0 BS |
4634 | { |
4635 | .name = "failcnt", | |
8c7c6e34 | 4636 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4637 | .trigger = mem_cgroup_reset, |
af36f906 | 4638 | .read = mem_cgroup_read, |
8cdea7c0 | 4639 | }, |
d2ceb9b7 KH |
4640 | { |
4641 | .name = "stat", | |
ab215884 | 4642 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 4643 | }, |
c1e862c1 KH |
4644 | { |
4645 | .name = "force_empty", | |
4646 | .trigger = mem_cgroup_force_empty_write, | |
4647 | }, | |
18f59ea7 BS |
4648 | { |
4649 | .name = "use_hierarchy", | |
4650 | .write_u64 = mem_cgroup_hierarchy_write, | |
4651 | .read_u64 = mem_cgroup_hierarchy_read, | |
4652 | }, | |
a7885eb8 KM |
4653 | { |
4654 | .name = "swappiness", | |
4655 | .read_u64 = mem_cgroup_swappiness_read, | |
4656 | .write_u64 = mem_cgroup_swappiness_write, | |
4657 | }, | |
7dc74be0 DN |
4658 | { |
4659 | .name = "move_charge_at_immigrate", | |
4660 | .read_u64 = mem_cgroup_move_charge_read, | |
4661 | .write_u64 = mem_cgroup_move_charge_write, | |
4662 | }, | |
9490ff27 KH |
4663 | { |
4664 | .name = "oom_control", | |
3c11ecf4 KH |
4665 | .read_map = mem_cgroup_oom_control_read, |
4666 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4667 | .register_event = mem_cgroup_oom_register_event, |
4668 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4669 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4670 | }, | |
406eb0c9 YH |
4671 | #ifdef CONFIG_NUMA |
4672 | { | |
4673 | .name = "numa_stat", | |
ab215884 | 4674 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
4675 | }, |
4676 | #endif | |
c255a458 | 4677 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4678 | { |
4679 | .name = "memsw.usage_in_bytes", | |
4680 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
af36f906 | 4681 | .read = mem_cgroup_read, |
9490ff27 KH |
4682 | .register_event = mem_cgroup_usage_register_event, |
4683 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4684 | }, |
4685 | { | |
4686 | .name = "memsw.max_usage_in_bytes", | |
4687 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4688 | .trigger = mem_cgroup_reset, | |
af36f906 | 4689 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4690 | }, |
4691 | { | |
4692 | .name = "memsw.limit_in_bytes", | |
4693 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4694 | .write_string = mem_cgroup_write, | |
af36f906 | 4695 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4696 | }, |
4697 | { | |
4698 | .name = "memsw.failcnt", | |
4699 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4700 | .trigger = mem_cgroup_reset, | |
af36f906 | 4701 | .read = mem_cgroup_read, |
8c7c6e34 | 4702 | }, |
8c7c6e34 | 4703 | #endif |
6bc10349 | 4704 | { }, /* terminate */ |
af36f906 | 4705 | }; |
8c7c6e34 | 4706 | |
c0ff4b85 | 4707 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4708 | { |
4709 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4710 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4711 | int zone, tmp = node; |
1ecaab2b KH |
4712 | /* |
4713 | * This routine is called against possible nodes. | |
4714 | * But it's BUG to call kmalloc() against offline node. | |
4715 | * | |
4716 | * TODO: this routine can waste much memory for nodes which will | |
4717 | * never be onlined. It's better to use memory hotplug callback | |
4718 | * function. | |
4719 | */ | |
41e3355d KH |
4720 | if (!node_state(node, N_NORMAL_MEMORY)) |
4721 | tmp = -1; | |
17295c88 | 4722 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4723 | if (!pn) |
4724 | return 1; | |
1ecaab2b | 4725 | |
1ecaab2b KH |
4726 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4727 | mz = &pn->zoneinfo[zone]; | |
7f5e86c2 | 4728 | lruvec_init(&mz->lruvec, &NODE_DATA(node)->node_zones[zone]); |
f64c3f54 | 4729 | mz->usage_in_excess = 0; |
4e416953 | 4730 | mz->on_tree = false; |
d79154bb | 4731 | mz->memcg = memcg; |
1ecaab2b | 4732 | } |
0a619e58 | 4733 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4734 | return 0; |
4735 | } | |
4736 | ||
c0ff4b85 | 4737 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4738 | { |
c0ff4b85 | 4739 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4740 | } |
4741 | ||
33327948 KH |
4742 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4743 | { | |
d79154bb | 4744 | struct mem_cgroup *memcg; |
c62b1a3b | 4745 | int size = sizeof(struct mem_cgroup); |
33327948 | 4746 | |
c62b1a3b | 4747 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4748 | if (size < PAGE_SIZE) |
d79154bb | 4749 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 4750 | else |
d79154bb | 4751 | memcg = vzalloc(size); |
33327948 | 4752 | |
d79154bb | 4753 | if (!memcg) |
e7bbcdf3 DC |
4754 | return NULL; |
4755 | ||
d79154bb HD |
4756 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4757 | if (!memcg->stat) | |
d2e61b8d | 4758 | goto out_free; |
d79154bb HD |
4759 | spin_lock_init(&memcg->pcp_counter_lock); |
4760 | return memcg; | |
d2e61b8d DC |
4761 | |
4762 | out_free: | |
4763 | if (size < PAGE_SIZE) | |
d79154bb | 4764 | kfree(memcg); |
d2e61b8d | 4765 | else |
d79154bb | 4766 | vfree(memcg); |
d2e61b8d | 4767 | return NULL; |
33327948 KH |
4768 | } |
4769 | ||
59927fb9 | 4770 | /* |
3afe36b1 | 4771 | * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU, |
59927fb9 HD |
4772 | * but in process context. The work_freeing structure is overlaid |
4773 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
4774 | */ | |
3afe36b1 | 4775 | static void free_work(struct work_struct *work) |
59927fb9 HD |
4776 | { |
4777 | struct mem_cgroup *memcg; | |
3afe36b1 | 4778 | int size = sizeof(struct mem_cgroup); |
59927fb9 HD |
4779 | |
4780 | memcg = container_of(work, struct mem_cgroup, work_freeing); | |
3f134619 GC |
4781 | /* |
4782 | * We need to make sure that (at least for now), the jump label | |
4783 | * destruction code runs outside of the cgroup lock. This is because | |
4784 | * get_online_cpus(), which is called from the static_branch update, | |
4785 | * can't be called inside the cgroup_lock. cpusets are the ones | |
4786 | * enforcing this dependency, so if they ever change, we might as well. | |
4787 | * | |
4788 | * schedule_work() will guarantee this happens. Be careful if you need | |
4789 | * to move this code around, and make sure it is outside | |
4790 | * the cgroup_lock. | |
4791 | */ | |
4792 | disarm_sock_keys(memcg); | |
3afe36b1 GC |
4793 | if (size < PAGE_SIZE) |
4794 | kfree(memcg); | |
4795 | else | |
4796 | vfree(memcg); | |
59927fb9 | 4797 | } |
3afe36b1 GC |
4798 | |
4799 | static void free_rcu(struct rcu_head *rcu_head) | |
59927fb9 HD |
4800 | { |
4801 | struct mem_cgroup *memcg; | |
4802 | ||
4803 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); | |
3afe36b1 | 4804 | INIT_WORK(&memcg->work_freeing, free_work); |
59927fb9 HD |
4805 | schedule_work(&memcg->work_freeing); |
4806 | } | |
4807 | ||
8c7c6e34 KH |
4808 | /* |
4809 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4810 | * (scanning all at force_empty is too costly...) | |
4811 | * | |
4812 | * Instead of clearing all references at force_empty, we remember | |
4813 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4814 | * it goes down to 0. | |
4815 | * | |
8c7c6e34 KH |
4816 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4817 | */ | |
4818 | ||
c0ff4b85 | 4819 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 4820 | { |
08e552c6 KH |
4821 | int node; |
4822 | ||
c0ff4b85 R |
4823 | mem_cgroup_remove_from_trees(memcg); |
4824 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 4825 | |
3ed28fa1 | 4826 | for_each_node(node) |
c0ff4b85 | 4827 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 4828 | |
c0ff4b85 | 4829 | free_percpu(memcg->stat); |
3afe36b1 | 4830 | call_rcu(&memcg->rcu_freeing, free_rcu); |
33327948 KH |
4831 | } |
4832 | ||
c0ff4b85 | 4833 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 4834 | { |
c0ff4b85 | 4835 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
4836 | } |
4837 | ||
c0ff4b85 | 4838 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 4839 | { |
c0ff4b85 R |
4840 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
4841 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4842 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
4843 | if (parent) |
4844 | mem_cgroup_put(parent); | |
4845 | } | |
8c7c6e34 KH |
4846 | } |
4847 | ||
c0ff4b85 | 4848 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 4849 | { |
c0ff4b85 | 4850 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
4851 | } |
4852 | ||
7bcc1bb1 DN |
4853 | /* |
4854 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4855 | */ | |
e1aab161 | 4856 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4857 | { |
c0ff4b85 | 4858 | if (!memcg->res.parent) |
7bcc1bb1 | 4859 | return NULL; |
c0ff4b85 | 4860 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 4861 | } |
e1aab161 | 4862 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4863 | |
c255a458 | 4864 | #ifdef CONFIG_MEMCG_SWAP |
c077719b KH |
4865 | static void __init enable_swap_cgroup(void) |
4866 | { | |
f8d66542 | 4867 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4868 | do_swap_account = 1; |
4869 | } | |
4870 | #else | |
4871 | static void __init enable_swap_cgroup(void) | |
4872 | { | |
4873 | } | |
4874 | #endif | |
4875 | ||
f64c3f54 BS |
4876 | static int mem_cgroup_soft_limit_tree_init(void) |
4877 | { | |
4878 | struct mem_cgroup_tree_per_node *rtpn; | |
4879 | struct mem_cgroup_tree_per_zone *rtpz; | |
4880 | int tmp, node, zone; | |
4881 | ||
3ed28fa1 | 4882 | for_each_node(node) { |
f64c3f54 BS |
4883 | tmp = node; |
4884 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4885 | tmp = -1; | |
4886 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4887 | if (!rtpn) | |
c3cecc68 | 4888 | goto err_cleanup; |
f64c3f54 BS |
4889 | |
4890 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4891 | ||
4892 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4893 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4894 | rtpz->rb_root = RB_ROOT; | |
4895 | spin_lock_init(&rtpz->lock); | |
4896 | } | |
4897 | } | |
4898 | return 0; | |
c3cecc68 MH |
4899 | |
4900 | err_cleanup: | |
3ed28fa1 | 4901 | for_each_node(node) { |
c3cecc68 MH |
4902 | if (!soft_limit_tree.rb_tree_per_node[node]) |
4903 | break; | |
4904 | kfree(soft_limit_tree.rb_tree_per_node[node]); | |
4905 | soft_limit_tree.rb_tree_per_node[node] = NULL; | |
4906 | } | |
4907 | return 1; | |
4908 | ||
f64c3f54 BS |
4909 | } |
4910 | ||
0eb253e2 | 4911 | static struct cgroup_subsys_state * __ref |
761b3ef5 | 4912 | mem_cgroup_create(struct cgroup *cont) |
8cdea7c0 | 4913 | { |
c0ff4b85 | 4914 | struct mem_cgroup *memcg, *parent; |
04046e1a | 4915 | long error = -ENOMEM; |
6d12e2d8 | 4916 | int node; |
8cdea7c0 | 4917 | |
c0ff4b85 R |
4918 | memcg = mem_cgroup_alloc(); |
4919 | if (!memcg) | |
04046e1a | 4920 | return ERR_PTR(error); |
78fb7466 | 4921 | |
3ed28fa1 | 4922 | for_each_node(node) |
c0ff4b85 | 4923 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4924 | goto free_out; |
f64c3f54 | 4925 | |
c077719b | 4926 | /* root ? */ |
28dbc4b6 | 4927 | if (cont->parent == NULL) { |
cdec2e42 | 4928 | int cpu; |
c077719b | 4929 | enable_swap_cgroup(); |
28dbc4b6 | 4930 | parent = NULL; |
f64c3f54 BS |
4931 | if (mem_cgroup_soft_limit_tree_init()) |
4932 | goto free_out; | |
a41c58a6 | 4933 | root_mem_cgroup = memcg; |
cdec2e42 KH |
4934 | for_each_possible_cpu(cpu) { |
4935 | struct memcg_stock_pcp *stock = | |
4936 | &per_cpu(memcg_stock, cpu); | |
4937 | INIT_WORK(&stock->work, drain_local_stock); | |
4938 | } | |
711d3d2c | 4939 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4940 | } else { |
28dbc4b6 | 4941 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
4942 | memcg->use_hierarchy = parent->use_hierarchy; |
4943 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 4944 | } |
28dbc4b6 | 4945 | |
18f59ea7 | 4946 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
4947 | res_counter_init(&memcg->res, &parent->res); |
4948 | res_counter_init(&memcg->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4949 | /* |
4950 | * We increment refcnt of the parent to ensure that we can | |
4951 | * safely access it on res_counter_charge/uncharge. | |
4952 | * This refcnt will be decremented when freeing this | |
4953 | * mem_cgroup(see mem_cgroup_put). | |
4954 | */ | |
4955 | mem_cgroup_get(parent); | |
18f59ea7 | 4956 | } else { |
c0ff4b85 R |
4957 | res_counter_init(&memcg->res, NULL); |
4958 | res_counter_init(&memcg->memsw, NULL); | |
18f59ea7 | 4959 | } |
c0ff4b85 R |
4960 | memcg->last_scanned_node = MAX_NUMNODES; |
4961 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 4962 | |
a7885eb8 | 4963 | if (parent) |
c0ff4b85 R |
4964 | memcg->swappiness = mem_cgroup_swappiness(parent); |
4965 | atomic_set(&memcg->refcnt, 1); | |
4966 | memcg->move_charge_at_immigrate = 0; | |
4967 | mutex_init(&memcg->thresholds_lock); | |
312734c0 | 4968 | spin_lock_init(&memcg->move_lock); |
cbe128e3 GC |
4969 | |
4970 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
4971 | if (error) { | |
4972 | /* | |
4973 | * We call put now because our (and parent's) refcnts | |
4974 | * are already in place. mem_cgroup_put() will internally | |
4975 | * call __mem_cgroup_free, so return directly | |
4976 | */ | |
4977 | mem_cgroup_put(memcg); | |
4978 | return ERR_PTR(error); | |
4979 | } | |
c0ff4b85 | 4980 | return &memcg->css; |
6d12e2d8 | 4981 | free_out: |
c0ff4b85 | 4982 | __mem_cgroup_free(memcg); |
04046e1a | 4983 | return ERR_PTR(error); |
8cdea7c0 BS |
4984 | } |
4985 | ||
761b3ef5 | 4986 | static int mem_cgroup_pre_destroy(struct cgroup *cont) |
df878fb0 | 4987 | { |
c0ff4b85 | 4988 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 4989 | |
c0ff4b85 | 4990 | return mem_cgroup_force_empty(memcg, false); |
df878fb0 KH |
4991 | } |
4992 | ||
761b3ef5 | 4993 | static void mem_cgroup_destroy(struct cgroup *cont) |
8cdea7c0 | 4994 | { |
c0ff4b85 | 4995 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 4996 | |
1d62e436 | 4997 | kmem_cgroup_destroy(memcg); |
d1a4c0b3 | 4998 | |
c0ff4b85 | 4999 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
5000 | } |
5001 | ||
02491447 | 5002 | #ifdef CONFIG_MMU |
7dc74be0 | 5003 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5004 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5005 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5006 | { |
854ffa8d DN |
5007 | int ret = 0; |
5008 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5009 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5010 | |
c0ff4b85 | 5011 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5012 | mc.precharge += count; |
5013 | /* we don't need css_get for root */ | |
5014 | return ret; | |
5015 | } | |
5016 | /* try to charge at once */ | |
5017 | if (count > 1) { | |
5018 | struct res_counter *dummy; | |
5019 | /* | |
c0ff4b85 | 5020 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5021 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5022 | * are still under the same cgroup_mutex. So we can postpone | |
5023 | * css_get(). | |
5024 | */ | |
c0ff4b85 | 5025 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5026 | goto one_by_one; |
c0ff4b85 | 5027 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5028 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5029 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5030 | goto one_by_one; |
5031 | } | |
5032 | mc.precharge += count; | |
854ffa8d DN |
5033 | return ret; |
5034 | } | |
5035 | one_by_one: | |
5036 | /* fall back to one by one charge */ | |
5037 | while (count--) { | |
5038 | if (signal_pending(current)) { | |
5039 | ret = -EINTR; | |
5040 | break; | |
5041 | } | |
5042 | if (!batch_count--) { | |
5043 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5044 | cond_resched(); | |
5045 | } | |
c0ff4b85 R |
5046 | ret = __mem_cgroup_try_charge(NULL, |
5047 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 5048 | if (ret) |
854ffa8d | 5049 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 5050 | return ret; |
854ffa8d DN |
5051 | mc.precharge++; |
5052 | } | |
4ffef5fe DN |
5053 | return ret; |
5054 | } | |
5055 | ||
5056 | /** | |
8d32ff84 | 5057 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5058 | * @vma: the vma the pte to be checked belongs |
5059 | * @addr: the address corresponding to the pte to be checked | |
5060 | * @ptent: the pte to be checked | |
02491447 | 5061 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5062 | * |
5063 | * Returns | |
5064 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5065 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5066 | * move charge. if @target is not NULL, the page is stored in target->page | |
5067 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5068 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5069 | * target for charge migration. if @target is not NULL, the entry is stored | |
5070 | * in target->ent. | |
4ffef5fe DN |
5071 | * |
5072 | * Called with pte lock held. | |
5073 | */ | |
4ffef5fe DN |
5074 | union mc_target { |
5075 | struct page *page; | |
02491447 | 5076 | swp_entry_t ent; |
4ffef5fe DN |
5077 | }; |
5078 | ||
4ffef5fe | 5079 | enum mc_target_type { |
8d32ff84 | 5080 | MC_TARGET_NONE = 0, |
4ffef5fe | 5081 | MC_TARGET_PAGE, |
02491447 | 5082 | MC_TARGET_SWAP, |
4ffef5fe DN |
5083 | }; |
5084 | ||
90254a65 DN |
5085 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5086 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5087 | { |
90254a65 | 5088 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5089 | |
90254a65 DN |
5090 | if (!page || !page_mapped(page)) |
5091 | return NULL; | |
5092 | if (PageAnon(page)) { | |
5093 | /* we don't move shared anon */ | |
4b91355e | 5094 | if (!move_anon()) |
90254a65 | 5095 | return NULL; |
87946a72 DN |
5096 | } else if (!move_file()) |
5097 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5098 | return NULL; |
5099 | if (!get_page_unless_zero(page)) | |
5100 | return NULL; | |
5101 | ||
5102 | return page; | |
5103 | } | |
5104 | ||
4b91355e | 5105 | #ifdef CONFIG_SWAP |
90254a65 DN |
5106 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5107 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5108 | { | |
90254a65 DN |
5109 | struct page *page = NULL; |
5110 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5111 | ||
5112 | if (!move_anon() || non_swap_entry(ent)) | |
5113 | return NULL; | |
4b91355e KH |
5114 | /* |
5115 | * Because lookup_swap_cache() updates some statistics counter, | |
5116 | * we call find_get_page() with swapper_space directly. | |
5117 | */ | |
5118 | page = find_get_page(&swapper_space, ent.val); | |
90254a65 DN |
5119 | if (do_swap_account) |
5120 | entry->val = ent.val; | |
5121 | ||
5122 | return page; | |
5123 | } | |
4b91355e KH |
5124 | #else |
5125 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5126 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5127 | { | |
5128 | return NULL; | |
5129 | } | |
5130 | #endif | |
90254a65 | 5131 | |
87946a72 DN |
5132 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5133 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5134 | { | |
5135 | struct page *page = NULL; | |
87946a72 DN |
5136 | struct address_space *mapping; |
5137 | pgoff_t pgoff; | |
5138 | ||
5139 | if (!vma->vm_file) /* anonymous vma */ | |
5140 | return NULL; | |
5141 | if (!move_file()) | |
5142 | return NULL; | |
5143 | ||
87946a72 DN |
5144 | mapping = vma->vm_file->f_mapping; |
5145 | if (pte_none(ptent)) | |
5146 | pgoff = linear_page_index(vma, addr); | |
5147 | else /* pte_file(ptent) is true */ | |
5148 | pgoff = pte_to_pgoff(ptent); | |
5149 | ||
5150 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5151 | page = find_get_page(mapping, pgoff); |
5152 | ||
5153 | #ifdef CONFIG_SWAP | |
5154 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5155 | if (radix_tree_exceptional_entry(page)) { | |
5156 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5157 | if (do_swap_account) |
aa3b1895 HD |
5158 | *entry = swap; |
5159 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5160 | } |
aa3b1895 | 5161 | #endif |
87946a72 DN |
5162 | return page; |
5163 | } | |
5164 | ||
8d32ff84 | 5165 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5166 | unsigned long addr, pte_t ptent, union mc_target *target) |
5167 | { | |
5168 | struct page *page = NULL; | |
5169 | struct page_cgroup *pc; | |
8d32ff84 | 5170 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5171 | swp_entry_t ent = { .val = 0 }; |
5172 | ||
5173 | if (pte_present(ptent)) | |
5174 | page = mc_handle_present_pte(vma, addr, ptent); | |
5175 | else if (is_swap_pte(ptent)) | |
5176 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5177 | else if (pte_none(ptent) || pte_file(ptent)) |
5178 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5179 | |
5180 | if (!page && !ent.val) | |
8d32ff84 | 5181 | return ret; |
02491447 DN |
5182 | if (page) { |
5183 | pc = lookup_page_cgroup(page); | |
5184 | /* | |
5185 | * Do only loose check w/o page_cgroup lock. | |
5186 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5187 | * the lock. | |
5188 | */ | |
5189 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5190 | ret = MC_TARGET_PAGE; | |
5191 | if (target) | |
5192 | target->page = page; | |
5193 | } | |
5194 | if (!ret || !target) | |
5195 | put_page(page); | |
5196 | } | |
90254a65 DN |
5197 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5198 | if (ent.val && !ret && | |
9fb4b7cc | 5199 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5200 | ret = MC_TARGET_SWAP; |
5201 | if (target) | |
5202 | target->ent = ent; | |
4ffef5fe | 5203 | } |
4ffef5fe DN |
5204 | return ret; |
5205 | } | |
5206 | ||
12724850 NH |
5207 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5208 | /* | |
5209 | * We don't consider swapping or file mapped pages because THP does not | |
5210 | * support them for now. | |
5211 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5212 | */ | |
5213 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5214 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5215 | { | |
5216 | struct page *page = NULL; | |
5217 | struct page_cgroup *pc; | |
5218 | enum mc_target_type ret = MC_TARGET_NONE; | |
5219 | ||
5220 | page = pmd_page(pmd); | |
5221 | VM_BUG_ON(!page || !PageHead(page)); | |
5222 | if (!move_anon()) | |
5223 | return ret; | |
5224 | pc = lookup_page_cgroup(page); | |
5225 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5226 | ret = MC_TARGET_PAGE; | |
5227 | if (target) { | |
5228 | get_page(page); | |
5229 | target->page = page; | |
5230 | } | |
5231 | } | |
5232 | return ret; | |
5233 | } | |
5234 | #else | |
5235 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5236 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5237 | { | |
5238 | return MC_TARGET_NONE; | |
5239 | } | |
5240 | #endif | |
5241 | ||
4ffef5fe DN |
5242 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5243 | unsigned long addr, unsigned long end, | |
5244 | struct mm_walk *walk) | |
5245 | { | |
5246 | struct vm_area_struct *vma = walk->private; | |
5247 | pte_t *pte; | |
5248 | spinlock_t *ptl; | |
5249 | ||
12724850 NH |
5250 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
5251 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
5252 | mc.precharge += HPAGE_PMD_NR; | |
5253 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5254 | return 0; |
12724850 | 5255 | } |
03319327 | 5256 | |
45f83cef AA |
5257 | if (pmd_trans_unstable(pmd)) |
5258 | return 0; | |
4ffef5fe DN |
5259 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5260 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5261 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5262 | mc.precharge++; /* increment precharge temporarily */ |
5263 | pte_unmap_unlock(pte - 1, ptl); | |
5264 | cond_resched(); | |
5265 | ||
7dc74be0 DN |
5266 | return 0; |
5267 | } | |
5268 | ||
4ffef5fe DN |
5269 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5270 | { | |
5271 | unsigned long precharge; | |
5272 | struct vm_area_struct *vma; | |
5273 | ||
dfe076b0 | 5274 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5275 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5276 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5277 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5278 | .mm = mm, | |
5279 | .private = vma, | |
5280 | }; | |
5281 | if (is_vm_hugetlb_page(vma)) | |
5282 | continue; | |
4ffef5fe DN |
5283 | walk_page_range(vma->vm_start, vma->vm_end, |
5284 | &mem_cgroup_count_precharge_walk); | |
5285 | } | |
dfe076b0 | 5286 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5287 | |
5288 | precharge = mc.precharge; | |
5289 | mc.precharge = 0; | |
5290 | ||
5291 | return precharge; | |
5292 | } | |
5293 | ||
4ffef5fe DN |
5294 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5295 | { | |
dfe076b0 DN |
5296 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5297 | ||
5298 | VM_BUG_ON(mc.moving_task); | |
5299 | mc.moving_task = current; | |
5300 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5301 | } |
5302 | ||
dfe076b0 DN |
5303 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5304 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5305 | { |
2bd9bb20 KH |
5306 | struct mem_cgroup *from = mc.from; |
5307 | struct mem_cgroup *to = mc.to; | |
5308 | ||
4ffef5fe | 5309 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5310 | if (mc.precharge) { |
5311 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5312 | mc.precharge = 0; | |
5313 | } | |
5314 | /* | |
5315 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5316 | * we must uncharge here. | |
5317 | */ | |
5318 | if (mc.moved_charge) { | |
5319 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5320 | mc.moved_charge = 0; | |
4ffef5fe | 5321 | } |
483c30b5 DN |
5322 | /* we must fixup refcnts and charges */ |
5323 | if (mc.moved_swap) { | |
483c30b5 DN |
5324 | /* uncharge swap account from the old cgroup */ |
5325 | if (!mem_cgroup_is_root(mc.from)) | |
5326 | res_counter_uncharge(&mc.from->memsw, | |
5327 | PAGE_SIZE * mc.moved_swap); | |
5328 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5329 | ||
5330 | if (!mem_cgroup_is_root(mc.to)) { | |
5331 | /* | |
5332 | * we charged both to->res and to->memsw, so we should | |
5333 | * uncharge to->res. | |
5334 | */ | |
5335 | res_counter_uncharge(&mc.to->res, | |
5336 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5337 | } |
5338 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5339 | mc.moved_swap = 0; |
5340 | } | |
dfe076b0 DN |
5341 | memcg_oom_recover(from); |
5342 | memcg_oom_recover(to); | |
5343 | wake_up_all(&mc.waitq); | |
5344 | } | |
5345 | ||
5346 | static void mem_cgroup_clear_mc(void) | |
5347 | { | |
5348 | struct mem_cgroup *from = mc.from; | |
5349 | ||
5350 | /* | |
5351 | * we must clear moving_task before waking up waiters at the end of | |
5352 | * task migration. | |
5353 | */ | |
5354 | mc.moving_task = NULL; | |
5355 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5356 | spin_lock(&mc.lock); |
4ffef5fe DN |
5357 | mc.from = NULL; |
5358 | mc.to = NULL; | |
2bd9bb20 | 5359 | spin_unlock(&mc.lock); |
32047e2a | 5360 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5361 | } |
5362 | ||
761b3ef5 LZ |
5363 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5364 | struct cgroup_taskset *tset) | |
7dc74be0 | 5365 | { |
2f7ee569 | 5366 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5367 | int ret = 0; |
c0ff4b85 | 5368 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5369 | |
c0ff4b85 | 5370 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5371 | struct mm_struct *mm; |
5372 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5373 | ||
c0ff4b85 | 5374 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5375 | |
5376 | mm = get_task_mm(p); | |
5377 | if (!mm) | |
5378 | return 0; | |
7dc74be0 | 5379 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5380 | if (mm->owner == p) { |
5381 | VM_BUG_ON(mc.from); | |
5382 | VM_BUG_ON(mc.to); | |
5383 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5384 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5385 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5386 | mem_cgroup_start_move(from); |
2bd9bb20 | 5387 | spin_lock(&mc.lock); |
4ffef5fe | 5388 | mc.from = from; |
c0ff4b85 | 5389 | mc.to = memcg; |
2bd9bb20 | 5390 | spin_unlock(&mc.lock); |
dfe076b0 | 5391 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5392 | |
5393 | ret = mem_cgroup_precharge_mc(mm); | |
5394 | if (ret) | |
5395 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5396 | } |
5397 | mmput(mm); | |
7dc74be0 DN |
5398 | } |
5399 | return ret; | |
5400 | } | |
5401 | ||
761b3ef5 LZ |
5402 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5403 | struct cgroup_taskset *tset) | |
7dc74be0 | 5404 | { |
4ffef5fe | 5405 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5406 | } |
5407 | ||
4ffef5fe DN |
5408 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5409 | unsigned long addr, unsigned long end, | |
5410 | struct mm_walk *walk) | |
7dc74be0 | 5411 | { |
4ffef5fe DN |
5412 | int ret = 0; |
5413 | struct vm_area_struct *vma = walk->private; | |
5414 | pte_t *pte; | |
5415 | spinlock_t *ptl; | |
12724850 NH |
5416 | enum mc_target_type target_type; |
5417 | union mc_target target; | |
5418 | struct page *page; | |
5419 | struct page_cgroup *pc; | |
4ffef5fe | 5420 | |
12724850 NH |
5421 | /* |
5422 | * We don't take compound_lock() here but no race with splitting thp | |
5423 | * happens because: | |
5424 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5425 | * under splitting, which means there's no concurrent thp split, | |
5426 | * - if another thread runs into split_huge_page() just after we | |
5427 | * entered this if-block, the thread must wait for page table lock | |
5428 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5429 | * part of thp split is not executed yet. | |
5430 | */ | |
5431 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 5432 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
5433 | spin_unlock(&vma->vm_mm->page_table_lock); |
5434 | return 0; | |
5435 | } | |
5436 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5437 | if (target_type == MC_TARGET_PAGE) { | |
5438 | page = target.page; | |
5439 | if (!isolate_lru_page(page)) { | |
5440 | pc = lookup_page_cgroup(page); | |
5441 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5442 | pc, mc.from, mc.to)) { |
12724850 NH |
5443 | mc.precharge -= HPAGE_PMD_NR; |
5444 | mc.moved_charge += HPAGE_PMD_NR; | |
5445 | } | |
5446 | putback_lru_page(page); | |
5447 | } | |
5448 | put_page(page); | |
5449 | } | |
5450 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5451 | return 0; |
12724850 NH |
5452 | } |
5453 | ||
45f83cef AA |
5454 | if (pmd_trans_unstable(pmd)) |
5455 | return 0; | |
4ffef5fe DN |
5456 | retry: |
5457 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5458 | for (; addr != end; addr += PAGE_SIZE) { | |
5459 | pte_t ptent = *(pte++); | |
02491447 | 5460 | swp_entry_t ent; |
4ffef5fe DN |
5461 | |
5462 | if (!mc.precharge) | |
5463 | break; | |
5464 | ||
8d32ff84 | 5465 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5466 | case MC_TARGET_PAGE: |
5467 | page = target.page; | |
5468 | if (isolate_lru_page(page)) | |
5469 | goto put; | |
5470 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5471 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5472 | mc.from, mc.to)) { |
4ffef5fe | 5473 | mc.precharge--; |
854ffa8d DN |
5474 | /* we uncharge from mc.from later. */ |
5475 | mc.moved_charge++; | |
4ffef5fe DN |
5476 | } |
5477 | putback_lru_page(page); | |
8d32ff84 | 5478 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5479 | put_page(page); |
5480 | break; | |
02491447 DN |
5481 | case MC_TARGET_SWAP: |
5482 | ent = target.ent; | |
e91cbb42 | 5483 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5484 | mc.precharge--; |
483c30b5 DN |
5485 | /* we fixup refcnts and charges later. */ |
5486 | mc.moved_swap++; | |
5487 | } | |
02491447 | 5488 | break; |
4ffef5fe DN |
5489 | default: |
5490 | break; | |
5491 | } | |
5492 | } | |
5493 | pte_unmap_unlock(pte - 1, ptl); | |
5494 | cond_resched(); | |
5495 | ||
5496 | if (addr != end) { | |
5497 | /* | |
5498 | * We have consumed all precharges we got in can_attach(). | |
5499 | * We try charge one by one, but don't do any additional | |
5500 | * charges to mc.to if we have failed in charge once in attach() | |
5501 | * phase. | |
5502 | */ | |
854ffa8d | 5503 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5504 | if (!ret) |
5505 | goto retry; | |
5506 | } | |
5507 | ||
5508 | return ret; | |
5509 | } | |
5510 | ||
5511 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5512 | { | |
5513 | struct vm_area_struct *vma; | |
5514 | ||
5515 | lru_add_drain_all(); | |
dfe076b0 DN |
5516 | retry: |
5517 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5518 | /* | |
5519 | * Someone who are holding the mmap_sem might be waiting in | |
5520 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5521 | * and retry. Because we cancel precharges, we might not be able | |
5522 | * to move enough charges, but moving charge is a best-effort | |
5523 | * feature anyway, so it wouldn't be a big problem. | |
5524 | */ | |
5525 | __mem_cgroup_clear_mc(); | |
5526 | cond_resched(); | |
5527 | goto retry; | |
5528 | } | |
4ffef5fe DN |
5529 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5530 | int ret; | |
5531 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5532 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5533 | .mm = mm, | |
5534 | .private = vma, | |
5535 | }; | |
5536 | if (is_vm_hugetlb_page(vma)) | |
5537 | continue; | |
4ffef5fe DN |
5538 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5539 | &mem_cgroup_move_charge_walk); | |
5540 | if (ret) | |
5541 | /* | |
5542 | * means we have consumed all precharges and failed in | |
5543 | * doing additional charge. Just abandon here. | |
5544 | */ | |
5545 | break; | |
5546 | } | |
dfe076b0 | 5547 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5548 | } |
5549 | ||
761b3ef5 LZ |
5550 | static void mem_cgroup_move_task(struct cgroup *cont, |
5551 | struct cgroup_taskset *tset) | |
67e465a7 | 5552 | { |
2f7ee569 | 5553 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5554 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5555 | |
dfe076b0 | 5556 | if (mm) { |
a433658c KM |
5557 | if (mc.to) |
5558 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5559 | mmput(mm); |
5560 | } | |
a433658c KM |
5561 | if (mc.to) |
5562 | mem_cgroup_clear_mc(); | |
67e465a7 | 5563 | } |
5cfb80a7 | 5564 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
5565 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5566 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5567 | { |
5568 | return 0; | |
5569 | } | |
761b3ef5 LZ |
5570 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5571 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5572 | { |
5573 | } | |
761b3ef5 LZ |
5574 | static void mem_cgroup_move_task(struct cgroup *cont, |
5575 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5576 | { |
5577 | } | |
5578 | #endif | |
67e465a7 | 5579 | |
8cdea7c0 BS |
5580 | struct cgroup_subsys mem_cgroup_subsys = { |
5581 | .name = "memory", | |
5582 | .subsys_id = mem_cgroup_subsys_id, | |
5583 | .create = mem_cgroup_create, | |
df878fb0 | 5584 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 | 5585 | .destroy = mem_cgroup_destroy, |
7dc74be0 DN |
5586 | .can_attach = mem_cgroup_can_attach, |
5587 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5588 | .attach = mem_cgroup_move_task, |
6bc10349 | 5589 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 5590 | .early_init = 0, |
04046e1a | 5591 | .use_id = 1, |
48ddbe19 | 5592 | .__DEPRECATED_clear_css_refs = true, |
8cdea7c0 | 5593 | }; |
c077719b | 5594 | |
c255a458 | 5595 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5596 | static int __init enable_swap_account(char *s) |
5597 | { | |
5598 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5599 | if (!strcmp(s, "1")) |
a42c390c | 5600 | really_do_swap_account = 1; |
a2c8990a | 5601 | else if (!strcmp(s, "0")) |
a42c390c MH |
5602 | really_do_swap_account = 0; |
5603 | return 1; | |
5604 | } | |
a2c8990a | 5605 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5606 | |
c077719b | 5607 | #endif |