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 | 407 | |
b2145145 WL |
408 | static inline |
409 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) | |
410 | { | |
411 | return container_of(s, struct mem_cgroup, css); | |
412 | } | |
413 | ||
e1aab161 | 414 | /* Writing them here to avoid exposing memcg's inner layout */ |
c255a458 | 415 | #ifdef CONFIG_MEMCG_KMEM |
e1aab161 | 416 | #include <net/sock.h> |
d1a4c0b3 | 417 | #include <net/ip.h> |
e1aab161 GC |
418 | |
419 | static bool mem_cgroup_is_root(struct mem_cgroup *memcg); | |
420 | void sock_update_memcg(struct sock *sk) | |
421 | { | |
376be5ff | 422 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 423 | struct mem_cgroup *memcg; |
3f134619 | 424 | struct cg_proto *cg_proto; |
e1aab161 GC |
425 | |
426 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
427 | ||
f3f511e1 GC |
428 | /* Socket cloning can throw us here with sk_cgrp already |
429 | * filled. It won't however, necessarily happen from | |
430 | * process context. So the test for root memcg given | |
431 | * the current task's memcg won't help us in this case. | |
432 | * | |
433 | * Respecting the original socket's memcg is a better | |
434 | * decision in this case. | |
435 | */ | |
436 | if (sk->sk_cgrp) { | |
437 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
438 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
439 | return; | |
440 | } | |
441 | ||
e1aab161 GC |
442 | rcu_read_lock(); |
443 | memcg = mem_cgroup_from_task(current); | |
3f134619 GC |
444 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
445 | if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) { | |
e1aab161 | 446 | mem_cgroup_get(memcg); |
3f134619 | 447 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
448 | } |
449 | rcu_read_unlock(); | |
450 | } | |
451 | } | |
452 | EXPORT_SYMBOL(sock_update_memcg); | |
453 | ||
454 | void sock_release_memcg(struct sock *sk) | |
455 | { | |
376be5ff | 456 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
457 | struct mem_cgroup *memcg; |
458 | WARN_ON(!sk->sk_cgrp->memcg); | |
459 | memcg = sk->sk_cgrp->memcg; | |
460 | mem_cgroup_put(memcg); | |
461 | } | |
462 | } | |
d1a4c0b3 | 463 | |
319d3b9c | 464 | #ifdef CONFIG_INET |
d1a4c0b3 GC |
465 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
466 | { | |
467 | if (!memcg || mem_cgroup_is_root(memcg)) | |
468 | return NULL; | |
469 | ||
470 | return &memcg->tcp_mem.cg_proto; | |
471 | } | |
472 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 473 | #endif /* CONFIG_INET */ |
c255a458 | 474 | #endif /* CONFIG_MEMCG_KMEM */ |
e1aab161 | 475 | |
c255a458 | 476 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
3f134619 GC |
477 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
478 | { | |
479 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
480 | return; | |
481 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
482 | } | |
483 | #else | |
484 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
485 | { | |
486 | } | |
487 | #endif | |
488 | ||
c0ff4b85 | 489 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 490 | |
f64c3f54 | 491 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 492 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 493 | { |
c0ff4b85 | 494 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
495 | } |
496 | ||
c0ff4b85 | 497 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 498 | { |
c0ff4b85 | 499 | return &memcg->css; |
d324236b WF |
500 | } |
501 | ||
f64c3f54 | 502 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 503 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 504 | { |
97a6c37b JW |
505 | int nid = page_to_nid(page); |
506 | int zid = page_zonenum(page); | |
f64c3f54 | 507 | |
c0ff4b85 | 508 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
509 | } |
510 | ||
511 | static struct mem_cgroup_tree_per_zone * | |
512 | soft_limit_tree_node_zone(int nid, int zid) | |
513 | { | |
514 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
515 | } | |
516 | ||
517 | static struct mem_cgroup_tree_per_zone * | |
518 | soft_limit_tree_from_page(struct page *page) | |
519 | { | |
520 | int nid = page_to_nid(page); | |
521 | int zid = page_zonenum(page); | |
522 | ||
523 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
524 | } | |
525 | ||
526 | static void | |
c0ff4b85 | 527 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 528 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
529 | struct mem_cgroup_tree_per_zone *mctz, |
530 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
531 | { |
532 | struct rb_node **p = &mctz->rb_root.rb_node; | |
533 | struct rb_node *parent = NULL; | |
534 | struct mem_cgroup_per_zone *mz_node; | |
535 | ||
536 | if (mz->on_tree) | |
537 | return; | |
538 | ||
ef8745c1 KH |
539 | mz->usage_in_excess = new_usage_in_excess; |
540 | if (!mz->usage_in_excess) | |
541 | return; | |
f64c3f54 BS |
542 | while (*p) { |
543 | parent = *p; | |
544 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
545 | tree_node); | |
546 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
547 | p = &(*p)->rb_left; | |
548 | /* | |
549 | * We can't avoid mem cgroups that are over their soft | |
550 | * limit by the same amount | |
551 | */ | |
552 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
553 | p = &(*p)->rb_right; | |
554 | } | |
555 | rb_link_node(&mz->tree_node, parent, p); | |
556 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
557 | mz->on_tree = true; | |
4e416953 BS |
558 | } |
559 | ||
560 | static void | |
c0ff4b85 | 561 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
562 | struct mem_cgroup_per_zone *mz, |
563 | struct mem_cgroup_tree_per_zone *mctz) | |
564 | { | |
565 | if (!mz->on_tree) | |
566 | return; | |
567 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
568 | mz->on_tree = false; | |
569 | } | |
570 | ||
f64c3f54 | 571 | static void |
c0ff4b85 | 572 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
573 | struct mem_cgroup_per_zone *mz, |
574 | struct mem_cgroup_tree_per_zone *mctz) | |
575 | { | |
576 | spin_lock(&mctz->lock); | |
c0ff4b85 | 577 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
578 | spin_unlock(&mctz->lock); |
579 | } | |
580 | ||
f64c3f54 | 581 | |
c0ff4b85 | 582 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 583 | { |
ef8745c1 | 584 | unsigned long long excess; |
f64c3f54 BS |
585 | struct mem_cgroup_per_zone *mz; |
586 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
587 | int nid = page_to_nid(page); |
588 | int zid = page_zonenum(page); | |
f64c3f54 BS |
589 | mctz = soft_limit_tree_from_page(page); |
590 | ||
591 | /* | |
4e649152 KH |
592 | * Necessary to update all ancestors when hierarchy is used. |
593 | * because their event counter is not touched. | |
f64c3f54 | 594 | */ |
c0ff4b85 R |
595 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
596 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
597 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
598 | /* |
599 | * We have to update the tree if mz is on RB-tree or | |
600 | * mem is over its softlimit. | |
601 | */ | |
ef8745c1 | 602 | if (excess || mz->on_tree) { |
4e649152 KH |
603 | spin_lock(&mctz->lock); |
604 | /* if on-tree, remove it */ | |
605 | if (mz->on_tree) | |
c0ff4b85 | 606 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 607 | /* |
ef8745c1 KH |
608 | * Insert again. mz->usage_in_excess will be updated. |
609 | * If excess is 0, no tree ops. | |
4e649152 | 610 | */ |
c0ff4b85 | 611 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
612 | spin_unlock(&mctz->lock); |
613 | } | |
f64c3f54 BS |
614 | } |
615 | } | |
616 | ||
c0ff4b85 | 617 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
618 | { |
619 | int node, zone; | |
620 | struct mem_cgroup_per_zone *mz; | |
621 | struct mem_cgroup_tree_per_zone *mctz; | |
622 | ||
3ed28fa1 | 623 | for_each_node(node) { |
f64c3f54 | 624 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 625 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 626 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 627 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
628 | } |
629 | } | |
630 | } | |
631 | ||
4e416953 BS |
632 | static struct mem_cgroup_per_zone * |
633 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
634 | { | |
635 | struct rb_node *rightmost = NULL; | |
26251eaf | 636 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
637 | |
638 | retry: | |
26251eaf | 639 | mz = NULL; |
4e416953 BS |
640 | rightmost = rb_last(&mctz->rb_root); |
641 | if (!rightmost) | |
642 | goto done; /* Nothing to reclaim from */ | |
643 | ||
644 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
645 | /* | |
646 | * Remove the node now but someone else can add it back, | |
647 | * we will to add it back at the end of reclaim to its correct | |
648 | * position in the tree. | |
649 | */ | |
d79154bb HD |
650 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
651 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
652 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
653 | goto retry; |
654 | done: | |
655 | return mz; | |
656 | } | |
657 | ||
658 | static struct mem_cgroup_per_zone * | |
659 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
660 | { | |
661 | struct mem_cgroup_per_zone *mz; | |
662 | ||
663 | spin_lock(&mctz->lock); | |
664 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
665 | spin_unlock(&mctz->lock); | |
666 | return mz; | |
667 | } | |
668 | ||
711d3d2c KH |
669 | /* |
670 | * Implementation Note: reading percpu statistics for memcg. | |
671 | * | |
672 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
673 | * synchronization to implement "quick" read. There are trade-off between | |
674 | * reading cost and precision of value. Then, we may have a chance to implement | |
675 | * a periodic synchronizion of counter in memcg's counter. | |
676 | * | |
677 | * But this _read() function is used for user interface now. The user accounts | |
678 | * memory usage by memory cgroup and he _always_ requires exact value because | |
679 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
680 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
681 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
682 | * | |
683 | * If there are kernel internal actions which can make use of some not-exact | |
684 | * value, and reading all cpu value can be performance bottleneck in some | |
685 | * common workload, threashold and synchonization as vmstat[] should be | |
686 | * implemented. | |
687 | */ | |
c0ff4b85 | 688 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 689 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 690 | { |
7a159cc9 | 691 | long val = 0; |
c62b1a3b | 692 | int cpu; |
c62b1a3b | 693 | |
711d3d2c KH |
694 | get_online_cpus(); |
695 | for_each_online_cpu(cpu) | |
c0ff4b85 | 696 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 697 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
698 | spin_lock(&memcg->pcp_counter_lock); |
699 | val += memcg->nocpu_base.count[idx]; | |
700 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
701 | #endif |
702 | put_online_cpus(); | |
c62b1a3b KH |
703 | return val; |
704 | } | |
705 | ||
c0ff4b85 | 706 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
707 | bool charge) |
708 | { | |
709 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 710 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
711 | } |
712 | ||
c0ff4b85 | 713 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
714 | enum mem_cgroup_events_index idx) |
715 | { | |
716 | unsigned long val = 0; | |
717 | int cpu; | |
718 | ||
719 | for_each_online_cpu(cpu) | |
c0ff4b85 | 720 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 721 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
722 | spin_lock(&memcg->pcp_counter_lock); |
723 | val += memcg->nocpu_base.events[idx]; | |
724 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
725 | #endif |
726 | return val; | |
727 | } | |
728 | ||
c0ff4b85 | 729 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 730 | bool anon, int nr_pages) |
d52aa412 | 731 | { |
c62b1a3b KH |
732 | preempt_disable(); |
733 | ||
b2402857 KH |
734 | /* |
735 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
736 | * counted as CACHE even if it's on ANON LRU. | |
737 | */ | |
738 | if (anon) | |
739 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 740 | nr_pages); |
d52aa412 | 741 | else |
b2402857 | 742 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 743 | nr_pages); |
55e462b0 | 744 | |
e401f176 KH |
745 | /* pagein of a big page is an event. So, ignore page size */ |
746 | if (nr_pages > 0) | |
c0ff4b85 | 747 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 748 | else { |
c0ff4b85 | 749 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
750 | nr_pages = -nr_pages; /* for event */ |
751 | } | |
e401f176 | 752 | |
13114716 | 753 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 754 | |
c62b1a3b | 755 | preempt_enable(); |
6d12e2d8 KH |
756 | } |
757 | ||
bb2a0de9 | 758 | unsigned long |
4d7dcca2 | 759 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
760 | { |
761 | struct mem_cgroup_per_zone *mz; | |
762 | ||
763 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
764 | return mz->lru_size[lru]; | |
765 | } | |
766 | ||
767 | static unsigned long | |
c0ff4b85 | 768 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 769 | unsigned int lru_mask) |
889976db YH |
770 | { |
771 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 772 | enum lru_list lru; |
bb2a0de9 KH |
773 | unsigned long ret = 0; |
774 | ||
c0ff4b85 | 775 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 776 | |
f156ab93 HD |
777 | for_each_lru(lru) { |
778 | if (BIT(lru) & lru_mask) | |
779 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
780 | } |
781 | return ret; | |
782 | } | |
783 | ||
784 | static unsigned long | |
c0ff4b85 | 785 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
786 | int nid, unsigned int lru_mask) |
787 | { | |
889976db YH |
788 | u64 total = 0; |
789 | int zid; | |
790 | ||
bb2a0de9 | 791 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
792 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
793 | nid, zid, lru_mask); | |
bb2a0de9 | 794 | |
889976db YH |
795 | return total; |
796 | } | |
bb2a0de9 | 797 | |
c0ff4b85 | 798 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 799 | unsigned int lru_mask) |
6d12e2d8 | 800 | { |
889976db | 801 | int nid; |
6d12e2d8 KH |
802 | u64 total = 0; |
803 | ||
bb2a0de9 | 804 | for_each_node_state(nid, N_HIGH_MEMORY) |
c0ff4b85 | 805 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 806 | return total; |
d52aa412 KH |
807 | } |
808 | ||
f53d7ce3 JW |
809 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
810 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
811 | { |
812 | unsigned long val, next; | |
813 | ||
13114716 | 814 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 815 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 816 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
817 | if ((long)next - (long)val < 0) { |
818 | switch (target) { | |
819 | case MEM_CGROUP_TARGET_THRESH: | |
820 | next = val + THRESHOLDS_EVENTS_TARGET; | |
821 | break; | |
822 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
823 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
824 | break; | |
825 | case MEM_CGROUP_TARGET_NUMAINFO: | |
826 | next = val + NUMAINFO_EVENTS_TARGET; | |
827 | break; | |
828 | default: | |
829 | break; | |
830 | } | |
831 | __this_cpu_write(memcg->stat->targets[target], next); | |
832 | return true; | |
7a159cc9 | 833 | } |
f53d7ce3 | 834 | return false; |
d2265e6f KH |
835 | } |
836 | ||
837 | /* | |
838 | * Check events in order. | |
839 | * | |
840 | */ | |
c0ff4b85 | 841 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 842 | { |
4799401f | 843 | preempt_disable(); |
d2265e6f | 844 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
845 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
846 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
847 | bool do_softlimit; |
848 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
849 | |
850 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
851 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
852 | #if MAX_NUMNODES > 1 | |
853 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
854 | MEM_CGROUP_TARGET_NUMAINFO); | |
855 | #endif | |
856 | preempt_enable(); | |
857 | ||
c0ff4b85 | 858 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 859 | if (unlikely(do_softlimit)) |
c0ff4b85 | 860 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 861 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 862 | if (unlikely(do_numainfo)) |
c0ff4b85 | 863 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 864 | #endif |
f53d7ce3 JW |
865 | } else |
866 | preempt_enable(); | |
d2265e6f KH |
867 | } |
868 | ||
d1a4c0b3 | 869 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 | 870 | { |
b2145145 WL |
871 | return mem_cgroup_from_css( |
872 | cgroup_subsys_state(cont, mem_cgroup_subsys_id)); | |
8cdea7c0 BS |
873 | } |
874 | ||
cf475ad2 | 875 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 876 | { |
31a78f23 BS |
877 | /* |
878 | * mm_update_next_owner() may clear mm->owner to NULL | |
879 | * if it races with swapoff, page migration, etc. | |
880 | * So this can be called with p == NULL. | |
881 | */ | |
882 | if (unlikely(!p)) | |
883 | return NULL; | |
884 | ||
b2145145 | 885 | return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
886 | } |
887 | ||
a433658c | 888 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 889 | { |
c0ff4b85 | 890 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
891 | |
892 | if (!mm) | |
893 | return NULL; | |
54595fe2 KH |
894 | /* |
895 | * Because we have no locks, mm->owner's may be being moved to other | |
896 | * cgroup. We use css_tryget() here even if this looks | |
897 | * pessimistic (rather than adding locks here). | |
898 | */ | |
899 | rcu_read_lock(); | |
900 | do { | |
c0ff4b85 R |
901 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
902 | if (unlikely(!memcg)) | |
54595fe2 | 903 | break; |
c0ff4b85 | 904 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 905 | rcu_read_unlock(); |
c0ff4b85 | 906 | return memcg; |
54595fe2 KH |
907 | } |
908 | ||
5660048c JW |
909 | /** |
910 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
911 | * @root: hierarchy root | |
912 | * @prev: previously returned memcg, NULL on first invocation | |
913 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
914 | * | |
915 | * Returns references to children of the hierarchy below @root, or | |
916 | * @root itself, or %NULL after a full round-trip. | |
917 | * | |
918 | * Caller must pass the return value in @prev on subsequent | |
919 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
920 | * to cancel a hierarchy walk before the round-trip is complete. | |
921 | * | |
922 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
923 | * divide up the memcgs in the hierarchy among all concurrent | |
924 | * reclaimers operating on the same zone and priority. | |
925 | */ | |
926 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
927 | struct mem_cgroup *prev, | |
928 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 929 | { |
9f3a0d09 JW |
930 | struct mem_cgroup *memcg = NULL; |
931 | int id = 0; | |
711d3d2c | 932 | |
5660048c JW |
933 | if (mem_cgroup_disabled()) |
934 | return NULL; | |
935 | ||
9f3a0d09 JW |
936 | if (!root) |
937 | root = root_mem_cgroup; | |
7d74b06f | 938 | |
9f3a0d09 JW |
939 | if (prev && !reclaim) |
940 | id = css_id(&prev->css); | |
14067bb3 | 941 | |
9f3a0d09 JW |
942 | if (prev && prev != root) |
943 | css_put(&prev->css); | |
14067bb3 | 944 | |
9f3a0d09 JW |
945 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
946 | if (prev) | |
947 | return NULL; | |
948 | return root; | |
949 | } | |
14067bb3 | 950 | |
9f3a0d09 | 951 | while (!memcg) { |
527a5ec9 | 952 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 953 | struct cgroup_subsys_state *css; |
711d3d2c | 954 | |
527a5ec9 JW |
955 | if (reclaim) { |
956 | int nid = zone_to_nid(reclaim->zone); | |
957 | int zid = zone_idx(reclaim->zone); | |
958 | struct mem_cgroup_per_zone *mz; | |
959 | ||
960 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
961 | iter = &mz->reclaim_iter[reclaim->priority]; | |
962 | if (prev && reclaim->generation != iter->generation) | |
963 | return NULL; | |
964 | id = iter->position; | |
965 | } | |
7d74b06f | 966 | |
9f3a0d09 JW |
967 | rcu_read_lock(); |
968 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
969 | if (css) { | |
970 | if (css == &root->css || css_tryget(css)) | |
b2145145 | 971 | memcg = mem_cgroup_from_css(css); |
9f3a0d09 JW |
972 | } else |
973 | id = 0; | |
14067bb3 | 974 | rcu_read_unlock(); |
14067bb3 | 975 | |
527a5ec9 JW |
976 | if (reclaim) { |
977 | iter->position = id; | |
978 | if (!css) | |
979 | iter->generation++; | |
980 | else if (!prev && memcg) | |
981 | reclaim->generation = iter->generation; | |
982 | } | |
9f3a0d09 JW |
983 | |
984 | if (prev && !css) | |
985 | return NULL; | |
986 | } | |
987 | return memcg; | |
14067bb3 | 988 | } |
7d74b06f | 989 | |
5660048c JW |
990 | /** |
991 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
992 | * @root: hierarchy root | |
993 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
994 | */ | |
995 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
996 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
997 | { |
998 | if (!root) | |
999 | root = root_mem_cgroup; | |
1000 | if (prev && prev != root) | |
1001 | css_put(&prev->css); | |
1002 | } | |
7d74b06f | 1003 | |
9f3a0d09 JW |
1004 | /* |
1005 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1006 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1007 | * be used for reference counting. | |
1008 | */ | |
1009 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1010 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1011 | iter != NULL; \ |
527a5ec9 | 1012 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1013 | |
9f3a0d09 | 1014 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1015 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1016 | iter != NULL; \ |
527a5ec9 | 1017 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1018 | |
c0ff4b85 | 1019 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
4b3bde4c | 1020 | { |
c0ff4b85 | 1021 | return (memcg == root_mem_cgroup); |
4b3bde4c BS |
1022 | } |
1023 | ||
456f998e YH |
1024 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
1025 | { | |
c0ff4b85 | 1026 | struct mem_cgroup *memcg; |
456f998e YH |
1027 | |
1028 | if (!mm) | |
1029 | return; | |
1030 | ||
1031 | rcu_read_lock(); | |
c0ff4b85 R |
1032 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1033 | if (unlikely(!memcg)) | |
456f998e YH |
1034 | goto out; |
1035 | ||
1036 | switch (idx) { | |
456f998e | 1037 | case PGFAULT: |
0e574a93 JW |
1038 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1039 | break; | |
1040 | case PGMAJFAULT: | |
1041 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1042 | break; |
1043 | default: | |
1044 | BUG(); | |
1045 | } | |
1046 | out: | |
1047 | rcu_read_unlock(); | |
1048 | } | |
1049 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
1050 | ||
925b7673 JW |
1051 | /** |
1052 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1053 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1054 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1055 | * |
1056 | * Returns the lru list vector holding pages for the given @zone and | |
1057 | * @mem. This can be the global zone lruvec, if the memory controller | |
1058 | * is disabled. | |
1059 | */ | |
1060 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1061 | struct mem_cgroup *memcg) | |
1062 | { | |
1063 | struct mem_cgroup_per_zone *mz; | |
1064 | ||
1065 | if (mem_cgroup_disabled()) | |
1066 | return &zone->lruvec; | |
1067 | ||
1068 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
1069 | return &mz->lruvec; | |
1070 | } | |
1071 | ||
08e552c6 KH |
1072 | /* |
1073 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1074 | * Operations are called by routine of global LRU independently from memcg. | |
1075 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1076 | * | |
1077 | * Changes to pc->mem_cgroup happens when | |
1078 | * 1. charge | |
1079 | * 2. moving account | |
1080 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1081 | * It is added to LRU before charge. | |
1082 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1083 | * When moving account, the page is not on LRU. It's isolated. | |
1084 | */ | |
4f98a2fe | 1085 | |
925b7673 | 1086 | /** |
fa9add64 | 1087 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1088 | * @page: the page |
fa9add64 | 1089 | * @zone: zone of the page |
925b7673 | 1090 | */ |
fa9add64 | 1091 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1092 | { |
08e552c6 | 1093 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1094 | struct mem_cgroup *memcg; |
1095 | struct page_cgroup *pc; | |
6d12e2d8 | 1096 | |
f8d66542 | 1097 | if (mem_cgroup_disabled()) |
925b7673 JW |
1098 | return &zone->lruvec; |
1099 | ||
08e552c6 | 1100 | pc = lookup_page_cgroup(page); |
38c5d72f | 1101 | memcg = pc->mem_cgroup; |
7512102c HD |
1102 | |
1103 | /* | |
fa9add64 | 1104 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1105 | * an uncharged page off lru does nothing to secure |
1106 | * its former mem_cgroup from sudden removal. | |
1107 | * | |
1108 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1109 | * under page_cgroup lock: between them, they make all uses | |
1110 | * of pc->mem_cgroup safe. | |
1111 | */ | |
fa9add64 | 1112 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1113 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1114 | ||
925b7673 | 1115 | mz = page_cgroup_zoneinfo(memcg, page); |
925b7673 | 1116 | return &mz->lruvec; |
08e552c6 | 1117 | } |
b69408e8 | 1118 | |
925b7673 | 1119 | /** |
fa9add64 HD |
1120 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1121 | * @lruvec: mem_cgroup per zone lru vector | |
1122 | * @lru: index of lru list the page is sitting on | |
1123 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1124 | * |
fa9add64 HD |
1125 | * This function must be called when a page is added to or removed from an |
1126 | * lru list. | |
3f58a829 | 1127 | */ |
fa9add64 HD |
1128 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1129 | int nr_pages) | |
3f58a829 MK |
1130 | { |
1131 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1132 | unsigned long *lru_size; |
3f58a829 MK |
1133 | |
1134 | if (mem_cgroup_disabled()) | |
1135 | return; | |
1136 | ||
fa9add64 HD |
1137 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1138 | lru_size = mz->lru_size + lru; | |
1139 | *lru_size += nr_pages; | |
1140 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1141 | } |
544122e5 | 1142 | |
3e92041d | 1143 | /* |
c0ff4b85 | 1144 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1145 | * hierarchy subtree |
1146 | */ | |
c3ac9a8a JW |
1147 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1148 | struct mem_cgroup *memcg) | |
3e92041d | 1149 | { |
91c63734 JW |
1150 | if (root_memcg == memcg) |
1151 | return true; | |
3a981f48 | 1152 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1153 | return false; |
c3ac9a8a JW |
1154 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1155 | } | |
1156 | ||
1157 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1158 | struct mem_cgroup *memcg) | |
1159 | { | |
1160 | bool ret; | |
1161 | ||
91c63734 | 1162 | rcu_read_lock(); |
c3ac9a8a | 1163 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1164 | rcu_read_unlock(); |
1165 | return ret; | |
3e92041d MH |
1166 | } |
1167 | ||
c0ff4b85 | 1168 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1169 | { |
1170 | int ret; | |
0b7f569e | 1171 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1172 | struct task_struct *p; |
4c4a2214 | 1173 | |
158e0a2d | 1174 | p = find_lock_task_mm(task); |
de077d22 DR |
1175 | if (p) { |
1176 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1177 | task_unlock(p); | |
1178 | } else { | |
1179 | /* | |
1180 | * All threads may have already detached their mm's, but the oom | |
1181 | * killer still needs to detect if they have already been oom | |
1182 | * killed to prevent needlessly killing additional tasks. | |
1183 | */ | |
1184 | task_lock(task); | |
1185 | curr = mem_cgroup_from_task(task); | |
1186 | if (curr) | |
1187 | css_get(&curr->css); | |
1188 | task_unlock(task); | |
1189 | } | |
0b7f569e KH |
1190 | if (!curr) |
1191 | return 0; | |
d31f56db | 1192 | /* |
c0ff4b85 | 1193 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1194 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1195 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1196 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1197 | */ |
c0ff4b85 | 1198 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1199 | css_put(&curr->css); |
4c4a2214 DR |
1200 | return ret; |
1201 | } | |
1202 | ||
c56d5c7d | 1203 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1204 | { |
9b272977 | 1205 | unsigned long inactive_ratio; |
14797e23 | 1206 | unsigned long inactive; |
9b272977 | 1207 | unsigned long active; |
c772be93 | 1208 | unsigned long gb; |
14797e23 | 1209 | |
4d7dcca2 HD |
1210 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1211 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1212 | |
c772be93 KM |
1213 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1214 | if (gb) | |
1215 | inactive_ratio = int_sqrt(10 * gb); | |
1216 | else | |
1217 | inactive_ratio = 1; | |
1218 | ||
9b272977 | 1219 | return inactive * inactive_ratio < active; |
14797e23 KM |
1220 | } |
1221 | ||
c56d5c7d | 1222 | int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 RR |
1223 | { |
1224 | unsigned long active; | |
1225 | unsigned long inactive; | |
1226 | ||
4d7dcca2 HD |
1227 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE); |
1228 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 RR |
1229 | |
1230 | return (active > inactive); | |
1231 | } | |
1232 | ||
6d61ef40 BS |
1233 | #define mem_cgroup_from_res_counter(counter, member) \ |
1234 | container_of(counter, struct mem_cgroup, member) | |
1235 | ||
19942822 | 1236 | /** |
9d11ea9f | 1237 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1238 | * @memcg: the memory cgroup |
19942822 | 1239 | * |
9d11ea9f | 1240 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1241 | * pages. |
19942822 | 1242 | */ |
c0ff4b85 | 1243 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1244 | { |
9d11ea9f JW |
1245 | unsigned long long margin; |
1246 | ||
c0ff4b85 | 1247 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1248 | if (do_swap_account) |
c0ff4b85 | 1249 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1250 | return margin >> PAGE_SHIFT; |
19942822 JW |
1251 | } |
1252 | ||
1f4c025b | 1253 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1254 | { |
1255 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1256 | |
1257 | /* root ? */ | |
1258 | if (cgrp->parent == NULL) | |
1259 | return vm_swappiness; | |
1260 | ||
bf1ff263 | 1261 | return memcg->swappiness; |
a7885eb8 KM |
1262 | } |
1263 | ||
619d094b KH |
1264 | /* |
1265 | * memcg->moving_account is used for checking possibility that some thread is | |
1266 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1267 | * a memcg, other threads should check memcg->moving_account under | |
1268 | * rcu_read_lock(), like this: | |
1269 | * | |
1270 | * CPU-A CPU-B | |
1271 | * rcu_read_lock() | |
1272 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1273 | * take heavy locks. | |
1274 | * synchronize_rcu() update something. | |
1275 | * rcu_read_unlock() | |
1276 | * start move here. | |
1277 | */ | |
4331f7d3 KH |
1278 | |
1279 | /* for quick checking without looking up memcg */ | |
1280 | atomic_t memcg_moving __read_mostly; | |
1281 | ||
c0ff4b85 | 1282 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1283 | { |
4331f7d3 | 1284 | atomic_inc(&memcg_moving); |
619d094b | 1285 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1286 | synchronize_rcu(); |
1287 | } | |
1288 | ||
c0ff4b85 | 1289 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1290 | { |
619d094b KH |
1291 | /* |
1292 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1293 | * We check NULL in callee rather than caller. | |
1294 | */ | |
4331f7d3 KH |
1295 | if (memcg) { |
1296 | atomic_dec(&memcg_moving); | |
619d094b | 1297 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1298 | } |
32047e2a | 1299 | } |
619d094b | 1300 | |
32047e2a KH |
1301 | /* |
1302 | * 2 routines for checking "mem" is under move_account() or not. | |
1303 | * | |
13fd1dd9 AM |
1304 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1305 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1306 | * pc->mem_cgroup may be overwritten. |
1307 | * | |
1308 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1309 | * under hierarchy of moving cgroups. This is for | |
1310 | * waiting at hith-memory prressure caused by "move". | |
1311 | */ | |
1312 | ||
13fd1dd9 | 1313 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1314 | { |
1315 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1316 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1317 | } |
4b534334 | 1318 | |
c0ff4b85 | 1319 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1320 | { |
2bd9bb20 KH |
1321 | struct mem_cgroup *from; |
1322 | struct mem_cgroup *to; | |
4b534334 | 1323 | bool ret = false; |
2bd9bb20 KH |
1324 | /* |
1325 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1326 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1327 | */ | |
1328 | spin_lock(&mc.lock); | |
1329 | from = mc.from; | |
1330 | to = mc.to; | |
1331 | if (!from) | |
1332 | goto unlock; | |
3e92041d | 1333 | |
c0ff4b85 R |
1334 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1335 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1336 | unlock: |
1337 | spin_unlock(&mc.lock); | |
4b534334 KH |
1338 | return ret; |
1339 | } | |
1340 | ||
c0ff4b85 | 1341 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1342 | { |
1343 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1344 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1345 | DEFINE_WAIT(wait); |
1346 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1347 | /* moving charge context might have finished. */ | |
1348 | if (mc.moving_task) | |
1349 | schedule(); | |
1350 | finish_wait(&mc.waitq, &wait); | |
1351 | return true; | |
1352 | } | |
1353 | } | |
1354 | return false; | |
1355 | } | |
1356 | ||
312734c0 KH |
1357 | /* |
1358 | * Take this lock when | |
1359 | * - a code tries to modify page's memcg while it's USED. | |
1360 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1361 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1362 | */ |
1363 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1364 | unsigned long *flags) | |
1365 | { | |
1366 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1367 | } | |
1368 | ||
1369 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1370 | unsigned long *flags) | |
1371 | { | |
1372 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1373 | } | |
1374 | ||
e222432b | 1375 | /** |
6a6135b6 | 1376 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1377 | * @memcg: The memory cgroup that went over limit |
1378 | * @p: Task that is going to be killed | |
1379 | * | |
1380 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1381 | * enabled | |
1382 | */ | |
1383 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1384 | { | |
1385 | struct cgroup *task_cgrp; | |
1386 | struct cgroup *mem_cgrp; | |
1387 | /* | |
1388 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1389 | * on the assumption that OOM is serialized for memory controller. | |
1390 | * If this assumption is broken, revisit this code. | |
1391 | */ | |
1392 | static char memcg_name[PATH_MAX]; | |
1393 | int ret; | |
1394 | ||
d31f56db | 1395 | if (!memcg || !p) |
e222432b BS |
1396 | return; |
1397 | ||
e222432b BS |
1398 | rcu_read_lock(); |
1399 | ||
1400 | mem_cgrp = memcg->css.cgroup; | |
1401 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1402 | ||
1403 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1404 | if (ret < 0) { | |
1405 | /* | |
1406 | * Unfortunately, we are unable to convert to a useful name | |
1407 | * But we'll still print out the usage information | |
1408 | */ | |
1409 | rcu_read_unlock(); | |
1410 | goto done; | |
1411 | } | |
1412 | rcu_read_unlock(); | |
1413 | ||
1414 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1415 | ||
1416 | rcu_read_lock(); | |
1417 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1418 | if (ret < 0) { | |
1419 | rcu_read_unlock(); | |
1420 | goto done; | |
1421 | } | |
1422 | rcu_read_unlock(); | |
1423 | ||
1424 | /* | |
1425 | * Continues from above, so we don't need an KERN_ level | |
1426 | */ | |
1427 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1428 | done: | |
1429 | ||
1430 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1431 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1432 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1433 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1434 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1435 | "failcnt %llu\n", | |
1436 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1437 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1438 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1439 | } | |
1440 | ||
81d39c20 KH |
1441 | /* |
1442 | * This function returns the number of memcg under hierarchy tree. Returns | |
1443 | * 1(self count) if no children. | |
1444 | */ | |
c0ff4b85 | 1445 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1446 | { |
1447 | int num = 0; | |
7d74b06f KH |
1448 | struct mem_cgroup *iter; |
1449 | ||
c0ff4b85 | 1450 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1451 | num++; |
81d39c20 KH |
1452 | return num; |
1453 | } | |
1454 | ||
a63d83f4 DR |
1455 | /* |
1456 | * Return the memory (and swap, if configured) limit for a memcg. | |
1457 | */ | |
9cbb78bb | 1458 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1459 | { |
1460 | u64 limit; | |
1461 | u64 memsw; | |
1462 | ||
f3e8eb70 JW |
1463 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1464 | limit += total_swap_pages << PAGE_SHIFT; | |
1465 | ||
a63d83f4 DR |
1466 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1467 | /* | |
1468 | * If memsw is finite and limits the amount of swap space available | |
1469 | * to this memcg, return that limit. | |
1470 | */ | |
1471 | return min(limit, memsw); | |
1472 | } | |
1473 | ||
876aafbf DR |
1474 | void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1475 | int order) | |
9cbb78bb DR |
1476 | { |
1477 | struct mem_cgroup *iter; | |
1478 | unsigned long chosen_points = 0; | |
1479 | unsigned long totalpages; | |
1480 | unsigned int points = 0; | |
1481 | struct task_struct *chosen = NULL; | |
1482 | ||
876aafbf DR |
1483 | /* |
1484 | * If current has a pending SIGKILL, then automatically select it. The | |
1485 | * goal is to allow it to allocate so that it may quickly exit and free | |
1486 | * its memory. | |
1487 | */ | |
1488 | if (fatal_signal_pending(current)) { | |
1489 | set_thread_flag(TIF_MEMDIE); | |
1490 | return; | |
1491 | } | |
1492 | ||
1493 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1494 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1495 | for_each_mem_cgroup_tree(iter, memcg) { | |
1496 | struct cgroup *cgroup = iter->css.cgroup; | |
1497 | struct cgroup_iter it; | |
1498 | struct task_struct *task; | |
1499 | ||
1500 | cgroup_iter_start(cgroup, &it); | |
1501 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
1502 | switch (oom_scan_process_thread(task, totalpages, NULL, | |
1503 | false)) { | |
1504 | case OOM_SCAN_SELECT: | |
1505 | if (chosen) | |
1506 | put_task_struct(chosen); | |
1507 | chosen = task; | |
1508 | chosen_points = ULONG_MAX; | |
1509 | get_task_struct(chosen); | |
1510 | /* fall through */ | |
1511 | case OOM_SCAN_CONTINUE: | |
1512 | continue; | |
1513 | case OOM_SCAN_ABORT: | |
1514 | cgroup_iter_end(cgroup, &it); | |
1515 | mem_cgroup_iter_break(memcg, iter); | |
1516 | if (chosen) | |
1517 | put_task_struct(chosen); | |
1518 | return; | |
1519 | case OOM_SCAN_OK: | |
1520 | break; | |
1521 | }; | |
1522 | points = oom_badness(task, memcg, NULL, totalpages); | |
1523 | if (points > chosen_points) { | |
1524 | if (chosen) | |
1525 | put_task_struct(chosen); | |
1526 | chosen = task; | |
1527 | chosen_points = points; | |
1528 | get_task_struct(chosen); | |
1529 | } | |
1530 | } | |
1531 | cgroup_iter_end(cgroup, &it); | |
1532 | } | |
1533 | ||
1534 | if (!chosen) | |
1535 | return; | |
1536 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1537 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1538 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1539 | } |
1540 | ||
5660048c JW |
1541 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1542 | gfp_t gfp_mask, | |
1543 | unsigned long flags) | |
1544 | { | |
1545 | unsigned long total = 0; | |
1546 | bool noswap = false; | |
1547 | int loop; | |
1548 | ||
1549 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1550 | noswap = true; | |
1551 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1552 | noswap = true; | |
1553 | ||
1554 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1555 | if (loop) | |
1556 | drain_all_stock_async(memcg); | |
1557 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1558 | /* | |
1559 | * Allow limit shrinkers, which are triggered directly | |
1560 | * by userspace, to catch signals and stop reclaim | |
1561 | * after minimal progress, regardless of the margin. | |
1562 | */ | |
1563 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1564 | break; | |
1565 | if (mem_cgroup_margin(memcg)) | |
1566 | break; | |
1567 | /* | |
1568 | * If nothing was reclaimed after two attempts, there | |
1569 | * may be no reclaimable pages in this hierarchy. | |
1570 | */ | |
1571 | if (loop && !total) | |
1572 | break; | |
1573 | } | |
1574 | return total; | |
1575 | } | |
1576 | ||
4d0c066d KH |
1577 | /** |
1578 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1579 | * @memcg: the target memcg |
4d0c066d KH |
1580 | * @nid: the node ID to be checked. |
1581 | * @noswap : specify true here if the user wants flle only information. | |
1582 | * | |
1583 | * This function returns whether the specified memcg contains any | |
1584 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1585 | * pages in the node. | |
1586 | */ | |
c0ff4b85 | 1587 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1588 | int nid, bool noswap) |
1589 | { | |
c0ff4b85 | 1590 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1591 | return true; |
1592 | if (noswap || !total_swap_pages) | |
1593 | return false; | |
c0ff4b85 | 1594 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1595 | return true; |
1596 | return false; | |
1597 | ||
1598 | } | |
889976db YH |
1599 | #if MAX_NUMNODES > 1 |
1600 | ||
1601 | /* | |
1602 | * Always updating the nodemask is not very good - even if we have an empty | |
1603 | * list or the wrong list here, we can start from some node and traverse all | |
1604 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1605 | * | |
1606 | */ | |
c0ff4b85 | 1607 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1608 | { |
1609 | int nid; | |
453a9bf3 KH |
1610 | /* |
1611 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1612 | * pagein/pageout changes since the last update. | |
1613 | */ | |
c0ff4b85 | 1614 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1615 | return; |
c0ff4b85 | 1616 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1617 | return; |
1618 | ||
889976db | 1619 | /* make a nodemask where this memcg uses memory from */ |
c0ff4b85 | 1620 | memcg->scan_nodes = node_states[N_HIGH_MEMORY]; |
889976db YH |
1621 | |
1622 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1623 | ||
c0ff4b85 R |
1624 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1625 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1626 | } |
453a9bf3 | 1627 | |
c0ff4b85 R |
1628 | atomic_set(&memcg->numainfo_events, 0); |
1629 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1630 | } |
1631 | ||
1632 | /* | |
1633 | * Selecting a node where we start reclaim from. Because what we need is just | |
1634 | * reducing usage counter, start from anywhere is O,K. Considering | |
1635 | * memory reclaim from current node, there are pros. and cons. | |
1636 | * | |
1637 | * Freeing memory from current node means freeing memory from a node which | |
1638 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1639 | * hit limits, it will see a contention on a node. But freeing from remote | |
1640 | * node means more costs for memory reclaim because of memory latency. | |
1641 | * | |
1642 | * Now, we use round-robin. Better algorithm is welcomed. | |
1643 | */ | |
c0ff4b85 | 1644 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1645 | { |
1646 | int node; | |
1647 | ||
c0ff4b85 R |
1648 | mem_cgroup_may_update_nodemask(memcg); |
1649 | node = memcg->last_scanned_node; | |
889976db | 1650 | |
c0ff4b85 | 1651 | node = next_node(node, memcg->scan_nodes); |
889976db | 1652 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1653 | node = first_node(memcg->scan_nodes); |
889976db YH |
1654 | /* |
1655 | * We call this when we hit limit, not when pages are added to LRU. | |
1656 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1657 | * memcg is too small and all pages are not on LRU. In that case, | |
1658 | * we use curret node. | |
1659 | */ | |
1660 | if (unlikely(node == MAX_NUMNODES)) | |
1661 | node = numa_node_id(); | |
1662 | ||
c0ff4b85 | 1663 | memcg->last_scanned_node = node; |
889976db YH |
1664 | return node; |
1665 | } | |
1666 | ||
4d0c066d KH |
1667 | /* |
1668 | * Check all nodes whether it contains reclaimable pages or not. | |
1669 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1670 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1671 | * enough new information. We need to do double check. | |
1672 | */ | |
6bbda35c | 1673 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1674 | { |
1675 | int nid; | |
1676 | ||
1677 | /* | |
1678 | * quick check...making use of scan_node. | |
1679 | * We can skip unused nodes. | |
1680 | */ | |
c0ff4b85 R |
1681 | if (!nodes_empty(memcg->scan_nodes)) { |
1682 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1683 | nid < MAX_NUMNODES; |
c0ff4b85 | 1684 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1685 | |
c0ff4b85 | 1686 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1687 | return true; |
1688 | } | |
1689 | } | |
1690 | /* | |
1691 | * Check rest of nodes. | |
1692 | */ | |
1693 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
c0ff4b85 | 1694 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1695 | continue; |
c0ff4b85 | 1696 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1697 | return true; |
1698 | } | |
1699 | return false; | |
1700 | } | |
1701 | ||
889976db | 1702 | #else |
c0ff4b85 | 1703 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1704 | { |
1705 | return 0; | |
1706 | } | |
4d0c066d | 1707 | |
6bbda35c | 1708 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1709 | { |
c0ff4b85 | 1710 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1711 | } |
889976db YH |
1712 | #endif |
1713 | ||
5660048c JW |
1714 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1715 | struct zone *zone, | |
1716 | gfp_t gfp_mask, | |
1717 | unsigned long *total_scanned) | |
6d61ef40 | 1718 | { |
9f3a0d09 | 1719 | struct mem_cgroup *victim = NULL; |
5660048c | 1720 | int total = 0; |
04046e1a | 1721 | int loop = 0; |
9d11ea9f | 1722 | unsigned long excess; |
185efc0f | 1723 | unsigned long nr_scanned; |
527a5ec9 JW |
1724 | struct mem_cgroup_reclaim_cookie reclaim = { |
1725 | .zone = zone, | |
1726 | .priority = 0, | |
1727 | }; | |
9d11ea9f | 1728 | |
c0ff4b85 | 1729 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1730 | |
4e416953 | 1731 | while (1) { |
527a5ec9 | 1732 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1733 | if (!victim) { |
04046e1a | 1734 | loop++; |
4e416953 BS |
1735 | if (loop >= 2) { |
1736 | /* | |
1737 | * If we have not been able to reclaim | |
1738 | * anything, it might because there are | |
1739 | * no reclaimable pages under this hierarchy | |
1740 | */ | |
5660048c | 1741 | if (!total) |
4e416953 | 1742 | break; |
4e416953 | 1743 | /* |
25985edc | 1744 | * We want to do more targeted reclaim. |
4e416953 BS |
1745 | * excess >> 2 is not to excessive so as to |
1746 | * reclaim too much, nor too less that we keep | |
1747 | * coming back to reclaim from this cgroup | |
1748 | */ | |
1749 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1750 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1751 | break; |
4e416953 | 1752 | } |
9f3a0d09 | 1753 | continue; |
4e416953 | 1754 | } |
5660048c | 1755 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1756 | continue; |
5660048c JW |
1757 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1758 | zone, &nr_scanned); | |
1759 | *total_scanned += nr_scanned; | |
1760 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1761 | break; |
6d61ef40 | 1762 | } |
9f3a0d09 | 1763 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1764 | return total; |
6d61ef40 BS |
1765 | } |
1766 | ||
867578cb KH |
1767 | /* |
1768 | * Check OOM-Killer is already running under our hierarchy. | |
1769 | * If someone is running, return false. | |
1af8efe9 | 1770 | * Has to be called with memcg_oom_lock |
867578cb | 1771 | */ |
c0ff4b85 | 1772 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1773 | { |
79dfdacc | 1774 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1775 | |
9f3a0d09 | 1776 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1777 | if (iter->oom_lock) { |
79dfdacc MH |
1778 | /* |
1779 | * this subtree of our hierarchy is already locked | |
1780 | * so we cannot give a lock. | |
1781 | */ | |
79dfdacc | 1782 | failed = iter; |
9f3a0d09 JW |
1783 | mem_cgroup_iter_break(memcg, iter); |
1784 | break; | |
23751be0 JW |
1785 | } else |
1786 | iter->oom_lock = true; | |
7d74b06f | 1787 | } |
867578cb | 1788 | |
79dfdacc | 1789 | if (!failed) |
23751be0 | 1790 | return true; |
79dfdacc MH |
1791 | |
1792 | /* | |
1793 | * OK, we failed to lock the whole subtree so we have to clean up | |
1794 | * what we set up to the failing subtree | |
1795 | */ | |
9f3a0d09 | 1796 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1797 | if (iter == failed) { |
9f3a0d09 JW |
1798 | mem_cgroup_iter_break(memcg, iter); |
1799 | break; | |
79dfdacc MH |
1800 | } |
1801 | iter->oom_lock = false; | |
1802 | } | |
23751be0 | 1803 | return false; |
a636b327 | 1804 | } |
0b7f569e | 1805 | |
79dfdacc | 1806 | /* |
1af8efe9 | 1807 | * Has to be called with memcg_oom_lock |
79dfdacc | 1808 | */ |
c0ff4b85 | 1809 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1810 | { |
7d74b06f KH |
1811 | struct mem_cgroup *iter; |
1812 | ||
c0ff4b85 | 1813 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1814 | iter->oom_lock = false; |
1815 | return 0; | |
1816 | } | |
1817 | ||
c0ff4b85 | 1818 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1819 | { |
1820 | struct mem_cgroup *iter; | |
1821 | ||
c0ff4b85 | 1822 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1823 | atomic_inc(&iter->under_oom); |
1824 | } | |
1825 | ||
c0ff4b85 | 1826 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1827 | { |
1828 | struct mem_cgroup *iter; | |
1829 | ||
867578cb KH |
1830 | /* |
1831 | * When a new child is created while the hierarchy is under oom, | |
1832 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1833 | * atomic_add_unless() here. | |
1834 | */ | |
c0ff4b85 | 1835 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1836 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1837 | } |
1838 | ||
1af8efe9 | 1839 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1840 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1841 | ||
dc98df5a | 1842 | struct oom_wait_info { |
d79154bb | 1843 | struct mem_cgroup *memcg; |
dc98df5a KH |
1844 | wait_queue_t wait; |
1845 | }; | |
1846 | ||
1847 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1848 | unsigned mode, int sync, void *arg) | |
1849 | { | |
d79154bb HD |
1850 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1851 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1852 | struct oom_wait_info *oom_wait_info; |
1853 | ||
1854 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1855 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1856 | |
dc98df5a | 1857 | /* |
d79154bb | 1858 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1859 | * Then we can use css_is_ancestor without taking care of RCU. |
1860 | */ | |
c0ff4b85 R |
1861 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1862 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1863 | return 0; |
dc98df5a KH |
1864 | return autoremove_wake_function(wait, mode, sync, arg); |
1865 | } | |
1866 | ||
c0ff4b85 | 1867 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1868 | { |
c0ff4b85 R |
1869 | /* for filtering, pass "memcg" as argument. */ |
1870 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1871 | } |
1872 | ||
c0ff4b85 | 1873 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1874 | { |
c0ff4b85 R |
1875 | if (memcg && atomic_read(&memcg->under_oom)) |
1876 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1877 | } |
1878 | ||
867578cb KH |
1879 | /* |
1880 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1881 | */ | |
6bbda35c KS |
1882 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
1883 | int order) | |
0b7f569e | 1884 | { |
dc98df5a | 1885 | struct oom_wait_info owait; |
3c11ecf4 | 1886 | bool locked, need_to_kill; |
867578cb | 1887 | |
d79154bb | 1888 | owait.memcg = memcg; |
dc98df5a KH |
1889 | owait.wait.flags = 0; |
1890 | owait.wait.func = memcg_oom_wake_function; | |
1891 | owait.wait.private = current; | |
1892 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1893 | need_to_kill = true; |
c0ff4b85 | 1894 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1895 | |
c0ff4b85 | 1896 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1897 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1898 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1899 | /* |
1900 | * Even if signal_pending(), we can't quit charge() loop without | |
1901 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1902 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1903 | */ | |
3c11ecf4 | 1904 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1905 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1906 | need_to_kill = false; |
1907 | if (locked) | |
c0ff4b85 | 1908 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1909 | spin_unlock(&memcg_oom_lock); |
867578cb | 1910 | |
3c11ecf4 KH |
1911 | if (need_to_kill) { |
1912 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 1913 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 1914 | } else { |
867578cb | 1915 | schedule(); |
dc98df5a | 1916 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1917 | } |
1af8efe9 | 1918 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1919 | if (locked) |
c0ff4b85 R |
1920 | mem_cgroup_oom_unlock(memcg); |
1921 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1922 | spin_unlock(&memcg_oom_lock); |
867578cb | 1923 | |
c0ff4b85 | 1924 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1925 | |
867578cb KH |
1926 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1927 | return false; | |
1928 | /* Give chance to dying process */ | |
715a5ee8 | 1929 | schedule_timeout_uninterruptible(1); |
867578cb | 1930 | return true; |
0b7f569e KH |
1931 | } |
1932 | ||
d69b042f BS |
1933 | /* |
1934 | * Currently used to update mapped file statistics, but the routine can be | |
1935 | * generalized to update other statistics as well. | |
32047e2a KH |
1936 | * |
1937 | * Notes: Race condition | |
1938 | * | |
1939 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1940 | * it tends to be costly. But considering some conditions, we doesn't need | |
1941 | * to do so _always_. | |
1942 | * | |
1943 | * Considering "charge", lock_page_cgroup() is not required because all | |
1944 | * file-stat operations happen after a page is attached to radix-tree. There | |
1945 | * are no race with "charge". | |
1946 | * | |
1947 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1948 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1949 | * if there are race with "uncharge". Statistics itself is properly handled | |
1950 | * by flags. | |
1951 | * | |
1952 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
1953 | * small, we check mm->moving_account and detect there are possibility of race |
1954 | * If there is, we take a lock. | |
d69b042f | 1955 | */ |
26174efd | 1956 | |
89c06bd5 KH |
1957 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
1958 | bool *locked, unsigned long *flags) | |
1959 | { | |
1960 | struct mem_cgroup *memcg; | |
1961 | struct page_cgroup *pc; | |
1962 | ||
1963 | pc = lookup_page_cgroup(page); | |
1964 | again: | |
1965 | memcg = pc->mem_cgroup; | |
1966 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
1967 | return; | |
1968 | /* | |
1969 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 1970 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 1971 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 1972 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 1973 | */ |
13fd1dd9 | 1974 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
1975 | return; |
1976 | ||
1977 | move_lock_mem_cgroup(memcg, flags); | |
1978 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
1979 | move_unlock_mem_cgroup(memcg, flags); | |
1980 | goto again; | |
1981 | } | |
1982 | *locked = true; | |
1983 | } | |
1984 | ||
1985 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
1986 | { | |
1987 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1988 | ||
1989 | /* | |
1990 | * It's guaranteed that pc->mem_cgroup never changes while | |
1991 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 1992 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
1993 | */ |
1994 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
1995 | } | |
1996 | ||
2a7106f2 GT |
1997 | void mem_cgroup_update_page_stat(struct page *page, |
1998 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 1999 | { |
c0ff4b85 | 2000 | struct mem_cgroup *memcg; |
32047e2a | 2001 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2002 | unsigned long uninitialized_var(flags); |
d69b042f | 2003 | |
cfa44946 | 2004 | if (mem_cgroup_disabled()) |
d69b042f | 2005 | return; |
89c06bd5 | 2006 | |
c0ff4b85 R |
2007 | memcg = pc->mem_cgroup; |
2008 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2009 | return; |
26174efd | 2010 | |
26174efd | 2011 | switch (idx) { |
2a7106f2 | 2012 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2013 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2014 | break; |
2015 | default: | |
2016 | BUG(); | |
8725d541 | 2017 | } |
d69b042f | 2018 | |
c0ff4b85 | 2019 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2020 | } |
26174efd | 2021 | |
cdec2e42 KH |
2022 | /* |
2023 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2024 | * TODO: maybe necessary to use big numbers in big irons. | |
2025 | */ | |
7ec99d62 | 2026 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2027 | struct memcg_stock_pcp { |
2028 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2029 | unsigned int nr_pages; |
cdec2e42 | 2030 | struct work_struct work; |
26fe6168 | 2031 | unsigned long flags; |
a0db00fc | 2032 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2033 | }; |
2034 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2035 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
2036 | |
2037 | /* | |
11c9ea4e | 2038 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
2039 | * from local stock and true is returned. If the stock is 0 or charges from a |
2040 | * cgroup which is not current target, returns false. This stock will be | |
2041 | * refilled. | |
2042 | */ | |
c0ff4b85 | 2043 | static bool consume_stock(struct mem_cgroup *memcg) |
cdec2e42 KH |
2044 | { |
2045 | struct memcg_stock_pcp *stock; | |
2046 | bool ret = true; | |
2047 | ||
2048 | stock = &get_cpu_var(memcg_stock); | |
c0ff4b85 | 2049 | if (memcg == stock->cached && stock->nr_pages) |
11c9ea4e | 2050 | stock->nr_pages--; |
cdec2e42 KH |
2051 | else /* need to call res_counter_charge */ |
2052 | ret = false; | |
2053 | put_cpu_var(memcg_stock); | |
2054 | return ret; | |
2055 | } | |
2056 | ||
2057 | /* | |
2058 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2059 | */ | |
2060 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2061 | { | |
2062 | struct mem_cgroup *old = stock->cached; | |
2063 | ||
11c9ea4e JW |
2064 | if (stock->nr_pages) { |
2065 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2066 | ||
2067 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2068 | if (do_swap_account) |
11c9ea4e JW |
2069 | res_counter_uncharge(&old->memsw, bytes); |
2070 | stock->nr_pages = 0; | |
cdec2e42 KH |
2071 | } |
2072 | stock->cached = NULL; | |
cdec2e42 KH |
2073 | } |
2074 | ||
2075 | /* | |
2076 | * This must be called under preempt disabled or must be called by | |
2077 | * a thread which is pinned to local cpu. | |
2078 | */ | |
2079 | static void drain_local_stock(struct work_struct *dummy) | |
2080 | { | |
2081 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2082 | drain_stock(stock); | |
26fe6168 | 2083 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2084 | } |
2085 | ||
2086 | /* | |
2087 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2088 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2089 | */ |
c0ff4b85 | 2090 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2091 | { |
2092 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2093 | ||
c0ff4b85 | 2094 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2095 | drain_stock(stock); |
c0ff4b85 | 2096 | stock->cached = memcg; |
cdec2e42 | 2097 | } |
11c9ea4e | 2098 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2099 | put_cpu_var(memcg_stock); |
2100 | } | |
2101 | ||
2102 | /* | |
c0ff4b85 | 2103 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2104 | * of the hierarchy under it. sync flag says whether we should block |
2105 | * until the work is done. | |
cdec2e42 | 2106 | */ |
c0ff4b85 | 2107 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2108 | { |
26fe6168 | 2109 | int cpu, curcpu; |
d38144b7 | 2110 | |
cdec2e42 | 2111 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2112 | get_online_cpus(); |
5af12d0e | 2113 | curcpu = get_cpu(); |
cdec2e42 KH |
2114 | for_each_online_cpu(cpu) { |
2115 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2116 | struct mem_cgroup *memcg; |
26fe6168 | 2117 | |
c0ff4b85 R |
2118 | memcg = stock->cached; |
2119 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2120 | continue; |
c0ff4b85 | 2121 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2122 | continue; |
d1a05b69 MH |
2123 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2124 | if (cpu == curcpu) | |
2125 | drain_local_stock(&stock->work); | |
2126 | else | |
2127 | schedule_work_on(cpu, &stock->work); | |
2128 | } | |
cdec2e42 | 2129 | } |
5af12d0e | 2130 | put_cpu(); |
d38144b7 MH |
2131 | |
2132 | if (!sync) | |
2133 | goto out; | |
2134 | ||
2135 | for_each_online_cpu(cpu) { | |
2136 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2137 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2138 | flush_work(&stock->work); |
2139 | } | |
2140 | out: | |
cdec2e42 | 2141 | put_online_cpus(); |
d38144b7 MH |
2142 | } |
2143 | ||
2144 | /* | |
2145 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2146 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2147 | * expects some charges will be back to res_counter later but cannot wait for | |
2148 | * it. | |
2149 | */ | |
c0ff4b85 | 2150 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2151 | { |
9f50fad6 MH |
2152 | /* |
2153 | * If someone calls draining, avoid adding more kworker runs. | |
2154 | */ | |
2155 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2156 | return; | |
c0ff4b85 | 2157 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2158 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2159 | } |
2160 | ||
2161 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2162 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2163 | { |
2164 | /* called when force_empty is called */ | |
9f50fad6 | 2165 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2166 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2167 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2168 | } |
2169 | ||
711d3d2c KH |
2170 | /* |
2171 | * This function drains percpu counter value from DEAD cpu and | |
2172 | * move it to local cpu. Note that this function can be preempted. | |
2173 | */ | |
c0ff4b85 | 2174 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2175 | { |
2176 | int i; | |
2177 | ||
c0ff4b85 | 2178 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2179 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2180 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2181 | |
c0ff4b85 R |
2182 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2183 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2184 | } |
e9f8974f | 2185 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2186 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2187 | |
c0ff4b85 R |
2188 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2189 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2190 | } |
c0ff4b85 | 2191 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2192 | } |
2193 | ||
2194 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2195 | unsigned long action, |
2196 | void *hcpu) | |
2197 | { | |
2198 | int cpu = (unsigned long)hcpu; | |
2199 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2200 | struct mem_cgroup *iter; |
cdec2e42 | 2201 | |
619d094b | 2202 | if (action == CPU_ONLINE) |
1489ebad | 2203 | return NOTIFY_OK; |
1489ebad | 2204 | |
d833049b | 2205 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2206 | return NOTIFY_OK; |
711d3d2c | 2207 | |
9f3a0d09 | 2208 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2209 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2210 | ||
cdec2e42 KH |
2211 | stock = &per_cpu(memcg_stock, cpu); |
2212 | drain_stock(stock); | |
2213 | return NOTIFY_OK; | |
2214 | } | |
2215 | ||
4b534334 KH |
2216 | |
2217 | /* See __mem_cgroup_try_charge() for details */ | |
2218 | enum { | |
2219 | CHARGE_OK, /* success */ | |
2220 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2221 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2222 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2223 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2224 | }; | |
2225 | ||
c0ff4b85 | 2226 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
7ec99d62 | 2227 | unsigned int nr_pages, bool oom_check) |
4b534334 | 2228 | { |
7ec99d62 | 2229 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2230 | struct mem_cgroup *mem_over_limit; |
2231 | struct res_counter *fail_res; | |
2232 | unsigned long flags = 0; | |
2233 | int ret; | |
2234 | ||
c0ff4b85 | 2235 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2236 | |
2237 | if (likely(!ret)) { | |
2238 | if (!do_swap_account) | |
2239 | return CHARGE_OK; | |
c0ff4b85 | 2240 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2241 | if (likely(!ret)) |
2242 | return CHARGE_OK; | |
2243 | ||
c0ff4b85 | 2244 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2245 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2246 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2247 | } else | |
2248 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2249 | /* |
7ec99d62 JW |
2250 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2251 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2252 | * |
2253 | * Never reclaim on behalf of optional batching, retry with a | |
2254 | * single page instead. | |
2255 | */ | |
7ec99d62 | 2256 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2257 | return CHARGE_RETRY; |
2258 | ||
2259 | if (!(gfp_mask & __GFP_WAIT)) | |
2260 | return CHARGE_WOULDBLOCK; | |
2261 | ||
5660048c | 2262 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2263 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2264 | return CHARGE_RETRY; |
4b534334 | 2265 | /* |
19942822 JW |
2266 | * Even though the limit is exceeded at this point, reclaim |
2267 | * may have been able to free some pages. Retry the charge | |
2268 | * before killing the task. | |
2269 | * | |
2270 | * Only for regular pages, though: huge pages are rather | |
2271 | * unlikely to succeed so close to the limit, and we fall back | |
2272 | * to regular pages anyway in case of failure. | |
4b534334 | 2273 | */ |
7ec99d62 | 2274 | if (nr_pages == 1 && ret) |
4b534334 KH |
2275 | return CHARGE_RETRY; |
2276 | ||
2277 | /* | |
2278 | * At task move, charge accounts can be doubly counted. So, it's | |
2279 | * better to wait until the end of task_move if something is going on. | |
2280 | */ | |
2281 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2282 | return CHARGE_RETRY; | |
2283 | ||
2284 | /* If we don't need to call oom-killer at el, return immediately */ | |
2285 | if (!oom_check) | |
2286 | return CHARGE_NOMEM; | |
2287 | /* check OOM */ | |
e845e199 | 2288 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2289 | return CHARGE_OOM_DIE; |
2290 | ||
2291 | return CHARGE_RETRY; | |
2292 | } | |
2293 | ||
f817ed48 | 2294 | /* |
38c5d72f KH |
2295 | * __mem_cgroup_try_charge() does |
2296 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2297 | * 2. update res_counter | |
2298 | * 3. call memory reclaim if necessary. | |
2299 | * | |
2300 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2301 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2302 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2303 | * as possible without any hazards. 2: all pages should have a valid | |
2304 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2305 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2306 | * | |
2307 | * So __mem_cgroup_try_charge() will return | |
2308 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2309 | * -ENOMEM ... charge failure because of resource limits. | |
2310 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2311 | * | |
2312 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2313 | * the oom-killer can be invoked. | |
8a9f3ccd | 2314 | */ |
f817ed48 | 2315 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2316 | gfp_t gfp_mask, |
7ec99d62 | 2317 | unsigned int nr_pages, |
c0ff4b85 | 2318 | struct mem_cgroup **ptr, |
7ec99d62 | 2319 | bool oom) |
8a9f3ccd | 2320 | { |
7ec99d62 | 2321 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2322 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2323 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2324 | int ret; |
a636b327 | 2325 | |
867578cb KH |
2326 | /* |
2327 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2328 | * in system level. So, allow to go ahead dying process in addition to | |
2329 | * MEMDIE process. | |
2330 | */ | |
2331 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2332 | || fatal_signal_pending(current))) | |
2333 | goto bypass; | |
a636b327 | 2334 | |
8a9f3ccd | 2335 | /* |
3be91277 HD |
2336 | * We always charge the cgroup the mm_struct belongs to. |
2337 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2338 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2339 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2340 | */ |
c0ff4b85 | 2341 | if (!*ptr && !mm) |
38c5d72f | 2342 | *ptr = root_mem_cgroup; |
f75ca962 | 2343 | again: |
c0ff4b85 R |
2344 | if (*ptr) { /* css should be a valid one */ |
2345 | memcg = *ptr; | |
c0ff4b85 | 2346 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2347 | goto done; |
c0ff4b85 | 2348 | if (nr_pages == 1 && consume_stock(memcg)) |
f75ca962 | 2349 | goto done; |
c0ff4b85 | 2350 | css_get(&memcg->css); |
4b534334 | 2351 | } else { |
f75ca962 | 2352 | struct task_struct *p; |
54595fe2 | 2353 | |
f75ca962 KH |
2354 | rcu_read_lock(); |
2355 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2356 | /* |
ebb76ce1 | 2357 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2358 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2359 | * race with swapoff. Then, we have small risk of mis-accouning. |
2360 | * But such kind of mis-account by race always happens because | |
2361 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2362 | * small race, here. | |
2363 | * (*) swapoff at el will charge against mm-struct not against | |
2364 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2365 | */ |
c0ff4b85 | 2366 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2367 | if (!memcg) |
2368 | memcg = root_mem_cgroup; | |
2369 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2370 | rcu_read_unlock(); |
2371 | goto done; | |
2372 | } | |
c0ff4b85 | 2373 | if (nr_pages == 1 && consume_stock(memcg)) { |
f75ca962 KH |
2374 | /* |
2375 | * It seems dagerous to access memcg without css_get(). | |
2376 | * But considering how consume_stok works, it's not | |
2377 | * necessary. If consume_stock success, some charges | |
2378 | * from this memcg are cached on this cpu. So, we | |
2379 | * don't need to call css_get()/css_tryget() before | |
2380 | * calling consume_stock(). | |
2381 | */ | |
2382 | rcu_read_unlock(); | |
2383 | goto done; | |
2384 | } | |
2385 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2386 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2387 | rcu_read_unlock(); |
2388 | goto again; | |
2389 | } | |
2390 | rcu_read_unlock(); | |
2391 | } | |
8a9f3ccd | 2392 | |
4b534334 KH |
2393 | do { |
2394 | bool oom_check; | |
7a81b88c | 2395 | |
4b534334 | 2396 | /* If killed, bypass charge */ |
f75ca962 | 2397 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2398 | css_put(&memcg->css); |
4b534334 | 2399 | goto bypass; |
f75ca962 | 2400 | } |
6d61ef40 | 2401 | |
4b534334 KH |
2402 | oom_check = false; |
2403 | if (oom && !nr_oom_retries) { | |
2404 | oom_check = true; | |
2405 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2406 | } |
66e1707b | 2407 | |
c0ff4b85 | 2408 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check); |
4b534334 KH |
2409 | switch (ret) { |
2410 | case CHARGE_OK: | |
2411 | break; | |
2412 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2413 | batch = nr_pages; |
c0ff4b85 R |
2414 | css_put(&memcg->css); |
2415 | memcg = NULL; | |
f75ca962 | 2416 | goto again; |
4b534334 | 2417 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2418 | css_put(&memcg->css); |
4b534334 KH |
2419 | goto nomem; |
2420 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2421 | if (!oom) { |
c0ff4b85 | 2422 | css_put(&memcg->css); |
867578cb | 2423 | goto nomem; |
f75ca962 | 2424 | } |
4b534334 KH |
2425 | /* If oom, we never return -ENOMEM */ |
2426 | nr_oom_retries--; | |
2427 | break; | |
2428 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2429 | css_put(&memcg->css); |
867578cb | 2430 | goto bypass; |
66e1707b | 2431 | } |
4b534334 KH |
2432 | } while (ret != CHARGE_OK); |
2433 | ||
7ec99d62 | 2434 | if (batch > nr_pages) |
c0ff4b85 R |
2435 | refill_stock(memcg, batch - nr_pages); |
2436 | css_put(&memcg->css); | |
0c3e73e8 | 2437 | done: |
c0ff4b85 | 2438 | *ptr = memcg; |
7a81b88c KH |
2439 | return 0; |
2440 | nomem: | |
c0ff4b85 | 2441 | *ptr = NULL; |
7a81b88c | 2442 | return -ENOMEM; |
867578cb | 2443 | bypass: |
38c5d72f KH |
2444 | *ptr = root_mem_cgroup; |
2445 | return -EINTR; | |
7a81b88c | 2446 | } |
8a9f3ccd | 2447 | |
a3032a2c DN |
2448 | /* |
2449 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2450 | * This function is for that and do uncharge, put css's refcnt. | |
2451 | * gotten by try_charge(). | |
2452 | */ | |
c0ff4b85 | 2453 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2454 | unsigned int nr_pages) |
a3032a2c | 2455 | { |
c0ff4b85 | 2456 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2457 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2458 | ||
c0ff4b85 | 2459 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2460 | if (do_swap_account) |
c0ff4b85 | 2461 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2462 | } |
854ffa8d DN |
2463 | } |
2464 | ||
d01dd17f KH |
2465 | /* |
2466 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2467 | * This is useful when moving usage to parent cgroup. | |
2468 | */ | |
2469 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2470 | unsigned int nr_pages) | |
2471 | { | |
2472 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2473 | ||
2474 | if (mem_cgroup_is_root(memcg)) | |
2475 | return; | |
2476 | ||
2477 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2478 | if (do_swap_account) | |
2479 | res_counter_uncharge_until(&memcg->memsw, | |
2480 | memcg->memsw.parent, bytes); | |
2481 | } | |
2482 | ||
a3b2d692 KH |
2483 | /* |
2484 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2485 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2486 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2487 | * called against removed memcg.) | |
a3b2d692 KH |
2488 | */ |
2489 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2490 | { | |
2491 | struct cgroup_subsys_state *css; | |
2492 | ||
2493 | /* ID 0 is unused ID */ | |
2494 | if (!id) | |
2495 | return NULL; | |
2496 | css = css_lookup(&mem_cgroup_subsys, id); | |
2497 | if (!css) | |
2498 | return NULL; | |
b2145145 | 2499 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2500 | } |
2501 | ||
e42d9d5d | 2502 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2503 | { |
c0ff4b85 | 2504 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2505 | struct page_cgroup *pc; |
a3b2d692 | 2506 | unsigned short id; |
b5a84319 KH |
2507 | swp_entry_t ent; |
2508 | ||
3c776e64 DN |
2509 | VM_BUG_ON(!PageLocked(page)); |
2510 | ||
3c776e64 | 2511 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2512 | lock_page_cgroup(pc); |
a3b2d692 | 2513 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2514 | memcg = pc->mem_cgroup; |
2515 | if (memcg && !css_tryget(&memcg->css)) | |
2516 | memcg = NULL; | |
e42d9d5d | 2517 | } else if (PageSwapCache(page)) { |
3c776e64 | 2518 | ent.val = page_private(page); |
9fb4b7cc | 2519 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2520 | rcu_read_lock(); |
c0ff4b85 R |
2521 | memcg = mem_cgroup_lookup(id); |
2522 | if (memcg && !css_tryget(&memcg->css)) | |
2523 | memcg = NULL; | |
a3b2d692 | 2524 | rcu_read_unlock(); |
3c776e64 | 2525 | } |
c0bd3f63 | 2526 | unlock_page_cgroup(pc); |
c0ff4b85 | 2527 | return memcg; |
b5a84319 KH |
2528 | } |
2529 | ||
c0ff4b85 | 2530 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2531 | struct page *page, |
7ec99d62 | 2532 | unsigned int nr_pages, |
9ce70c02 HD |
2533 | enum charge_type ctype, |
2534 | bool lrucare) | |
7a81b88c | 2535 | { |
ce587e65 | 2536 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2537 | struct zone *uninitialized_var(zone); |
fa9add64 | 2538 | struct lruvec *lruvec; |
9ce70c02 | 2539 | bool was_on_lru = false; |
b2402857 | 2540 | bool anon; |
9ce70c02 | 2541 | |
ca3e0214 | 2542 | lock_page_cgroup(pc); |
90deb788 | 2543 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2544 | /* |
2545 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2546 | * accessed by any other context at this point. | |
2547 | */ | |
9ce70c02 HD |
2548 | |
2549 | /* | |
2550 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2551 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2552 | */ | |
2553 | if (lrucare) { | |
2554 | zone = page_zone(page); | |
2555 | spin_lock_irq(&zone->lru_lock); | |
2556 | if (PageLRU(page)) { | |
fa9add64 | 2557 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2558 | ClearPageLRU(page); |
fa9add64 | 2559 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2560 | was_on_lru = true; |
2561 | } | |
2562 | } | |
2563 | ||
c0ff4b85 | 2564 | pc->mem_cgroup = memcg; |
261fb61a KH |
2565 | /* |
2566 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2567 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2568 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2569 | * before USED bit, we need memory barrier here. | |
2570 | * See mem_cgroup_add_lru_list(), etc. | |
2571 | */ | |
08e552c6 | 2572 | smp_wmb(); |
b2402857 | 2573 | SetPageCgroupUsed(pc); |
3be91277 | 2574 | |
9ce70c02 HD |
2575 | if (lrucare) { |
2576 | if (was_on_lru) { | |
fa9add64 | 2577 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2578 | VM_BUG_ON(PageLRU(page)); |
2579 | SetPageLRU(page); | |
fa9add64 | 2580 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2581 | } |
2582 | spin_unlock_irq(&zone->lru_lock); | |
2583 | } | |
2584 | ||
41326c17 | 2585 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2586 | anon = true; |
2587 | else | |
2588 | anon = false; | |
2589 | ||
2590 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2591 | unlock_page_cgroup(pc); |
9ce70c02 | 2592 | |
430e4863 KH |
2593 | /* |
2594 | * "charge_statistics" updated event counter. Then, check it. | |
2595 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2596 | * if they exceeds softlimit. | |
2597 | */ | |
c0ff4b85 | 2598 | memcg_check_events(memcg, page); |
7a81b88c | 2599 | } |
66e1707b | 2600 | |
ca3e0214 KH |
2601 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2602 | ||
a0db00fc | 2603 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
2604 | /* |
2605 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2606 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2607 | * charge/uncharge will be never happen and move_account() is done under | |
2608 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2609 | */ |
e94c8a9c | 2610 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
2611 | { |
2612 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
2613 | struct page_cgroup *pc; |
2614 | int i; | |
ca3e0214 | 2615 | |
3d37c4a9 KH |
2616 | if (mem_cgroup_disabled()) |
2617 | return; | |
e94c8a9c KH |
2618 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
2619 | pc = head_pc + i; | |
2620 | pc->mem_cgroup = head_pc->mem_cgroup; | |
2621 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
2622 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2623 | } | |
ca3e0214 | 2624 | } |
12d27107 | 2625 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2626 | |
f817ed48 | 2627 | /** |
de3638d9 | 2628 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2629 | * @page: the page |
7ec99d62 | 2630 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2631 | * @pc: page_cgroup of the page. |
2632 | * @from: mem_cgroup which the page is moved from. | |
2633 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2634 | * | |
2635 | * The caller must confirm following. | |
08e552c6 | 2636 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2637 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2638 | * |
2f3479b1 KH |
2639 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2640 | * from old cgroup. | |
f817ed48 | 2641 | */ |
7ec99d62 JW |
2642 | static int mem_cgroup_move_account(struct page *page, |
2643 | unsigned int nr_pages, | |
2644 | struct page_cgroup *pc, | |
2645 | struct mem_cgroup *from, | |
2f3479b1 | 2646 | struct mem_cgroup *to) |
f817ed48 | 2647 | { |
de3638d9 JW |
2648 | unsigned long flags; |
2649 | int ret; | |
b2402857 | 2650 | bool anon = PageAnon(page); |
987eba66 | 2651 | |
f817ed48 | 2652 | VM_BUG_ON(from == to); |
5564e88b | 2653 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2654 | /* |
2655 | * The page is isolated from LRU. So, collapse function | |
2656 | * will not handle this page. But page splitting can happen. | |
2657 | * Do this check under compound_page_lock(). The caller should | |
2658 | * hold it. | |
2659 | */ | |
2660 | ret = -EBUSY; | |
7ec99d62 | 2661 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2662 | goto out; |
2663 | ||
2664 | lock_page_cgroup(pc); | |
2665 | ||
2666 | ret = -EINVAL; | |
2667 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2668 | goto unlock; | |
2669 | ||
312734c0 | 2670 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 2671 | |
2ff76f11 | 2672 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
2673 | /* Update mapped_file data for mem_cgroup */ |
2674 | preempt_disable(); | |
2675 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2676 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2677 | preempt_enable(); | |
d69b042f | 2678 | } |
b2402857 | 2679 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
d69b042f | 2680 | |
854ffa8d | 2681 | /* caller should have done css_get */ |
08e552c6 | 2682 | pc->mem_cgroup = to; |
b2402857 | 2683 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
88703267 KH |
2684 | /* |
2685 | * We charges against "to" which may not have any tasks. Then, "to" | |
2686 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2687 | * this function is just force_empty() and move charge, so it's |
25985edc | 2688 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2689 | * status here. |
88703267 | 2690 | */ |
312734c0 | 2691 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
2692 | ret = 0; |
2693 | unlock: | |
57f9fd7d | 2694 | unlock_page_cgroup(pc); |
d2265e6f KH |
2695 | /* |
2696 | * check events | |
2697 | */ | |
5564e88b JW |
2698 | memcg_check_events(to, page); |
2699 | memcg_check_events(from, page); | |
de3638d9 | 2700 | out: |
f817ed48 KH |
2701 | return ret; |
2702 | } | |
2703 | ||
2ef37d3f MH |
2704 | /** |
2705 | * mem_cgroup_move_parent - moves page to the parent group | |
2706 | * @page: the page to move | |
2707 | * @pc: page_cgroup of the page | |
2708 | * @child: page's cgroup | |
2709 | * | |
2710 | * move charges to its parent or the root cgroup if the group has no | |
2711 | * parent (aka use_hierarchy==0). | |
2712 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
2713 | * mem_cgroup_move_account fails) the failure is always temporary and | |
2714 | * it signals a race with a page removal/uncharge or migration. In the | |
2715 | * first case the page is on the way out and it will vanish from the LRU | |
2716 | * on the next attempt and the call should be retried later. | |
2717 | * Isolation from the LRU fails only if page has been isolated from | |
2718 | * the LRU since we looked at it and that usually means either global | |
2719 | * reclaim or migration going on. The page will either get back to the | |
2720 | * LRU or vanish. | |
2721 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
2722 | * (!PageCgroupUsed) or moved to a different group. The page will | |
2723 | * disappear in the next attempt. | |
f817ed48 | 2724 | */ |
5564e88b JW |
2725 | static int mem_cgroup_move_parent(struct page *page, |
2726 | struct page_cgroup *pc, | |
6068bf01 | 2727 | struct mem_cgroup *child) |
f817ed48 | 2728 | { |
f817ed48 | 2729 | struct mem_cgroup *parent; |
7ec99d62 | 2730 | unsigned int nr_pages; |
4be4489f | 2731 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2732 | int ret; |
2733 | ||
d8423011 | 2734 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 2735 | |
57f9fd7d DN |
2736 | ret = -EBUSY; |
2737 | if (!get_page_unless_zero(page)) | |
2738 | goto out; | |
2739 | if (isolate_lru_page(page)) | |
2740 | goto put; | |
52dbb905 | 2741 | |
7ec99d62 | 2742 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2743 | |
cc926f78 KH |
2744 | parent = parent_mem_cgroup(child); |
2745 | /* | |
2746 | * If no parent, move charges to root cgroup. | |
2747 | */ | |
2748 | if (!parent) | |
2749 | parent = root_mem_cgroup; | |
f817ed48 | 2750 | |
2ef37d3f MH |
2751 | if (nr_pages > 1) { |
2752 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 2753 | flags = compound_lock_irqsave(page); |
2ef37d3f | 2754 | } |
987eba66 | 2755 | |
cc926f78 | 2756 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 2757 | pc, child, parent); |
cc926f78 KH |
2758 | if (!ret) |
2759 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 2760 | |
7ec99d62 | 2761 | if (nr_pages > 1) |
987eba66 | 2762 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 2763 | putback_lru_page(page); |
57f9fd7d | 2764 | put: |
40d58138 | 2765 | put_page(page); |
57f9fd7d | 2766 | out: |
f817ed48 KH |
2767 | return ret; |
2768 | } | |
2769 | ||
7a81b88c KH |
2770 | /* |
2771 | * Charge the memory controller for page usage. | |
2772 | * Return | |
2773 | * 0 if the charge was successful | |
2774 | * < 0 if the cgroup is over its limit | |
2775 | */ | |
2776 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2777 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2778 | { |
c0ff4b85 | 2779 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2780 | unsigned int nr_pages = 1; |
8493ae43 | 2781 | bool oom = true; |
7a81b88c | 2782 | int ret; |
ec168510 | 2783 | |
37c2ac78 | 2784 | if (PageTransHuge(page)) { |
7ec99d62 | 2785 | nr_pages <<= compound_order(page); |
37c2ac78 | 2786 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2787 | /* |
2788 | * Never OOM-kill a process for a huge page. The | |
2789 | * fault handler will fall back to regular pages. | |
2790 | */ | |
2791 | oom = false; | |
37c2ac78 | 2792 | } |
7a81b88c | 2793 | |
c0ff4b85 | 2794 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 2795 | if (ret == -ENOMEM) |
7a81b88c | 2796 | return ret; |
ce587e65 | 2797 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 2798 | return 0; |
8a9f3ccd BS |
2799 | } |
2800 | ||
7a81b88c KH |
2801 | int mem_cgroup_newpage_charge(struct page *page, |
2802 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2803 | { |
f8d66542 | 2804 | if (mem_cgroup_disabled()) |
cede86ac | 2805 | return 0; |
7a0524cf JW |
2806 | VM_BUG_ON(page_mapped(page)); |
2807 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
2808 | VM_BUG_ON(!mm); | |
217bc319 | 2809 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 2810 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
2811 | } |
2812 | ||
54595fe2 KH |
2813 | /* |
2814 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2815 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2816 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2817 | * "commit()" or removed by "cancel()" |
2818 | */ | |
0435a2fd JW |
2819 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2820 | struct page *page, | |
2821 | gfp_t mask, | |
2822 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 2823 | { |
c0ff4b85 | 2824 | struct mem_cgroup *memcg; |
90deb788 | 2825 | struct page_cgroup *pc; |
54595fe2 | 2826 | int ret; |
8c7c6e34 | 2827 | |
90deb788 JW |
2828 | pc = lookup_page_cgroup(page); |
2829 | /* | |
2830 | * Every swap fault against a single page tries to charge the | |
2831 | * page, bail as early as possible. shmem_unuse() encounters | |
2832 | * already charged pages, too. The USED bit is protected by | |
2833 | * the page lock, which serializes swap cache removal, which | |
2834 | * in turn serializes uncharging. | |
2835 | */ | |
2836 | if (PageCgroupUsed(pc)) | |
2837 | return 0; | |
8c7c6e34 KH |
2838 | if (!do_swap_account) |
2839 | goto charge_cur_mm; | |
c0ff4b85 R |
2840 | memcg = try_get_mem_cgroup_from_page(page); |
2841 | if (!memcg) | |
54595fe2 | 2842 | goto charge_cur_mm; |
72835c86 JW |
2843 | *memcgp = memcg; |
2844 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 2845 | css_put(&memcg->css); |
38c5d72f KH |
2846 | if (ret == -EINTR) |
2847 | ret = 0; | |
54595fe2 | 2848 | return ret; |
8c7c6e34 | 2849 | charge_cur_mm: |
38c5d72f KH |
2850 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
2851 | if (ret == -EINTR) | |
2852 | ret = 0; | |
2853 | return ret; | |
8c7c6e34 KH |
2854 | } |
2855 | ||
0435a2fd JW |
2856 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
2857 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
2858 | { | |
2859 | *memcgp = NULL; | |
2860 | if (mem_cgroup_disabled()) | |
2861 | return 0; | |
bdf4f4d2 JW |
2862 | /* |
2863 | * A racing thread's fault, or swapoff, may have already | |
2864 | * updated the pte, and even removed page from swap cache: in | |
2865 | * those cases unuse_pte()'s pte_same() test will fail; but | |
2866 | * there's also a KSM case which does need to charge the page. | |
2867 | */ | |
2868 | if (!PageSwapCache(page)) { | |
2869 | int ret; | |
2870 | ||
2871 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
2872 | if (ret == -EINTR) | |
2873 | ret = 0; | |
2874 | return ret; | |
2875 | } | |
0435a2fd JW |
2876 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
2877 | } | |
2878 | ||
827a03d2 JW |
2879 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
2880 | { | |
2881 | if (mem_cgroup_disabled()) | |
2882 | return; | |
2883 | if (!memcg) | |
2884 | return; | |
2885 | __mem_cgroup_cancel_charge(memcg, 1); | |
2886 | } | |
2887 | ||
83aae4c7 | 2888 | static void |
72835c86 | 2889 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 2890 | enum charge_type ctype) |
7a81b88c | 2891 | { |
f8d66542 | 2892 | if (mem_cgroup_disabled()) |
7a81b88c | 2893 | return; |
72835c86 | 2894 | if (!memcg) |
7a81b88c | 2895 | return; |
72835c86 | 2896 | cgroup_exclude_rmdir(&memcg->css); |
5a6475a4 | 2897 | |
ce587e65 | 2898 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
2899 | /* |
2900 | * Now swap is on-memory. This means this page may be | |
2901 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2902 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2903 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2904 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2905 | */ |
03f3c433 | 2906 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2907 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 2908 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 2909 | } |
88703267 KH |
2910 | /* |
2911 | * At swapin, we may charge account against cgroup which has no tasks. | |
2912 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2913 | * In that case, we need to call pre_destroy() again. check it here. | |
2914 | */ | |
72835c86 | 2915 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
7a81b88c KH |
2916 | } |
2917 | ||
72835c86 JW |
2918 | void mem_cgroup_commit_charge_swapin(struct page *page, |
2919 | struct mem_cgroup *memcg) | |
83aae4c7 | 2920 | { |
72835c86 | 2921 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 2922 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
2923 | } |
2924 | ||
827a03d2 JW |
2925 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2926 | gfp_t gfp_mask) | |
7a81b88c | 2927 | { |
827a03d2 JW |
2928 | struct mem_cgroup *memcg = NULL; |
2929 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2930 | int ret; | |
2931 | ||
f8d66542 | 2932 | if (mem_cgroup_disabled()) |
827a03d2 JW |
2933 | return 0; |
2934 | if (PageCompound(page)) | |
2935 | return 0; | |
2936 | ||
827a03d2 JW |
2937 | if (!PageSwapCache(page)) |
2938 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
2939 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
2940 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
2941 | gfp_mask, &memcg); | |
827a03d2 JW |
2942 | if (!ret) |
2943 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
2944 | } | |
2945 | return ret; | |
7a81b88c KH |
2946 | } |
2947 | ||
c0ff4b85 | 2948 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2949 | unsigned int nr_pages, |
2950 | const enum charge_type ctype) | |
569b846d KH |
2951 | { |
2952 | struct memcg_batch_info *batch = NULL; | |
2953 | bool uncharge_memsw = true; | |
7ec99d62 | 2954 | |
569b846d KH |
2955 | /* If swapout, usage of swap doesn't decrease */ |
2956 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2957 | uncharge_memsw = false; | |
569b846d KH |
2958 | |
2959 | batch = ¤t->memcg_batch; | |
2960 | /* | |
2961 | * In usual, we do css_get() when we remember memcg pointer. | |
2962 | * But in this case, we keep res->usage until end of a series of | |
2963 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2964 | */ | |
2965 | if (!batch->memcg) | |
c0ff4b85 | 2966 | batch->memcg = memcg; |
3c11ecf4 KH |
2967 | /* |
2968 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2969 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2970 | * the same cgroup and we have chance to coalesce uncharges. |
2971 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2972 | * because we want to do uncharge as soon as possible. | |
2973 | */ | |
2974 | ||
2975 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2976 | goto direct_uncharge; | |
2977 | ||
7ec99d62 | 2978 | if (nr_pages > 1) |
ec168510 AA |
2979 | goto direct_uncharge; |
2980 | ||
569b846d KH |
2981 | /* |
2982 | * In typical case, batch->memcg == mem. This means we can | |
2983 | * merge a series of uncharges to an uncharge of res_counter. | |
2984 | * If not, we uncharge res_counter ony by one. | |
2985 | */ | |
c0ff4b85 | 2986 | if (batch->memcg != memcg) |
569b846d KH |
2987 | goto direct_uncharge; |
2988 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2989 | batch->nr_pages++; |
569b846d | 2990 | if (uncharge_memsw) |
7ffd4ca7 | 2991 | batch->memsw_nr_pages++; |
569b846d KH |
2992 | return; |
2993 | direct_uncharge: | |
c0ff4b85 | 2994 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 2995 | if (uncharge_memsw) |
c0ff4b85 R |
2996 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
2997 | if (unlikely(batch->memcg != memcg)) | |
2998 | memcg_oom_recover(memcg); | |
569b846d | 2999 | } |
7a81b88c | 3000 | |
8a9f3ccd | 3001 | /* |
69029cd5 | 3002 | * uncharge if !page_mapped(page) |
8a9f3ccd | 3003 | */ |
8c7c6e34 | 3004 | static struct mem_cgroup * |
0030f535 JW |
3005 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
3006 | bool end_migration) | |
8a9f3ccd | 3007 | { |
c0ff4b85 | 3008 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
3009 | unsigned int nr_pages = 1; |
3010 | struct page_cgroup *pc; | |
b2402857 | 3011 | bool anon; |
8a9f3ccd | 3012 | |
f8d66542 | 3013 | if (mem_cgroup_disabled()) |
8c7c6e34 | 3014 | return NULL; |
4077960e | 3015 | |
0c59b89c | 3016 | VM_BUG_ON(PageSwapCache(page)); |
d13d1443 | 3017 | |
37c2ac78 | 3018 | if (PageTransHuge(page)) { |
7ec99d62 | 3019 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
3020 | VM_BUG_ON(!PageTransHuge(page)); |
3021 | } | |
8697d331 | 3022 | /* |
3c541e14 | 3023 | * Check if our page_cgroup is valid |
8697d331 | 3024 | */ |
52d4b9ac | 3025 | pc = lookup_page_cgroup(page); |
cfa44946 | 3026 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 3027 | return NULL; |
b9c565d5 | 3028 | |
52d4b9ac | 3029 | lock_page_cgroup(pc); |
d13d1443 | 3030 | |
c0ff4b85 | 3031 | memcg = pc->mem_cgroup; |
8c7c6e34 | 3032 | |
d13d1443 KH |
3033 | if (!PageCgroupUsed(pc)) |
3034 | goto unlock_out; | |
3035 | ||
b2402857 KH |
3036 | anon = PageAnon(page); |
3037 | ||
d13d1443 | 3038 | switch (ctype) { |
41326c17 | 3039 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
3040 | /* |
3041 | * Generally PageAnon tells if it's the anon statistics to be | |
3042 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
3043 | * used before page reached the stage of being marked PageAnon. | |
3044 | */ | |
b2402857 KH |
3045 | anon = true; |
3046 | /* fallthrough */ | |
8a9478ca | 3047 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 3048 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
3049 | if (page_mapped(page)) |
3050 | goto unlock_out; | |
3051 | /* | |
3052 | * Pages under migration may not be uncharged. But | |
3053 | * end_migration() /must/ be the one uncharging the | |
3054 | * unused post-migration page and so it has to call | |
3055 | * here with the migration bit still set. See the | |
3056 | * res_counter handling below. | |
3057 | */ | |
3058 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
3059 | goto unlock_out; |
3060 | break; | |
3061 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
3062 | if (!PageAnon(page)) { /* Shared memory */ | |
3063 | if (page->mapping && !page_is_file_cache(page)) | |
3064 | goto unlock_out; | |
3065 | } else if (page_mapped(page)) /* Anon */ | |
3066 | goto unlock_out; | |
3067 | break; | |
3068 | default: | |
3069 | break; | |
52d4b9ac | 3070 | } |
d13d1443 | 3071 | |
b2402857 | 3072 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 3073 | |
52d4b9ac | 3074 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3075 | /* |
3076 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3077 | * freed from LRU. This is safe because uncharged page is expected not | |
3078 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3079 | * special functions. | |
3080 | */ | |
b9c565d5 | 3081 | |
52d4b9ac | 3082 | unlock_page_cgroup(pc); |
f75ca962 | 3083 | /* |
c0ff4b85 | 3084 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
3085 | * will never be freed. |
3086 | */ | |
c0ff4b85 | 3087 | memcg_check_events(memcg, page); |
f75ca962 | 3088 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
3089 | mem_cgroup_swap_statistics(memcg, true); |
3090 | mem_cgroup_get(memcg); | |
f75ca962 | 3091 | } |
0030f535 JW |
3092 | /* |
3093 | * Migration does not charge the res_counter for the | |
3094 | * replacement page, so leave it alone when phasing out the | |
3095 | * page that is unused after the migration. | |
3096 | */ | |
3097 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 3098 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 3099 | |
c0ff4b85 | 3100 | return memcg; |
d13d1443 KH |
3101 | |
3102 | unlock_out: | |
3103 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3104 | return NULL; |
3c541e14 BS |
3105 | } |
3106 | ||
69029cd5 KH |
3107 | void mem_cgroup_uncharge_page(struct page *page) |
3108 | { | |
52d4b9ac KH |
3109 | /* early check. */ |
3110 | if (page_mapped(page)) | |
3111 | return; | |
40f23a21 | 3112 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
0c59b89c JW |
3113 | if (PageSwapCache(page)) |
3114 | return; | |
0030f535 | 3115 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
3116 | } |
3117 | ||
3118 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3119 | { | |
3120 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3121 | VM_BUG_ON(page->mapping); |
0030f535 | 3122 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
3123 | } |
3124 | ||
569b846d KH |
3125 | /* |
3126 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3127 | * In that cases, pages are freed continuously and we can expect pages | |
3128 | * are in the same memcg. All these calls itself limits the number of | |
3129 | * pages freed at once, then uncharge_start/end() is called properly. | |
3130 | * This may be called prural(2) times in a context, | |
3131 | */ | |
3132 | ||
3133 | void mem_cgroup_uncharge_start(void) | |
3134 | { | |
3135 | current->memcg_batch.do_batch++; | |
3136 | /* We can do nest. */ | |
3137 | if (current->memcg_batch.do_batch == 1) { | |
3138 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3139 | current->memcg_batch.nr_pages = 0; |
3140 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3141 | } |
3142 | } | |
3143 | ||
3144 | void mem_cgroup_uncharge_end(void) | |
3145 | { | |
3146 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3147 | ||
3148 | if (!batch->do_batch) | |
3149 | return; | |
3150 | ||
3151 | batch->do_batch--; | |
3152 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3153 | return; | |
3154 | ||
3155 | if (!batch->memcg) | |
3156 | return; | |
3157 | /* | |
3158 | * This "batch->memcg" is valid without any css_get/put etc... | |
3159 | * bacause we hide charges behind us. | |
3160 | */ | |
7ffd4ca7 JW |
3161 | if (batch->nr_pages) |
3162 | res_counter_uncharge(&batch->memcg->res, | |
3163 | batch->nr_pages * PAGE_SIZE); | |
3164 | if (batch->memsw_nr_pages) | |
3165 | res_counter_uncharge(&batch->memcg->memsw, | |
3166 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3167 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3168 | /* forget this pointer (for sanity check) */ |
3169 | batch->memcg = NULL; | |
3170 | } | |
3171 | ||
e767e056 | 3172 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3173 | /* |
e767e056 | 3174 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3175 | * memcg information is recorded to swap_cgroup of "ent" |
3176 | */ | |
8a9478ca KH |
3177 | void |
3178 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3179 | { |
3180 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3181 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3182 | ||
3183 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3184 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3185 | ||
0030f535 | 3186 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 3187 | |
f75ca962 KH |
3188 | /* |
3189 | * record memcg information, if swapout && memcg != NULL, | |
3190 | * mem_cgroup_get() was called in uncharge(). | |
3191 | */ | |
3192 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3193 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3194 | } |
e767e056 | 3195 | #endif |
8c7c6e34 | 3196 | |
c255a458 | 3197 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
3198 | /* |
3199 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3200 | * uncharge "memsw" account. | |
3201 | */ | |
3202 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3203 | { |
8c7c6e34 | 3204 | struct mem_cgroup *memcg; |
a3b2d692 | 3205 | unsigned short id; |
8c7c6e34 KH |
3206 | |
3207 | if (!do_swap_account) | |
3208 | return; | |
3209 | ||
a3b2d692 KH |
3210 | id = swap_cgroup_record(ent, 0); |
3211 | rcu_read_lock(); | |
3212 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3213 | if (memcg) { |
a3b2d692 KH |
3214 | /* |
3215 | * We uncharge this because swap is freed. | |
3216 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3217 | */ | |
0c3e73e8 | 3218 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3219 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3220 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3221 | mem_cgroup_put(memcg); |
3222 | } | |
a3b2d692 | 3223 | rcu_read_unlock(); |
d13d1443 | 3224 | } |
02491447 DN |
3225 | |
3226 | /** | |
3227 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3228 | * @entry: swap entry to be moved | |
3229 | * @from: mem_cgroup which the entry is moved from | |
3230 | * @to: mem_cgroup which the entry is moved to | |
3231 | * | |
3232 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3233 | * as the mem_cgroup's id of @from. | |
3234 | * | |
3235 | * Returns 0 on success, -EINVAL on failure. | |
3236 | * | |
3237 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3238 | * both res and memsw, and called css_get(). | |
3239 | */ | |
3240 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3241 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3242 | { |
3243 | unsigned short old_id, new_id; | |
3244 | ||
3245 | old_id = css_id(&from->css); | |
3246 | new_id = css_id(&to->css); | |
3247 | ||
3248 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3249 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3250 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3251 | /* |
483c30b5 DN |
3252 | * This function is only called from task migration context now. |
3253 | * It postpones res_counter and refcount handling till the end | |
3254 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3255 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3256 | * because if the process that has been moved to @to does | |
3257 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3258 | */ |
02491447 | 3259 | mem_cgroup_get(to); |
02491447 DN |
3260 | return 0; |
3261 | } | |
3262 | return -EINVAL; | |
3263 | } | |
3264 | #else | |
3265 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3266 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3267 | { |
3268 | return -EINVAL; | |
3269 | } | |
8c7c6e34 | 3270 | #endif |
d13d1443 | 3271 | |
ae41be37 | 3272 | /* |
01b1ae63 KH |
3273 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3274 | * page belongs to. | |
ae41be37 | 3275 | */ |
0030f535 JW |
3276 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
3277 | struct mem_cgroup **memcgp) | |
ae41be37 | 3278 | { |
c0ff4b85 | 3279 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3280 | struct page_cgroup *pc; |
ac39cf8c | 3281 | enum charge_type ctype; |
8869b8f6 | 3282 | |
72835c86 | 3283 | *memcgp = NULL; |
56039efa | 3284 | |
ec168510 | 3285 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3286 | if (mem_cgroup_disabled()) |
0030f535 | 3287 | return; |
4077960e | 3288 | |
52d4b9ac KH |
3289 | pc = lookup_page_cgroup(page); |
3290 | lock_page_cgroup(pc); | |
3291 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3292 | memcg = pc->mem_cgroup; |
3293 | css_get(&memcg->css); | |
ac39cf8c | 3294 | /* |
3295 | * At migrating an anonymous page, its mapcount goes down | |
3296 | * to 0 and uncharge() will be called. But, even if it's fully | |
3297 | * unmapped, migration may fail and this page has to be | |
3298 | * charged again. We set MIGRATION flag here and delay uncharge | |
3299 | * until end_migration() is called | |
3300 | * | |
3301 | * Corner Case Thinking | |
3302 | * A) | |
3303 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3304 | * while migration was ongoing. | |
3305 | * If unmap finds the old page, uncharge() of it will be delayed | |
3306 | * until end_migration(). If unmap finds a new page, it's | |
3307 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3308 | * finds swap_migration_entry, the new page will not be mapped | |
3309 | * and end_migration() will find it(mapcount==0). | |
3310 | * | |
3311 | * B) | |
3312 | * When the old page was mapped but migraion fails, the kernel | |
3313 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3314 | * if mapcount goes down to 0. We can do remap successfully | |
3315 | * without charging it again. | |
3316 | * | |
3317 | * C) | |
3318 | * The "old" page is under lock_page() until the end of | |
3319 | * migration, so, the old page itself will not be swapped-out. | |
3320 | * If the new page is swapped out before end_migraton, our | |
3321 | * hook to usual swap-out path will catch the event. | |
3322 | */ | |
3323 | if (PageAnon(page)) | |
3324 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3325 | } |
52d4b9ac | 3326 | unlock_page_cgroup(pc); |
ac39cf8c | 3327 | /* |
3328 | * If the page is not charged at this point, | |
3329 | * we return here. | |
3330 | */ | |
c0ff4b85 | 3331 | if (!memcg) |
0030f535 | 3332 | return; |
01b1ae63 | 3333 | |
72835c86 | 3334 | *memcgp = memcg; |
ac39cf8c | 3335 | /* |
3336 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3337 | * is called before end_migration, we can catch all events on this new | |
3338 | * page. In the case new page is migrated but not remapped, new page's | |
3339 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3340 | */ | |
ac39cf8c | 3341 | if (PageAnon(page)) |
41326c17 | 3342 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 3343 | else |
62ba7442 | 3344 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
3345 | /* |
3346 | * The page is committed to the memcg, but it's not actually | |
3347 | * charged to the res_counter since we plan on replacing the | |
3348 | * old one and only one page is going to be left afterwards. | |
3349 | */ | |
ce587e65 | 3350 | __mem_cgroup_commit_charge(memcg, newpage, 1, ctype, false); |
ae41be37 | 3351 | } |
8869b8f6 | 3352 | |
69029cd5 | 3353 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3354 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3355 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3356 | { |
ac39cf8c | 3357 | struct page *used, *unused; |
01b1ae63 | 3358 | struct page_cgroup *pc; |
b2402857 | 3359 | bool anon; |
01b1ae63 | 3360 | |
c0ff4b85 | 3361 | if (!memcg) |
01b1ae63 | 3362 | return; |
ac39cf8c | 3363 | /* blocks rmdir() */ |
c0ff4b85 | 3364 | cgroup_exclude_rmdir(&memcg->css); |
50de1dd9 | 3365 | if (!migration_ok) { |
ac39cf8c | 3366 | used = oldpage; |
3367 | unused = newpage; | |
01b1ae63 | 3368 | } else { |
ac39cf8c | 3369 | used = newpage; |
01b1ae63 KH |
3370 | unused = oldpage; |
3371 | } | |
0030f535 | 3372 | anon = PageAnon(used); |
7d188958 JW |
3373 | __mem_cgroup_uncharge_common(unused, |
3374 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
3375 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
3376 | true); | |
0030f535 | 3377 | css_put(&memcg->css); |
69029cd5 | 3378 | /* |
ac39cf8c | 3379 | * We disallowed uncharge of pages under migration because mapcount |
3380 | * of the page goes down to zero, temporarly. | |
3381 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3382 | */ |
ac39cf8c | 3383 | pc = lookup_page_cgroup(oldpage); |
3384 | lock_page_cgroup(pc); | |
3385 | ClearPageCgroupMigration(pc); | |
3386 | unlock_page_cgroup(pc); | |
ac39cf8c | 3387 | |
01b1ae63 | 3388 | /* |
ac39cf8c | 3389 | * If a page is a file cache, radix-tree replacement is very atomic |
3390 | * and we can skip this check. When it was an Anon page, its mapcount | |
3391 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3392 | * uncharged yet. There are several case but page->mapcount check | |
3393 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3394 | * check. (see prepare_charge() also) | |
69029cd5 | 3395 | */ |
b2402857 | 3396 | if (anon) |
ac39cf8c | 3397 | mem_cgroup_uncharge_page(used); |
88703267 | 3398 | /* |
ac39cf8c | 3399 | * At migration, we may charge account against cgroup which has no |
3400 | * tasks. | |
88703267 KH |
3401 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3402 | * In that case, we need to call pre_destroy() again. check it here. | |
3403 | */ | |
c0ff4b85 | 3404 | cgroup_release_and_wakeup_rmdir(&memcg->css); |
ae41be37 | 3405 | } |
78fb7466 | 3406 | |
ab936cbc KH |
3407 | /* |
3408 | * At replace page cache, newpage is not under any memcg but it's on | |
3409 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
3410 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
3411 | */ | |
3412 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
3413 | struct page *newpage) | |
3414 | { | |
bde05d1c | 3415 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 3416 | struct page_cgroup *pc; |
ab936cbc | 3417 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
3418 | |
3419 | if (mem_cgroup_disabled()) | |
3420 | return; | |
3421 | ||
3422 | pc = lookup_page_cgroup(oldpage); | |
3423 | /* fix accounting on old pages */ | |
3424 | lock_page_cgroup(pc); | |
bde05d1c HD |
3425 | if (PageCgroupUsed(pc)) { |
3426 | memcg = pc->mem_cgroup; | |
3427 | mem_cgroup_charge_statistics(memcg, false, -1); | |
3428 | ClearPageCgroupUsed(pc); | |
3429 | } | |
ab936cbc KH |
3430 | unlock_page_cgroup(pc); |
3431 | ||
bde05d1c HD |
3432 | /* |
3433 | * When called from shmem_replace_page(), in some cases the | |
3434 | * oldpage has already been charged, and in some cases not. | |
3435 | */ | |
3436 | if (!memcg) | |
3437 | return; | |
ab936cbc KH |
3438 | /* |
3439 | * Even if newpage->mapping was NULL before starting replacement, | |
3440 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
3441 | * LRU while we overwrite pc->mem_cgroup. | |
3442 | */ | |
ce587e65 | 3443 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
3444 | } |
3445 | ||
f212ad7c DN |
3446 | #ifdef CONFIG_DEBUG_VM |
3447 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3448 | { | |
3449 | struct page_cgroup *pc; | |
3450 | ||
3451 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3452 | /* |
3453 | * Can be NULL while feeding pages into the page allocator for | |
3454 | * the first time, i.e. during boot or memory hotplug; | |
3455 | * or when mem_cgroup_disabled(). | |
3456 | */ | |
f212ad7c DN |
3457 | if (likely(pc) && PageCgroupUsed(pc)) |
3458 | return pc; | |
3459 | return NULL; | |
3460 | } | |
3461 | ||
3462 | bool mem_cgroup_bad_page_check(struct page *page) | |
3463 | { | |
3464 | if (mem_cgroup_disabled()) | |
3465 | return false; | |
3466 | ||
3467 | return lookup_page_cgroup_used(page) != NULL; | |
3468 | } | |
3469 | ||
3470 | void mem_cgroup_print_bad_page(struct page *page) | |
3471 | { | |
3472 | struct page_cgroup *pc; | |
3473 | ||
3474 | pc = lookup_page_cgroup_used(page); | |
3475 | if (pc) { | |
90b3feae | 3476 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
f212ad7c | 3477 | pc, pc->flags, pc->mem_cgroup); |
f212ad7c DN |
3478 | } |
3479 | } | |
3480 | #endif | |
3481 | ||
8c7c6e34 KH |
3482 | static DEFINE_MUTEX(set_limit_mutex); |
3483 | ||
d38d2a75 | 3484 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3485 | unsigned long long val) |
628f4235 | 3486 | { |
81d39c20 | 3487 | int retry_count; |
3c11ecf4 | 3488 | u64 memswlimit, memlimit; |
628f4235 | 3489 | int ret = 0; |
81d39c20 KH |
3490 | int children = mem_cgroup_count_children(memcg); |
3491 | u64 curusage, oldusage; | |
3c11ecf4 | 3492 | int enlarge; |
81d39c20 KH |
3493 | |
3494 | /* | |
3495 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3496 | * is depends on callers. We set our retry-count to be function | |
3497 | * of # of children which we should visit in this loop. | |
3498 | */ | |
3499 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3500 | ||
3501 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3502 | |
3c11ecf4 | 3503 | enlarge = 0; |
8c7c6e34 | 3504 | while (retry_count) { |
628f4235 KH |
3505 | if (signal_pending(current)) { |
3506 | ret = -EINTR; | |
3507 | break; | |
3508 | } | |
8c7c6e34 KH |
3509 | /* |
3510 | * Rather than hide all in some function, I do this in | |
3511 | * open coded manner. You see what this really does. | |
aaad153e | 3512 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3513 | */ |
3514 | mutex_lock(&set_limit_mutex); | |
3515 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3516 | if (memswlimit < val) { | |
3517 | ret = -EINVAL; | |
3518 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3519 | break; |
3520 | } | |
3c11ecf4 KH |
3521 | |
3522 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3523 | if (memlimit < val) | |
3524 | enlarge = 1; | |
3525 | ||
8c7c6e34 | 3526 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3527 | if (!ret) { |
3528 | if (memswlimit == val) | |
3529 | memcg->memsw_is_minimum = true; | |
3530 | else | |
3531 | memcg->memsw_is_minimum = false; | |
3532 | } | |
8c7c6e34 KH |
3533 | mutex_unlock(&set_limit_mutex); |
3534 | ||
3535 | if (!ret) | |
3536 | break; | |
3537 | ||
5660048c JW |
3538 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3539 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3540 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3541 | /* Usage is reduced ? */ | |
3542 | if (curusage >= oldusage) | |
3543 | retry_count--; | |
3544 | else | |
3545 | oldusage = curusage; | |
8c7c6e34 | 3546 | } |
3c11ecf4 KH |
3547 | if (!ret && enlarge) |
3548 | memcg_oom_recover(memcg); | |
14797e23 | 3549 | |
8c7c6e34 KH |
3550 | return ret; |
3551 | } | |
3552 | ||
338c8431 LZ |
3553 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3554 | unsigned long long val) | |
8c7c6e34 | 3555 | { |
81d39c20 | 3556 | int retry_count; |
3c11ecf4 | 3557 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3558 | int children = mem_cgroup_count_children(memcg); |
3559 | int ret = -EBUSY; | |
3c11ecf4 | 3560 | int enlarge = 0; |
8c7c6e34 | 3561 | |
81d39c20 KH |
3562 | /* see mem_cgroup_resize_res_limit */ |
3563 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3564 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3565 | while (retry_count) { |
3566 | if (signal_pending(current)) { | |
3567 | ret = -EINTR; | |
3568 | break; | |
3569 | } | |
3570 | /* | |
3571 | * Rather than hide all in some function, I do this in | |
3572 | * open coded manner. You see what this really does. | |
aaad153e | 3573 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3574 | */ |
3575 | mutex_lock(&set_limit_mutex); | |
3576 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3577 | if (memlimit > val) { | |
3578 | ret = -EINVAL; | |
3579 | mutex_unlock(&set_limit_mutex); | |
3580 | break; | |
3581 | } | |
3c11ecf4 KH |
3582 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3583 | if (memswlimit < val) | |
3584 | enlarge = 1; | |
8c7c6e34 | 3585 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3586 | if (!ret) { |
3587 | if (memlimit == val) | |
3588 | memcg->memsw_is_minimum = true; | |
3589 | else | |
3590 | memcg->memsw_is_minimum = false; | |
3591 | } | |
8c7c6e34 KH |
3592 | mutex_unlock(&set_limit_mutex); |
3593 | ||
3594 | if (!ret) | |
3595 | break; | |
3596 | ||
5660048c JW |
3597 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3598 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3599 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3600 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3601 | /* Usage is reduced ? */ |
8c7c6e34 | 3602 | if (curusage >= oldusage) |
628f4235 | 3603 | retry_count--; |
81d39c20 KH |
3604 | else |
3605 | oldusage = curusage; | |
628f4235 | 3606 | } |
3c11ecf4 KH |
3607 | if (!ret && enlarge) |
3608 | memcg_oom_recover(memcg); | |
628f4235 KH |
3609 | return ret; |
3610 | } | |
3611 | ||
4e416953 | 3612 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3613 | gfp_t gfp_mask, |
3614 | unsigned long *total_scanned) | |
4e416953 BS |
3615 | { |
3616 | unsigned long nr_reclaimed = 0; | |
3617 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3618 | unsigned long reclaimed; | |
3619 | int loop = 0; | |
3620 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3621 | unsigned long long excess; |
0ae5e89c | 3622 | unsigned long nr_scanned; |
4e416953 BS |
3623 | |
3624 | if (order > 0) | |
3625 | return 0; | |
3626 | ||
00918b6a | 3627 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3628 | /* |
3629 | * This loop can run a while, specially if mem_cgroup's continuously | |
3630 | * keep exceeding their soft limit and putting the system under | |
3631 | * pressure | |
3632 | */ | |
3633 | do { | |
3634 | if (next_mz) | |
3635 | mz = next_mz; | |
3636 | else | |
3637 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3638 | if (!mz) | |
3639 | break; | |
3640 | ||
0ae5e89c | 3641 | nr_scanned = 0; |
d79154bb | 3642 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 3643 | gfp_mask, &nr_scanned); |
4e416953 | 3644 | nr_reclaimed += reclaimed; |
0ae5e89c | 3645 | *total_scanned += nr_scanned; |
4e416953 BS |
3646 | spin_lock(&mctz->lock); |
3647 | ||
3648 | /* | |
3649 | * If we failed to reclaim anything from this memory cgroup | |
3650 | * it is time to move on to the next cgroup | |
3651 | */ | |
3652 | next_mz = NULL; | |
3653 | if (!reclaimed) { | |
3654 | do { | |
3655 | /* | |
3656 | * Loop until we find yet another one. | |
3657 | * | |
3658 | * By the time we get the soft_limit lock | |
3659 | * again, someone might have aded the | |
3660 | * group back on the RB tree. Iterate to | |
3661 | * make sure we get a different mem. | |
3662 | * mem_cgroup_largest_soft_limit_node returns | |
3663 | * NULL if no other cgroup is present on | |
3664 | * the tree | |
3665 | */ | |
3666 | next_mz = | |
3667 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3668 | if (next_mz == mz) |
d79154bb | 3669 | css_put(&next_mz->memcg->css); |
39cc98f1 | 3670 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3671 | break; |
3672 | } while (1); | |
3673 | } | |
d79154bb HD |
3674 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
3675 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
3676 | /* |
3677 | * One school of thought says that we should not add | |
3678 | * back the node to the tree if reclaim returns 0. | |
3679 | * But our reclaim could return 0, simply because due | |
3680 | * to priority we are exposing a smaller subset of | |
3681 | * memory to reclaim from. Consider this as a longer | |
3682 | * term TODO. | |
3683 | */ | |
ef8745c1 | 3684 | /* If excess == 0, no tree ops */ |
d79154bb | 3685 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 3686 | spin_unlock(&mctz->lock); |
d79154bb | 3687 | css_put(&mz->memcg->css); |
4e416953 BS |
3688 | loop++; |
3689 | /* | |
3690 | * Could not reclaim anything and there are no more | |
3691 | * mem cgroups to try or we seem to be looping without | |
3692 | * reclaiming anything. | |
3693 | */ | |
3694 | if (!nr_reclaimed && | |
3695 | (next_mz == NULL || | |
3696 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3697 | break; | |
3698 | } while (!nr_reclaimed); | |
3699 | if (next_mz) | |
d79154bb | 3700 | css_put(&next_mz->memcg->css); |
4e416953 BS |
3701 | return nr_reclaimed; |
3702 | } | |
3703 | ||
2ef37d3f MH |
3704 | /** |
3705 | * mem_cgroup_force_empty_list - clears LRU of a group | |
3706 | * @memcg: group to clear | |
3707 | * @node: NUMA node | |
3708 | * @zid: zone id | |
3709 | * @lru: lru to to clear | |
3710 | * | |
3c935d18 | 3711 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
3712 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
3713 | * group. | |
cc847582 | 3714 | */ |
2ef37d3f | 3715 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3716 | int node, int zid, enum lru_list lru) |
cc847582 | 3717 | { |
08e552c6 | 3718 | struct mem_cgroup_per_zone *mz; |
2ef37d3f | 3719 | unsigned long flags; |
072c56c1 | 3720 | struct list_head *list; |
925b7673 JW |
3721 | struct page *busy; |
3722 | struct zone *zone; | |
072c56c1 | 3723 | |
08e552c6 | 3724 | zone = &NODE_DATA(node)->node_zones[zid]; |
c0ff4b85 | 3725 | mz = mem_cgroup_zoneinfo(memcg, node, zid); |
6290df54 | 3726 | list = &mz->lruvec.lists[lru]; |
cc847582 | 3727 | |
f817ed48 | 3728 | busy = NULL; |
2ef37d3f | 3729 | do { |
925b7673 | 3730 | struct page_cgroup *pc; |
5564e88b JW |
3731 | struct page *page; |
3732 | ||
08e552c6 | 3733 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3734 | if (list_empty(list)) { |
08e552c6 | 3735 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3736 | break; |
f817ed48 | 3737 | } |
925b7673 JW |
3738 | page = list_entry(list->prev, struct page, lru); |
3739 | if (busy == page) { | |
3740 | list_move(&page->lru, list); | |
648bcc77 | 3741 | busy = NULL; |
08e552c6 | 3742 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3743 | continue; |
3744 | } | |
08e552c6 | 3745 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3746 | |
925b7673 | 3747 | pc = lookup_page_cgroup(page); |
5564e88b | 3748 | |
3c935d18 | 3749 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 3750 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 3751 | busy = page; |
f817ed48 KH |
3752 | cond_resched(); |
3753 | } else | |
3754 | busy = NULL; | |
2ef37d3f | 3755 | } while (!list_empty(list)); |
cc847582 KH |
3756 | } |
3757 | ||
3758 | /* | |
c26251f9 MH |
3759 | * make mem_cgroup's charge to be 0 if there is no task by moving |
3760 | * all the charges and pages to the parent. | |
cc847582 | 3761 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
3762 | * |
3763 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 3764 | */ |
c26251f9 | 3765 | static int mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 3766 | { |
c0ff4b85 | 3767 | struct cgroup *cgrp = memcg->css.cgroup; |
c26251f9 | 3768 | int node, zid; |
8869b8f6 | 3769 | |
fce66477 | 3770 | do { |
c1e862c1 | 3771 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
c26251f9 | 3772 | return -EBUSY; |
52d4b9ac KH |
3773 | /* This is for making all *used* pages to be on LRU. */ |
3774 | lru_add_drain_all(); | |
c0ff4b85 | 3775 | drain_all_stock_sync(memcg); |
c0ff4b85 | 3776 | mem_cgroup_start_move(memcg); |
299b4eaa | 3777 | for_each_node_state(node, N_HIGH_MEMORY) { |
2ef37d3f | 3778 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3779 | enum lru_list lru; |
3780 | for_each_lru(lru) { | |
2ef37d3f | 3781 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3782 | node, zid, lru); |
f817ed48 | 3783 | } |
1ecaab2b | 3784 | } |
f817ed48 | 3785 | } |
c0ff4b85 R |
3786 | mem_cgroup_end_move(memcg); |
3787 | memcg_oom_recover(memcg); | |
52d4b9ac | 3788 | cond_resched(); |
c26251f9 | 3789 | |
2ef37d3f MH |
3790 | /* |
3791 | * This is a safety check because mem_cgroup_force_empty_list | |
3792 | * could have raced with mem_cgroup_replace_page_cache callers | |
3793 | * so the lru seemed empty but the page could have been added | |
3794 | * right after the check. RES_USAGE should be safe as we always | |
3795 | * charge before adding to the LRU. | |
3796 | */ | |
3797 | } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0); | |
3798 | ||
3799 | return 0; | |
c26251f9 MH |
3800 | } |
3801 | ||
3802 | /* | |
3803 | * Reclaims as many pages from the given memcg as possible and moves | |
3804 | * the rest to the parent. | |
3805 | * | |
3806 | * Caller is responsible for holding css reference for memcg. | |
3807 | */ | |
3808 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3809 | { | |
3810 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
3811 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 3812 | |
c1e862c1 | 3813 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
3814 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3815 | return -EBUSY; | |
3816 | ||
c1e862c1 KH |
3817 | /* we call try-to-free pages for make this cgroup empty */ |
3818 | lru_add_drain_all(); | |
f817ed48 | 3819 | /* try to free all pages in this cgroup */ |
569530fb | 3820 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 3821 | int progress; |
c1e862c1 | 3822 | |
c26251f9 MH |
3823 | if (signal_pending(current)) |
3824 | return -EINTR; | |
3825 | ||
c0ff4b85 | 3826 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3827 | false); |
c1e862c1 | 3828 | if (!progress) { |
f817ed48 | 3829 | nr_retries--; |
c1e862c1 | 3830 | /* maybe some writeback is necessary */ |
8aa7e847 | 3831 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3832 | } |
f817ed48 KH |
3833 | |
3834 | } | |
08e552c6 | 3835 | lru_add_drain(); |
c26251f9 | 3836 | return mem_cgroup_reparent_charges(memcg); |
cc847582 KH |
3837 | } |
3838 | ||
6bbda35c | 3839 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 | 3840 | { |
c26251f9 MH |
3841 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
3842 | int ret; | |
3843 | ||
d8423011 MH |
3844 | if (mem_cgroup_is_root(memcg)) |
3845 | return -EINVAL; | |
c26251f9 MH |
3846 | css_get(&memcg->css); |
3847 | ret = mem_cgroup_force_empty(memcg); | |
3848 | css_put(&memcg->css); | |
3849 | ||
3850 | return ret; | |
c1e862c1 KH |
3851 | } |
3852 | ||
3853 | ||
18f59ea7 BS |
3854 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3855 | { | |
3856 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3857 | } | |
3858 | ||
3859 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3860 | u64 val) | |
3861 | { | |
3862 | int retval = 0; | |
c0ff4b85 | 3863 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3864 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3865 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3866 | |
3867 | if (parent) | |
c0ff4b85 | 3868 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3869 | |
3870 | cgroup_lock(); | |
567fb435 GC |
3871 | |
3872 | if (memcg->use_hierarchy == val) | |
3873 | goto out; | |
3874 | ||
18f59ea7 | 3875 | /* |
af901ca1 | 3876 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3877 | * in the child subtrees. If it is unset, then the change can |
3878 | * occur, provided the current cgroup has no children. | |
3879 | * | |
3880 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3881 | * set if there are no children. | |
3882 | */ | |
c0ff4b85 | 3883 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3884 | (val == 1 || val == 0)) { |
3885 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3886 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3887 | else |
3888 | retval = -EBUSY; | |
3889 | } else | |
3890 | retval = -EINVAL; | |
567fb435 GC |
3891 | |
3892 | out: | |
18f59ea7 BS |
3893 | cgroup_unlock(); |
3894 | ||
3895 | return retval; | |
3896 | } | |
3897 | ||
0c3e73e8 | 3898 | |
c0ff4b85 | 3899 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3900 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3901 | { |
7d74b06f | 3902 | struct mem_cgroup *iter; |
7a159cc9 | 3903 | long val = 0; |
0c3e73e8 | 3904 | |
7a159cc9 | 3905 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3906 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3907 | val += mem_cgroup_read_stat(iter, idx); |
3908 | ||
3909 | if (val < 0) /* race ? */ | |
3910 | val = 0; | |
3911 | return val; | |
0c3e73e8 BS |
3912 | } |
3913 | ||
c0ff4b85 | 3914 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3915 | { |
7d74b06f | 3916 | u64 val; |
104f3928 | 3917 | |
c0ff4b85 | 3918 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 3919 | if (!swap) |
65c64ce8 | 3920 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3921 | else |
65c64ce8 | 3922 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
3923 | } |
3924 | ||
c0ff4b85 R |
3925 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3926 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3927 | |
7d74b06f | 3928 | if (swap) |
bff6bb83 | 3929 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
3930 | |
3931 | return val << PAGE_SHIFT; | |
3932 | } | |
3933 | ||
af36f906 TH |
3934 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
3935 | struct file *file, char __user *buf, | |
3936 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 3937 | { |
c0ff4b85 | 3938 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 3939 | char str[64]; |
104f3928 | 3940 | u64 val; |
af36f906 | 3941 | int type, name, len; |
8c7c6e34 KH |
3942 | |
3943 | type = MEMFILE_TYPE(cft->private); | |
3944 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3945 | |
3946 | if (!do_swap_account && type == _MEMSWAP) | |
3947 | return -EOPNOTSUPP; | |
3948 | ||
8c7c6e34 KH |
3949 | switch (type) { |
3950 | case _MEM: | |
104f3928 | 3951 | if (name == RES_USAGE) |
c0ff4b85 | 3952 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3953 | else |
c0ff4b85 | 3954 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3955 | break; |
3956 | case _MEMSWAP: | |
104f3928 | 3957 | if (name == RES_USAGE) |
c0ff4b85 | 3958 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 3959 | else |
c0ff4b85 | 3960 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 KH |
3961 | break; |
3962 | default: | |
3963 | BUG(); | |
8c7c6e34 | 3964 | } |
af36f906 TH |
3965 | |
3966 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
3967 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 3968 | } |
628f4235 KH |
3969 | /* |
3970 | * The user of this function is... | |
3971 | * RES_LIMIT. | |
3972 | */ | |
856c13aa PM |
3973 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3974 | const char *buffer) | |
8cdea7c0 | 3975 | { |
628f4235 | 3976 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3977 | int type, name; |
628f4235 KH |
3978 | unsigned long long val; |
3979 | int ret; | |
3980 | ||
8c7c6e34 KH |
3981 | type = MEMFILE_TYPE(cft->private); |
3982 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3983 | |
3984 | if (!do_swap_account && type == _MEMSWAP) | |
3985 | return -EOPNOTSUPP; | |
3986 | ||
8c7c6e34 | 3987 | switch (name) { |
628f4235 | 3988 | case RES_LIMIT: |
4b3bde4c BS |
3989 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3990 | ret = -EINVAL; | |
3991 | break; | |
3992 | } | |
628f4235 KH |
3993 | /* This function does all necessary parse...reuse it */ |
3994 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3995 | if (ret) |
3996 | break; | |
3997 | if (type == _MEM) | |
628f4235 | 3998 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3999 | else |
4000 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 4001 | break; |
296c81d8 BS |
4002 | case RES_SOFT_LIMIT: |
4003 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
4004 | if (ret) | |
4005 | break; | |
4006 | /* | |
4007 | * For memsw, soft limits are hard to implement in terms | |
4008 | * of semantics, for now, we support soft limits for | |
4009 | * control without swap | |
4010 | */ | |
4011 | if (type == _MEM) | |
4012 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
4013 | else | |
4014 | ret = -EINVAL; | |
4015 | break; | |
628f4235 KH |
4016 | default: |
4017 | ret = -EINVAL; /* should be BUG() ? */ | |
4018 | break; | |
4019 | } | |
4020 | return ret; | |
8cdea7c0 BS |
4021 | } |
4022 | ||
fee7b548 KH |
4023 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
4024 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
4025 | { | |
4026 | struct cgroup *cgroup; | |
4027 | unsigned long long min_limit, min_memsw_limit, tmp; | |
4028 | ||
4029 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4030 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4031 | cgroup = memcg->css.cgroup; | |
4032 | if (!memcg->use_hierarchy) | |
4033 | goto out; | |
4034 | ||
4035 | while (cgroup->parent) { | |
4036 | cgroup = cgroup->parent; | |
4037 | memcg = mem_cgroup_from_cont(cgroup); | |
4038 | if (!memcg->use_hierarchy) | |
4039 | break; | |
4040 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4041 | min_limit = min(min_limit, tmp); | |
4042 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4043 | min_memsw_limit = min(min_memsw_limit, tmp); | |
4044 | } | |
4045 | out: | |
4046 | *mem_limit = min_limit; | |
4047 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
4048 | } |
4049 | ||
29f2a4da | 4050 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 4051 | { |
af36f906 | 4052 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 4053 | int type, name; |
c84872e1 | 4054 | |
8c7c6e34 KH |
4055 | type = MEMFILE_TYPE(event); |
4056 | name = MEMFILE_ATTR(event); | |
af36f906 TH |
4057 | |
4058 | if (!do_swap_account && type == _MEMSWAP) | |
4059 | return -EOPNOTSUPP; | |
4060 | ||
8c7c6e34 | 4061 | switch (name) { |
29f2a4da | 4062 | case RES_MAX_USAGE: |
8c7c6e34 | 4063 | if (type == _MEM) |
c0ff4b85 | 4064 | res_counter_reset_max(&memcg->res); |
8c7c6e34 | 4065 | else |
c0ff4b85 | 4066 | res_counter_reset_max(&memcg->memsw); |
29f2a4da PE |
4067 | break; |
4068 | case RES_FAILCNT: | |
8c7c6e34 | 4069 | if (type == _MEM) |
c0ff4b85 | 4070 | res_counter_reset_failcnt(&memcg->res); |
8c7c6e34 | 4071 | else |
c0ff4b85 | 4072 | res_counter_reset_failcnt(&memcg->memsw); |
29f2a4da PE |
4073 | break; |
4074 | } | |
f64c3f54 | 4075 | |
85cc59db | 4076 | return 0; |
c84872e1 PE |
4077 | } |
4078 | ||
7dc74be0 DN |
4079 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
4080 | struct cftype *cft) | |
4081 | { | |
4082 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
4083 | } | |
4084 | ||
02491447 | 4085 | #ifdef CONFIG_MMU |
7dc74be0 DN |
4086 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
4087 | struct cftype *cft, u64 val) | |
4088 | { | |
c0ff4b85 | 4089 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
4090 | |
4091 | if (val >= (1 << NR_MOVE_TYPE)) | |
4092 | return -EINVAL; | |
4093 | /* | |
4094 | * We check this value several times in both in can_attach() and | |
4095 | * attach(), so we need cgroup lock to prevent this value from being | |
4096 | * inconsistent. | |
4097 | */ | |
4098 | cgroup_lock(); | |
c0ff4b85 | 4099 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4100 | cgroup_unlock(); |
4101 | ||
4102 | return 0; | |
4103 | } | |
02491447 DN |
4104 | #else |
4105 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4106 | struct cftype *cft, u64 val) | |
4107 | { | |
4108 | return -ENOSYS; | |
4109 | } | |
4110 | #endif | |
7dc74be0 | 4111 | |
406eb0c9 | 4112 | #ifdef CONFIG_NUMA |
ab215884 | 4113 | static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft, |
fada52ca | 4114 | struct seq_file *m) |
406eb0c9 YH |
4115 | { |
4116 | int nid; | |
4117 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4118 | unsigned long node_nr; | |
d79154bb | 4119 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 4120 | |
d79154bb | 4121 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 YH |
4122 | seq_printf(m, "total=%lu", total_nr); |
4123 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4124 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
4125 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4126 | } | |
4127 | seq_putc(m, '\n'); | |
4128 | ||
d79154bb | 4129 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 YH |
4130 | seq_printf(m, "file=%lu", file_nr); |
4131 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4132 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4133 | LRU_ALL_FILE); |
406eb0c9 YH |
4134 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4135 | } | |
4136 | seq_putc(m, '\n'); | |
4137 | ||
d79154bb | 4138 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 YH |
4139 | seq_printf(m, "anon=%lu", anon_nr); |
4140 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4141 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4142 | LRU_ALL_ANON); |
406eb0c9 YH |
4143 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4144 | } | |
4145 | seq_putc(m, '\n'); | |
4146 | ||
d79154bb | 4147 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4148 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4149 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
d79154bb | 4150 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4151 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4152 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4153 | } | |
4154 | seq_putc(m, '\n'); | |
4155 | return 0; | |
4156 | } | |
4157 | #endif /* CONFIG_NUMA */ | |
4158 | ||
af7c4b0e JW |
4159 | static const char * const mem_cgroup_lru_names[] = { |
4160 | "inactive_anon", | |
4161 | "active_anon", | |
4162 | "inactive_file", | |
4163 | "active_file", | |
4164 | "unevictable", | |
4165 | }; | |
4166 | ||
4167 | static inline void mem_cgroup_lru_names_not_uptodate(void) | |
4168 | { | |
4169 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4170 | } | |
4171 | ||
ab215884 | 4172 | static int memcg_stat_show(struct cgroup *cont, struct cftype *cft, |
78ccf5b5 | 4173 | struct seq_file *m) |
d2ceb9b7 | 4174 | { |
d79154bb | 4175 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af7c4b0e JW |
4176 | struct mem_cgroup *mi; |
4177 | unsigned int i; | |
406eb0c9 | 4178 | |
af7c4b0e | 4179 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4180 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4181 | continue; |
af7c4b0e JW |
4182 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4183 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4184 | } |
7b854121 | 4185 | |
af7c4b0e JW |
4186 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4187 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4188 | mem_cgroup_read_events(memcg, i)); | |
4189 | ||
4190 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4191 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4192 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4193 | ||
14067bb3 | 4194 | /* Hierarchical information */ |
fee7b548 KH |
4195 | { |
4196 | unsigned long long limit, memsw_limit; | |
d79154bb | 4197 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 4198 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 4199 | if (do_swap_account) |
78ccf5b5 JW |
4200 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4201 | memsw_limit); | |
fee7b548 | 4202 | } |
7f016ee8 | 4203 | |
af7c4b0e JW |
4204 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4205 | long long val = 0; | |
4206 | ||
bff6bb83 | 4207 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4208 | continue; |
af7c4b0e JW |
4209 | for_each_mem_cgroup_tree(mi, memcg) |
4210 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4211 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4212 | } | |
4213 | ||
4214 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4215 | unsigned long long val = 0; | |
4216 | ||
4217 | for_each_mem_cgroup_tree(mi, memcg) | |
4218 | val += mem_cgroup_read_events(mi, i); | |
4219 | seq_printf(m, "total_%s %llu\n", | |
4220 | mem_cgroup_events_names[i], val); | |
4221 | } | |
4222 | ||
4223 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4224 | unsigned long long val = 0; | |
4225 | ||
4226 | for_each_mem_cgroup_tree(mi, memcg) | |
4227 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4228 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4229 | } |
14067bb3 | 4230 | |
7f016ee8 | 4231 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4232 | { |
4233 | int nid, zid; | |
4234 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4235 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4236 | unsigned long recent_rotated[2] = {0, 0}; |
4237 | unsigned long recent_scanned[2] = {0, 0}; | |
4238 | ||
4239 | for_each_online_node(nid) | |
4240 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 4241 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 4242 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4243 | |
89abfab1 HD |
4244 | recent_rotated[0] += rstat->recent_rotated[0]; |
4245 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4246 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4247 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4248 | } |
78ccf5b5 JW |
4249 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4250 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4251 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4252 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4253 | } |
4254 | #endif | |
4255 | ||
d2ceb9b7 KH |
4256 | return 0; |
4257 | } | |
4258 | ||
a7885eb8 KM |
4259 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4260 | { | |
4261 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4262 | ||
1f4c025b | 4263 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4264 | } |
4265 | ||
4266 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4267 | u64 val) | |
4268 | { | |
4269 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4270 | struct mem_cgroup *parent; | |
068b38c1 | 4271 | |
a7885eb8 KM |
4272 | if (val > 100) |
4273 | return -EINVAL; | |
4274 | ||
4275 | if (cgrp->parent == NULL) | |
4276 | return -EINVAL; | |
4277 | ||
4278 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4279 | |
4280 | cgroup_lock(); | |
4281 | ||
a7885eb8 KM |
4282 | /* If under hierarchy, only empty-root can set this value */ |
4283 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4284 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4285 | cgroup_unlock(); | |
a7885eb8 | 4286 | return -EINVAL; |
068b38c1 | 4287 | } |
a7885eb8 | 4288 | |
a7885eb8 | 4289 | memcg->swappiness = val; |
a7885eb8 | 4290 | |
068b38c1 LZ |
4291 | cgroup_unlock(); |
4292 | ||
a7885eb8 KM |
4293 | return 0; |
4294 | } | |
4295 | ||
2e72b634 KS |
4296 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4297 | { | |
4298 | struct mem_cgroup_threshold_ary *t; | |
4299 | u64 usage; | |
4300 | int i; | |
4301 | ||
4302 | rcu_read_lock(); | |
4303 | if (!swap) | |
2c488db2 | 4304 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4305 | else |
2c488db2 | 4306 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4307 | |
4308 | if (!t) | |
4309 | goto unlock; | |
4310 | ||
4311 | usage = mem_cgroup_usage(memcg, swap); | |
4312 | ||
4313 | /* | |
748dad36 | 4314 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4315 | * If it's not true, a threshold was crossed after last |
4316 | * call of __mem_cgroup_threshold(). | |
4317 | */ | |
5407a562 | 4318 | i = t->current_threshold; |
2e72b634 KS |
4319 | |
4320 | /* | |
4321 | * Iterate backward over array of thresholds starting from | |
4322 | * current_threshold and check if a threshold is crossed. | |
4323 | * If none of thresholds below usage is crossed, we read | |
4324 | * only one element of the array here. | |
4325 | */ | |
4326 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4327 | eventfd_signal(t->entries[i].eventfd, 1); | |
4328 | ||
4329 | /* i = current_threshold + 1 */ | |
4330 | i++; | |
4331 | ||
4332 | /* | |
4333 | * Iterate forward over array of thresholds starting from | |
4334 | * current_threshold+1 and check if a threshold is crossed. | |
4335 | * If none of thresholds above usage is crossed, we read | |
4336 | * only one element of the array here. | |
4337 | */ | |
4338 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4339 | eventfd_signal(t->entries[i].eventfd, 1); | |
4340 | ||
4341 | /* Update current_threshold */ | |
5407a562 | 4342 | t->current_threshold = i - 1; |
2e72b634 KS |
4343 | unlock: |
4344 | rcu_read_unlock(); | |
4345 | } | |
4346 | ||
4347 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4348 | { | |
ad4ca5f4 KS |
4349 | while (memcg) { |
4350 | __mem_cgroup_threshold(memcg, false); | |
4351 | if (do_swap_account) | |
4352 | __mem_cgroup_threshold(memcg, true); | |
4353 | ||
4354 | memcg = parent_mem_cgroup(memcg); | |
4355 | } | |
2e72b634 KS |
4356 | } |
4357 | ||
4358 | static int compare_thresholds(const void *a, const void *b) | |
4359 | { | |
4360 | const struct mem_cgroup_threshold *_a = a; | |
4361 | const struct mem_cgroup_threshold *_b = b; | |
4362 | ||
4363 | return _a->threshold - _b->threshold; | |
4364 | } | |
4365 | ||
c0ff4b85 | 4366 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4367 | { |
4368 | struct mem_cgroup_eventfd_list *ev; | |
4369 | ||
c0ff4b85 | 4370 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4371 | eventfd_signal(ev->eventfd, 1); |
4372 | return 0; | |
4373 | } | |
4374 | ||
c0ff4b85 | 4375 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4376 | { |
7d74b06f KH |
4377 | struct mem_cgroup *iter; |
4378 | ||
c0ff4b85 | 4379 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4380 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4381 | } |
4382 | ||
4383 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4384 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4385 | { |
4386 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4387 | struct mem_cgroup_thresholds *thresholds; |
4388 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4389 | int type = MEMFILE_TYPE(cft->private); |
4390 | u64 threshold, usage; | |
2c488db2 | 4391 | int i, size, ret; |
2e72b634 KS |
4392 | |
4393 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4394 | if (ret) | |
4395 | return ret; | |
4396 | ||
4397 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4398 | |
2e72b634 | 4399 | if (type == _MEM) |
2c488db2 | 4400 | thresholds = &memcg->thresholds; |
2e72b634 | 4401 | else if (type == _MEMSWAP) |
2c488db2 | 4402 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4403 | else |
4404 | BUG(); | |
4405 | ||
4406 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4407 | ||
4408 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4409 | if (thresholds->primary) |
2e72b634 KS |
4410 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4411 | ||
2c488db2 | 4412 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4413 | |
4414 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4415 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4416 | GFP_KERNEL); |
2c488db2 | 4417 | if (!new) { |
2e72b634 KS |
4418 | ret = -ENOMEM; |
4419 | goto unlock; | |
4420 | } | |
2c488db2 | 4421 | new->size = size; |
2e72b634 KS |
4422 | |
4423 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4424 | if (thresholds->primary) { |
4425 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4426 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4427 | } |
4428 | ||
2e72b634 | 4429 | /* Add new threshold */ |
2c488db2 KS |
4430 | new->entries[size - 1].eventfd = eventfd; |
4431 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4432 | |
4433 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4434 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4435 | compare_thresholds, NULL); |
4436 | ||
4437 | /* Find current threshold */ | |
2c488db2 | 4438 | new->current_threshold = -1; |
2e72b634 | 4439 | for (i = 0; i < size; i++) { |
748dad36 | 4440 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4441 | /* |
2c488db2 KS |
4442 | * new->current_threshold will not be used until |
4443 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4444 | * it here. |
4445 | */ | |
2c488db2 | 4446 | ++new->current_threshold; |
748dad36 SZ |
4447 | } else |
4448 | break; | |
2e72b634 KS |
4449 | } |
4450 | ||
2c488db2 KS |
4451 | /* Free old spare buffer and save old primary buffer as spare */ |
4452 | kfree(thresholds->spare); | |
4453 | thresholds->spare = thresholds->primary; | |
4454 | ||
4455 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4456 | |
907860ed | 4457 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4458 | synchronize_rcu(); |
4459 | ||
2e72b634 KS |
4460 | unlock: |
4461 | mutex_unlock(&memcg->thresholds_lock); | |
4462 | ||
4463 | return ret; | |
4464 | } | |
4465 | ||
907860ed | 4466 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4467 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4468 | { |
4469 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4470 | struct mem_cgroup_thresholds *thresholds; |
4471 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4472 | int type = MEMFILE_TYPE(cft->private); |
4473 | u64 usage; | |
2c488db2 | 4474 | int i, j, size; |
2e72b634 KS |
4475 | |
4476 | mutex_lock(&memcg->thresholds_lock); | |
4477 | if (type == _MEM) | |
2c488db2 | 4478 | thresholds = &memcg->thresholds; |
2e72b634 | 4479 | else if (type == _MEMSWAP) |
2c488db2 | 4480 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4481 | else |
4482 | BUG(); | |
4483 | ||
371528ca AV |
4484 | if (!thresholds->primary) |
4485 | goto unlock; | |
4486 | ||
2e72b634 KS |
4487 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
4488 | ||
4489 | /* Check if a threshold crossed before removing */ | |
4490 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4491 | ||
4492 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4493 | size = 0; |
4494 | for (i = 0; i < thresholds->primary->size; i++) { | |
4495 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4496 | size++; |
4497 | } | |
4498 | ||
2c488db2 | 4499 | new = thresholds->spare; |
907860ed | 4500 | |
2e72b634 KS |
4501 | /* Set thresholds array to NULL if we don't have thresholds */ |
4502 | if (!size) { | |
2c488db2 KS |
4503 | kfree(new); |
4504 | new = NULL; | |
907860ed | 4505 | goto swap_buffers; |
2e72b634 KS |
4506 | } |
4507 | ||
2c488db2 | 4508 | new->size = size; |
2e72b634 KS |
4509 | |
4510 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4511 | new->current_threshold = -1; |
4512 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4513 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4514 | continue; |
4515 | ||
2c488db2 | 4516 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4517 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4518 | /* |
2c488db2 | 4519 | * new->current_threshold will not be used |
2e72b634 KS |
4520 | * until rcu_assign_pointer(), so it's safe to increment |
4521 | * it here. | |
4522 | */ | |
2c488db2 | 4523 | ++new->current_threshold; |
2e72b634 KS |
4524 | } |
4525 | j++; | |
4526 | } | |
4527 | ||
907860ed | 4528 | swap_buffers: |
2c488db2 KS |
4529 | /* Swap primary and spare array */ |
4530 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4531 | /* If all events are unregistered, free the spare array */ |
4532 | if (!new) { | |
4533 | kfree(thresholds->spare); | |
4534 | thresholds->spare = NULL; | |
4535 | } | |
4536 | ||
2c488db2 | 4537 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4538 | |
907860ed | 4539 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4540 | synchronize_rcu(); |
371528ca | 4541 | unlock: |
2e72b634 | 4542 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4543 | } |
c1e862c1 | 4544 | |
9490ff27 KH |
4545 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4546 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4547 | { | |
4548 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4549 | struct mem_cgroup_eventfd_list *event; | |
4550 | int type = MEMFILE_TYPE(cft->private); | |
4551 | ||
4552 | BUG_ON(type != _OOM_TYPE); | |
4553 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4554 | if (!event) | |
4555 | return -ENOMEM; | |
4556 | ||
1af8efe9 | 4557 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4558 | |
4559 | event->eventfd = eventfd; | |
4560 | list_add(&event->list, &memcg->oom_notify); | |
4561 | ||
4562 | /* already in OOM ? */ | |
79dfdacc | 4563 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4564 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4565 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4566 | |
4567 | return 0; | |
4568 | } | |
4569 | ||
907860ed | 4570 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4571 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4572 | { | |
c0ff4b85 | 4573 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 KH |
4574 | struct mem_cgroup_eventfd_list *ev, *tmp; |
4575 | int type = MEMFILE_TYPE(cft->private); | |
4576 | ||
4577 | BUG_ON(type != _OOM_TYPE); | |
4578 | ||
1af8efe9 | 4579 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4580 | |
c0ff4b85 | 4581 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4582 | if (ev->eventfd == eventfd) { |
4583 | list_del(&ev->list); | |
4584 | kfree(ev); | |
4585 | } | |
4586 | } | |
4587 | ||
1af8efe9 | 4588 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4589 | } |
4590 | ||
3c11ecf4 KH |
4591 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4592 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4593 | { | |
c0ff4b85 | 4594 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4595 | |
c0ff4b85 | 4596 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4597 | |
c0ff4b85 | 4598 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4599 | cb->fill(cb, "under_oom", 1); |
4600 | else | |
4601 | cb->fill(cb, "under_oom", 0); | |
4602 | return 0; | |
4603 | } | |
4604 | ||
3c11ecf4 KH |
4605 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4606 | struct cftype *cft, u64 val) | |
4607 | { | |
c0ff4b85 | 4608 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4609 | struct mem_cgroup *parent; |
4610 | ||
4611 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4612 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4613 | return -EINVAL; | |
4614 | ||
4615 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4616 | ||
4617 | cgroup_lock(); | |
4618 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4619 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4620 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4621 | cgroup_unlock(); |
4622 | return -EINVAL; | |
4623 | } | |
c0ff4b85 | 4624 | memcg->oom_kill_disable = val; |
4d845ebf | 4625 | if (!val) |
c0ff4b85 | 4626 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4627 | cgroup_unlock(); |
4628 | return 0; | |
4629 | } | |
4630 | ||
c255a458 | 4631 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4632 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4633 | { |
1d62e436 | 4634 | return mem_cgroup_sockets_init(memcg, ss); |
e5671dfa GC |
4635 | }; |
4636 | ||
1d62e436 | 4637 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 | 4638 | { |
1d62e436 | 4639 | mem_cgroup_sockets_destroy(memcg); |
d1a4c0b3 | 4640 | } |
e5671dfa | 4641 | #else |
cbe128e3 | 4642 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4643 | { |
4644 | return 0; | |
4645 | } | |
d1a4c0b3 | 4646 | |
1d62e436 | 4647 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 GC |
4648 | { |
4649 | } | |
e5671dfa GC |
4650 | #endif |
4651 | ||
8cdea7c0 BS |
4652 | static struct cftype mem_cgroup_files[] = { |
4653 | { | |
0eea1030 | 4654 | .name = "usage_in_bytes", |
8c7c6e34 | 4655 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 4656 | .read = mem_cgroup_read, |
9490ff27 KH |
4657 | .register_event = mem_cgroup_usage_register_event, |
4658 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4659 | }, |
c84872e1 PE |
4660 | { |
4661 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4662 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4663 | .trigger = mem_cgroup_reset, |
af36f906 | 4664 | .read = mem_cgroup_read, |
c84872e1 | 4665 | }, |
8cdea7c0 | 4666 | { |
0eea1030 | 4667 | .name = "limit_in_bytes", |
8c7c6e34 | 4668 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4669 | .write_string = mem_cgroup_write, |
af36f906 | 4670 | .read = mem_cgroup_read, |
8cdea7c0 | 4671 | }, |
296c81d8 BS |
4672 | { |
4673 | .name = "soft_limit_in_bytes", | |
4674 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4675 | .write_string = mem_cgroup_write, | |
af36f906 | 4676 | .read = mem_cgroup_read, |
296c81d8 | 4677 | }, |
8cdea7c0 BS |
4678 | { |
4679 | .name = "failcnt", | |
8c7c6e34 | 4680 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4681 | .trigger = mem_cgroup_reset, |
af36f906 | 4682 | .read = mem_cgroup_read, |
8cdea7c0 | 4683 | }, |
d2ceb9b7 KH |
4684 | { |
4685 | .name = "stat", | |
ab215884 | 4686 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 4687 | }, |
c1e862c1 KH |
4688 | { |
4689 | .name = "force_empty", | |
4690 | .trigger = mem_cgroup_force_empty_write, | |
4691 | }, | |
18f59ea7 BS |
4692 | { |
4693 | .name = "use_hierarchy", | |
4694 | .write_u64 = mem_cgroup_hierarchy_write, | |
4695 | .read_u64 = mem_cgroup_hierarchy_read, | |
4696 | }, | |
a7885eb8 KM |
4697 | { |
4698 | .name = "swappiness", | |
4699 | .read_u64 = mem_cgroup_swappiness_read, | |
4700 | .write_u64 = mem_cgroup_swappiness_write, | |
4701 | }, | |
7dc74be0 DN |
4702 | { |
4703 | .name = "move_charge_at_immigrate", | |
4704 | .read_u64 = mem_cgroup_move_charge_read, | |
4705 | .write_u64 = mem_cgroup_move_charge_write, | |
4706 | }, | |
9490ff27 KH |
4707 | { |
4708 | .name = "oom_control", | |
3c11ecf4 KH |
4709 | .read_map = mem_cgroup_oom_control_read, |
4710 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4711 | .register_event = mem_cgroup_oom_register_event, |
4712 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4713 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4714 | }, | |
406eb0c9 YH |
4715 | #ifdef CONFIG_NUMA |
4716 | { | |
4717 | .name = "numa_stat", | |
ab215884 | 4718 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
4719 | }, |
4720 | #endif | |
c255a458 | 4721 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4722 | { |
4723 | .name = "memsw.usage_in_bytes", | |
4724 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
af36f906 | 4725 | .read = mem_cgroup_read, |
9490ff27 KH |
4726 | .register_event = mem_cgroup_usage_register_event, |
4727 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4728 | }, |
4729 | { | |
4730 | .name = "memsw.max_usage_in_bytes", | |
4731 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4732 | .trigger = mem_cgroup_reset, | |
af36f906 | 4733 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4734 | }, |
4735 | { | |
4736 | .name = "memsw.limit_in_bytes", | |
4737 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4738 | .write_string = mem_cgroup_write, | |
af36f906 | 4739 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4740 | }, |
4741 | { | |
4742 | .name = "memsw.failcnt", | |
4743 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4744 | .trigger = mem_cgroup_reset, | |
af36f906 | 4745 | .read = mem_cgroup_read, |
8c7c6e34 | 4746 | }, |
8c7c6e34 | 4747 | #endif |
6bc10349 | 4748 | { }, /* terminate */ |
af36f906 | 4749 | }; |
8c7c6e34 | 4750 | |
c0ff4b85 | 4751 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4752 | { |
4753 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4754 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4755 | int zone, tmp = node; |
1ecaab2b KH |
4756 | /* |
4757 | * This routine is called against possible nodes. | |
4758 | * But it's BUG to call kmalloc() against offline node. | |
4759 | * | |
4760 | * TODO: this routine can waste much memory for nodes which will | |
4761 | * never be onlined. It's better to use memory hotplug callback | |
4762 | * function. | |
4763 | */ | |
41e3355d KH |
4764 | if (!node_state(node, N_NORMAL_MEMORY)) |
4765 | tmp = -1; | |
17295c88 | 4766 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4767 | if (!pn) |
4768 | return 1; | |
1ecaab2b | 4769 | |
1ecaab2b KH |
4770 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4771 | mz = &pn->zoneinfo[zone]; | |
7f5e86c2 | 4772 | lruvec_init(&mz->lruvec, &NODE_DATA(node)->node_zones[zone]); |
f64c3f54 | 4773 | mz->usage_in_excess = 0; |
4e416953 | 4774 | mz->on_tree = false; |
d79154bb | 4775 | mz->memcg = memcg; |
1ecaab2b | 4776 | } |
0a619e58 | 4777 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4778 | return 0; |
4779 | } | |
4780 | ||
c0ff4b85 | 4781 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4782 | { |
c0ff4b85 | 4783 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4784 | } |
4785 | ||
33327948 KH |
4786 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4787 | { | |
d79154bb | 4788 | struct mem_cgroup *memcg; |
c62b1a3b | 4789 | int size = sizeof(struct mem_cgroup); |
33327948 | 4790 | |
c62b1a3b | 4791 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4792 | if (size < PAGE_SIZE) |
d79154bb | 4793 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 4794 | else |
d79154bb | 4795 | memcg = vzalloc(size); |
33327948 | 4796 | |
d79154bb | 4797 | if (!memcg) |
e7bbcdf3 DC |
4798 | return NULL; |
4799 | ||
d79154bb HD |
4800 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4801 | if (!memcg->stat) | |
d2e61b8d | 4802 | goto out_free; |
d79154bb HD |
4803 | spin_lock_init(&memcg->pcp_counter_lock); |
4804 | return memcg; | |
d2e61b8d DC |
4805 | |
4806 | out_free: | |
4807 | if (size < PAGE_SIZE) | |
d79154bb | 4808 | kfree(memcg); |
d2e61b8d | 4809 | else |
d79154bb | 4810 | vfree(memcg); |
d2e61b8d | 4811 | return NULL; |
33327948 KH |
4812 | } |
4813 | ||
59927fb9 | 4814 | /* |
3afe36b1 | 4815 | * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU, |
59927fb9 HD |
4816 | * but in process context. The work_freeing structure is overlaid |
4817 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
4818 | */ | |
3afe36b1 | 4819 | static void free_work(struct work_struct *work) |
59927fb9 HD |
4820 | { |
4821 | struct mem_cgroup *memcg; | |
3afe36b1 | 4822 | int size = sizeof(struct mem_cgroup); |
59927fb9 HD |
4823 | |
4824 | memcg = container_of(work, struct mem_cgroup, work_freeing); | |
3f134619 GC |
4825 | /* |
4826 | * We need to make sure that (at least for now), the jump label | |
4827 | * destruction code runs outside of the cgroup lock. This is because | |
4828 | * get_online_cpus(), which is called from the static_branch update, | |
4829 | * can't be called inside the cgroup_lock. cpusets are the ones | |
4830 | * enforcing this dependency, so if they ever change, we might as well. | |
4831 | * | |
4832 | * schedule_work() will guarantee this happens. Be careful if you need | |
4833 | * to move this code around, and make sure it is outside | |
4834 | * the cgroup_lock. | |
4835 | */ | |
4836 | disarm_sock_keys(memcg); | |
3afe36b1 GC |
4837 | if (size < PAGE_SIZE) |
4838 | kfree(memcg); | |
4839 | else | |
4840 | vfree(memcg); | |
59927fb9 | 4841 | } |
3afe36b1 GC |
4842 | |
4843 | static void free_rcu(struct rcu_head *rcu_head) | |
59927fb9 HD |
4844 | { |
4845 | struct mem_cgroup *memcg; | |
4846 | ||
4847 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); | |
3afe36b1 | 4848 | INIT_WORK(&memcg->work_freeing, free_work); |
59927fb9 HD |
4849 | schedule_work(&memcg->work_freeing); |
4850 | } | |
4851 | ||
8c7c6e34 KH |
4852 | /* |
4853 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4854 | * (scanning all at force_empty is too costly...) | |
4855 | * | |
4856 | * Instead of clearing all references at force_empty, we remember | |
4857 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4858 | * it goes down to 0. | |
4859 | * | |
8c7c6e34 KH |
4860 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4861 | */ | |
4862 | ||
c0ff4b85 | 4863 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 4864 | { |
08e552c6 KH |
4865 | int node; |
4866 | ||
c0ff4b85 R |
4867 | mem_cgroup_remove_from_trees(memcg); |
4868 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 4869 | |
3ed28fa1 | 4870 | for_each_node(node) |
c0ff4b85 | 4871 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 4872 | |
c0ff4b85 | 4873 | free_percpu(memcg->stat); |
3afe36b1 | 4874 | call_rcu(&memcg->rcu_freeing, free_rcu); |
33327948 KH |
4875 | } |
4876 | ||
c0ff4b85 | 4877 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 4878 | { |
c0ff4b85 | 4879 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
4880 | } |
4881 | ||
c0ff4b85 | 4882 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 4883 | { |
c0ff4b85 R |
4884 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
4885 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4886 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
4887 | if (parent) |
4888 | mem_cgroup_put(parent); | |
4889 | } | |
8c7c6e34 KH |
4890 | } |
4891 | ||
c0ff4b85 | 4892 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 4893 | { |
c0ff4b85 | 4894 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
4895 | } |
4896 | ||
7bcc1bb1 DN |
4897 | /* |
4898 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4899 | */ | |
e1aab161 | 4900 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 4901 | { |
c0ff4b85 | 4902 | if (!memcg->res.parent) |
7bcc1bb1 | 4903 | return NULL; |
c0ff4b85 | 4904 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 4905 | } |
e1aab161 | 4906 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 4907 | |
c255a458 | 4908 | #ifdef CONFIG_MEMCG_SWAP |
c077719b KH |
4909 | static void __init enable_swap_cgroup(void) |
4910 | { | |
f8d66542 | 4911 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4912 | do_swap_account = 1; |
4913 | } | |
4914 | #else | |
4915 | static void __init enable_swap_cgroup(void) | |
4916 | { | |
4917 | } | |
4918 | #endif | |
4919 | ||
f64c3f54 BS |
4920 | static int mem_cgroup_soft_limit_tree_init(void) |
4921 | { | |
4922 | struct mem_cgroup_tree_per_node *rtpn; | |
4923 | struct mem_cgroup_tree_per_zone *rtpz; | |
4924 | int tmp, node, zone; | |
4925 | ||
3ed28fa1 | 4926 | for_each_node(node) { |
f64c3f54 BS |
4927 | tmp = node; |
4928 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4929 | tmp = -1; | |
4930 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4931 | if (!rtpn) | |
c3cecc68 | 4932 | goto err_cleanup; |
f64c3f54 BS |
4933 | |
4934 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4935 | ||
4936 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4937 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4938 | rtpz->rb_root = RB_ROOT; | |
4939 | spin_lock_init(&rtpz->lock); | |
4940 | } | |
4941 | } | |
4942 | return 0; | |
c3cecc68 MH |
4943 | |
4944 | err_cleanup: | |
3ed28fa1 | 4945 | for_each_node(node) { |
c3cecc68 MH |
4946 | if (!soft_limit_tree.rb_tree_per_node[node]) |
4947 | break; | |
4948 | kfree(soft_limit_tree.rb_tree_per_node[node]); | |
4949 | soft_limit_tree.rb_tree_per_node[node] = NULL; | |
4950 | } | |
4951 | return 1; | |
4952 | ||
f64c3f54 BS |
4953 | } |
4954 | ||
0eb253e2 | 4955 | static struct cgroup_subsys_state * __ref |
761b3ef5 | 4956 | mem_cgroup_create(struct cgroup *cont) |
8cdea7c0 | 4957 | { |
c0ff4b85 | 4958 | struct mem_cgroup *memcg, *parent; |
04046e1a | 4959 | long error = -ENOMEM; |
6d12e2d8 | 4960 | int node; |
8cdea7c0 | 4961 | |
c0ff4b85 R |
4962 | memcg = mem_cgroup_alloc(); |
4963 | if (!memcg) | |
04046e1a | 4964 | return ERR_PTR(error); |
78fb7466 | 4965 | |
3ed28fa1 | 4966 | for_each_node(node) |
c0ff4b85 | 4967 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4968 | goto free_out; |
f64c3f54 | 4969 | |
c077719b | 4970 | /* root ? */ |
28dbc4b6 | 4971 | if (cont->parent == NULL) { |
cdec2e42 | 4972 | int cpu; |
c077719b | 4973 | enable_swap_cgroup(); |
28dbc4b6 | 4974 | parent = NULL; |
f64c3f54 BS |
4975 | if (mem_cgroup_soft_limit_tree_init()) |
4976 | goto free_out; | |
a41c58a6 | 4977 | root_mem_cgroup = memcg; |
cdec2e42 KH |
4978 | for_each_possible_cpu(cpu) { |
4979 | struct memcg_stock_pcp *stock = | |
4980 | &per_cpu(memcg_stock, cpu); | |
4981 | INIT_WORK(&stock->work, drain_local_stock); | |
4982 | } | |
711d3d2c | 4983 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4984 | } else { |
28dbc4b6 | 4985 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
4986 | memcg->use_hierarchy = parent->use_hierarchy; |
4987 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 4988 | } |
28dbc4b6 | 4989 | |
18f59ea7 | 4990 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
4991 | res_counter_init(&memcg->res, &parent->res); |
4992 | res_counter_init(&memcg->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4993 | /* |
4994 | * We increment refcnt of the parent to ensure that we can | |
4995 | * safely access it on res_counter_charge/uncharge. | |
4996 | * This refcnt will be decremented when freeing this | |
4997 | * mem_cgroup(see mem_cgroup_put). | |
4998 | */ | |
4999 | mem_cgroup_get(parent); | |
18f59ea7 | 5000 | } else { |
c0ff4b85 R |
5001 | res_counter_init(&memcg->res, NULL); |
5002 | res_counter_init(&memcg->memsw, NULL); | |
18f59ea7 | 5003 | } |
c0ff4b85 R |
5004 | memcg->last_scanned_node = MAX_NUMNODES; |
5005 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 5006 | |
a7885eb8 | 5007 | if (parent) |
c0ff4b85 R |
5008 | memcg->swappiness = mem_cgroup_swappiness(parent); |
5009 | atomic_set(&memcg->refcnt, 1); | |
5010 | memcg->move_charge_at_immigrate = 0; | |
5011 | mutex_init(&memcg->thresholds_lock); | |
312734c0 | 5012 | spin_lock_init(&memcg->move_lock); |
cbe128e3 GC |
5013 | |
5014 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
5015 | if (error) { | |
5016 | /* | |
5017 | * We call put now because our (and parent's) refcnts | |
5018 | * are already in place. mem_cgroup_put() will internally | |
5019 | * call __mem_cgroup_free, so return directly | |
5020 | */ | |
5021 | mem_cgroup_put(memcg); | |
5022 | return ERR_PTR(error); | |
5023 | } | |
c0ff4b85 | 5024 | return &memcg->css; |
6d12e2d8 | 5025 | free_out: |
c0ff4b85 | 5026 | __mem_cgroup_free(memcg); |
04046e1a | 5027 | return ERR_PTR(error); |
8cdea7c0 BS |
5028 | } |
5029 | ||
761b3ef5 | 5030 | static int mem_cgroup_pre_destroy(struct cgroup *cont) |
df878fb0 | 5031 | { |
c0ff4b85 | 5032 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c26251f9 | 5033 | int ret; |
ec64f515 | 5034 | |
c26251f9 MH |
5035 | css_get(&memcg->css); |
5036 | ret = mem_cgroup_reparent_charges(memcg); | |
5037 | css_put(&memcg->css); | |
5038 | ||
5039 | return ret; | |
df878fb0 KH |
5040 | } |
5041 | ||
761b3ef5 | 5042 | static void mem_cgroup_destroy(struct cgroup *cont) |
8cdea7c0 | 5043 | { |
c0ff4b85 | 5044 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 5045 | |
1d62e436 | 5046 | kmem_cgroup_destroy(memcg); |
d1a4c0b3 | 5047 | |
c0ff4b85 | 5048 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
5049 | } |
5050 | ||
02491447 | 5051 | #ifdef CONFIG_MMU |
7dc74be0 | 5052 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5053 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5054 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5055 | { |
854ffa8d DN |
5056 | int ret = 0; |
5057 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5058 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5059 | |
c0ff4b85 | 5060 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5061 | mc.precharge += count; |
5062 | /* we don't need css_get for root */ | |
5063 | return ret; | |
5064 | } | |
5065 | /* try to charge at once */ | |
5066 | if (count > 1) { | |
5067 | struct res_counter *dummy; | |
5068 | /* | |
c0ff4b85 | 5069 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5070 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5071 | * are still under the same cgroup_mutex. So we can postpone | |
5072 | * css_get(). | |
5073 | */ | |
c0ff4b85 | 5074 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5075 | goto one_by_one; |
c0ff4b85 | 5076 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5077 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5078 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5079 | goto one_by_one; |
5080 | } | |
5081 | mc.precharge += count; | |
854ffa8d DN |
5082 | return ret; |
5083 | } | |
5084 | one_by_one: | |
5085 | /* fall back to one by one charge */ | |
5086 | while (count--) { | |
5087 | if (signal_pending(current)) { | |
5088 | ret = -EINTR; | |
5089 | break; | |
5090 | } | |
5091 | if (!batch_count--) { | |
5092 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5093 | cond_resched(); | |
5094 | } | |
c0ff4b85 R |
5095 | ret = __mem_cgroup_try_charge(NULL, |
5096 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 5097 | if (ret) |
854ffa8d | 5098 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 5099 | return ret; |
854ffa8d DN |
5100 | mc.precharge++; |
5101 | } | |
4ffef5fe DN |
5102 | return ret; |
5103 | } | |
5104 | ||
5105 | /** | |
8d32ff84 | 5106 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5107 | * @vma: the vma the pte to be checked belongs |
5108 | * @addr: the address corresponding to the pte to be checked | |
5109 | * @ptent: the pte to be checked | |
02491447 | 5110 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5111 | * |
5112 | * Returns | |
5113 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5114 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5115 | * move charge. if @target is not NULL, the page is stored in target->page | |
5116 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5117 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5118 | * target for charge migration. if @target is not NULL, the entry is stored | |
5119 | * in target->ent. | |
4ffef5fe DN |
5120 | * |
5121 | * Called with pte lock held. | |
5122 | */ | |
4ffef5fe DN |
5123 | union mc_target { |
5124 | struct page *page; | |
02491447 | 5125 | swp_entry_t ent; |
4ffef5fe DN |
5126 | }; |
5127 | ||
4ffef5fe | 5128 | enum mc_target_type { |
8d32ff84 | 5129 | MC_TARGET_NONE = 0, |
4ffef5fe | 5130 | MC_TARGET_PAGE, |
02491447 | 5131 | MC_TARGET_SWAP, |
4ffef5fe DN |
5132 | }; |
5133 | ||
90254a65 DN |
5134 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5135 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5136 | { |
90254a65 | 5137 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5138 | |
90254a65 DN |
5139 | if (!page || !page_mapped(page)) |
5140 | return NULL; | |
5141 | if (PageAnon(page)) { | |
5142 | /* we don't move shared anon */ | |
4b91355e | 5143 | if (!move_anon()) |
90254a65 | 5144 | return NULL; |
87946a72 DN |
5145 | } else if (!move_file()) |
5146 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5147 | return NULL; |
5148 | if (!get_page_unless_zero(page)) | |
5149 | return NULL; | |
5150 | ||
5151 | return page; | |
5152 | } | |
5153 | ||
4b91355e | 5154 | #ifdef CONFIG_SWAP |
90254a65 DN |
5155 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5156 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5157 | { | |
90254a65 DN |
5158 | struct page *page = NULL; |
5159 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5160 | ||
5161 | if (!move_anon() || non_swap_entry(ent)) | |
5162 | return NULL; | |
4b91355e KH |
5163 | /* |
5164 | * Because lookup_swap_cache() updates some statistics counter, | |
5165 | * we call find_get_page() with swapper_space directly. | |
5166 | */ | |
5167 | page = find_get_page(&swapper_space, ent.val); | |
90254a65 DN |
5168 | if (do_swap_account) |
5169 | entry->val = ent.val; | |
5170 | ||
5171 | return page; | |
5172 | } | |
4b91355e KH |
5173 | #else |
5174 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5175 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5176 | { | |
5177 | return NULL; | |
5178 | } | |
5179 | #endif | |
90254a65 | 5180 | |
87946a72 DN |
5181 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5182 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5183 | { | |
5184 | struct page *page = NULL; | |
87946a72 DN |
5185 | struct address_space *mapping; |
5186 | pgoff_t pgoff; | |
5187 | ||
5188 | if (!vma->vm_file) /* anonymous vma */ | |
5189 | return NULL; | |
5190 | if (!move_file()) | |
5191 | return NULL; | |
5192 | ||
87946a72 DN |
5193 | mapping = vma->vm_file->f_mapping; |
5194 | if (pte_none(ptent)) | |
5195 | pgoff = linear_page_index(vma, addr); | |
5196 | else /* pte_file(ptent) is true */ | |
5197 | pgoff = pte_to_pgoff(ptent); | |
5198 | ||
5199 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5200 | page = find_get_page(mapping, pgoff); |
5201 | ||
5202 | #ifdef CONFIG_SWAP | |
5203 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5204 | if (radix_tree_exceptional_entry(page)) { | |
5205 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5206 | if (do_swap_account) |
aa3b1895 HD |
5207 | *entry = swap; |
5208 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5209 | } |
aa3b1895 | 5210 | #endif |
87946a72 DN |
5211 | return page; |
5212 | } | |
5213 | ||
8d32ff84 | 5214 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5215 | unsigned long addr, pte_t ptent, union mc_target *target) |
5216 | { | |
5217 | struct page *page = NULL; | |
5218 | struct page_cgroup *pc; | |
8d32ff84 | 5219 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5220 | swp_entry_t ent = { .val = 0 }; |
5221 | ||
5222 | if (pte_present(ptent)) | |
5223 | page = mc_handle_present_pte(vma, addr, ptent); | |
5224 | else if (is_swap_pte(ptent)) | |
5225 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5226 | else if (pte_none(ptent) || pte_file(ptent)) |
5227 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5228 | |
5229 | if (!page && !ent.val) | |
8d32ff84 | 5230 | return ret; |
02491447 DN |
5231 | if (page) { |
5232 | pc = lookup_page_cgroup(page); | |
5233 | /* | |
5234 | * Do only loose check w/o page_cgroup lock. | |
5235 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5236 | * the lock. | |
5237 | */ | |
5238 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5239 | ret = MC_TARGET_PAGE; | |
5240 | if (target) | |
5241 | target->page = page; | |
5242 | } | |
5243 | if (!ret || !target) | |
5244 | put_page(page); | |
5245 | } | |
90254a65 DN |
5246 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5247 | if (ent.val && !ret && | |
9fb4b7cc | 5248 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5249 | ret = MC_TARGET_SWAP; |
5250 | if (target) | |
5251 | target->ent = ent; | |
4ffef5fe | 5252 | } |
4ffef5fe DN |
5253 | return ret; |
5254 | } | |
5255 | ||
12724850 NH |
5256 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5257 | /* | |
5258 | * We don't consider swapping or file mapped pages because THP does not | |
5259 | * support them for now. | |
5260 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5261 | */ | |
5262 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5263 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5264 | { | |
5265 | struct page *page = NULL; | |
5266 | struct page_cgroup *pc; | |
5267 | enum mc_target_type ret = MC_TARGET_NONE; | |
5268 | ||
5269 | page = pmd_page(pmd); | |
5270 | VM_BUG_ON(!page || !PageHead(page)); | |
5271 | if (!move_anon()) | |
5272 | return ret; | |
5273 | pc = lookup_page_cgroup(page); | |
5274 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5275 | ret = MC_TARGET_PAGE; | |
5276 | if (target) { | |
5277 | get_page(page); | |
5278 | target->page = page; | |
5279 | } | |
5280 | } | |
5281 | return ret; | |
5282 | } | |
5283 | #else | |
5284 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5285 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5286 | { | |
5287 | return MC_TARGET_NONE; | |
5288 | } | |
5289 | #endif | |
5290 | ||
4ffef5fe DN |
5291 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5292 | unsigned long addr, unsigned long end, | |
5293 | struct mm_walk *walk) | |
5294 | { | |
5295 | struct vm_area_struct *vma = walk->private; | |
5296 | pte_t *pte; | |
5297 | spinlock_t *ptl; | |
5298 | ||
12724850 NH |
5299 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
5300 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
5301 | mc.precharge += HPAGE_PMD_NR; | |
5302 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5303 | return 0; |
12724850 | 5304 | } |
03319327 | 5305 | |
45f83cef AA |
5306 | if (pmd_trans_unstable(pmd)) |
5307 | return 0; | |
4ffef5fe DN |
5308 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5309 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5310 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5311 | mc.precharge++; /* increment precharge temporarily */ |
5312 | pte_unmap_unlock(pte - 1, ptl); | |
5313 | cond_resched(); | |
5314 | ||
7dc74be0 DN |
5315 | return 0; |
5316 | } | |
5317 | ||
4ffef5fe DN |
5318 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5319 | { | |
5320 | unsigned long precharge; | |
5321 | struct vm_area_struct *vma; | |
5322 | ||
dfe076b0 | 5323 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5324 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5325 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5326 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5327 | .mm = mm, | |
5328 | .private = vma, | |
5329 | }; | |
5330 | if (is_vm_hugetlb_page(vma)) | |
5331 | continue; | |
4ffef5fe DN |
5332 | walk_page_range(vma->vm_start, vma->vm_end, |
5333 | &mem_cgroup_count_precharge_walk); | |
5334 | } | |
dfe076b0 | 5335 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5336 | |
5337 | precharge = mc.precharge; | |
5338 | mc.precharge = 0; | |
5339 | ||
5340 | return precharge; | |
5341 | } | |
5342 | ||
4ffef5fe DN |
5343 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5344 | { | |
dfe076b0 DN |
5345 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5346 | ||
5347 | VM_BUG_ON(mc.moving_task); | |
5348 | mc.moving_task = current; | |
5349 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5350 | } |
5351 | ||
dfe076b0 DN |
5352 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5353 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5354 | { |
2bd9bb20 KH |
5355 | struct mem_cgroup *from = mc.from; |
5356 | struct mem_cgroup *to = mc.to; | |
5357 | ||
4ffef5fe | 5358 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5359 | if (mc.precharge) { |
5360 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5361 | mc.precharge = 0; | |
5362 | } | |
5363 | /* | |
5364 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5365 | * we must uncharge here. | |
5366 | */ | |
5367 | if (mc.moved_charge) { | |
5368 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5369 | mc.moved_charge = 0; | |
4ffef5fe | 5370 | } |
483c30b5 DN |
5371 | /* we must fixup refcnts and charges */ |
5372 | if (mc.moved_swap) { | |
483c30b5 DN |
5373 | /* uncharge swap account from the old cgroup */ |
5374 | if (!mem_cgroup_is_root(mc.from)) | |
5375 | res_counter_uncharge(&mc.from->memsw, | |
5376 | PAGE_SIZE * mc.moved_swap); | |
5377 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5378 | ||
5379 | if (!mem_cgroup_is_root(mc.to)) { | |
5380 | /* | |
5381 | * we charged both to->res and to->memsw, so we should | |
5382 | * uncharge to->res. | |
5383 | */ | |
5384 | res_counter_uncharge(&mc.to->res, | |
5385 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5386 | } |
5387 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5388 | mc.moved_swap = 0; |
5389 | } | |
dfe076b0 DN |
5390 | memcg_oom_recover(from); |
5391 | memcg_oom_recover(to); | |
5392 | wake_up_all(&mc.waitq); | |
5393 | } | |
5394 | ||
5395 | static void mem_cgroup_clear_mc(void) | |
5396 | { | |
5397 | struct mem_cgroup *from = mc.from; | |
5398 | ||
5399 | /* | |
5400 | * we must clear moving_task before waking up waiters at the end of | |
5401 | * task migration. | |
5402 | */ | |
5403 | mc.moving_task = NULL; | |
5404 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5405 | spin_lock(&mc.lock); |
4ffef5fe DN |
5406 | mc.from = NULL; |
5407 | mc.to = NULL; | |
2bd9bb20 | 5408 | spin_unlock(&mc.lock); |
32047e2a | 5409 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5410 | } |
5411 | ||
761b3ef5 LZ |
5412 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5413 | struct cgroup_taskset *tset) | |
7dc74be0 | 5414 | { |
2f7ee569 | 5415 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5416 | int ret = 0; |
c0ff4b85 | 5417 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5418 | |
c0ff4b85 | 5419 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5420 | struct mm_struct *mm; |
5421 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5422 | ||
c0ff4b85 | 5423 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5424 | |
5425 | mm = get_task_mm(p); | |
5426 | if (!mm) | |
5427 | return 0; | |
7dc74be0 | 5428 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5429 | if (mm->owner == p) { |
5430 | VM_BUG_ON(mc.from); | |
5431 | VM_BUG_ON(mc.to); | |
5432 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5433 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5434 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5435 | mem_cgroup_start_move(from); |
2bd9bb20 | 5436 | spin_lock(&mc.lock); |
4ffef5fe | 5437 | mc.from = from; |
c0ff4b85 | 5438 | mc.to = memcg; |
2bd9bb20 | 5439 | spin_unlock(&mc.lock); |
dfe076b0 | 5440 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5441 | |
5442 | ret = mem_cgroup_precharge_mc(mm); | |
5443 | if (ret) | |
5444 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5445 | } |
5446 | mmput(mm); | |
7dc74be0 DN |
5447 | } |
5448 | return ret; | |
5449 | } | |
5450 | ||
761b3ef5 LZ |
5451 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5452 | struct cgroup_taskset *tset) | |
7dc74be0 | 5453 | { |
4ffef5fe | 5454 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5455 | } |
5456 | ||
4ffef5fe DN |
5457 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5458 | unsigned long addr, unsigned long end, | |
5459 | struct mm_walk *walk) | |
7dc74be0 | 5460 | { |
4ffef5fe DN |
5461 | int ret = 0; |
5462 | struct vm_area_struct *vma = walk->private; | |
5463 | pte_t *pte; | |
5464 | spinlock_t *ptl; | |
12724850 NH |
5465 | enum mc_target_type target_type; |
5466 | union mc_target target; | |
5467 | struct page *page; | |
5468 | struct page_cgroup *pc; | |
4ffef5fe | 5469 | |
12724850 NH |
5470 | /* |
5471 | * We don't take compound_lock() here but no race with splitting thp | |
5472 | * happens because: | |
5473 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5474 | * under splitting, which means there's no concurrent thp split, | |
5475 | * - if another thread runs into split_huge_page() just after we | |
5476 | * entered this if-block, the thread must wait for page table lock | |
5477 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5478 | * part of thp split is not executed yet. | |
5479 | */ | |
5480 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 5481 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
5482 | spin_unlock(&vma->vm_mm->page_table_lock); |
5483 | return 0; | |
5484 | } | |
5485 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5486 | if (target_type == MC_TARGET_PAGE) { | |
5487 | page = target.page; | |
5488 | if (!isolate_lru_page(page)) { | |
5489 | pc = lookup_page_cgroup(page); | |
5490 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5491 | pc, mc.from, mc.to)) { |
12724850 NH |
5492 | mc.precharge -= HPAGE_PMD_NR; |
5493 | mc.moved_charge += HPAGE_PMD_NR; | |
5494 | } | |
5495 | putback_lru_page(page); | |
5496 | } | |
5497 | put_page(page); | |
5498 | } | |
5499 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5500 | return 0; |
12724850 NH |
5501 | } |
5502 | ||
45f83cef AA |
5503 | if (pmd_trans_unstable(pmd)) |
5504 | return 0; | |
4ffef5fe DN |
5505 | retry: |
5506 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5507 | for (; addr != end; addr += PAGE_SIZE) { | |
5508 | pte_t ptent = *(pte++); | |
02491447 | 5509 | swp_entry_t ent; |
4ffef5fe DN |
5510 | |
5511 | if (!mc.precharge) | |
5512 | break; | |
5513 | ||
8d32ff84 | 5514 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5515 | case MC_TARGET_PAGE: |
5516 | page = target.page; | |
5517 | if (isolate_lru_page(page)) | |
5518 | goto put; | |
5519 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5520 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5521 | mc.from, mc.to)) { |
4ffef5fe | 5522 | mc.precharge--; |
854ffa8d DN |
5523 | /* we uncharge from mc.from later. */ |
5524 | mc.moved_charge++; | |
4ffef5fe DN |
5525 | } |
5526 | putback_lru_page(page); | |
8d32ff84 | 5527 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5528 | put_page(page); |
5529 | break; | |
02491447 DN |
5530 | case MC_TARGET_SWAP: |
5531 | ent = target.ent; | |
e91cbb42 | 5532 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5533 | mc.precharge--; |
483c30b5 DN |
5534 | /* we fixup refcnts and charges later. */ |
5535 | mc.moved_swap++; | |
5536 | } | |
02491447 | 5537 | break; |
4ffef5fe DN |
5538 | default: |
5539 | break; | |
5540 | } | |
5541 | } | |
5542 | pte_unmap_unlock(pte - 1, ptl); | |
5543 | cond_resched(); | |
5544 | ||
5545 | if (addr != end) { | |
5546 | /* | |
5547 | * We have consumed all precharges we got in can_attach(). | |
5548 | * We try charge one by one, but don't do any additional | |
5549 | * charges to mc.to if we have failed in charge once in attach() | |
5550 | * phase. | |
5551 | */ | |
854ffa8d | 5552 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5553 | if (!ret) |
5554 | goto retry; | |
5555 | } | |
5556 | ||
5557 | return ret; | |
5558 | } | |
5559 | ||
5560 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5561 | { | |
5562 | struct vm_area_struct *vma; | |
5563 | ||
5564 | lru_add_drain_all(); | |
dfe076b0 DN |
5565 | retry: |
5566 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5567 | /* | |
5568 | * Someone who are holding the mmap_sem might be waiting in | |
5569 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5570 | * and retry. Because we cancel precharges, we might not be able | |
5571 | * to move enough charges, but moving charge is a best-effort | |
5572 | * feature anyway, so it wouldn't be a big problem. | |
5573 | */ | |
5574 | __mem_cgroup_clear_mc(); | |
5575 | cond_resched(); | |
5576 | goto retry; | |
5577 | } | |
4ffef5fe DN |
5578 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5579 | int ret; | |
5580 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5581 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5582 | .mm = mm, | |
5583 | .private = vma, | |
5584 | }; | |
5585 | if (is_vm_hugetlb_page(vma)) | |
5586 | continue; | |
4ffef5fe DN |
5587 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5588 | &mem_cgroup_move_charge_walk); | |
5589 | if (ret) | |
5590 | /* | |
5591 | * means we have consumed all precharges and failed in | |
5592 | * doing additional charge. Just abandon here. | |
5593 | */ | |
5594 | break; | |
5595 | } | |
dfe076b0 | 5596 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5597 | } |
5598 | ||
761b3ef5 LZ |
5599 | static void mem_cgroup_move_task(struct cgroup *cont, |
5600 | struct cgroup_taskset *tset) | |
67e465a7 | 5601 | { |
2f7ee569 | 5602 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5603 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5604 | |
dfe076b0 | 5605 | if (mm) { |
a433658c KM |
5606 | if (mc.to) |
5607 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5608 | mmput(mm); |
5609 | } | |
a433658c KM |
5610 | if (mc.to) |
5611 | mem_cgroup_clear_mc(); | |
67e465a7 | 5612 | } |
5cfb80a7 | 5613 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
5614 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5615 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5616 | { |
5617 | return 0; | |
5618 | } | |
761b3ef5 LZ |
5619 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5620 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5621 | { |
5622 | } | |
761b3ef5 LZ |
5623 | static void mem_cgroup_move_task(struct cgroup *cont, |
5624 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5625 | { |
5626 | } | |
5627 | #endif | |
67e465a7 | 5628 | |
8cdea7c0 BS |
5629 | struct cgroup_subsys mem_cgroup_subsys = { |
5630 | .name = "memory", | |
5631 | .subsys_id = mem_cgroup_subsys_id, | |
5632 | .create = mem_cgroup_create, | |
df878fb0 | 5633 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 | 5634 | .destroy = mem_cgroup_destroy, |
7dc74be0 DN |
5635 | .can_attach = mem_cgroup_can_attach, |
5636 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5637 | .attach = mem_cgroup_move_task, |
6bc10349 | 5638 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 5639 | .early_init = 0, |
04046e1a | 5640 | .use_id = 1, |
8cdea7c0 | 5641 | }; |
c077719b | 5642 | |
c255a458 | 5643 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5644 | static int __init enable_swap_account(char *s) |
5645 | { | |
5646 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5647 | if (!strcmp(s, "1")) |
a42c390c | 5648 | really_do_swap_account = 1; |
a2c8990a | 5649 | else if (!strcmp(s, "0")) |
a42c390c MH |
5650 | really_do_swap_account = 0; |
5651 | return 1; | |
5652 | } | |
a2c8990a | 5653 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5654 | |
c077719b | 5655 | #endif |