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 | 53 | #include <net/sock.h> |
4bd2c1ee | 54 | #include <net/ip.h> |
d1a4c0b3 | 55 | #include <net/tcp_memcontrol.h> |
8cdea7c0 | 56 | |
8697d331 BS |
57 | #include <asm/uaccess.h> |
58 | ||
cc8e970c KM |
59 | #include <trace/events/vmscan.h> |
60 | ||
a181b0e8 | 61 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
68ae564b DR |
62 | EXPORT_SYMBOL(mem_cgroup_subsys); |
63 | ||
a181b0e8 | 64 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 65 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 66 | |
c255a458 | 67 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 68 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 69 | int do_swap_account __read_mostly; |
a42c390c MH |
70 | |
71 | /* for remember boot option*/ | |
c255a458 | 72 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
73 | static int really_do_swap_account __initdata = 1; |
74 | #else | |
75 | static int really_do_swap_account __initdata = 0; | |
76 | #endif | |
77 | ||
c077719b | 78 | #else |
a0db00fc | 79 | #define do_swap_account 0 |
c077719b KH |
80 | #endif |
81 | ||
82 | ||
d52aa412 KH |
83 | /* |
84 | * Statistics for memory cgroup. | |
85 | */ | |
86 | enum mem_cgroup_stat_index { | |
87 | /* | |
88 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
89 | */ | |
90 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 91 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 92 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
bff6bb83 | 93 | MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */ |
d52aa412 KH |
94 | MEM_CGROUP_STAT_NSTATS, |
95 | }; | |
96 | ||
af7c4b0e JW |
97 | static const char * const mem_cgroup_stat_names[] = { |
98 | "cache", | |
99 | "rss", | |
100 | "mapped_file", | |
101 | "swap", | |
102 | }; | |
103 | ||
e9f8974f JW |
104 | enum mem_cgroup_events_index { |
105 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
106 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
107 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
108 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
109 | MEM_CGROUP_EVENTS_NSTATS, |
110 | }; | |
af7c4b0e JW |
111 | |
112 | static const char * const mem_cgroup_events_names[] = { | |
113 | "pgpgin", | |
114 | "pgpgout", | |
115 | "pgfault", | |
116 | "pgmajfault", | |
117 | }; | |
118 | ||
7a159cc9 JW |
119 | /* |
120 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
121 | * it will be incremated by the number of pages. This counter is used for | |
122 | * for trigger some periodic events. This is straightforward and better | |
123 | * than using jiffies etc. to handle periodic memcg event. | |
124 | */ | |
125 | enum mem_cgroup_events_target { | |
126 | MEM_CGROUP_TARGET_THRESH, | |
127 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 128 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
129 | MEM_CGROUP_NTARGETS, |
130 | }; | |
a0db00fc KS |
131 | #define THRESHOLDS_EVENTS_TARGET 128 |
132 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
133 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 134 | |
d52aa412 | 135 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 136 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 137 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 138 | unsigned long nr_page_events; |
7a159cc9 | 139 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
140 | }; |
141 | ||
527a5ec9 JW |
142 | struct mem_cgroup_reclaim_iter { |
143 | /* css_id of the last scanned hierarchy member */ | |
144 | int position; | |
145 | /* scan generation, increased every round-trip */ | |
146 | unsigned int generation; | |
147 | }; | |
148 | ||
6d12e2d8 KH |
149 | /* |
150 | * per-zone information in memory controller. | |
151 | */ | |
6d12e2d8 | 152 | struct mem_cgroup_per_zone { |
6290df54 | 153 | struct lruvec lruvec; |
1eb49272 | 154 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 155 | |
527a5ec9 JW |
156 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
157 | ||
f64c3f54 BS |
158 | struct rb_node tree_node; /* RB tree node */ |
159 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
160 | /* the soft limit is exceeded*/ | |
161 | bool on_tree; | |
d79154bb | 162 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 163 | /* use container_of */ |
6d12e2d8 | 164 | }; |
6d12e2d8 KH |
165 | |
166 | struct mem_cgroup_per_node { | |
167 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
168 | }; | |
169 | ||
170 | struct mem_cgroup_lru_info { | |
171 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
172 | }; | |
173 | ||
f64c3f54 BS |
174 | /* |
175 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
176 | * their hierarchy representation | |
177 | */ | |
178 | ||
179 | struct mem_cgroup_tree_per_zone { | |
180 | struct rb_root rb_root; | |
181 | spinlock_t lock; | |
182 | }; | |
183 | ||
184 | struct mem_cgroup_tree_per_node { | |
185 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
186 | }; | |
187 | ||
188 | struct mem_cgroup_tree { | |
189 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
190 | }; | |
191 | ||
192 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
193 | ||
2e72b634 KS |
194 | struct mem_cgroup_threshold { |
195 | struct eventfd_ctx *eventfd; | |
196 | u64 threshold; | |
197 | }; | |
198 | ||
9490ff27 | 199 | /* For threshold */ |
2e72b634 | 200 | struct mem_cgroup_threshold_ary { |
748dad36 | 201 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 202 | int current_threshold; |
2e72b634 KS |
203 | /* Size of entries[] */ |
204 | unsigned int size; | |
205 | /* Array of thresholds */ | |
206 | struct mem_cgroup_threshold entries[0]; | |
207 | }; | |
2c488db2 KS |
208 | |
209 | struct mem_cgroup_thresholds { | |
210 | /* Primary thresholds array */ | |
211 | struct mem_cgroup_threshold_ary *primary; | |
212 | /* | |
213 | * Spare threshold array. | |
214 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
215 | * It must be able to store at least primary->size - 1 entries. | |
216 | */ | |
217 | struct mem_cgroup_threshold_ary *spare; | |
218 | }; | |
219 | ||
9490ff27 KH |
220 | /* for OOM */ |
221 | struct mem_cgroup_eventfd_list { | |
222 | struct list_head list; | |
223 | struct eventfd_ctx *eventfd; | |
224 | }; | |
2e72b634 | 225 | |
c0ff4b85 R |
226 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
227 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 228 | |
8cdea7c0 BS |
229 | /* |
230 | * The memory controller data structure. The memory controller controls both | |
231 | * page cache and RSS per cgroup. We would eventually like to provide | |
232 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
233 | * to help the administrator determine what knobs to tune. | |
234 | * | |
235 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
236 | * we hit the water mark. May be even add a low water mark, such that |
237 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
238 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
239 | */ |
240 | struct mem_cgroup { | |
241 | struct cgroup_subsys_state css; | |
242 | /* | |
243 | * the counter to account for memory usage | |
244 | */ | |
245 | struct res_counter res; | |
59927fb9 HD |
246 | |
247 | union { | |
248 | /* | |
249 | * the counter to account for mem+swap usage. | |
250 | */ | |
251 | struct res_counter memsw; | |
252 | ||
253 | /* | |
254 | * rcu_freeing is used only when freeing struct mem_cgroup, | |
255 | * so put it into a union to avoid wasting more memory. | |
256 | * It must be disjoint from the css field. It could be | |
257 | * in a union with the res field, but res plays a much | |
258 | * larger part in mem_cgroup life than memsw, and might | |
259 | * be of interest, even at time of free, when debugging. | |
260 | * So share rcu_head with the less interesting memsw. | |
261 | */ | |
262 | struct rcu_head rcu_freeing; | |
263 | /* | |
3afe36b1 GC |
264 | * We also need some space for a worker in deferred freeing. |
265 | * By the time we call it, rcu_freeing is no longer in use. | |
59927fb9 HD |
266 | */ |
267 | struct work_struct work_freeing; | |
268 | }; | |
269 | ||
510fc4e1 GC |
270 | /* |
271 | * the counter to account for kernel memory usage. | |
272 | */ | |
273 | struct res_counter kmem; | |
78fb7466 PE |
274 | /* |
275 | * Per cgroup active and inactive list, similar to the | |
276 | * per zone LRU lists. | |
78fb7466 | 277 | */ |
6d12e2d8 | 278 | struct mem_cgroup_lru_info info; |
889976db YH |
279 | int last_scanned_node; |
280 | #if MAX_NUMNODES > 1 | |
281 | nodemask_t scan_nodes; | |
453a9bf3 KH |
282 | atomic_t numainfo_events; |
283 | atomic_t numainfo_updating; | |
889976db | 284 | #endif |
18f59ea7 BS |
285 | /* |
286 | * Should the accounting and control be hierarchical, per subtree? | |
287 | */ | |
288 | bool use_hierarchy; | |
510fc4e1 | 289 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
290 | |
291 | bool oom_lock; | |
292 | atomic_t under_oom; | |
293 | ||
8c7c6e34 | 294 | atomic_t refcnt; |
14797e23 | 295 | |
1f4c025b | 296 | int swappiness; |
3c11ecf4 KH |
297 | /* OOM-Killer disable */ |
298 | int oom_kill_disable; | |
a7885eb8 | 299 | |
22a668d7 KH |
300 | /* set when res.limit == memsw.limit */ |
301 | bool memsw_is_minimum; | |
302 | ||
2e72b634 KS |
303 | /* protect arrays of thresholds */ |
304 | struct mutex thresholds_lock; | |
305 | ||
306 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 307 | struct mem_cgroup_thresholds thresholds; |
907860ed | 308 | |
2e72b634 | 309 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 310 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 311 | |
9490ff27 KH |
312 | /* For oom notifier event fd */ |
313 | struct list_head oom_notify; | |
185efc0f | 314 | |
7dc74be0 DN |
315 | /* |
316 | * Should we move charges of a task when a task is moved into this | |
317 | * mem_cgroup ? And what type of charges should we move ? | |
318 | */ | |
319 | unsigned long move_charge_at_immigrate; | |
619d094b KH |
320 | /* |
321 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
322 | */ | |
323 | atomic_t moving_account; | |
312734c0 KH |
324 | /* taken only while moving_account > 0 */ |
325 | spinlock_t move_lock; | |
d52aa412 | 326 | /* |
c62b1a3b | 327 | * percpu counter. |
d52aa412 | 328 | */ |
3a7951b4 | 329 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
330 | /* |
331 | * used when a cpu is offlined or other synchronizations | |
332 | * See mem_cgroup_read_stat(). | |
333 | */ | |
334 | struct mem_cgroup_stat_cpu nocpu_base; | |
335 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 336 | |
4bd2c1ee | 337 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
338 | struct tcp_memcontrol tcp_mem; |
339 | #endif | |
8cdea7c0 BS |
340 | }; |
341 | ||
510fc4e1 GC |
342 | /* internal only representation about the status of kmem accounting. */ |
343 | enum { | |
344 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
345 | }; | |
346 | ||
347 | #define KMEM_ACCOUNTED_MASK (1 << KMEM_ACCOUNTED_ACTIVE) | |
348 | ||
349 | #ifdef CONFIG_MEMCG_KMEM | |
350 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
351 | { | |
352 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
353 | } | |
354 | #endif | |
355 | ||
7dc74be0 DN |
356 | /* Stuffs for move charges at task migration. */ |
357 | /* | |
358 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
359 | * left-shifted bitmap of these types. | |
360 | */ | |
361 | enum move_type { | |
4ffef5fe | 362 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 363 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
364 | NR_MOVE_TYPE, |
365 | }; | |
366 | ||
4ffef5fe DN |
367 | /* "mc" and its members are protected by cgroup_mutex */ |
368 | static struct move_charge_struct { | |
b1dd693e | 369 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
370 | struct mem_cgroup *from; |
371 | struct mem_cgroup *to; | |
372 | unsigned long precharge; | |
854ffa8d | 373 | unsigned long moved_charge; |
483c30b5 | 374 | unsigned long moved_swap; |
8033b97c DN |
375 | struct task_struct *moving_task; /* a task moving charges */ |
376 | wait_queue_head_t waitq; /* a waitq for other context */ | |
377 | } mc = { | |
2bd9bb20 | 378 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
379 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
380 | }; | |
4ffef5fe | 381 | |
90254a65 DN |
382 | static bool move_anon(void) |
383 | { | |
384 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
385 | &mc.to->move_charge_at_immigrate); | |
386 | } | |
387 | ||
87946a72 DN |
388 | static bool move_file(void) |
389 | { | |
390 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
391 | &mc.to->move_charge_at_immigrate); | |
392 | } | |
393 | ||
4e416953 BS |
394 | /* |
395 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
396 | * limit reclaim to prevent infinite loops, if they ever occur. | |
397 | */ | |
a0db00fc KS |
398 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
399 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 | |
4e416953 | 400 | |
217bc319 KH |
401 | enum charge_type { |
402 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 403 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 404 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 405 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
406 | NR_CHARGE_TYPE, |
407 | }; | |
408 | ||
8c7c6e34 | 409 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
410 | enum res_type { |
411 | _MEM, | |
412 | _MEMSWAP, | |
413 | _OOM_TYPE, | |
510fc4e1 | 414 | _KMEM, |
86ae53e1 GC |
415 | }; |
416 | ||
a0db00fc KS |
417 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
418 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 419 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
420 | /* Used for OOM nofiier */ |
421 | #define OOM_CONTROL (0) | |
8c7c6e34 | 422 | |
75822b44 BS |
423 | /* |
424 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
425 | */ | |
426 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
427 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
428 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
429 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
430 | ||
c0ff4b85 R |
431 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
432 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 | 433 | |
b2145145 WL |
434 | static inline |
435 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) | |
436 | { | |
437 | return container_of(s, struct mem_cgroup, css); | |
438 | } | |
439 | ||
7ffc0edc MH |
440 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
441 | { | |
442 | return (memcg == root_mem_cgroup); | |
443 | } | |
444 | ||
e1aab161 | 445 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 446 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 447 | |
e1aab161 GC |
448 | void sock_update_memcg(struct sock *sk) |
449 | { | |
376be5ff | 450 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 451 | struct mem_cgroup *memcg; |
3f134619 | 452 | struct cg_proto *cg_proto; |
e1aab161 GC |
453 | |
454 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
455 | ||
f3f511e1 GC |
456 | /* Socket cloning can throw us here with sk_cgrp already |
457 | * filled. It won't however, necessarily happen from | |
458 | * process context. So the test for root memcg given | |
459 | * the current task's memcg won't help us in this case. | |
460 | * | |
461 | * Respecting the original socket's memcg is a better | |
462 | * decision in this case. | |
463 | */ | |
464 | if (sk->sk_cgrp) { | |
465 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
466 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
467 | return; | |
468 | } | |
469 | ||
e1aab161 GC |
470 | rcu_read_lock(); |
471 | memcg = mem_cgroup_from_task(current); | |
3f134619 GC |
472 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
473 | if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) { | |
e1aab161 | 474 | mem_cgroup_get(memcg); |
3f134619 | 475 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
476 | } |
477 | rcu_read_unlock(); | |
478 | } | |
479 | } | |
480 | EXPORT_SYMBOL(sock_update_memcg); | |
481 | ||
482 | void sock_release_memcg(struct sock *sk) | |
483 | { | |
376be5ff | 484 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
485 | struct mem_cgroup *memcg; |
486 | WARN_ON(!sk->sk_cgrp->memcg); | |
487 | memcg = sk->sk_cgrp->memcg; | |
488 | mem_cgroup_put(memcg); | |
489 | } | |
490 | } | |
d1a4c0b3 GC |
491 | |
492 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
493 | { | |
494 | if (!memcg || mem_cgroup_is_root(memcg)) | |
495 | return NULL; | |
496 | ||
497 | return &memcg->tcp_mem.cg_proto; | |
498 | } | |
499 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 500 | |
3f134619 GC |
501 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
502 | { | |
503 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
504 | return; | |
505 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
506 | } | |
507 | #else | |
508 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
509 | { | |
510 | } | |
511 | #endif | |
512 | ||
c0ff4b85 | 513 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 514 | |
f64c3f54 | 515 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 516 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 517 | { |
c0ff4b85 | 518 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
519 | } |
520 | ||
c0ff4b85 | 521 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 522 | { |
c0ff4b85 | 523 | return &memcg->css; |
d324236b WF |
524 | } |
525 | ||
f64c3f54 | 526 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 527 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 528 | { |
97a6c37b JW |
529 | int nid = page_to_nid(page); |
530 | int zid = page_zonenum(page); | |
f64c3f54 | 531 | |
c0ff4b85 | 532 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
533 | } |
534 | ||
535 | static struct mem_cgroup_tree_per_zone * | |
536 | soft_limit_tree_node_zone(int nid, int zid) | |
537 | { | |
538 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
539 | } | |
540 | ||
541 | static struct mem_cgroup_tree_per_zone * | |
542 | soft_limit_tree_from_page(struct page *page) | |
543 | { | |
544 | int nid = page_to_nid(page); | |
545 | int zid = page_zonenum(page); | |
546 | ||
547 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
548 | } | |
549 | ||
550 | static void | |
c0ff4b85 | 551 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 552 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
553 | struct mem_cgroup_tree_per_zone *mctz, |
554 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
555 | { |
556 | struct rb_node **p = &mctz->rb_root.rb_node; | |
557 | struct rb_node *parent = NULL; | |
558 | struct mem_cgroup_per_zone *mz_node; | |
559 | ||
560 | if (mz->on_tree) | |
561 | return; | |
562 | ||
ef8745c1 KH |
563 | mz->usage_in_excess = new_usage_in_excess; |
564 | if (!mz->usage_in_excess) | |
565 | return; | |
f64c3f54 BS |
566 | while (*p) { |
567 | parent = *p; | |
568 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
569 | tree_node); | |
570 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
571 | p = &(*p)->rb_left; | |
572 | /* | |
573 | * We can't avoid mem cgroups that are over their soft | |
574 | * limit by the same amount | |
575 | */ | |
576 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
577 | p = &(*p)->rb_right; | |
578 | } | |
579 | rb_link_node(&mz->tree_node, parent, p); | |
580 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
581 | mz->on_tree = true; | |
4e416953 BS |
582 | } |
583 | ||
584 | static void | |
c0ff4b85 | 585 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
586 | struct mem_cgroup_per_zone *mz, |
587 | struct mem_cgroup_tree_per_zone *mctz) | |
588 | { | |
589 | if (!mz->on_tree) | |
590 | return; | |
591 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
592 | mz->on_tree = false; | |
593 | } | |
594 | ||
f64c3f54 | 595 | static void |
c0ff4b85 | 596 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
597 | struct mem_cgroup_per_zone *mz, |
598 | struct mem_cgroup_tree_per_zone *mctz) | |
599 | { | |
600 | spin_lock(&mctz->lock); | |
c0ff4b85 | 601 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
602 | spin_unlock(&mctz->lock); |
603 | } | |
604 | ||
f64c3f54 | 605 | |
c0ff4b85 | 606 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 607 | { |
ef8745c1 | 608 | unsigned long long excess; |
f64c3f54 BS |
609 | struct mem_cgroup_per_zone *mz; |
610 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
611 | int nid = page_to_nid(page); |
612 | int zid = page_zonenum(page); | |
f64c3f54 BS |
613 | mctz = soft_limit_tree_from_page(page); |
614 | ||
615 | /* | |
4e649152 KH |
616 | * Necessary to update all ancestors when hierarchy is used. |
617 | * because their event counter is not touched. | |
f64c3f54 | 618 | */ |
c0ff4b85 R |
619 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
620 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
621 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
622 | /* |
623 | * We have to update the tree if mz is on RB-tree or | |
624 | * mem is over its softlimit. | |
625 | */ | |
ef8745c1 | 626 | if (excess || mz->on_tree) { |
4e649152 KH |
627 | spin_lock(&mctz->lock); |
628 | /* if on-tree, remove it */ | |
629 | if (mz->on_tree) | |
c0ff4b85 | 630 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 631 | /* |
ef8745c1 KH |
632 | * Insert again. mz->usage_in_excess will be updated. |
633 | * If excess is 0, no tree ops. | |
4e649152 | 634 | */ |
c0ff4b85 | 635 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
636 | spin_unlock(&mctz->lock); |
637 | } | |
f64c3f54 BS |
638 | } |
639 | } | |
640 | ||
c0ff4b85 | 641 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
642 | { |
643 | int node, zone; | |
644 | struct mem_cgroup_per_zone *mz; | |
645 | struct mem_cgroup_tree_per_zone *mctz; | |
646 | ||
3ed28fa1 | 647 | for_each_node(node) { |
f64c3f54 | 648 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 649 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 650 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 651 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
652 | } |
653 | } | |
654 | } | |
655 | ||
4e416953 BS |
656 | static struct mem_cgroup_per_zone * |
657 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
658 | { | |
659 | struct rb_node *rightmost = NULL; | |
26251eaf | 660 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
661 | |
662 | retry: | |
26251eaf | 663 | mz = NULL; |
4e416953 BS |
664 | rightmost = rb_last(&mctz->rb_root); |
665 | if (!rightmost) | |
666 | goto done; /* Nothing to reclaim from */ | |
667 | ||
668 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
669 | /* | |
670 | * Remove the node now but someone else can add it back, | |
671 | * we will to add it back at the end of reclaim to its correct | |
672 | * position in the tree. | |
673 | */ | |
d79154bb HD |
674 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
675 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
676 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
677 | goto retry; |
678 | done: | |
679 | return mz; | |
680 | } | |
681 | ||
682 | static struct mem_cgroup_per_zone * | |
683 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
684 | { | |
685 | struct mem_cgroup_per_zone *mz; | |
686 | ||
687 | spin_lock(&mctz->lock); | |
688 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
689 | spin_unlock(&mctz->lock); | |
690 | return mz; | |
691 | } | |
692 | ||
711d3d2c KH |
693 | /* |
694 | * Implementation Note: reading percpu statistics for memcg. | |
695 | * | |
696 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
697 | * synchronization to implement "quick" read. There are trade-off between | |
698 | * reading cost and precision of value. Then, we may have a chance to implement | |
699 | * a periodic synchronizion of counter in memcg's counter. | |
700 | * | |
701 | * But this _read() function is used for user interface now. The user accounts | |
702 | * memory usage by memory cgroup and he _always_ requires exact value because | |
703 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
704 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
705 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
706 | * | |
707 | * If there are kernel internal actions which can make use of some not-exact | |
708 | * value, and reading all cpu value can be performance bottleneck in some | |
709 | * common workload, threashold and synchonization as vmstat[] should be | |
710 | * implemented. | |
711 | */ | |
c0ff4b85 | 712 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 713 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 714 | { |
7a159cc9 | 715 | long val = 0; |
c62b1a3b | 716 | int cpu; |
c62b1a3b | 717 | |
711d3d2c KH |
718 | get_online_cpus(); |
719 | for_each_online_cpu(cpu) | |
c0ff4b85 | 720 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 721 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
722 | spin_lock(&memcg->pcp_counter_lock); |
723 | val += memcg->nocpu_base.count[idx]; | |
724 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
725 | #endif |
726 | put_online_cpus(); | |
c62b1a3b KH |
727 | return val; |
728 | } | |
729 | ||
c0ff4b85 | 730 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
731 | bool charge) |
732 | { | |
733 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 734 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
735 | } |
736 | ||
c0ff4b85 | 737 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
738 | enum mem_cgroup_events_index idx) |
739 | { | |
740 | unsigned long val = 0; | |
741 | int cpu; | |
742 | ||
743 | for_each_online_cpu(cpu) | |
c0ff4b85 | 744 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 745 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
746 | spin_lock(&memcg->pcp_counter_lock); |
747 | val += memcg->nocpu_base.events[idx]; | |
748 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
749 | #endif |
750 | return val; | |
751 | } | |
752 | ||
c0ff4b85 | 753 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 754 | bool anon, int nr_pages) |
d52aa412 | 755 | { |
c62b1a3b KH |
756 | preempt_disable(); |
757 | ||
b2402857 KH |
758 | /* |
759 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
760 | * counted as CACHE even if it's on ANON LRU. | |
761 | */ | |
762 | if (anon) | |
763 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 764 | nr_pages); |
d52aa412 | 765 | else |
b2402857 | 766 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 767 | nr_pages); |
55e462b0 | 768 | |
e401f176 KH |
769 | /* pagein of a big page is an event. So, ignore page size */ |
770 | if (nr_pages > 0) | |
c0ff4b85 | 771 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 772 | else { |
c0ff4b85 | 773 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
774 | nr_pages = -nr_pages; /* for event */ |
775 | } | |
e401f176 | 776 | |
13114716 | 777 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 778 | |
c62b1a3b | 779 | preempt_enable(); |
6d12e2d8 KH |
780 | } |
781 | ||
bb2a0de9 | 782 | unsigned long |
4d7dcca2 | 783 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
784 | { |
785 | struct mem_cgroup_per_zone *mz; | |
786 | ||
787 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
788 | return mz->lru_size[lru]; | |
789 | } | |
790 | ||
791 | static unsigned long | |
c0ff4b85 | 792 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 793 | unsigned int lru_mask) |
889976db YH |
794 | { |
795 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 796 | enum lru_list lru; |
bb2a0de9 KH |
797 | unsigned long ret = 0; |
798 | ||
c0ff4b85 | 799 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 800 | |
f156ab93 HD |
801 | for_each_lru(lru) { |
802 | if (BIT(lru) & lru_mask) | |
803 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
804 | } |
805 | return ret; | |
806 | } | |
807 | ||
808 | static unsigned long | |
c0ff4b85 | 809 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
810 | int nid, unsigned int lru_mask) |
811 | { | |
889976db YH |
812 | u64 total = 0; |
813 | int zid; | |
814 | ||
bb2a0de9 | 815 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
816 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
817 | nid, zid, lru_mask); | |
bb2a0de9 | 818 | |
889976db YH |
819 | return total; |
820 | } | |
bb2a0de9 | 821 | |
c0ff4b85 | 822 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 823 | unsigned int lru_mask) |
6d12e2d8 | 824 | { |
889976db | 825 | int nid; |
6d12e2d8 KH |
826 | u64 total = 0; |
827 | ||
31aaea4a | 828 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 829 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 830 | return total; |
d52aa412 KH |
831 | } |
832 | ||
f53d7ce3 JW |
833 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
834 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
835 | { |
836 | unsigned long val, next; | |
837 | ||
13114716 | 838 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 839 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 840 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
841 | if ((long)next - (long)val < 0) { |
842 | switch (target) { | |
843 | case MEM_CGROUP_TARGET_THRESH: | |
844 | next = val + THRESHOLDS_EVENTS_TARGET; | |
845 | break; | |
846 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
847 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
848 | break; | |
849 | case MEM_CGROUP_TARGET_NUMAINFO: | |
850 | next = val + NUMAINFO_EVENTS_TARGET; | |
851 | break; | |
852 | default: | |
853 | break; | |
854 | } | |
855 | __this_cpu_write(memcg->stat->targets[target], next); | |
856 | return true; | |
7a159cc9 | 857 | } |
f53d7ce3 | 858 | return false; |
d2265e6f KH |
859 | } |
860 | ||
861 | /* | |
862 | * Check events in order. | |
863 | * | |
864 | */ | |
c0ff4b85 | 865 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 866 | { |
4799401f | 867 | preempt_disable(); |
d2265e6f | 868 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
869 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
870 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
871 | bool do_softlimit; |
872 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
873 | |
874 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
875 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
876 | #if MAX_NUMNODES > 1 | |
877 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
878 | MEM_CGROUP_TARGET_NUMAINFO); | |
879 | #endif | |
880 | preempt_enable(); | |
881 | ||
c0ff4b85 | 882 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 883 | if (unlikely(do_softlimit)) |
c0ff4b85 | 884 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 885 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 886 | if (unlikely(do_numainfo)) |
c0ff4b85 | 887 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 888 | #endif |
f53d7ce3 JW |
889 | } else |
890 | preempt_enable(); | |
d2265e6f KH |
891 | } |
892 | ||
d1a4c0b3 | 893 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 | 894 | { |
b2145145 WL |
895 | return mem_cgroup_from_css( |
896 | cgroup_subsys_state(cont, mem_cgroup_subsys_id)); | |
8cdea7c0 BS |
897 | } |
898 | ||
cf475ad2 | 899 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 900 | { |
31a78f23 BS |
901 | /* |
902 | * mm_update_next_owner() may clear mm->owner to NULL | |
903 | * if it races with swapoff, page migration, etc. | |
904 | * So this can be called with p == NULL. | |
905 | */ | |
906 | if (unlikely(!p)) | |
907 | return NULL; | |
908 | ||
b2145145 | 909 | return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
910 | } |
911 | ||
a433658c | 912 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 913 | { |
c0ff4b85 | 914 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
915 | |
916 | if (!mm) | |
917 | return NULL; | |
54595fe2 KH |
918 | /* |
919 | * Because we have no locks, mm->owner's may be being moved to other | |
920 | * cgroup. We use css_tryget() here even if this looks | |
921 | * pessimistic (rather than adding locks here). | |
922 | */ | |
923 | rcu_read_lock(); | |
924 | do { | |
c0ff4b85 R |
925 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
926 | if (unlikely(!memcg)) | |
54595fe2 | 927 | break; |
c0ff4b85 | 928 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 929 | rcu_read_unlock(); |
c0ff4b85 | 930 | return memcg; |
54595fe2 KH |
931 | } |
932 | ||
5660048c JW |
933 | /** |
934 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
935 | * @root: hierarchy root | |
936 | * @prev: previously returned memcg, NULL on first invocation | |
937 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
938 | * | |
939 | * Returns references to children of the hierarchy below @root, or | |
940 | * @root itself, or %NULL after a full round-trip. | |
941 | * | |
942 | * Caller must pass the return value in @prev on subsequent | |
943 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
944 | * to cancel a hierarchy walk before the round-trip is complete. | |
945 | * | |
946 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
947 | * divide up the memcgs in the hierarchy among all concurrent | |
948 | * reclaimers operating on the same zone and priority. | |
949 | */ | |
950 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
951 | struct mem_cgroup *prev, | |
952 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 953 | { |
9f3a0d09 JW |
954 | struct mem_cgroup *memcg = NULL; |
955 | int id = 0; | |
711d3d2c | 956 | |
5660048c JW |
957 | if (mem_cgroup_disabled()) |
958 | return NULL; | |
959 | ||
9f3a0d09 JW |
960 | if (!root) |
961 | root = root_mem_cgroup; | |
7d74b06f | 962 | |
9f3a0d09 JW |
963 | if (prev && !reclaim) |
964 | id = css_id(&prev->css); | |
14067bb3 | 965 | |
9f3a0d09 JW |
966 | if (prev && prev != root) |
967 | css_put(&prev->css); | |
14067bb3 | 968 | |
9f3a0d09 JW |
969 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
970 | if (prev) | |
971 | return NULL; | |
972 | return root; | |
973 | } | |
14067bb3 | 974 | |
9f3a0d09 | 975 | while (!memcg) { |
527a5ec9 | 976 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
9f3a0d09 | 977 | struct cgroup_subsys_state *css; |
711d3d2c | 978 | |
527a5ec9 JW |
979 | if (reclaim) { |
980 | int nid = zone_to_nid(reclaim->zone); | |
981 | int zid = zone_idx(reclaim->zone); | |
982 | struct mem_cgroup_per_zone *mz; | |
983 | ||
984 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
985 | iter = &mz->reclaim_iter[reclaim->priority]; | |
986 | if (prev && reclaim->generation != iter->generation) | |
987 | return NULL; | |
988 | id = iter->position; | |
989 | } | |
7d74b06f | 990 | |
9f3a0d09 JW |
991 | rcu_read_lock(); |
992 | css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id); | |
993 | if (css) { | |
994 | if (css == &root->css || css_tryget(css)) | |
b2145145 | 995 | memcg = mem_cgroup_from_css(css); |
9f3a0d09 JW |
996 | } else |
997 | id = 0; | |
14067bb3 | 998 | rcu_read_unlock(); |
14067bb3 | 999 | |
527a5ec9 JW |
1000 | if (reclaim) { |
1001 | iter->position = id; | |
1002 | if (!css) | |
1003 | iter->generation++; | |
1004 | else if (!prev && memcg) | |
1005 | reclaim->generation = iter->generation; | |
1006 | } | |
9f3a0d09 JW |
1007 | |
1008 | if (prev && !css) | |
1009 | return NULL; | |
1010 | } | |
1011 | return memcg; | |
14067bb3 | 1012 | } |
7d74b06f | 1013 | |
5660048c JW |
1014 | /** |
1015 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1016 | * @root: hierarchy root | |
1017 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1018 | */ | |
1019 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1020 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1021 | { |
1022 | if (!root) | |
1023 | root = root_mem_cgroup; | |
1024 | if (prev && prev != root) | |
1025 | css_put(&prev->css); | |
1026 | } | |
7d74b06f | 1027 | |
9f3a0d09 JW |
1028 | /* |
1029 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1030 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1031 | * be used for reference counting. | |
1032 | */ | |
1033 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1034 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1035 | iter != NULL; \ |
527a5ec9 | 1036 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1037 | |
9f3a0d09 | 1038 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1039 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1040 | iter != NULL; \ |
527a5ec9 | 1041 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1042 | |
68ae564b | 1043 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1044 | { |
c0ff4b85 | 1045 | struct mem_cgroup *memcg; |
456f998e | 1046 | |
456f998e | 1047 | rcu_read_lock(); |
c0ff4b85 R |
1048 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1049 | if (unlikely(!memcg)) | |
456f998e YH |
1050 | goto out; |
1051 | ||
1052 | switch (idx) { | |
456f998e | 1053 | case PGFAULT: |
0e574a93 JW |
1054 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1055 | break; | |
1056 | case PGMAJFAULT: | |
1057 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1058 | break; |
1059 | default: | |
1060 | BUG(); | |
1061 | } | |
1062 | out: | |
1063 | rcu_read_unlock(); | |
1064 | } | |
68ae564b | 1065 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1066 | |
925b7673 JW |
1067 | /** |
1068 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1069 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1070 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1071 | * |
1072 | * Returns the lru list vector holding pages for the given @zone and | |
1073 | * @mem. This can be the global zone lruvec, if the memory controller | |
1074 | * is disabled. | |
1075 | */ | |
1076 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1077 | struct mem_cgroup *memcg) | |
1078 | { | |
1079 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1080 | struct lruvec *lruvec; |
925b7673 | 1081 | |
bea8c150 HD |
1082 | if (mem_cgroup_disabled()) { |
1083 | lruvec = &zone->lruvec; | |
1084 | goto out; | |
1085 | } | |
925b7673 JW |
1086 | |
1087 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1088 | lruvec = &mz->lruvec; |
1089 | out: | |
1090 | /* | |
1091 | * Since a node can be onlined after the mem_cgroup was created, | |
1092 | * we have to be prepared to initialize lruvec->zone here; | |
1093 | * and if offlined then reonlined, we need to reinitialize it. | |
1094 | */ | |
1095 | if (unlikely(lruvec->zone != zone)) | |
1096 | lruvec->zone = zone; | |
1097 | return lruvec; | |
925b7673 JW |
1098 | } |
1099 | ||
08e552c6 KH |
1100 | /* |
1101 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1102 | * Operations are called by routine of global LRU independently from memcg. | |
1103 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1104 | * | |
1105 | * Changes to pc->mem_cgroup happens when | |
1106 | * 1. charge | |
1107 | * 2. moving account | |
1108 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1109 | * It is added to LRU before charge. | |
1110 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1111 | * When moving account, the page is not on LRU. It's isolated. | |
1112 | */ | |
4f98a2fe | 1113 | |
925b7673 | 1114 | /** |
fa9add64 | 1115 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1116 | * @page: the page |
fa9add64 | 1117 | * @zone: zone of the page |
925b7673 | 1118 | */ |
fa9add64 | 1119 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1120 | { |
08e552c6 | 1121 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1122 | struct mem_cgroup *memcg; |
1123 | struct page_cgroup *pc; | |
bea8c150 | 1124 | struct lruvec *lruvec; |
6d12e2d8 | 1125 | |
bea8c150 HD |
1126 | if (mem_cgroup_disabled()) { |
1127 | lruvec = &zone->lruvec; | |
1128 | goto out; | |
1129 | } | |
925b7673 | 1130 | |
08e552c6 | 1131 | pc = lookup_page_cgroup(page); |
38c5d72f | 1132 | memcg = pc->mem_cgroup; |
7512102c HD |
1133 | |
1134 | /* | |
fa9add64 | 1135 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1136 | * an uncharged page off lru does nothing to secure |
1137 | * its former mem_cgroup from sudden removal. | |
1138 | * | |
1139 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1140 | * under page_cgroup lock: between them, they make all uses | |
1141 | * of pc->mem_cgroup safe. | |
1142 | */ | |
fa9add64 | 1143 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1144 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1145 | ||
925b7673 | 1146 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1147 | lruvec = &mz->lruvec; |
1148 | out: | |
1149 | /* | |
1150 | * Since a node can be onlined after the mem_cgroup was created, | |
1151 | * we have to be prepared to initialize lruvec->zone here; | |
1152 | * and if offlined then reonlined, we need to reinitialize it. | |
1153 | */ | |
1154 | if (unlikely(lruvec->zone != zone)) | |
1155 | lruvec->zone = zone; | |
1156 | return lruvec; | |
08e552c6 | 1157 | } |
b69408e8 | 1158 | |
925b7673 | 1159 | /** |
fa9add64 HD |
1160 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1161 | * @lruvec: mem_cgroup per zone lru vector | |
1162 | * @lru: index of lru list the page is sitting on | |
1163 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1164 | * |
fa9add64 HD |
1165 | * This function must be called when a page is added to or removed from an |
1166 | * lru list. | |
3f58a829 | 1167 | */ |
fa9add64 HD |
1168 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1169 | int nr_pages) | |
3f58a829 MK |
1170 | { |
1171 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1172 | unsigned long *lru_size; |
3f58a829 MK |
1173 | |
1174 | if (mem_cgroup_disabled()) | |
1175 | return; | |
1176 | ||
fa9add64 HD |
1177 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1178 | lru_size = mz->lru_size + lru; | |
1179 | *lru_size += nr_pages; | |
1180 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1181 | } |
544122e5 | 1182 | |
3e92041d | 1183 | /* |
c0ff4b85 | 1184 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1185 | * hierarchy subtree |
1186 | */ | |
c3ac9a8a JW |
1187 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1188 | struct mem_cgroup *memcg) | |
3e92041d | 1189 | { |
91c63734 JW |
1190 | if (root_memcg == memcg) |
1191 | return true; | |
3a981f48 | 1192 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1193 | return false; |
c3ac9a8a JW |
1194 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1195 | } | |
1196 | ||
1197 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1198 | struct mem_cgroup *memcg) | |
1199 | { | |
1200 | bool ret; | |
1201 | ||
91c63734 | 1202 | rcu_read_lock(); |
c3ac9a8a | 1203 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1204 | rcu_read_unlock(); |
1205 | return ret; | |
3e92041d MH |
1206 | } |
1207 | ||
c0ff4b85 | 1208 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1209 | { |
1210 | int ret; | |
0b7f569e | 1211 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1212 | struct task_struct *p; |
4c4a2214 | 1213 | |
158e0a2d | 1214 | p = find_lock_task_mm(task); |
de077d22 DR |
1215 | if (p) { |
1216 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1217 | task_unlock(p); | |
1218 | } else { | |
1219 | /* | |
1220 | * All threads may have already detached their mm's, but the oom | |
1221 | * killer still needs to detect if they have already been oom | |
1222 | * killed to prevent needlessly killing additional tasks. | |
1223 | */ | |
1224 | task_lock(task); | |
1225 | curr = mem_cgroup_from_task(task); | |
1226 | if (curr) | |
1227 | css_get(&curr->css); | |
1228 | task_unlock(task); | |
1229 | } | |
0b7f569e KH |
1230 | if (!curr) |
1231 | return 0; | |
d31f56db | 1232 | /* |
c0ff4b85 | 1233 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1234 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1235 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1236 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1237 | */ |
c0ff4b85 | 1238 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1239 | css_put(&curr->css); |
4c4a2214 DR |
1240 | return ret; |
1241 | } | |
1242 | ||
c56d5c7d | 1243 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1244 | { |
9b272977 | 1245 | unsigned long inactive_ratio; |
14797e23 | 1246 | unsigned long inactive; |
9b272977 | 1247 | unsigned long active; |
c772be93 | 1248 | unsigned long gb; |
14797e23 | 1249 | |
4d7dcca2 HD |
1250 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1251 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1252 | |
c772be93 KM |
1253 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1254 | if (gb) | |
1255 | inactive_ratio = int_sqrt(10 * gb); | |
1256 | else | |
1257 | inactive_ratio = 1; | |
1258 | ||
9b272977 | 1259 | return inactive * inactive_ratio < active; |
14797e23 KM |
1260 | } |
1261 | ||
c56d5c7d | 1262 | int mem_cgroup_inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 RR |
1263 | { |
1264 | unsigned long active; | |
1265 | unsigned long inactive; | |
1266 | ||
4d7dcca2 HD |
1267 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_FILE); |
1268 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 RR |
1269 | |
1270 | return (active > inactive); | |
1271 | } | |
1272 | ||
6d61ef40 BS |
1273 | #define mem_cgroup_from_res_counter(counter, member) \ |
1274 | container_of(counter, struct mem_cgroup, member) | |
1275 | ||
19942822 | 1276 | /** |
9d11ea9f | 1277 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1278 | * @memcg: the memory cgroup |
19942822 | 1279 | * |
9d11ea9f | 1280 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1281 | * pages. |
19942822 | 1282 | */ |
c0ff4b85 | 1283 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1284 | { |
9d11ea9f JW |
1285 | unsigned long long margin; |
1286 | ||
c0ff4b85 | 1287 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1288 | if (do_swap_account) |
c0ff4b85 | 1289 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1290 | return margin >> PAGE_SHIFT; |
19942822 JW |
1291 | } |
1292 | ||
1f4c025b | 1293 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1294 | { |
1295 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1296 | |
1297 | /* root ? */ | |
1298 | if (cgrp->parent == NULL) | |
1299 | return vm_swappiness; | |
1300 | ||
bf1ff263 | 1301 | return memcg->swappiness; |
a7885eb8 KM |
1302 | } |
1303 | ||
619d094b KH |
1304 | /* |
1305 | * memcg->moving_account is used for checking possibility that some thread is | |
1306 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1307 | * a memcg, other threads should check memcg->moving_account under | |
1308 | * rcu_read_lock(), like this: | |
1309 | * | |
1310 | * CPU-A CPU-B | |
1311 | * rcu_read_lock() | |
1312 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1313 | * take heavy locks. | |
1314 | * synchronize_rcu() update something. | |
1315 | * rcu_read_unlock() | |
1316 | * start move here. | |
1317 | */ | |
4331f7d3 KH |
1318 | |
1319 | /* for quick checking without looking up memcg */ | |
1320 | atomic_t memcg_moving __read_mostly; | |
1321 | ||
c0ff4b85 | 1322 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1323 | { |
4331f7d3 | 1324 | atomic_inc(&memcg_moving); |
619d094b | 1325 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1326 | synchronize_rcu(); |
1327 | } | |
1328 | ||
c0ff4b85 | 1329 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1330 | { |
619d094b KH |
1331 | /* |
1332 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1333 | * We check NULL in callee rather than caller. | |
1334 | */ | |
4331f7d3 KH |
1335 | if (memcg) { |
1336 | atomic_dec(&memcg_moving); | |
619d094b | 1337 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1338 | } |
32047e2a | 1339 | } |
619d094b | 1340 | |
32047e2a KH |
1341 | /* |
1342 | * 2 routines for checking "mem" is under move_account() or not. | |
1343 | * | |
13fd1dd9 AM |
1344 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1345 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1346 | * pc->mem_cgroup may be overwritten. |
1347 | * | |
1348 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1349 | * under hierarchy of moving cgroups. This is for | |
1350 | * waiting at hith-memory prressure caused by "move". | |
1351 | */ | |
1352 | ||
13fd1dd9 | 1353 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1354 | { |
1355 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1356 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1357 | } |
4b534334 | 1358 | |
c0ff4b85 | 1359 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1360 | { |
2bd9bb20 KH |
1361 | struct mem_cgroup *from; |
1362 | struct mem_cgroup *to; | |
4b534334 | 1363 | bool ret = false; |
2bd9bb20 KH |
1364 | /* |
1365 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1366 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1367 | */ | |
1368 | spin_lock(&mc.lock); | |
1369 | from = mc.from; | |
1370 | to = mc.to; | |
1371 | if (!from) | |
1372 | goto unlock; | |
3e92041d | 1373 | |
c0ff4b85 R |
1374 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1375 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1376 | unlock: |
1377 | spin_unlock(&mc.lock); | |
4b534334 KH |
1378 | return ret; |
1379 | } | |
1380 | ||
c0ff4b85 | 1381 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1382 | { |
1383 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1384 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1385 | DEFINE_WAIT(wait); |
1386 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1387 | /* moving charge context might have finished. */ | |
1388 | if (mc.moving_task) | |
1389 | schedule(); | |
1390 | finish_wait(&mc.waitq, &wait); | |
1391 | return true; | |
1392 | } | |
1393 | } | |
1394 | return false; | |
1395 | } | |
1396 | ||
312734c0 KH |
1397 | /* |
1398 | * Take this lock when | |
1399 | * - a code tries to modify page's memcg while it's USED. | |
1400 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1401 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1402 | */ |
1403 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1404 | unsigned long *flags) | |
1405 | { | |
1406 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1407 | } | |
1408 | ||
1409 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1410 | unsigned long *flags) | |
1411 | { | |
1412 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1413 | } | |
1414 | ||
e222432b | 1415 | /** |
6a6135b6 | 1416 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1417 | * @memcg: The memory cgroup that went over limit |
1418 | * @p: Task that is going to be killed | |
1419 | * | |
1420 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1421 | * enabled | |
1422 | */ | |
1423 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1424 | { | |
1425 | struct cgroup *task_cgrp; | |
1426 | struct cgroup *mem_cgrp; | |
1427 | /* | |
1428 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1429 | * on the assumption that OOM is serialized for memory controller. | |
1430 | * If this assumption is broken, revisit this code. | |
1431 | */ | |
1432 | static char memcg_name[PATH_MAX]; | |
1433 | int ret; | |
1434 | ||
d31f56db | 1435 | if (!memcg || !p) |
e222432b BS |
1436 | return; |
1437 | ||
e222432b BS |
1438 | rcu_read_lock(); |
1439 | ||
1440 | mem_cgrp = memcg->css.cgroup; | |
1441 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1442 | ||
1443 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1444 | if (ret < 0) { | |
1445 | /* | |
1446 | * Unfortunately, we are unable to convert to a useful name | |
1447 | * But we'll still print out the usage information | |
1448 | */ | |
1449 | rcu_read_unlock(); | |
1450 | goto done; | |
1451 | } | |
1452 | rcu_read_unlock(); | |
1453 | ||
1454 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1455 | ||
1456 | rcu_read_lock(); | |
1457 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1458 | if (ret < 0) { | |
1459 | rcu_read_unlock(); | |
1460 | goto done; | |
1461 | } | |
1462 | rcu_read_unlock(); | |
1463 | ||
1464 | /* | |
1465 | * Continues from above, so we don't need an KERN_ level | |
1466 | */ | |
1467 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1468 | done: | |
1469 | ||
1470 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1471 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1472 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1473 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1474 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1475 | "failcnt %llu\n", | |
1476 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1477 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1478 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
510fc4e1 GC |
1479 | printk(KERN_INFO "kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
1480 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, | |
1481 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1482 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
e222432b BS |
1483 | } |
1484 | ||
81d39c20 KH |
1485 | /* |
1486 | * This function returns the number of memcg under hierarchy tree. Returns | |
1487 | * 1(self count) if no children. | |
1488 | */ | |
c0ff4b85 | 1489 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1490 | { |
1491 | int num = 0; | |
7d74b06f KH |
1492 | struct mem_cgroup *iter; |
1493 | ||
c0ff4b85 | 1494 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1495 | num++; |
81d39c20 KH |
1496 | return num; |
1497 | } | |
1498 | ||
a63d83f4 DR |
1499 | /* |
1500 | * Return the memory (and swap, if configured) limit for a memcg. | |
1501 | */ | |
9cbb78bb | 1502 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1503 | { |
1504 | u64 limit; | |
a63d83f4 | 1505 | |
f3e8eb70 | 1506 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1507 | |
a63d83f4 | 1508 | /* |
9a5a8f19 | 1509 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1510 | */ |
9a5a8f19 MH |
1511 | if (mem_cgroup_swappiness(memcg)) { |
1512 | u64 memsw; | |
1513 | ||
1514 | limit += total_swap_pages << PAGE_SHIFT; | |
1515 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1516 | ||
1517 | /* | |
1518 | * If memsw is finite and limits the amount of swap space | |
1519 | * available to this memcg, return that limit. | |
1520 | */ | |
1521 | limit = min(limit, memsw); | |
1522 | } | |
1523 | ||
1524 | return limit; | |
a63d83f4 DR |
1525 | } |
1526 | ||
19965460 DR |
1527 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1528 | int order) | |
9cbb78bb DR |
1529 | { |
1530 | struct mem_cgroup *iter; | |
1531 | unsigned long chosen_points = 0; | |
1532 | unsigned long totalpages; | |
1533 | unsigned int points = 0; | |
1534 | struct task_struct *chosen = NULL; | |
1535 | ||
876aafbf DR |
1536 | /* |
1537 | * If current has a pending SIGKILL, then automatically select it. The | |
1538 | * goal is to allow it to allocate so that it may quickly exit and free | |
1539 | * its memory. | |
1540 | */ | |
1541 | if (fatal_signal_pending(current)) { | |
1542 | set_thread_flag(TIF_MEMDIE); | |
1543 | return; | |
1544 | } | |
1545 | ||
1546 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1547 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1548 | for_each_mem_cgroup_tree(iter, memcg) { | |
1549 | struct cgroup *cgroup = iter->css.cgroup; | |
1550 | struct cgroup_iter it; | |
1551 | struct task_struct *task; | |
1552 | ||
1553 | cgroup_iter_start(cgroup, &it); | |
1554 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
1555 | switch (oom_scan_process_thread(task, totalpages, NULL, | |
1556 | false)) { | |
1557 | case OOM_SCAN_SELECT: | |
1558 | if (chosen) | |
1559 | put_task_struct(chosen); | |
1560 | chosen = task; | |
1561 | chosen_points = ULONG_MAX; | |
1562 | get_task_struct(chosen); | |
1563 | /* fall through */ | |
1564 | case OOM_SCAN_CONTINUE: | |
1565 | continue; | |
1566 | case OOM_SCAN_ABORT: | |
1567 | cgroup_iter_end(cgroup, &it); | |
1568 | mem_cgroup_iter_break(memcg, iter); | |
1569 | if (chosen) | |
1570 | put_task_struct(chosen); | |
1571 | return; | |
1572 | case OOM_SCAN_OK: | |
1573 | break; | |
1574 | }; | |
1575 | points = oom_badness(task, memcg, NULL, totalpages); | |
1576 | if (points > chosen_points) { | |
1577 | if (chosen) | |
1578 | put_task_struct(chosen); | |
1579 | chosen = task; | |
1580 | chosen_points = points; | |
1581 | get_task_struct(chosen); | |
1582 | } | |
1583 | } | |
1584 | cgroup_iter_end(cgroup, &it); | |
1585 | } | |
1586 | ||
1587 | if (!chosen) | |
1588 | return; | |
1589 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1590 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1591 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1592 | } |
1593 | ||
5660048c JW |
1594 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1595 | gfp_t gfp_mask, | |
1596 | unsigned long flags) | |
1597 | { | |
1598 | unsigned long total = 0; | |
1599 | bool noswap = false; | |
1600 | int loop; | |
1601 | ||
1602 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1603 | noswap = true; | |
1604 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1605 | noswap = true; | |
1606 | ||
1607 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1608 | if (loop) | |
1609 | drain_all_stock_async(memcg); | |
1610 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1611 | /* | |
1612 | * Allow limit shrinkers, which are triggered directly | |
1613 | * by userspace, to catch signals and stop reclaim | |
1614 | * after minimal progress, regardless of the margin. | |
1615 | */ | |
1616 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1617 | break; | |
1618 | if (mem_cgroup_margin(memcg)) | |
1619 | break; | |
1620 | /* | |
1621 | * If nothing was reclaimed after two attempts, there | |
1622 | * may be no reclaimable pages in this hierarchy. | |
1623 | */ | |
1624 | if (loop && !total) | |
1625 | break; | |
1626 | } | |
1627 | return total; | |
1628 | } | |
1629 | ||
4d0c066d KH |
1630 | /** |
1631 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1632 | * @memcg: the target memcg |
4d0c066d KH |
1633 | * @nid: the node ID to be checked. |
1634 | * @noswap : specify true here if the user wants flle only information. | |
1635 | * | |
1636 | * This function returns whether the specified memcg contains any | |
1637 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1638 | * pages in the node. | |
1639 | */ | |
c0ff4b85 | 1640 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1641 | int nid, bool noswap) |
1642 | { | |
c0ff4b85 | 1643 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1644 | return true; |
1645 | if (noswap || !total_swap_pages) | |
1646 | return false; | |
c0ff4b85 | 1647 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1648 | return true; |
1649 | return false; | |
1650 | ||
1651 | } | |
889976db YH |
1652 | #if MAX_NUMNODES > 1 |
1653 | ||
1654 | /* | |
1655 | * Always updating the nodemask is not very good - even if we have an empty | |
1656 | * list or the wrong list here, we can start from some node and traverse all | |
1657 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1658 | * | |
1659 | */ | |
c0ff4b85 | 1660 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1661 | { |
1662 | int nid; | |
453a9bf3 KH |
1663 | /* |
1664 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1665 | * pagein/pageout changes since the last update. | |
1666 | */ | |
c0ff4b85 | 1667 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1668 | return; |
c0ff4b85 | 1669 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1670 | return; |
1671 | ||
889976db | 1672 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1673 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1674 | |
31aaea4a | 1675 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1676 | |
c0ff4b85 R |
1677 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1678 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1679 | } |
453a9bf3 | 1680 | |
c0ff4b85 R |
1681 | atomic_set(&memcg->numainfo_events, 0); |
1682 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1683 | } |
1684 | ||
1685 | /* | |
1686 | * Selecting a node where we start reclaim from. Because what we need is just | |
1687 | * reducing usage counter, start from anywhere is O,K. Considering | |
1688 | * memory reclaim from current node, there are pros. and cons. | |
1689 | * | |
1690 | * Freeing memory from current node means freeing memory from a node which | |
1691 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1692 | * hit limits, it will see a contention on a node. But freeing from remote | |
1693 | * node means more costs for memory reclaim because of memory latency. | |
1694 | * | |
1695 | * Now, we use round-robin. Better algorithm is welcomed. | |
1696 | */ | |
c0ff4b85 | 1697 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1698 | { |
1699 | int node; | |
1700 | ||
c0ff4b85 R |
1701 | mem_cgroup_may_update_nodemask(memcg); |
1702 | node = memcg->last_scanned_node; | |
889976db | 1703 | |
c0ff4b85 | 1704 | node = next_node(node, memcg->scan_nodes); |
889976db | 1705 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1706 | node = first_node(memcg->scan_nodes); |
889976db YH |
1707 | /* |
1708 | * We call this when we hit limit, not when pages are added to LRU. | |
1709 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1710 | * memcg is too small and all pages are not on LRU. In that case, | |
1711 | * we use curret node. | |
1712 | */ | |
1713 | if (unlikely(node == MAX_NUMNODES)) | |
1714 | node = numa_node_id(); | |
1715 | ||
c0ff4b85 | 1716 | memcg->last_scanned_node = node; |
889976db YH |
1717 | return node; |
1718 | } | |
1719 | ||
4d0c066d KH |
1720 | /* |
1721 | * Check all nodes whether it contains reclaimable pages or not. | |
1722 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1723 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1724 | * enough new information. We need to do double check. | |
1725 | */ | |
6bbda35c | 1726 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1727 | { |
1728 | int nid; | |
1729 | ||
1730 | /* | |
1731 | * quick check...making use of scan_node. | |
1732 | * We can skip unused nodes. | |
1733 | */ | |
c0ff4b85 R |
1734 | if (!nodes_empty(memcg->scan_nodes)) { |
1735 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1736 | nid < MAX_NUMNODES; |
c0ff4b85 | 1737 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1738 | |
c0ff4b85 | 1739 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1740 | return true; |
1741 | } | |
1742 | } | |
1743 | /* | |
1744 | * Check rest of nodes. | |
1745 | */ | |
31aaea4a | 1746 | for_each_node_state(nid, N_MEMORY) { |
c0ff4b85 | 1747 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 1748 | continue; |
c0ff4b85 | 1749 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1750 | return true; |
1751 | } | |
1752 | return false; | |
1753 | } | |
1754 | ||
889976db | 1755 | #else |
c0ff4b85 | 1756 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1757 | { |
1758 | return 0; | |
1759 | } | |
4d0c066d | 1760 | |
6bbda35c | 1761 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 1762 | { |
c0ff4b85 | 1763 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 1764 | } |
889976db YH |
1765 | #endif |
1766 | ||
5660048c JW |
1767 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1768 | struct zone *zone, | |
1769 | gfp_t gfp_mask, | |
1770 | unsigned long *total_scanned) | |
6d61ef40 | 1771 | { |
9f3a0d09 | 1772 | struct mem_cgroup *victim = NULL; |
5660048c | 1773 | int total = 0; |
04046e1a | 1774 | int loop = 0; |
9d11ea9f | 1775 | unsigned long excess; |
185efc0f | 1776 | unsigned long nr_scanned; |
527a5ec9 JW |
1777 | struct mem_cgroup_reclaim_cookie reclaim = { |
1778 | .zone = zone, | |
1779 | .priority = 0, | |
1780 | }; | |
9d11ea9f | 1781 | |
c0ff4b85 | 1782 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 1783 | |
4e416953 | 1784 | while (1) { |
527a5ec9 | 1785 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 1786 | if (!victim) { |
04046e1a | 1787 | loop++; |
4e416953 BS |
1788 | if (loop >= 2) { |
1789 | /* | |
1790 | * If we have not been able to reclaim | |
1791 | * anything, it might because there are | |
1792 | * no reclaimable pages under this hierarchy | |
1793 | */ | |
5660048c | 1794 | if (!total) |
4e416953 | 1795 | break; |
4e416953 | 1796 | /* |
25985edc | 1797 | * We want to do more targeted reclaim. |
4e416953 BS |
1798 | * excess >> 2 is not to excessive so as to |
1799 | * reclaim too much, nor too less that we keep | |
1800 | * coming back to reclaim from this cgroup | |
1801 | */ | |
1802 | if (total >= (excess >> 2) || | |
9f3a0d09 | 1803 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 1804 | break; |
4e416953 | 1805 | } |
9f3a0d09 | 1806 | continue; |
4e416953 | 1807 | } |
5660048c | 1808 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 1809 | continue; |
5660048c JW |
1810 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1811 | zone, &nr_scanned); | |
1812 | *total_scanned += nr_scanned; | |
1813 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 1814 | break; |
6d61ef40 | 1815 | } |
9f3a0d09 | 1816 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 1817 | return total; |
6d61ef40 BS |
1818 | } |
1819 | ||
867578cb KH |
1820 | /* |
1821 | * Check OOM-Killer is already running under our hierarchy. | |
1822 | * If someone is running, return false. | |
1af8efe9 | 1823 | * Has to be called with memcg_oom_lock |
867578cb | 1824 | */ |
c0ff4b85 | 1825 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 1826 | { |
79dfdacc | 1827 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1828 | |
9f3a0d09 | 1829 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1830 | if (iter->oom_lock) { |
79dfdacc MH |
1831 | /* |
1832 | * this subtree of our hierarchy is already locked | |
1833 | * so we cannot give a lock. | |
1834 | */ | |
79dfdacc | 1835 | failed = iter; |
9f3a0d09 JW |
1836 | mem_cgroup_iter_break(memcg, iter); |
1837 | break; | |
23751be0 JW |
1838 | } else |
1839 | iter->oom_lock = true; | |
7d74b06f | 1840 | } |
867578cb | 1841 | |
79dfdacc | 1842 | if (!failed) |
23751be0 | 1843 | return true; |
79dfdacc MH |
1844 | |
1845 | /* | |
1846 | * OK, we failed to lock the whole subtree so we have to clean up | |
1847 | * what we set up to the failing subtree | |
1848 | */ | |
9f3a0d09 | 1849 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 1850 | if (iter == failed) { |
9f3a0d09 JW |
1851 | mem_cgroup_iter_break(memcg, iter); |
1852 | break; | |
79dfdacc MH |
1853 | } |
1854 | iter->oom_lock = false; | |
1855 | } | |
23751be0 | 1856 | return false; |
a636b327 | 1857 | } |
0b7f569e | 1858 | |
79dfdacc | 1859 | /* |
1af8efe9 | 1860 | * Has to be called with memcg_oom_lock |
79dfdacc | 1861 | */ |
c0ff4b85 | 1862 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1863 | { |
7d74b06f KH |
1864 | struct mem_cgroup *iter; |
1865 | ||
c0ff4b85 | 1866 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1867 | iter->oom_lock = false; |
1868 | return 0; | |
1869 | } | |
1870 | ||
c0ff4b85 | 1871 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1872 | { |
1873 | struct mem_cgroup *iter; | |
1874 | ||
c0ff4b85 | 1875 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
1876 | atomic_inc(&iter->under_oom); |
1877 | } | |
1878 | ||
c0ff4b85 | 1879 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1880 | { |
1881 | struct mem_cgroup *iter; | |
1882 | ||
867578cb KH |
1883 | /* |
1884 | * When a new child is created while the hierarchy is under oom, | |
1885 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1886 | * atomic_add_unless() here. | |
1887 | */ | |
c0ff4b85 | 1888 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1889 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1890 | } |
1891 | ||
1af8efe9 | 1892 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1893 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1894 | ||
dc98df5a | 1895 | struct oom_wait_info { |
d79154bb | 1896 | struct mem_cgroup *memcg; |
dc98df5a KH |
1897 | wait_queue_t wait; |
1898 | }; | |
1899 | ||
1900 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1901 | unsigned mode, int sync, void *arg) | |
1902 | { | |
d79154bb HD |
1903 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1904 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1905 | struct oom_wait_info *oom_wait_info; |
1906 | ||
1907 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1908 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1909 | |
dc98df5a | 1910 | /* |
d79154bb | 1911 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
1912 | * Then we can use css_is_ancestor without taking care of RCU. |
1913 | */ | |
c0ff4b85 R |
1914 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
1915 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 1916 | return 0; |
dc98df5a KH |
1917 | return autoremove_wake_function(wait, mode, sync, arg); |
1918 | } | |
1919 | ||
c0ff4b85 | 1920 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 1921 | { |
c0ff4b85 R |
1922 | /* for filtering, pass "memcg" as argument. */ |
1923 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
1924 | } |
1925 | ||
c0ff4b85 | 1926 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1927 | { |
c0ff4b85 R |
1928 | if (memcg && atomic_read(&memcg->under_oom)) |
1929 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
1930 | } |
1931 | ||
867578cb KH |
1932 | /* |
1933 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1934 | */ | |
6bbda35c KS |
1935 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
1936 | int order) | |
0b7f569e | 1937 | { |
dc98df5a | 1938 | struct oom_wait_info owait; |
3c11ecf4 | 1939 | bool locked, need_to_kill; |
867578cb | 1940 | |
d79154bb | 1941 | owait.memcg = memcg; |
dc98df5a KH |
1942 | owait.wait.flags = 0; |
1943 | owait.wait.func = memcg_oom_wake_function; | |
1944 | owait.wait.private = current; | |
1945 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1946 | need_to_kill = true; |
c0ff4b85 | 1947 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 1948 | |
c0ff4b85 | 1949 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 1950 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1951 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
1952 | /* |
1953 | * Even if signal_pending(), we can't quit charge() loop without | |
1954 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1955 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1956 | */ | |
3c11ecf4 | 1957 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 1958 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
1959 | need_to_kill = false; |
1960 | if (locked) | |
c0ff4b85 | 1961 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 1962 | spin_unlock(&memcg_oom_lock); |
867578cb | 1963 | |
3c11ecf4 KH |
1964 | if (need_to_kill) { |
1965 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 1966 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 1967 | } else { |
867578cb | 1968 | schedule(); |
dc98df5a | 1969 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1970 | } |
1af8efe9 | 1971 | spin_lock(&memcg_oom_lock); |
79dfdacc | 1972 | if (locked) |
c0ff4b85 R |
1973 | mem_cgroup_oom_unlock(memcg); |
1974 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 1975 | spin_unlock(&memcg_oom_lock); |
867578cb | 1976 | |
c0ff4b85 | 1977 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 1978 | |
867578cb KH |
1979 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1980 | return false; | |
1981 | /* Give chance to dying process */ | |
715a5ee8 | 1982 | schedule_timeout_uninterruptible(1); |
867578cb | 1983 | return true; |
0b7f569e KH |
1984 | } |
1985 | ||
d69b042f BS |
1986 | /* |
1987 | * Currently used to update mapped file statistics, but the routine can be | |
1988 | * generalized to update other statistics as well. | |
32047e2a KH |
1989 | * |
1990 | * Notes: Race condition | |
1991 | * | |
1992 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1993 | * it tends to be costly. But considering some conditions, we doesn't need | |
1994 | * to do so _always_. | |
1995 | * | |
1996 | * Considering "charge", lock_page_cgroup() is not required because all | |
1997 | * file-stat operations happen after a page is attached to radix-tree. There | |
1998 | * are no race with "charge". | |
1999 | * | |
2000 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2001 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2002 | * if there are race with "uncharge". Statistics itself is properly handled | |
2003 | * by flags. | |
2004 | * | |
2005 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2006 | * small, we check mm->moving_account and detect there are possibility of race |
2007 | * If there is, we take a lock. | |
d69b042f | 2008 | */ |
26174efd | 2009 | |
89c06bd5 KH |
2010 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2011 | bool *locked, unsigned long *flags) | |
2012 | { | |
2013 | struct mem_cgroup *memcg; | |
2014 | struct page_cgroup *pc; | |
2015 | ||
2016 | pc = lookup_page_cgroup(page); | |
2017 | again: | |
2018 | memcg = pc->mem_cgroup; | |
2019 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2020 | return; | |
2021 | /* | |
2022 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2023 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2024 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2025 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2026 | */ |
13fd1dd9 | 2027 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2028 | return; |
2029 | ||
2030 | move_lock_mem_cgroup(memcg, flags); | |
2031 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2032 | move_unlock_mem_cgroup(memcg, flags); | |
2033 | goto again; | |
2034 | } | |
2035 | *locked = true; | |
2036 | } | |
2037 | ||
2038 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2039 | { | |
2040 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2041 | ||
2042 | /* | |
2043 | * It's guaranteed that pc->mem_cgroup never changes while | |
2044 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2045 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2046 | */ |
2047 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2048 | } | |
2049 | ||
2a7106f2 GT |
2050 | void mem_cgroup_update_page_stat(struct page *page, |
2051 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 2052 | { |
c0ff4b85 | 2053 | struct mem_cgroup *memcg; |
32047e2a | 2054 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2055 | unsigned long uninitialized_var(flags); |
d69b042f | 2056 | |
cfa44946 | 2057 | if (mem_cgroup_disabled()) |
d69b042f | 2058 | return; |
89c06bd5 | 2059 | |
c0ff4b85 R |
2060 | memcg = pc->mem_cgroup; |
2061 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2062 | return; |
26174efd | 2063 | |
26174efd | 2064 | switch (idx) { |
2a7106f2 | 2065 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2066 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2067 | break; |
2068 | default: | |
2069 | BUG(); | |
8725d541 | 2070 | } |
d69b042f | 2071 | |
c0ff4b85 | 2072 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2073 | } |
26174efd | 2074 | |
cdec2e42 KH |
2075 | /* |
2076 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2077 | * TODO: maybe necessary to use big numbers in big irons. | |
2078 | */ | |
7ec99d62 | 2079 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2080 | struct memcg_stock_pcp { |
2081 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2082 | unsigned int nr_pages; |
cdec2e42 | 2083 | struct work_struct work; |
26fe6168 | 2084 | unsigned long flags; |
a0db00fc | 2085 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2086 | }; |
2087 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2088 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2089 | |
a0956d54 SS |
2090 | /** |
2091 | * consume_stock: Try to consume stocked charge on this cpu. | |
2092 | * @memcg: memcg to consume from. | |
2093 | * @nr_pages: how many pages to charge. | |
2094 | * | |
2095 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2096 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2097 | * service an allocation will refill the stock. | |
2098 | * | |
2099 | * returns true if successful, false otherwise. | |
cdec2e42 | 2100 | */ |
a0956d54 | 2101 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2102 | { |
2103 | struct memcg_stock_pcp *stock; | |
2104 | bool ret = true; | |
2105 | ||
a0956d54 SS |
2106 | if (nr_pages > CHARGE_BATCH) |
2107 | return false; | |
2108 | ||
cdec2e42 | 2109 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2110 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2111 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2112 | else /* need to call res_counter_charge */ |
2113 | ret = false; | |
2114 | put_cpu_var(memcg_stock); | |
2115 | return ret; | |
2116 | } | |
2117 | ||
2118 | /* | |
2119 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2120 | */ | |
2121 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2122 | { | |
2123 | struct mem_cgroup *old = stock->cached; | |
2124 | ||
11c9ea4e JW |
2125 | if (stock->nr_pages) { |
2126 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2127 | ||
2128 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2129 | if (do_swap_account) |
11c9ea4e JW |
2130 | res_counter_uncharge(&old->memsw, bytes); |
2131 | stock->nr_pages = 0; | |
cdec2e42 KH |
2132 | } |
2133 | stock->cached = NULL; | |
cdec2e42 KH |
2134 | } |
2135 | ||
2136 | /* | |
2137 | * This must be called under preempt disabled or must be called by | |
2138 | * a thread which is pinned to local cpu. | |
2139 | */ | |
2140 | static void drain_local_stock(struct work_struct *dummy) | |
2141 | { | |
2142 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2143 | drain_stock(stock); | |
26fe6168 | 2144 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2145 | } |
2146 | ||
2147 | /* | |
2148 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2149 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2150 | */ |
c0ff4b85 | 2151 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2152 | { |
2153 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2154 | ||
c0ff4b85 | 2155 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2156 | drain_stock(stock); |
c0ff4b85 | 2157 | stock->cached = memcg; |
cdec2e42 | 2158 | } |
11c9ea4e | 2159 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2160 | put_cpu_var(memcg_stock); |
2161 | } | |
2162 | ||
2163 | /* | |
c0ff4b85 | 2164 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2165 | * of the hierarchy under it. sync flag says whether we should block |
2166 | * until the work is done. | |
cdec2e42 | 2167 | */ |
c0ff4b85 | 2168 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2169 | { |
26fe6168 | 2170 | int cpu, curcpu; |
d38144b7 | 2171 | |
cdec2e42 | 2172 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2173 | get_online_cpus(); |
5af12d0e | 2174 | curcpu = get_cpu(); |
cdec2e42 KH |
2175 | for_each_online_cpu(cpu) { |
2176 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2177 | struct mem_cgroup *memcg; |
26fe6168 | 2178 | |
c0ff4b85 R |
2179 | memcg = stock->cached; |
2180 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2181 | continue; |
c0ff4b85 | 2182 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2183 | continue; |
d1a05b69 MH |
2184 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2185 | if (cpu == curcpu) | |
2186 | drain_local_stock(&stock->work); | |
2187 | else | |
2188 | schedule_work_on(cpu, &stock->work); | |
2189 | } | |
cdec2e42 | 2190 | } |
5af12d0e | 2191 | put_cpu(); |
d38144b7 MH |
2192 | |
2193 | if (!sync) | |
2194 | goto out; | |
2195 | ||
2196 | for_each_online_cpu(cpu) { | |
2197 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2198 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2199 | flush_work(&stock->work); |
2200 | } | |
2201 | out: | |
cdec2e42 | 2202 | put_online_cpus(); |
d38144b7 MH |
2203 | } |
2204 | ||
2205 | /* | |
2206 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2207 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2208 | * expects some charges will be back to res_counter later but cannot wait for | |
2209 | * it. | |
2210 | */ | |
c0ff4b85 | 2211 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2212 | { |
9f50fad6 MH |
2213 | /* |
2214 | * If someone calls draining, avoid adding more kworker runs. | |
2215 | */ | |
2216 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2217 | return; | |
c0ff4b85 | 2218 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2219 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2220 | } |
2221 | ||
2222 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2223 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2224 | { |
2225 | /* called when force_empty is called */ | |
9f50fad6 | 2226 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2227 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2228 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2229 | } |
2230 | ||
711d3d2c KH |
2231 | /* |
2232 | * This function drains percpu counter value from DEAD cpu and | |
2233 | * move it to local cpu. Note that this function can be preempted. | |
2234 | */ | |
c0ff4b85 | 2235 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2236 | { |
2237 | int i; | |
2238 | ||
c0ff4b85 | 2239 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2240 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2241 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2242 | |
c0ff4b85 R |
2243 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2244 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2245 | } |
e9f8974f | 2246 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2247 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2248 | |
c0ff4b85 R |
2249 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2250 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2251 | } |
c0ff4b85 | 2252 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2253 | } |
2254 | ||
2255 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2256 | unsigned long action, |
2257 | void *hcpu) | |
2258 | { | |
2259 | int cpu = (unsigned long)hcpu; | |
2260 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2261 | struct mem_cgroup *iter; |
cdec2e42 | 2262 | |
619d094b | 2263 | if (action == CPU_ONLINE) |
1489ebad | 2264 | return NOTIFY_OK; |
1489ebad | 2265 | |
d833049b | 2266 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2267 | return NOTIFY_OK; |
711d3d2c | 2268 | |
9f3a0d09 | 2269 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2270 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2271 | ||
cdec2e42 KH |
2272 | stock = &per_cpu(memcg_stock, cpu); |
2273 | drain_stock(stock); | |
2274 | return NOTIFY_OK; | |
2275 | } | |
2276 | ||
4b534334 KH |
2277 | |
2278 | /* See __mem_cgroup_try_charge() for details */ | |
2279 | enum { | |
2280 | CHARGE_OK, /* success */ | |
2281 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2282 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2283 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2284 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2285 | }; | |
2286 | ||
c0ff4b85 | 2287 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 SS |
2288 | unsigned int nr_pages, unsigned int min_pages, |
2289 | bool oom_check) | |
4b534334 | 2290 | { |
7ec99d62 | 2291 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2292 | struct mem_cgroup *mem_over_limit; |
2293 | struct res_counter *fail_res; | |
2294 | unsigned long flags = 0; | |
2295 | int ret; | |
2296 | ||
c0ff4b85 | 2297 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2298 | |
2299 | if (likely(!ret)) { | |
2300 | if (!do_swap_account) | |
2301 | return CHARGE_OK; | |
c0ff4b85 | 2302 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2303 | if (likely(!ret)) |
2304 | return CHARGE_OK; | |
2305 | ||
c0ff4b85 | 2306 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2307 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2308 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2309 | } else | |
2310 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2311 | /* |
9221edb7 JW |
2312 | * Never reclaim on behalf of optional batching, retry with a |
2313 | * single page instead. | |
2314 | */ | |
4c9c5359 | 2315 | if (nr_pages > min_pages) |
4b534334 KH |
2316 | return CHARGE_RETRY; |
2317 | ||
2318 | if (!(gfp_mask & __GFP_WAIT)) | |
2319 | return CHARGE_WOULDBLOCK; | |
2320 | ||
4c9c5359 SS |
2321 | if (gfp_mask & __GFP_NORETRY) |
2322 | return CHARGE_NOMEM; | |
2323 | ||
5660048c | 2324 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2325 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2326 | return CHARGE_RETRY; |
4b534334 | 2327 | /* |
19942822 JW |
2328 | * Even though the limit is exceeded at this point, reclaim |
2329 | * may have been able to free some pages. Retry the charge | |
2330 | * before killing the task. | |
2331 | * | |
2332 | * Only for regular pages, though: huge pages are rather | |
2333 | * unlikely to succeed so close to the limit, and we fall back | |
2334 | * to regular pages anyway in case of failure. | |
4b534334 | 2335 | */ |
4c9c5359 | 2336 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2337 | return CHARGE_RETRY; |
2338 | ||
2339 | /* | |
2340 | * At task move, charge accounts can be doubly counted. So, it's | |
2341 | * better to wait until the end of task_move if something is going on. | |
2342 | */ | |
2343 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2344 | return CHARGE_RETRY; | |
2345 | ||
2346 | /* If we don't need to call oom-killer at el, return immediately */ | |
2347 | if (!oom_check) | |
2348 | return CHARGE_NOMEM; | |
2349 | /* check OOM */ | |
e845e199 | 2350 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2351 | return CHARGE_OOM_DIE; |
2352 | ||
2353 | return CHARGE_RETRY; | |
2354 | } | |
2355 | ||
f817ed48 | 2356 | /* |
38c5d72f KH |
2357 | * __mem_cgroup_try_charge() does |
2358 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2359 | * 2. update res_counter | |
2360 | * 3. call memory reclaim if necessary. | |
2361 | * | |
2362 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2363 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2364 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2365 | * as possible without any hazards. 2: all pages should have a valid | |
2366 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2367 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2368 | * | |
2369 | * So __mem_cgroup_try_charge() will return | |
2370 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2371 | * -ENOMEM ... charge failure because of resource limits. | |
2372 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2373 | * | |
2374 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2375 | * the oom-killer can be invoked. | |
8a9f3ccd | 2376 | */ |
f817ed48 | 2377 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2378 | gfp_t gfp_mask, |
7ec99d62 | 2379 | unsigned int nr_pages, |
c0ff4b85 | 2380 | struct mem_cgroup **ptr, |
7ec99d62 | 2381 | bool oom) |
8a9f3ccd | 2382 | { |
7ec99d62 | 2383 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2384 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2385 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2386 | int ret; |
a636b327 | 2387 | |
867578cb KH |
2388 | /* |
2389 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2390 | * in system level. So, allow to go ahead dying process in addition to | |
2391 | * MEMDIE process. | |
2392 | */ | |
2393 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2394 | || fatal_signal_pending(current))) | |
2395 | goto bypass; | |
a636b327 | 2396 | |
8a9f3ccd | 2397 | /* |
3be91277 HD |
2398 | * We always charge the cgroup the mm_struct belongs to. |
2399 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2400 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2401 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2402 | */ |
c0ff4b85 | 2403 | if (!*ptr && !mm) |
38c5d72f | 2404 | *ptr = root_mem_cgroup; |
f75ca962 | 2405 | again: |
c0ff4b85 R |
2406 | if (*ptr) { /* css should be a valid one */ |
2407 | memcg = *ptr; | |
c0ff4b85 | 2408 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2409 | goto done; |
a0956d54 | 2410 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2411 | goto done; |
c0ff4b85 | 2412 | css_get(&memcg->css); |
4b534334 | 2413 | } else { |
f75ca962 | 2414 | struct task_struct *p; |
54595fe2 | 2415 | |
f75ca962 KH |
2416 | rcu_read_lock(); |
2417 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2418 | /* |
ebb76ce1 | 2419 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2420 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2421 | * race with swapoff. Then, we have small risk of mis-accouning. |
2422 | * But such kind of mis-account by race always happens because | |
2423 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2424 | * small race, here. | |
2425 | * (*) swapoff at el will charge against mm-struct not against | |
2426 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2427 | */ |
c0ff4b85 | 2428 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2429 | if (!memcg) |
2430 | memcg = root_mem_cgroup; | |
2431 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2432 | rcu_read_unlock(); |
2433 | goto done; | |
2434 | } | |
a0956d54 | 2435 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2436 | /* |
2437 | * It seems dagerous to access memcg without css_get(). | |
2438 | * But considering how consume_stok works, it's not | |
2439 | * necessary. If consume_stock success, some charges | |
2440 | * from this memcg are cached on this cpu. So, we | |
2441 | * don't need to call css_get()/css_tryget() before | |
2442 | * calling consume_stock(). | |
2443 | */ | |
2444 | rcu_read_unlock(); | |
2445 | goto done; | |
2446 | } | |
2447 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2448 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2449 | rcu_read_unlock(); |
2450 | goto again; | |
2451 | } | |
2452 | rcu_read_unlock(); | |
2453 | } | |
8a9f3ccd | 2454 | |
4b534334 KH |
2455 | do { |
2456 | bool oom_check; | |
7a81b88c | 2457 | |
4b534334 | 2458 | /* If killed, bypass charge */ |
f75ca962 | 2459 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2460 | css_put(&memcg->css); |
4b534334 | 2461 | goto bypass; |
f75ca962 | 2462 | } |
6d61ef40 | 2463 | |
4b534334 KH |
2464 | oom_check = false; |
2465 | if (oom && !nr_oom_retries) { | |
2466 | oom_check = true; | |
2467 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2468 | } |
66e1707b | 2469 | |
4c9c5359 SS |
2470 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages, |
2471 | oom_check); | |
4b534334 KH |
2472 | switch (ret) { |
2473 | case CHARGE_OK: | |
2474 | break; | |
2475 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2476 | batch = nr_pages; |
c0ff4b85 R |
2477 | css_put(&memcg->css); |
2478 | memcg = NULL; | |
f75ca962 | 2479 | goto again; |
4b534334 | 2480 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2481 | css_put(&memcg->css); |
4b534334 KH |
2482 | goto nomem; |
2483 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2484 | if (!oom) { |
c0ff4b85 | 2485 | css_put(&memcg->css); |
867578cb | 2486 | goto nomem; |
f75ca962 | 2487 | } |
4b534334 KH |
2488 | /* If oom, we never return -ENOMEM */ |
2489 | nr_oom_retries--; | |
2490 | break; | |
2491 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2492 | css_put(&memcg->css); |
867578cb | 2493 | goto bypass; |
66e1707b | 2494 | } |
4b534334 KH |
2495 | } while (ret != CHARGE_OK); |
2496 | ||
7ec99d62 | 2497 | if (batch > nr_pages) |
c0ff4b85 R |
2498 | refill_stock(memcg, batch - nr_pages); |
2499 | css_put(&memcg->css); | |
0c3e73e8 | 2500 | done: |
c0ff4b85 | 2501 | *ptr = memcg; |
7a81b88c KH |
2502 | return 0; |
2503 | nomem: | |
c0ff4b85 | 2504 | *ptr = NULL; |
7a81b88c | 2505 | return -ENOMEM; |
867578cb | 2506 | bypass: |
38c5d72f KH |
2507 | *ptr = root_mem_cgroup; |
2508 | return -EINTR; | |
7a81b88c | 2509 | } |
8a9f3ccd | 2510 | |
a3032a2c DN |
2511 | /* |
2512 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2513 | * This function is for that and do uncharge, put css's refcnt. | |
2514 | * gotten by try_charge(). | |
2515 | */ | |
c0ff4b85 | 2516 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2517 | unsigned int nr_pages) |
a3032a2c | 2518 | { |
c0ff4b85 | 2519 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2520 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2521 | ||
c0ff4b85 | 2522 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2523 | if (do_swap_account) |
c0ff4b85 | 2524 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2525 | } |
854ffa8d DN |
2526 | } |
2527 | ||
d01dd17f KH |
2528 | /* |
2529 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2530 | * This is useful when moving usage to parent cgroup. | |
2531 | */ | |
2532 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2533 | unsigned int nr_pages) | |
2534 | { | |
2535 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2536 | ||
2537 | if (mem_cgroup_is_root(memcg)) | |
2538 | return; | |
2539 | ||
2540 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2541 | if (do_swap_account) | |
2542 | res_counter_uncharge_until(&memcg->memsw, | |
2543 | memcg->memsw.parent, bytes); | |
2544 | } | |
2545 | ||
a3b2d692 KH |
2546 | /* |
2547 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2548 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2549 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2550 | * called against removed memcg.) | |
a3b2d692 KH |
2551 | */ |
2552 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2553 | { | |
2554 | struct cgroup_subsys_state *css; | |
2555 | ||
2556 | /* ID 0 is unused ID */ | |
2557 | if (!id) | |
2558 | return NULL; | |
2559 | css = css_lookup(&mem_cgroup_subsys, id); | |
2560 | if (!css) | |
2561 | return NULL; | |
b2145145 | 2562 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2563 | } |
2564 | ||
e42d9d5d | 2565 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2566 | { |
c0ff4b85 | 2567 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2568 | struct page_cgroup *pc; |
a3b2d692 | 2569 | unsigned short id; |
b5a84319 KH |
2570 | swp_entry_t ent; |
2571 | ||
3c776e64 DN |
2572 | VM_BUG_ON(!PageLocked(page)); |
2573 | ||
3c776e64 | 2574 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2575 | lock_page_cgroup(pc); |
a3b2d692 | 2576 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2577 | memcg = pc->mem_cgroup; |
2578 | if (memcg && !css_tryget(&memcg->css)) | |
2579 | memcg = NULL; | |
e42d9d5d | 2580 | } else if (PageSwapCache(page)) { |
3c776e64 | 2581 | ent.val = page_private(page); |
9fb4b7cc | 2582 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2583 | rcu_read_lock(); |
c0ff4b85 R |
2584 | memcg = mem_cgroup_lookup(id); |
2585 | if (memcg && !css_tryget(&memcg->css)) | |
2586 | memcg = NULL; | |
a3b2d692 | 2587 | rcu_read_unlock(); |
3c776e64 | 2588 | } |
c0bd3f63 | 2589 | unlock_page_cgroup(pc); |
c0ff4b85 | 2590 | return memcg; |
b5a84319 KH |
2591 | } |
2592 | ||
c0ff4b85 | 2593 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2594 | struct page *page, |
7ec99d62 | 2595 | unsigned int nr_pages, |
9ce70c02 HD |
2596 | enum charge_type ctype, |
2597 | bool lrucare) | |
7a81b88c | 2598 | { |
ce587e65 | 2599 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2600 | struct zone *uninitialized_var(zone); |
fa9add64 | 2601 | struct lruvec *lruvec; |
9ce70c02 | 2602 | bool was_on_lru = false; |
b2402857 | 2603 | bool anon; |
9ce70c02 | 2604 | |
ca3e0214 | 2605 | lock_page_cgroup(pc); |
90deb788 | 2606 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2607 | /* |
2608 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2609 | * accessed by any other context at this point. | |
2610 | */ | |
9ce70c02 HD |
2611 | |
2612 | /* | |
2613 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2614 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2615 | */ | |
2616 | if (lrucare) { | |
2617 | zone = page_zone(page); | |
2618 | spin_lock_irq(&zone->lru_lock); | |
2619 | if (PageLRU(page)) { | |
fa9add64 | 2620 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2621 | ClearPageLRU(page); |
fa9add64 | 2622 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2623 | was_on_lru = true; |
2624 | } | |
2625 | } | |
2626 | ||
c0ff4b85 | 2627 | pc->mem_cgroup = memcg; |
261fb61a KH |
2628 | /* |
2629 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2630 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2631 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2632 | * before USED bit, we need memory barrier here. | |
2633 | * See mem_cgroup_add_lru_list(), etc. | |
2634 | */ | |
08e552c6 | 2635 | smp_wmb(); |
b2402857 | 2636 | SetPageCgroupUsed(pc); |
3be91277 | 2637 | |
9ce70c02 HD |
2638 | if (lrucare) { |
2639 | if (was_on_lru) { | |
fa9add64 | 2640 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2641 | VM_BUG_ON(PageLRU(page)); |
2642 | SetPageLRU(page); | |
fa9add64 | 2643 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2644 | } |
2645 | spin_unlock_irq(&zone->lru_lock); | |
2646 | } | |
2647 | ||
41326c17 | 2648 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2649 | anon = true; |
2650 | else | |
2651 | anon = false; | |
2652 | ||
2653 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2654 | unlock_page_cgroup(pc); |
9ce70c02 | 2655 | |
430e4863 KH |
2656 | /* |
2657 | * "charge_statistics" updated event counter. Then, check it. | |
2658 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2659 | * if they exceeds softlimit. | |
2660 | */ | |
c0ff4b85 | 2661 | memcg_check_events(memcg, page); |
7a81b88c | 2662 | } |
66e1707b | 2663 | |
ca3e0214 KH |
2664 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2665 | ||
a0db00fc | 2666 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
2667 | /* |
2668 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
2669 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
2670 | * charge/uncharge will be never happen and move_account() is done under | |
2671 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 2672 | */ |
e94c8a9c | 2673 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
2674 | { |
2675 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
2676 | struct page_cgroup *pc; |
2677 | int i; | |
ca3e0214 | 2678 | |
3d37c4a9 KH |
2679 | if (mem_cgroup_disabled()) |
2680 | return; | |
e94c8a9c KH |
2681 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
2682 | pc = head_pc + i; | |
2683 | pc->mem_cgroup = head_pc->mem_cgroup; | |
2684 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
2685 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2686 | } | |
ca3e0214 | 2687 | } |
12d27107 | 2688 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2689 | |
f817ed48 | 2690 | /** |
de3638d9 | 2691 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2692 | * @page: the page |
7ec99d62 | 2693 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2694 | * @pc: page_cgroup of the page. |
2695 | * @from: mem_cgroup which the page is moved from. | |
2696 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
2697 | * | |
2698 | * The caller must confirm following. | |
08e552c6 | 2699 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2700 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2701 | * |
2f3479b1 KH |
2702 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
2703 | * from old cgroup. | |
f817ed48 | 2704 | */ |
7ec99d62 JW |
2705 | static int mem_cgroup_move_account(struct page *page, |
2706 | unsigned int nr_pages, | |
2707 | struct page_cgroup *pc, | |
2708 | struct mem_cgroup *from, | |
2f3479b1 | 2709 | struct mem_cgroup *to) |
f817ed48 | 2710 | { |
de3638d9 JW |
2711 | unsigned long flags; |
2712 | int ret; | |
b2402857 | 2713 | bool anon = PageAnon(page); |
987eba66 | 2714 | |
f817ed48 | 2715 | VM_BUG_ON(from == to); |
5564e88b | 2716 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2717 | /* |
2718 | * The page is isolated from LRU. So, collapse function | |
2719 | * will not handle this page. But page splitting can happen. | |
2720 | * Do this check under compound_page_lock(). The caller should | |
2721 | * hold it. | |
2722 | */ | |
2723 | ret = -EBUSY; | |
7ec99d62 | 2724 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2725 | goto out; |
2726 | ||
2727 | lock_page_cgroup(pc); | |
2728 | ||
2729 | ret = -EINVAL; | |
2730 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2731 | goto unlock; | |
2732 | ||
312734c0 | 2733 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 2734 | |
2ff76f11 | 2735 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
2736 | /* Update mapped_file data for mem_cgroup */ |
2737 | preempt_disable(); | |
2738 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2739 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2740 | preempt_enable(); | |
d69b042f | 2741 | } |
b2402857 | 2742 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
d69b042f | 2743 | |
854ffa8d | 2744 | /* caller should have done css_get */ |
08e552c6 | 2745 | pc->mem_cgroup = to; |
b2402857 | 2746 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
312734c0 | 2747 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
2748 | ret = 0; |
2749 | unlock: | |
57f9fd7d | 2750 | unlock_page_cgroup(pc); |
d2265e6f KH |
2751 | /* |
2752 | * check events | |
2753 | */ | |
5564e88b JW |
2754 | memcg_check_events(to, page); |
2755 | memcg_check_events(from, page); | |
de3638d9 | 2756 | out: |
f817ed48 KH |
2757 | return ret; |
2758 | } | |
2759 | ||
2ef37d3f MH |
2760 | /** |
2761 | * mem_cgroup_move_parent - moves page to the parent group | |
2762 | * @page: the page to move | |
2763 | * @pc: page_cgroup of the page | |
2764 | * @child: page's cgroup | |
2765 | * | |
2766 | * move charges to its parent or the root cgroup if the group has no | |
2767 | * parent (aka use_hierarchy==0). | |
2768 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
2769 | * mem_cgroup_move_account fails) the failure is always temporary and | |
2770 | * it signals a race with a page removal/uncharge or migration. In the | |
2771 | * first case the page is on the way out and it will vanish from the LRU | |
2772 | * on the next attempt and the call should be retried later. | |
2773 | * Isolation from the LRU fails only if page has been isolated from | |
2774 | * the LRU since we looked at it and that usually means either global | |
2775 | * reclaim or migration going on. The page will either get back to the | |
2776 | * LRU or vanish. | |
2777 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
2778 | * (!PageCgroupUsed) or moved to a different group. The page will | |
2779 | * disappear in the next attempt. | |
f817ed48 | 2780 | */ |
5564e88b JW |
2781 | static int mem_cgroup_move_parent(struct page *page, |
2782 | struct page_cgroup *pc, | |
6068bf01 | 2783 | struct mem_cgroup *child) |
f817ed48 | 2784 | { |
f817ed48 | 2785 | struct mem_cgroup *parent; |
7ec99d62 | 2786 | unsigned int nr_pages; |
4be4489f | 2787 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2788 | int ret; |
2789 | ||
d8423011 | 2790 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 2791 | |
57f9fd7d DN |
2792 | ret = -EBUSY; |
2793 | if (!get_page_unless_zero(page)) | |
2794 | goto out; | |
2795 | if (isolate_lru_page(page)) | |
2796 | goto put; | |
52dbb905 | 2797 | |
7ec99d62 | 2798 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2799 | |
cc926f78 KH |
2800 | parent = parent_mem_cgroup(child); |
2801 | /* | |
2802 | * If no parent, move charges to root cgroup. | |
2803 | */ | |
2804 | if (!parent) | |
2805 | parent = root_mem_cgroup; | |
f817ed48 | 2806 | |
2ef37d3f MH |
2807 | if (nr_pages > 1) { |
2808 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 2809 | flags = compound_lock_irqsave(page); |
2ef37d3f | 2810 | } |
987eba66 | 2811 | |
cc926f78 | 2812 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 2813 | pc, child, parent); |
cc926f78 KH |
2814 | if (!ret) |
2815 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 2816 | |
7ec99d62 | 2817 | if (nr_pages > 1) |
987eba66 | 2818 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 2819 | putback_lru_page(page); |
57f9fd7d | 2820 | put: |
40d58138 | 2821 | put_page(page); |
57f9fd7d | 2822 | out: |
f817ed48 KH |
2823 | return ret; |
2824 | } | |
2825 | ||
7a81b88c KH |
2826 | /* |
2827 | * Charge the memory controller for page usage. | |
2828 | * Return | |
2829 | * 0 if the charge was successful | |
2830 | * < 0 if the cgroup is over its limit | |
2831 | */ | |
2832 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2833 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2834 | { |
c0ff4b85 | 2835 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 2836 | unsigned int nr_pages = 1; |
8493ae43 | 2837 | bool oom = true; |
7a81b88c | 2838 | int ret; |
ec168510 | 2839 | |
37c2ac78 | 2840 | if (PageTransHuge(page)) { |
7ec99d62 | 2841 | nr_pages <<= compound_order(page); |
37c2ac78 | 2842 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2843 | /* |
2844 | * Never OOM-kill a process for a huge page. The | |
2845 | * fault handler will fall back to regular pages. | |
2846 | */ | |
2847 | oom = false; | |
37c2ac78 | 2848 | } |
7a81b88c | 2849 | |
c0ff4b85 | 2850 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 2851 | if (ret == -ENOMEM) |
7a81b88c | 2852 | return ret; |
ce587e65 | 2853 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 2854 | return 0; |
8a9f3ccd BS |
2855 | } |
2856 | ||
7a81b88c KH |
2857 | int mem_cgroup_newpage_charge(struct page *page, |
2858 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2859 | { |
f8d66542 | 2860 | if (mem_cgroup_disabled()) |
cede86ac | 2861 | return 0; |
7a0524cf JW |
2862 | VM_BUG_ON(page_mapped(page)); |
2863 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
2864 | VM_BUG_ON(!mm); | |
217bc319 | 2865 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 2866 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
2867 | } |
2868 | ||
54595fe2 KH |
2869 | /* |
2870 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2871 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2872 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2873 | * "commit()" or removed by "cancel()" |
2874 | */ | |
0435a2fd JW |
2875 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2876 | struct page *page, | |
2877 | gfp_t mask, | |
2878 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 2879 | { |
c0ff4b85 | 2880 | struct mem_cgroup *memcg; |
90deb788 | 2881 | struct page_cgroup *pc; |
54595fe2 | 2882 | int ret; |
8c7c6e34 | 2883 | |
90deb788 JW |
2884 | pc = lookup_page_cgroup(page); |
2885 | /* | |
2886 | * Every swap fault against a single page tries to charge the | |
2887 | * page, bail as early as possible. shmem_unuse() encounters | |
2888 | * already charged pages, too. The USED bit is protected by | |
2889 | * the page lock, which serializes swap cache removal, which | |
2890 | * in turn serializes uncharging. | |
2891 | */ | |
2892 | if (PageCgroupUsed(pc)) | |
2893 | return 0; | |
8c7c6e34 KH |
2894 | if (!do_swap_account) |
2895 | goto charge_cur_mm; | |
c0ff4b85 R |
2896 | memcg = try_get_mem_cgroup_from_page(page); |
2897 | if (!memcg) | |
54595fe2 | 2898 | goto charge_cur_mm; |
72835c86 JW |
2899 | *memcgp = memcg; |
2900 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 2901 | css_put(&memcg->css); |
38c5d72f KH |
2902 | if (ret == -EINTR) |
2903 | ret = 0; | |
54595fe2 | 2904 | return ret; |
8c7c6e34 | 2905 | charge_cur_mm: |
38c5d72f KH |
2906 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
2907 | if (ret == -EINTR) | |
2908 | ret = 0; | |
2909 | return ret; | |
8c7c6e34 KH |
2910 | } |
2911 | ||
0435a2fd JW |
2912 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
2913 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
2914 | { | |
2915 | *memcgp = NULL; | |
2916 | if (mem_cgroup_disabled()) | |
2917 | return 0; | |
bdf4f4d2 JW |
2918 | /* |
2919 | * A racing thread's fault, or swapoff, may have already | |
2920 | * updated the pte, and even removed page from swap cache: in | |
2921 | * those cases unuse_pte()'s pte_same() test will fail; but | |
2922 | * there's also a KSM case which does need to charge the page. | |
2923 | */ | |
2924 | if (!PageSwapCache(page)) { | |
2925 | int ret; | |
2926 | ||
2927 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
2928 | if (ret == -EINTR) | |
2929 | ret = 0; | |
2930 | return ret; | |
2931 | } | |
0435a2fd JW |
2932 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
2933 | } | |
2934 | ||
827a03d2 JW |
2935 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
2936 | { | |
2937 | if (mem_cgroup_disabled()) | |
2938 | return; | |
2939 | if (!memcg) | |
2940 | return; | |
2941 | __mem_cgroup_cancel_charge(memcg, 1); | |
2942 | } | |
2943 | ||
83aae4c7 | 2944 | static void |
72835c86 | 2945 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 2946 | enum charge_type ctype) |
7a81b88c | 2947 | { |
f8d66542 | 2948 | if (mem_cgroup_disabled()) |
7a81b88c | 2949 | return; |
72835c86 | 2950 | if (!memcg) |
7a81b88c | 2951 | return; |
5a6475a4 | 2952 | |
ce587e65 | 2953 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
2954 | /* |
2955 | * Now swap is on-memory. This means this page may be | |
2956 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2957 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2958 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2959 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2960 | */ |
03f3c433 | 2961 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2962 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 2963 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 2964 | } |
7a81b88c KH |
2965 | } |
2966 | ||
72835c86 JW |
2967 | void mem_cgroup_commit_charge_swapin(struct page *page, |
2968 | struct mem_cgroup *memcg) | |
83aae4c7 | 2969 | { |
72835c86 | 2970 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 2971 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
2972 | } |
2973 | ||
827a03d2 JW |
2974 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2975 | gfp_t gfp_mask) | |
7a81b88c | 2976 | { |
827a03d2 JW |
2977 | struct mem_cgroup *memcg = NULL; |
2978 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2979 | int ret; | |
2980 | ||
f8d66542 | 2981 | if (mem_cgroup_disabled()) |
827a03d2 JW |
2982 | return 0; |
2983 | if (PageCompound(page)) | |
2984 | return 0; | |
2985 | ||
827a03d2 JW |
2986 | if (!PageSwapCache(page)) |
2987 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
2988 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
2989 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
2990 | gfp_mask, &memcg); | |
827a03d2 JW |
2991 | if (!ret) |
2992 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
2993 | } | |
2994 | return ret; | |
7a81b88c KH |
2995 | } |
2996 | ||
c0ff4b85 | 2997 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
2998 | unsigned int nr_pages, |
2999 | const enum charge_type ctype) | |
569b846d KH |
3000 | { |
3001 | struct memcg_batch_info *batch = NULL; | |
3002 | bool uncharge_memsw = true; | |
7ec99d62 | 3003 | |
569b846d KH |
3004 | /* If swapout, usage of swap doesn't decrease */ |
3005 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
3006 | uncharge_memsw = false; | |
569b846d KH |
3007 | |
3008 | batch = ¤t->memcg_batch; | |
3009 | /* | |
3010 | * In usual, we do css_get() when we remember memcg pointer. | |
3011 | * But in this case, we keep res->usage until end of a series of | |
3012 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
3013 | */ | |
3014 | if (!batch->memcg) | |
c0ff4b85 | 3015 | batch->memcg = memcg; |
3c11ecf4 KH |
3016 | /* |
3017 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 3018 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
3019 | * the same cgroup and we have chance to coalesce uncharges. |
3020 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
3021 | * because we want to do uncharge as soon as possible. | |
3022 | */ | |
3023 | ||
3024 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
3025 | goto direct_uncharge; | |
3026 | ||
7ec99d62 | 3027 | if (nr_pages > 1) |
ec168510 AA |
3028 | goto direct_uncharge; |
3029 | ||
569b846d KH |
3030 | /* |
3031 | * In typical case, batch->memcg == mem. This means we can | |
3032 | * merge a series of uncharges to an uncharge of res_counter. | |
3033 | * If not, we uncharge res_counter ony by one. | |
3034 | */ | |
c0ff4b85 | 3035 | if (batch->memcg != memcg) |
569b846d KH |
3036 | goto direct_uncharge; |
3037 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 3038 | batch->nr_pages++; |
569b846d | 3039 | if (uncharge_memsw) |
7ffd4ca7 | 3040 | batch->memsw_nr_pages++; |
569b846d KH |
3041 | return; |
3042 | direct_uncharge: | |
c0ff4b85 | 3043 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 3044 | if (uncharge_memsw) |
c0ff4b85 R |
3045 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
3046 | if (unlikely(batch->memcg != memcg)) | |
3047 | memcg_oom_recover(memcg); | |
569b846d | 3048 | } |
7a81b88c | 3049 | |
8a9f3ccd | 3050 | /* |
69029cd5 | 3051 | * uncharge if !page_mapped(page) |
8a9f3ccd | 3052 | */ |
8c7c6e34 | 3053 | static struct mem_cgroup * |
0030f535 JW |
3054 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
3055 | bool end_migration) | |
8a9f3ccd | 3056 | { |
c0ff4b85 | 3057 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
3058 | unsigned int nr_pages = 1; |
3059 | struct page_cgroup *pc; | |
b2402857 | 3060 | bool anon; |
8a9f3ccd | 3061 | |
f8d66542 | 3062 | if (mem_cgroup_disabled()) |
8c7c6e34 | 3063 | return NULL; |
4077960e | 3064 | |
0c59b89c | 3065 | VM_BUG_ON(PageSwapCache(page)); |
d13d1443 | 3066 | |
37c2ac78 | 3067 | if (PageTransHuge(page)) { |
7ec99d62 | 3068 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
3069 | VM_BUG_ON(!PageTransHuge(page)); |
3070 | } | |
8697d331 | 3071 | /* |
3c541e14 | 3072 | * Check if our page_cgroup is valid |
8697d331 | 3073 | */ |
52d4b9ac | 3074 | pc = lookup_page_cgroup(page); |
cfa44946 | 3075 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 3076 | return NULL; |
b9c565d5 | 3077 | |
52d4b9ac | 3078 | lock_page_cgroup(pc); |
d13d1443 | 3079 | |
c0ff4b85 | 3080 | memcg = pc->mem_cgroup; |
8c7c6e34 | 3081 | |
d13d1443 KH |
3082 | if (!PageCgroupUsed(pc)) |
3083 | goto unlock_out; | |
3084 | ||
b2402857 KH |
3085 | anon = PageAnon(page); |
3086 | ||
d13d1443 | 3087 | switch (ctype) { |
41326c17 | 3088 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
3089 | /* |
3090 | * Generally PageAnon tells if it's the anon statistics to be | |
3091 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
3092 | * used before page reached the stage of being marked PageAnon. | |
3093 | */ | |
b2402857 KH |
3094 | anon = true; |
3095 | /* fallthrough */ | |
8a9478ca | 3096 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 3097 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
3098 | if (page_mapped(page)) |
3099 | goto unlock_out; | |
3100 | /* | |
3101 | * Pages under migration may not be uncharged. But | |
3102 | * end_migration() /must/ be the one uncharging the | |
3103 | * unused post-migration page and so it has to call | |
3104 | * here with the migration bit still set. See the | |
3105 | * res_counter handling below. | |
3106 | */ | |
3107 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
3108 | goto unlock_out; |
3109 | break; | |
3110 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
3111 | if (!PageAnon(page)) { /* Shared memory */ | |
3112 | if (page->mapping && !page_is_file_cache(page)) | |
3113 | goto unlock_out; | |
3114 | } else if (page_mapped(page)) /* Anon */ | |
3115 | goto unlock_out; | |
3116 | break; | |
3117 | default: | |
3118 | break; | |
52d4b9ac | 3119 | } |
d13d1443 | 3120 | |
b2402857 | 3121 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 3122 | |
52d4b9ac | 3123 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3124 | /* |
3125 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3126 | * freed from LRU. This is safe because uncharged page is expected not | |
3127 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3128 | * special functions. | |
3129 | */ | |
b9c565d5 | 3130 | |
52d4b9ac | 3131 | unlock_page_cgroup(pc); |
f75ca962 | 3132 | /* |
c0ff4b85 | 3133 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
3134 | * will never be freed. |
3135 | */ | |
c0ff4b85 | 3136 | memcg_check_events(memcg, page); |
f75ca962 | 3137 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
3138 | mem_cgroup_swap_statistics(memcg, true); |
3139 | mem_cgroup_get(memcg); | |
f75ca962 | 3140 | } |
0030f535 JW |
3141 | /* |
3142 | * Migration does not charge the res_counter for the | |
3143 | * replacement page, so leave it alone when phasing out the | |
3144 | * page that is unused after the migration. | |
3145 | */ | |
3146 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 3147 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 3148 | |
c0ff4b85 | 3149 | return memcg; |
d13d1443 KH |
3150 | |
3151 | unlock_out: | |
3152 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3153 | return NULL; |
3c541e14 BS |
3154 | } |
3155 | ||
69029cd5 KH |
3156 | void mem_cgroup_uncharge_page(struct page *page) |
3157 | { | |
52d4b9ac KH |
3158 | /* early check. */ |
3159 | if (page_mapped(page)) | |
3160 | return; | |
40f23a21 | 3161 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
0c59b89c JW |
3162 | if (PageSwapCache(page)) |
3163 | return; | |
0030f535 | 3164 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
3165 | } |
3166 | ||
3167 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3168 | { | |
3169 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3170 | VM_BUG_ON(page->mapping); |
0030f535 | 3171 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
3172 | } |
3173 | ||
569b846d KH |
3174 | /* |
3175 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3176 | * In that cases, pages are freed continuously and we can expect pages | |
3177 | * are in the same memcg. All these calls itself limits the number of | |
3178 | * pages freed at once, then uncharge_start/end() is called properly. | |
3179 | * This may be called prural(2) times in a context, | |
3180 | */ | |
3181 | ||
3182 | void mem_cgroup_uncharge_start(void) | |
3183 | { | |
3184 | current->memcg_batch.do_batch++; | |
3185 | /* We can do nest. */ | |
3186 | if (current->memcg_batch.do_batch == 1) { | |
3187 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3188 | current->memcg_batch.nr_pages = 0; |
3189 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3190 | } |
3191 | } | |
3192 | ||
3193 | void mem_cgroup_uncharge_end(void) | |
3194 | { | |
3195 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3196 | ||
3197 | if (!batch->do_batch) | |
3198 | return; | |
3199 | ||
3200 | batch->do_batch--; | |
3201 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3202 | return; | |
3203 | ||
3204 | if (!batch->memcg) | |
3205 | return; | |
3206 | /* | |
3207 | * This "batch->memcg" is valid without any css_get/put etc... | |
3208 | * bacause we hide charges behind us. | |
3209 | */ | |
7ffd4ca7 JW |
3210 | if (batch->nr_pages) |
3211 | res_counter_uncharge(&batch->memcg->res, | |
3212 | batch->nr_pages * PAGE_SIZE); | |
3213 | if (batch->memsw_nr_pages) | |
3214 | res_counter_uncharge(&batch->memcg->memsw, | |
3215 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3216 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3217 | /* forget this pointer (for sanity check) */ |
3218 | batch->memcg = NULL; | |
3219 | } | |
3220 | ||
e767e056 | 3221 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3222 | /* |
e767e056 | 3223 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3224 | * memcg information is recorded to swap_cgroup of "ent" |
3225 | */ | |
8a9478ca KH |
3226 | void |
3227 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3228 | { |
3229 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3230 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3231 | ||
3232 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3233 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3234 | ||
0030f535 | 3235 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 3236 | |
f75ca962 KH |
3237 | /* |
3238 | * record memcg information, if swapout && memcg != NULL, | |
3239 | * mem_cgroup_get() was called in uncharge(). | |
3240 | */ | |
3241 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3242 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3243 | } |
e767e056 | 3244 | #endif |
8c7c6e34 | 3245 | |
c255a458 | 3246 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
3247 | /* |
3248 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3249 | * uncharge "memsw" account. | |
3250 | */ | |
3251 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3252 | { |
8c7c6e34 | 3253 | struct mem_cgroup *memcg; |
a3b2d692 | 3254 | unsigned short id; |
8c7c6e34 KH |
3255 | |
3256 | if (!do_swap_account) | |
3257 | return; | |
3258 | ||
a3b2d692 KH |
3259 | id = swap_cgroup_record(ent, 0); |
3260 | rcu_read_lock(); | |
3261 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3262 | if (memcg) { |
a3b2d692 KH |
3263 | /* |
3264 | * We uncharge this because swap is freed. | |
3265 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3266 | */ | |
0c3e73e8 | 3267 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3268 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3269 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3270 | mem_cgroup_put(memcg); |
3271 | } | |
a3b2d692 | 3272 | rcu_read_unlock(); |
d13d1443 | 3273 | } |
02491447 DN |
3274 | |
3275 | /** | |
3276 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3277 | * @entry: swap entry to be moved | |
3278 | * @from: mem_cgroup which the entry is moved from | |
3279 | * @to: mem_cgroup which the entry is moved to | |
3280 | * | |
3281 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3282 | * as the mem_cgroup's id of @from. | |
3283 | * | |
3284 | * Returns 0 on success, -EINVAL on failure. | |
3285 | * | |
3286 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3287 | * both res and memsw, and called css_get(). | |
3288 | */ | |
3289 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3290 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3291 | { |
3292 | unsigned short old_id, new_id; | |
3293 | ||
3294 | old_id = css_id(&from->css); | |
3295 | new_id = css_id(&to->css); | |
3296 | ||
3297 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3298 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3299 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3300 | /* |
483c30b5 DN |
3301 | * This function is only called from task migration context now. |
3302 | * It postpones res_counter and refcount handling till the end | |
3303 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3304 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3305 | * because if the process that has been moved to @to does | |
3306 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3307 | */ |
02491447 | 3308 | mem_cgroup_get(to); |
02491447 DN |
3309 | return 0; |
3310 | } | |
3311 | return -EINVAL; | |
3312 | } | |
3313 | #else | |
3314 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3315 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3316 | { |
3317 | return -EINVAL; | |
3318 | } | |
8c7c6e34 | 3319 | #endif |
d13d1443 | 3320 | |
ae41be37 | 3321 | /* |
01b1ae63 KH |
3322 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3323 | * page belongs to. | |
ae41be37 | 3324 | */ |
0030f535 JW |
3325 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
3326 | struct mem_cgroup **memcgp) | |
ae41be37 | 3327 | { |
c0ff4b85 | 3328 | struct mem_cgroup *memcg = NULL; |
b32967ff | 3329 | unsigned int nr_pages = 1; |
7ec99d62 | 3330 | struct page_cgroup *pc; |
ac39cf8c | 3331 | enum charge_type ctype; |
8869b8f6 | 3332 | |
72835c86 | 3333 | *memcgp = NULL; |
56039efa | 3334 | |
f8d66542 | 3335 | if (mem_cgroup_disabled()) |
0030f535 | 3336 | return; |
4077960e | 3337 | |
b32967ff MG |
3338 | if (PageTransHuge(page)) |
3339 | nr_pages <<= compound_order(page); | |
3340 | ||
52d4b9ac KH |
3341 | pc = lookup_page_cgroup(page); |
3342 | lock_page_cgroup(pc); | |
3343 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
3344 | memcg = pc->mem_cgroup; |
3345 | css_get(&memcg->css); | |
ac39cf8c | 3346 | /* |
3347 | * At migrating an anonymous page, its mapcount goes down | |
3348 | * to 0 and uncharge() will be called. But, even if it's fully | |
3349 | * unmapped, migration may fail and this page has to be | |
3350 | * charged again. We set MIGRATION flag here and delay uncharge | |
3351 | * until end_migration() is called | |
3352 | * | |
3353 | * Corner Case Thinking | |
3354 | * A) | |
3355 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3356 | * while migration was ongoing. | |
3357 | * If unmap finds the old page, uncharge() of it will be delayed | |
3358 | * until end_migration(). If unmap finds a new page, it's | |
3359 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3360 | * finds swap_migration_entry, the new page will not be mapped | |
3361 | * and end_migration() will find it(mapcount==0). | |
3362 | * | |
3363 | * B) | |
3364 | * When the old page was mapped but migraion fails, the kernel | |
3365 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3366 | * if mapcount goes down to 0. We can do remap successfully | |
3367 | * without charging it again. | |
3368 | * | |
3369 | * C) | |
3370 | * The "old" page is under lock_page() until the end of | |
3371 | * migration, so, the old page itself will not be swapped-out. | |
3372 | * If the new page is swapped out before end_migraton, our | |
3373 | * hook to usual swap-out path will catch the event. | |
3374 | */ | |
3375 | if (PageAnon(page)) | |
3376 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3377 | } |
52d4b9ac | 3378 | unlock_page_cgroup(pc); |
ac39cf8c | 3379 | /* |
3380 | * If the page is not charged at this point, | |
3381 | * we return here. | |
3382 | */ | |
c0ff4b85 | 3383 | if (!memcg) |
0030f535 | 3384 | return; |
01b1ae63 | 3385 | |
72835c86 | 3386 | *memcgp = memcg; |
ac39cf8c | 3387 | /* |
3388 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3389 | * is called before end_migration, we can catch all events on this new | |
3390 | * page. In the case new page is migrated but not remapped, new page's | |
3391 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3392 | */ | |
ac39cf8c | 3393 | if (PageAnon(page)) |
41326c17 | 3394 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 3395 | else |
62ba7442 | 3396 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
3397 | /* |
3398 | * The page is committed to the memcg, but it's not actually | |
3399 | * charged to the res_counter since we plan on replacing the | |
3400 | * old one and only one page is going to be left afterwards. | |
3401 | */ | |
b32967ff | 3402 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 3403 | } |
8869b8f6 | 3404 | |
69029cd5 | 3405 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 3406 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 3407 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3408 | { |
ac39cf8c | 3409 | struct page *used, *unused; |
01b1ae63 | 3410 | struct page_cgroup *pc; |
b2402857 | 3411 | bool anon; |
01b1ae63 | 3412 | |
c0ff4b85 | 3413 | if (!memcg) |
01b1ae63 | 3414 | return; |
b25ed609 | 3415 | |
50de1dd9 | 3416 | if (!migration_ok) { |
ac39cf8c | 3417 | used = oldpage; |
3418 | unused = newpage; | |
01b1ae63 | 3419 | } else { |
ac39cf8c | 3420 | used = newpage; |
01b1ae63 KH |
3421 | unused = oldpage; |
3422 | } | |
0030f535 | 3423 | anon = PageAnon(used); |
7d188958 JW |
3424 | __mem_cgroup_uncharge_common(unused, |
3425 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
3426 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
3427 | true); | |
0030f535 | 3428 | css_put(&memcg->css); |
69029cd5 | 3429 | /* |
ac39cf8c | 3430 | * We disallowed uncharge of pages under migration because mapcount |
3431 | * of the page goes down to zero, temporarly. | |
3432 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3433 | */ |
ac39cf8c | 3434 | pc = lookup_page_cgroup(oldpage); |
3435 | lock_page_cgroup(pc); | |
3436 | ClearPageCgroupMigration(pc); | |
3437 | unlock_page_cgroup(pc); | |
ac39cf8c | 3438 | |
01b1ae63 | 3439 | /* |
ac39cf8c | 3440 | * If a page is a file cache, radix-tree replacement is very atomic |
3441 | * and we can skip this check. When it was an Anon page, its mapcount | |
3442 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3443 | * uncharged yet. There are several case but page->mapcount check | |
3444 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3445 | * check. (see prepare_charge() also) | |
69029cd5 | 3446 | */ |
b2402857 | 3447 | if (anon) |
ac39cf8c | 3448 | mem_cgroup_uncharge_page(used); |
ae41be37 | 3449 | } |
78fb7466 | 3450 | |
ab936cbc KH |
3451 | /* |
3452 | * At replace page cache, newpage is not under any memcg but it's on | |
3453 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
3454 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
3455 | */ | |
3456 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
3457 | struct page *newpage) | |
3458 | { | |
bde05d1c | 3459 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 3460 | struct page_cgroup *pc; |
ab936cbc | 3461 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
3462 | |
3463 | if (mem_cgroup_disabled()) | |
3464 | return; | |
3465 | ||
3466 | pc = lookup_page_cgroup(oldpage); | |
3467 | /* fix accounting on old pages */ | |
3468 | lock_page_cgroup(pc); | |
bde05d1c HD |
3469 | if (PageCgroupUsed(pc)) { |
3470 | memcg = pc->mem_cgroup; | |
3471 | mem_cgroup_charge_statistics(memcg, false, -1); | |
3472 | ClearPageCgroupUsed(pc); | |
3473 | } | |
ab936cbc KH |
3474 | unlock_page_cgroup(pc); |
3475 | ||
bde05d1c HD |
3476 | /* |
3477 | * When called from shmem_replace_page(), in some cases the | |
3478 | * oldpage has already been charged, and in some cases not. | |
3479 | */ | |
3480 | if (!memcg) | |
3481 | return; | |
ab936cbc KH |
3482 | /* |
3483 | * Even if newpage->mapping was NULL before starting replacement, | |
3484 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
3485 | * LRU while we overwrite pc->mem_cgroup. | |
3486 | */ | |
ce587e65 | 3487 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
3488 | } |
3489 | ||
f212ad7c DN |
3490 | #ifdef CONFIG_DEBUG_VM |
3491 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3492 | { | |
3493 | struct page_cgroup *pc; | |
3494 | ||
3495 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
3496 | /* |
3497 | * Can be NULL while feeding pages into the page allocator for | |
3498 | * the first time, i.e. during boot or memory hotplug; | |
3499 | * or when mem_cgroup_disabled(). | |
3500 | */ | |
f212ad7c DN |
3501 | if (likely(pc) && PageCgroupUsed(pc)) |
3502 | return pc; | |
3503 | return NULL; | |
3504 | } | |
3505 | ||
3506 | bool mem_cgroup_bad_page_check(struct page *page) | |
3507 | { | |
3508 | if (mem_cgroup_disabled()) | |
3509 | return false; | |
3510 | ||
3511 | return lookup_page_cgroup_used(page) != NULL; | |
3512 | } | |
3513 | ||
3514 | void mem_cgroup_print_bad_page(struct page *page) | |
3515 | { | |
3516 | struct page_cgroup *pc; | |
3517 | ||
3518 | pc = lookup_page_cgroup_used(page); | |
3519 | if (pc) { | |
90b3feae | 3520 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
f212ad7c | 3521 | pc, pc->flags, pc->mem_cgroup); |
f212ad7c DN |
3522 | } |
3523 | } | |
3524 | #endif | |
3525 | ||
8c7c6e34 KH |
3526 | static DEFINE_MUTEX(set_limit_mutex); |
3527 | ||
d38d2a75 | 3528 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3529 | unsigned long long val) |
628f4235 | 3530 | { |
81d39c20 | 3531 | int retry_count; |
3c11ecf4 | 3532 | u64 memswlimit, memlimit; |
628f4235 | 3533 | int ret = 0; |
81d39c20 KH |
3534 | int children = mem_cgroup_count_children(memcg); |
3535 | u64 curusage, oldusage; | |
3c11ecf4 | 3536 | int enlarge; |
81d39c20 KH |
3537 | |
3538 | /* | |
3539 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3540 | * is depends on callers. We set our retry-count to be function | |
3541 | * of # of children which we should visit in this loop. | |
3542 | */ | |
3543 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3544 | ||
3545 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3546 | |
3c11ecf4 | 3547 | enlarge = 0; |
8c7c6e34 | 3548 | while (retry_count) { |
628f4235 KH |
3549 | if (signal_pending(current)) { |
3550 | ret = -EINTR; | |
3551 | break; | |
3552 | } | |
8c7c6e34 KH |
3553 | /* |
3554 | * Rather than hide all in some function, I do this in | |
3555 | * open coded manner. You see what this really does. | |
aaad153e | 3556 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3557 | */ |
3558 | mutex_lock(&set_limit_mutex); | |
3559 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3560 | if (memswlimit < val) { | |
3561 | ret = -EINVAL; | |
3562 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3563 | break; |
3564 | } | |
3c11ecf4 KH |
3565 | |
3566 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3567 | if (memlimit < val) | |
3568 | enlarge = 1; | |
3569 | ||
8c7c6e34 | 3570 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3571 | if (!ret) { |
3572 | if (memswlimit == val) | |
3573 | memcg->memsw_is_minimum = true; | |
3574 | else | |
3575 | memcg->memsw_is_minimum = false; | |
3576 | } | |
8c7c6e34 KH |
3577 | mutex_unlock(&set_limit_mutex); |
3578 | ||
3579 | if (!ret) | |
3580 | break; | |
3581 | ||
5660048c JW |
3582 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3583 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
3584 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3585 | /* Usage is reduced ? */ | |
3586 | if (curusage >= oldusage) | |
3587 | retry_count--; | |
3588 | else | |
3589 | oldusage = curusage; | |
8c7c6e34 | 3590 | } |
3c11ecf4 KH |
3591 | if (!ret && enlarge) |
3592 | memcg_oom_recover(memcg); | |
14797e23 | 3593 | |
8c7c6e34 KH |
3594 | return ret; |
3595 | } | |
3596 | ||
338c8431 LZ |
3597 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3598 | unsigned long long val) | |
8c7c6e34 | 3599 | { |
81d39c20 | 3600 | int retry_count; |
3c11ecf4 | 3601 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3602 | int children = mem_cgroup_count_children(memcg); |
3603 | int ret = -EBUSY; | |
3c11ecf4 | 3604 | int enlarge = 0; |
8c7c6e34 | 3605 | |
81d39c20 KH |
3606 | /* see mem_cgroup_resize_res_limit */ |
3607 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3608 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3609 | while (retry_count) { |
3610 | if (signal_pending(current)) { | |
3611 | ret = -EINTR; | |
3612 | break; | |
3613 | } | |
3614 | /* | |
3615 | * Rather than hide all in some function, I do this in | |
3616 | * open coded manner. You see what this really does. | |
aaad153e | 3617 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
3618 | */ |
3619 | mutex_lock(&set_limit_mutex); | |
3620 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3621 | if (memlimit > val) { | |
3622 | ret = -EINVAL; | |
3623 | mutex_unlock(&set_limit_mutex); | |
3624 | break; | |
3625 | } | |
3c11ecf4 KH |
3626 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3627 | if (memswlimit < val) | |
3628 | enlarge = 1; | |
8c7c6e34 | 3629 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3630 | if (!ret) { |
3631 | if (memlimit == val) | |
3632 | memcg->memsw_is_minimum = true; | |
3633 | else | |
3634 | memcg->memsw_is_minimum = false; | |
3635 | } | |
8c7c6e34 KH |
3636 | mutex_unlock(&set_limit_mutex); |
3637 | ||
3638 | if (!ret) | |
3639 | break; | |
3640 | ||
5660048c JW |
3641 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
3642 | MEM_CGROUP_RECLAIM_NOSWAP | | |
3643 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3644 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3645 | /* Usage is reduced ? */ |
8c7c6e34 | 3646 | if (curusage >= oldusage) |
628f4235 | 3647 | retry_count--; |
81d39c20 KH |
3648 | else |
3649 | oldusage = curusage; | |
628f4235 | 3650 | } |
3c11ecf4 KH |
3651 | if (!ret && enlarge) |
3652 | memcg_oom_recover(memcg); | |
628f4235 KH |
3653 | return ret; |
3654 | } | |
3655 | ||
4e416953 | 3656 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3657 | gfp_t gfp_mask, |
3658 | unsigned long *total_scanned) | |
4e416953 BS |
3659 | { |
3660 | unsigned long nr_reclaimed = 0; | |
3661 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3662 | unsigned long reclaimed; | |
3663 | int loop = 0; | |
3664 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3665 | unsigned long long excess; |
0ae5e89c | 3666 | unsigned long nr_scanned; |
4e416953 BS |
3667 | |
3668 | if (order > 0) | |
3669 | return 0; | |
3670 | ||
00918b6a | 3671 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3672 | /* |
3673 | * This loop can run a while, specially if mem_cgroup's continuously | |
3674 | * keep exceeding their soft limit and putting the system under | |
3675 | * pressure | |
3676 | */ | |
3677 | do { | |
3678 | if (next_mz) | |
3679 | mz = next_mz; | |
3680 | else | |
3681 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3682 | if (!mz) | |
3683 | break; | |
3684 | ||
0ae5e89c | 3685 | nr_scanned = 0; |
d79154bb | 3686 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 3687 | gfp_mask, &nr_scanned); |
4e416953 | 3688 | nr_reclaimed += reclaimed; |
0ae5e89c | 3689 | *total_scanned += nr_scanned; |
4e416953 BS |
3690 | spin_lock(&mctz->lock); |
3691 | ||
3692 | /* | |
3693 | * If we failed to reclaim anything from this memory cgroup | |
3694 | * it is time to move on to the next cgroup | |
3695 | */ | |
3696 | next_mz = NULL; | |
3697 | if (!reclaimed) { | |
3698 | do { | |
3699 | /* | |
3700 | * Loop until we find yet another one. | |
3701 | * | |
3702 | * By the time we get the soft_limit lock | |
3703 | * again, someone might have aded the | |
3704 | * group back on the RB tree. Iterate to | |
3705 | * make sure we get a different mem. | |
3706 | * mem_cgroup_largest_soft_limit_node returns | |
3707 | * NULL if no other cgroup is present on | |
3708 | * the tree | |
3709 | */ | |
3710 | next_mz = | |
3711 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3712 | if (next_mz == mz) |
d79154bb | 3713 | css_put(&next_mz->memcg->css); |
39cc98f1 | 3714 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3715 | break; |
3716 | } while (1); | |
3717 | } | |
d79154bb HD |
3718 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
3719 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
3720 | /* |
3721 | * One school of thought says that we should not add | |
3722 | * back the node to the tree if reclaim returns 0. | |
3723 | * But our reclaim could return 0, simply because due | |
3724 | * to priority we are exposing a smaller subset of | |
3725 | * memory to reclaim from. Consider this as a longer | |
3726 | * term TODO. | |
3727 | */ | |
ef8745c1 | 3728 | /* If excess == 0, no tree ops */ |
d79154bb | 3729 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 3730 | spin_unlock(&mctz->lock); |
d79154bb | 3731 | css_put(&mz->memcg->css); |
4e416953 BS |
3732 | loop++; |
3733 | /* | |
3734 | * Could not reclaim anything and there are no more | |
3735 | * mem cgroups to try or we seem to be looping without | |
3736 | * reclaiming anything. | |
3737 | */ | |
3738 | if (!nr_reclaimed && | |
3739 | (next_mz == NULL || | |
3740 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3741 | break; | |
3742 | } while (!nr_reclaimed); | |
3743 | if (next_mz) | |
d79154bb | 3744 | css_put(&next_mz->memcg->css); |
4e416953 BS |
3745 | return nr_reclaimed; |
3746 | } | |
3747 | ||
2ef37d3f MH |
3748 | /** |
3749 | * mem_cgroup_force_empty_list - clears LRU of a group | |
3750 | * @memcg: group to clear | |
3751 | * @node: NUMA node | |
3752 | * @zid: zone id | |
3753 | * @lru: lru to to clear | |
3754 | * | |
3c935d18 | 3755 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
3756 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
3757 | * group. | |
cc847582 | 3758 | */ |
2ef37d3f | 3759 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 3760 | int node, int zid, enum lru_list lru) |
cc847582 | 3761 | { |
bea8c150 | 3762 | struct lruvec *lruvec; |
2ef37d3f | 3763 | unsigned long flags; |
072c56c1 | 3764 | struct list_head *list; |
925b7673 JW |
3765 | struct page *busy; |
3766 | struct zone *zone; | |
072c56c1 | 3767 | |
08e552c6 | 3768 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
3769 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
3770 | list = &lruvec->lists[lru]; | |
cc847582 | 3771 | |
f817ed48 | 3772 | busy = NULL; |
2ef37d3f | 3773 | do { |
925b7673 | 3774 | struct page_cgroup *pc; |
5564e88b JW |
3775 | struct page *page; |
3776 | ||
08e552c6 | 3777 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3778 | if (list_empty(list)) { |
08e552c6 | 3779 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3780 | break; |
f817ed48 | 3781 | } |
925b7673 JW |
3782 | page = list_entry(list->prev, struct page, lru); |
3783 | if (busy == page) { | |
3784 | list_move(&page->lru, list); | |
648bcc77 | 3785 | busy = NULL; |
08e552c6 | 3786 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3787 | continue; |
3788 | } | |
08e552c6 | 3789 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3790 | |
925b7673 | 3791 | pc = lookup_page_cgroup(page); |
5564e88b | 3792 | |
3c935d18 | 3793 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 3794 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 3795 | busy = page; |
f817ed48 KH |
3796 | cond_resched(); |
3797 | } else | |
3798 | busy = NULL; | |
2ef37d3f | 3799 | } while (!list_empty(list)); |
cc847582 KH |
3800 | } |
3801 | ||
3802 | /* | |
c26251f9 MH |
3803 | * make mem_cgroup's charge to be 0 if there is no task by moving |
3804 | * all the charges and pages to the parent. | |
cc847582 | 3805 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
3806 | * |
3807 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 3808 | */ |
ab5196c2 | 3809 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 3810 | { |
c26251f9 | 3811 | int node, zid; |
f817ed48 | 3812 | |
fce66477 | 3813 | do { |
52d4b9ac KH |
3814 | /* This is for making all *used* pages to be on LRU. */ |
3815 | lru_add_drain_all(); | |
c0ff4b85 | 3816 | drain_all_stock_sync(memcg); |
c0ff4b85 | 3817 | mem_cgroup_start_move(memcg); |
31aaea4a | 3818 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 3819 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
3820 | enum lru_list lru; |
3821 | for_each_lru(lru) { | |
2ef37d3f | 3822 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 3823 | node, zid, lru); |
f817ed48 | 3824 | } |
1ecaab2b | 3825 | } |
f817ed48 | 3826 | } |
c0ff4b85 R |
3827 | mem_cgroup_end_move(memcg); |
3828 | memcg_oom_recover(memcg); | |
52d4b9ac | 3829 | cond_resched(); |
f817ed48 | 3830 | |
2ef37d3f MH |
3831 | /* |
3832 | * This is a safety check because mem_cgroup_force_empty_list | |
3833 | * could have raced with mem_cgroup_replace_page_cache callers | |
3834 | * so the lru seemed empty but the page could have been added | |
3835 | * right after the check. RES_USAGE should be safe as we always | |
3836 | * charge before adding to the LRU. | |
3837 | */ | |
3838 | } while (res_counter_read_u64(&memcg->res, RES_USAGE) > 0); | |
c26251f9 MH |
3839 | } |
3840 | ||
3841 | /* | |
3842 | * Reclaims as many pages from the given memcg as possible and moves | |
3843 | * the rest to the parent. | |
3844 | * | |
3845 | * Caller is responsible for holding css reference for memcg. | |
3846 | */ | |
3847 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3848 | { | |
3849 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
3850 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 3851 | |
c1e862c1 | 3852 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
3853 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3854 | return -EBUSY; | |
3855 | ||
c1e862c1 KH |
3856 | /* we call try-to-free pages for make this cgroup empty */ |
3857 | lru_add_drain_all(); | |
f817ed48 | 3858 | /* try to free all pages in this cgroup */ |
569530fb | 3859 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 3860 | int progress; |
c1e862c1 | 3861 | |
c26251f9 MH |
3862 | if (signal_pending(current)) |
3863 | return -EINTR; | |
3864 | ||
c0ff4b85 | 3865 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 3866 | false); |
c1e862c1 | 3867 | if (!progress) { |
f817ed48 | 3868 | nr_retries--; |
c1e862c1 | 3869 | /* maybe some writeback is necessary */ |
8aa7e847 | 3870 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3871 | } |
f817ed48 KH |
3872 | |
3873 | } | |
08e552c6 | 3874 | lru_add_drain(); |
ab5196c2 MH |
3875 | mem_cgroup_reparent_charges(memcg); |
3876 | ||
3877 | return 0; | |
cc847582 KH |
3878 | } |
3879 | ||
6bbda35c | 3880 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 | 3881 | { |
c26251f9 MH |
3882 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
3883 | int ret; | |
3884 | ||
d8423011 MH |
3885 | if (mem_cgroup_is_root(memcg)) |
3886 | return -EINVAL; | |
c26251f9 MH |
3887 | css_get(&memcg->css); |
3888 | ret = mem_cgroup_force_empty(memcg); | |
3889 | css_put(&memcg->css); | |
3890 | ||
3891 | return ret; | |
c1e862c1 KH |
3892 | } |
3893 | ||
3894 | ||
18f59ea7 BS |
3895 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3896 | { | |
3897 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3898 | } | |
3899 | ||
3900 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3901 | u64 val) | |
3902 | { | |
3903 | int retval = 0; | |
c0ff4b85 | 3904 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 3905 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 3906 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
3907 | |
3908 | if (parent) | |
c0ff4b85 | 3909 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 BS |
3910 | |
3911 | cgroup_lock(); | |
567fb435 GC |
3912 | |
3913 | if (memcg->use_hierarchy == val) | |
3914 | goto out; | |
3915 | ||
18f59ea7 | 3916 | /* |
af901ca1 | 3917 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3918 | * in the child subtrees. If it is unset, then the change can |
3919 | * occur, provided the current cgroup has no children. | |
3920 | * | |
3921 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3922 | * set if there are no children. | |
3923 | */ | |
c0ff4b85 | 3924 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 BS |
3925 | (val == 1 || val == 0)) { |
3926 | if (list_empty(&cont->children)) | |
c0ff4b85 | 3927 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3928 | else |
3929 | retval = -EBUSY; | |
3930 | } else | |
3931 | retval = -EINVAL; | |
567fb435 GC |
3932 | |
3933 | out: | |
18f59ea7 BS |
3934 | cgroup_unlock(); |
3935 | ||
3936 | return retval; | |
3937 | } | |
3938 | ||
0c3e73e8 | 3939 | |
c0ff4b85 | 3940 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 3941 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 3942 | { |
7d74b06f | 3943 | struct mem_cgroup *iter; |
7a159cc9 | 3944 | long val = 0; |
0c3e73e8 | 3945 | |
7a159cc9 | 3946 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 3947 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
3948 | val += mem_cgroup_read_stat(iter, idx); |
3949 | ||
3950 | if (val < 0) /* race ? */ | |
3951 | val = 0; | |
3952 | return val; | |
0c3e73e8 BS |
3953 | } |
3954 | ||
c0ff4b85 | 3955 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 3956 | { |
7d74b06f | 3957 | u64 val; |
104f3928 | 3958 | |
c0ff4b85 | 3959 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 3960 | if (!swap) |
65c64ce8 | 3961 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 3962 | else |
65c64ce8 | 3963 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
3964 | } |
3965 | ||
c0ff4b85 R |
3966 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
3967 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3968 | |
7d74b06f | 3969 | if (swap) |
bff6bb83 | 3970 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
3971 | |
3972 | return val << PAGE_SHIFT; | |
3973 | } | |
3974 | ||
af36f906 TH |
3975 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
3976 | struct file *file, char __user *buf, | |
3977 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 3978 | { |
c0ff4b85 | 3979 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 3980 | char str[64]; |
104f3928 | 3981 | u64 val; |
86ae53e1 GC |
3982 | int name, len; |
3983 | enum res_type type; | |
8c7c6e34 KH |
3984 | |
3985 | type = MEMFILE_TYPE(cft->private); | |
3986 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
3987 | |
3988 | if (!do_swap_account && type == _MEMSWAP) | |
3989 | return -EOPNOTSUPP; | |
3990 | ||
8c7c6e34 KH |
3991 | switch (type) { |
3992 | case _MEM: | |
104f3928 | 3993 | if (name == RES_USAGE) |
c0ff4b85 | 3994 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 3995 | else |
c0ff4b85 | 3996 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
3997 | break; |
3998 | case _MEMSWAP: | |
104f3928 | 3999 | if (name == RES_USAGE) |
c0ff4b85 | 4000 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 4001 | else |
c0ff4b85 | 4002 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 4003 | break; |
510fc4e1 GC |
4004 | case _KMEM: |
4005 | val = res_counter_read_u64(&memcg->kmem, name); | |
4006 | break; | |
8c7c6e34 KH |
4007 | default: |
4008 | BUG(); | |
8c7c6e34 | 4009 | } |
af36f906 TH |
4010 | |
4011 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
4012 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 4013 | } |
510fc4e1 GC |
4014 | |
4015 | static int memcg_update_kmem_limit(struct cgroup *cont, u64 val) | |
4016 | { | |
4017 | int ret = -EINVAL; | |
4018 | #ifdef CONFIG_MEMCG_KMEM | |
4019 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); | |
4020 | /* | |
4021 | * For simplicity, we won't allow this to be disabled. It also can't | |
4022 | * be changed if the cgroup has children already, or if tasks had | |
4023 | * already joined. | |
4024 | * | |
4025 | * If tasks join before we set the limit, a person looking at | |
4026 | * kmem.usage_in_bytes will have no way to determine when it took | |
4027 | * place, which makes the value quite meaningless. | |
4028 | * | |
4029 | * After it first became limited, changes in the value of the limit are | |
4030 | * of course permitted. | |
4031 | * | |
4032 | * Taking the cgroup_lock is really offensive, but it is so far the only | |
4033 | * way to guarantee that no children will appear. There are plenty of | |
4034 | * other offenders, and they should all go away. Fine grained locking | |
4035 | * is probably the way to go here. When we are fully hierarchical, we | |
4036 | * can also get rid of the use_hierarchy check. | |
4037 | */ | |
4038 | cgroup_lock(); | |
4039 | mutex_lock(&set_limit_mutex); | |
4040 | if (!memcg->kmem_account_flags && val != RESOURCE_MAX) { | |
4041 | if (cgroup_task_count(cont) || (memcg->use_hierarchy && | |
4042 | !list_empty(&cont->children))) { | |
4043 | ret = -EBUSY; | |
4044 | goto out; | |
4045 | } | |
4046 | ret = res_counter_set_limit(&memcg->kmem, val); | |
4047 | VM_BUG_ON(ret); | |
4048 | ||
4049 | memcg_kmem_set_active(memcg); | |
4050 | } else | |
4051 | ret = res_counter_set_limit(&memcg->kmem, val); | |
4052 | out: | |
4053 | mutex_unlock(&set_limit_mutex); | |
4054 | cgroup_unlock(); | |
4055 | #endif | |
4056 | return ret; | |
4057 | } | |
4058 | ||
4059 | static void memcg_propagate_kmem(struct mem_cgroup *memcg) | |
4060 | { | |
4061 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
4062 | if (!parent) | |
4063 | return; | |
4064 | memcg->kmem_account_flags = parent->kmem_account_flags; | |
4065 | } | |
4066 | ||
628f4235 KH |
4067 | /* |
4068 | * The user of this function is... | |
4069 | * RES_LIMIT. | |
4070 | */ | |
856c13aa PM |
4071 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
4072 | const char *buffer) | |
8cdea7c0 | 4073 | { |
628f4235 | 4074 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
4075 | enum res_type type; |
4076 | int name; | |
628f4235 KH |
4077 | unsigned long long val; |
4078 | int ret; | |
4079 | ||
8c7c6e34 KH |
4080 | type = MEMFILE_TYPE(cft->private); |
4081 | name = MEMFILE_ATTR(cft->private); | |
af36f906 TH |
4082 | |
4083 | if (!do_swap_account && type == _MEMSWAP) | |
4084 | return -EOPNOTSUPP; | |
4085 | ||
8c7c6e34 | 4086 | switch (name) { |
628f4235 | 4087 | case RES_LIMIT: |
4b3bde4c BS |
4088 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
4089 | ret = -EINVAL; | |
4090 | break; | |
4091 | } | |
628f4235 KH |
4092 | /* This function does all necessary parse...reuse it */ |
4093 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
4094 | if (ret) |
4095 | break; | |
4096 | if (type == _MEM) | |
628f4235 | 4097 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 4098 | else if (type == _MEMSWAP) |
8c7c6e34 | 4099 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 GC |
4100 | else if (type == _KMEM) |
4101 | ret = memcg_update_kmem_limit(cont, val); | |
4102 | else | |
4103 | return -EINVAL; | |
628f4235 | 4104 | break; |
296c81d8 BS |
4105 | case RES_SOFT_LIMIT: |
4106 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
4107 | if (ret) | |
4108 | break; | |
4109 | /* | |
4110 | * For memsw, soft limits are hard to implement in terms | |
4111 | * of semantics, for now, we support soft limits for | |
4112 | * control without swap | |
4113 | */ | |
4114 | if (type == _MEM) | |
4115 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
4116 | else | |
4117 | ret = -EINVAL; | |
4118 | break; | |
628f4235 KH |
4119 | default: |
4120 | ret = -EINVAL; /* should be BUG() ? */ | |
4121 | break; | |
4122 | } | |
4123 | return ret; | |
8cdea7c0 BS |
4124 | } |
4125 | ||
fee7b548 KH |
4126 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
4127 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
4128 | { | |
4129 | struct cgroup *cgroup; | |
4130 | unsigned long long min_limit, min_memsw_limit, tmp; | |
4131 | ||
4132 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4133 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4134 | cgroup = memcg->css.cgroup; | |
4135 | if (!memcg->use_hierarchy) | |
4136 | goto out; | |
4137 | ||
4138 | while (cgroup->parent) { | |
4139 | cgroup = cgroup->parent; | |
4140 | memcg = mem_cgroup_from_cont(cgroup); | |
4141 | if (!memcg->use_hierarchy) | |
4142 | break; | |
4143 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4144 | min_limit = min(min_limit, tmp); | |
4145 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4146 | min_memsw_limit = min(min_memsw_limit, tmp); | |
4147 | } | |
4148 | out: | |
4149 | *mem_limit = min_limit; | |
4150 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
4151 | } |
4152 | ||
29f2a4da | 4153 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 4154 | { |
af36f906 | 4155 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
4156 | int name; |
4157 | enum res_type type; | |
c84872e1 | 4158 | |
8c7c6e34 KH |
4159 | type = MEMFILE_TYPE(event); |
4160 | name = MEMFILE_ATTR(event); | |
af36f906 TH |
4161 | |
4162 | if (!do_swap_account && type == _MEMSWAP) | |
4163 | return -EOPNOTSUPP; | |
4164 | ||
8c7c6e34 | 4165 | switch (name) { |
29f2a4da | 4166 | case RES_MAX_USAGE: |
8c7c6e34 | 4167 | if (type == _MEM) |
c0ff4b85 | 4168 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 4169 | else if (type == _MEMSWAP) |
c0ff4b85 | 4170 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
4171 | else if (type == _KMEM) |
4172 | res_counter_reset_max(&memcg->kmem); | |
4173 | else | |
4174 | return -EINVAL; | |
29f2a4da PE |
4175 | break; |
4176 | case RES_FAILCNT: | |
8c7c6e34 | 4177 | if (type == _MEM) |
c0ff4b85 | 4178 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 4179 | else if (type == _MEMSWAP) |
c0ff4b85 | 4180 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
4181 | else if (type == _KMEM) |
4182 | res_counter_reset_failcnt(&memcg->kmem); | |
4183 | else | |
4184 | return -EINVAL; | |
29f2a4da PE |
4185 | break; |
4186 | } | |
f64c3f54 | 4187 | |
85cc59db | 4188 | return 0; |
c84872e1 PE |
4189 | } |
4190 | ||
7dc74be0 DN |
4191 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
4192 | struct cftype *cft) | |
4193 | { | |
4194 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
4195 | } | |
4196 | ||
02491447 | 4197 | #ifdef CONFIG_MMU |
7dc74be0 DN |
4198 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
4199 | struct cftype *cft, u64 val) | |
4200 | { | |
c0ff4b85 | 4201 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
4202 | |
4203 | if (val >= (1 << NR_MOVE_TYPE)) | |
4204 | return -EINVAL; | |
4205 | /* | |
4206 | * We check this value several times in both in can_attach() and | |
4207 | * attach(), so we need cgroup lock to prevent this value from being | |
4208 | * inconsistent. | |
4209 | */ | |
4210 | cgroup_lock(); | |
c0ff4b85 | 4211 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
4212 | cgroup_unlock(); |
4213 | ||
4214 | return 0; | |
4215 | } | |
02491447 DN |
4216 | #else |
4217 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4218 | struct cftype *cft, u64 val) | |
4219 | { | |
4220 | return -ENOSYS; | |
4221 | } | |
4222 | #endif | |
7dc74be0 | 4223 | |
406eb0c9 | 4224 | #ifdef CONFIG_NUMA |
ab215884 | 4225 | static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft, |
fada52ca | 4226 | struct seq_file *m) |
406eb0c9 YH |
4227 | { |
4228 | int nid; | |
4229 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4230 | unsigned long node_nr; | |
d79154bb | 4231 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 4232 | |
d79154bb | 4233 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 4234 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 4235 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 4236 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
4237 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4238 | } | |
4239 | seq_putc(m, '\n'); | |
4240 | ||
d79154bb | 4241 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 4242 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 4243 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 4244 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4245 | LRU_ALL_FILE); |
406eb0c9 YH |
4246 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4247 | } | |
4248 | seq_putc(m, '\n'); | |
4249 | ||
d79154bb | 4250 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 4251 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 4252 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 4253 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4254 | LRU_ALL_ANON); |
406eb0c9 YH |
4255 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4256 | } | |
4257 | seq_putc(m, '\n'); | |
4258 | ||
d79154bb | 4259 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 4260 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 4261 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 4262 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 4263 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4264 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4265 | } | |
4266 | seq_putc(m, '\n'); | |
4267 | return 0; | |
4268 | } | |
4269 | #endif /* CONFIG_NUMA */ | |
4270 | ||
af7c4b0e JW |
4271 | static const char * const mem_cgroup_lru_names[] = { |
4272 | "inactive_anon", | |
4273 | "active_anon", | |
4274 | "inactive_file", | |
4275 | "active_file", | |
4276 | "unevictable", | |
4277 | }; | |
4278 | ||
4279 | static inline void mem_cgroup_lru_names_not_uptodate(void) | |
4280 | { | |
4281 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
4282 | } | |
4283 | ||
ab215884 | 4284 | static int memcg_stat_show(struct cgroup *cont, struct cftype *cft, |
78ccf5b5 | 4285 | struct seq_file *m) |
d2ceb9b7 | 4286 | { |
d79154bb | 4287 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af7c4b0e JW |
4288 | struct mem_cgroup *mi; |
4289 | unsigned int i; | |
406eb0c9 | 4290 | |
af7c4b0e | 4291 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 4292 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4293 | continue; |
af7c4b0e JW |
4294 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
4295 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 4296 | } |
7b854121 | 4297 | |
af7c4b0e JW |
4298 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
4299 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
4300 | mem_cgroup_read_events(memcg, i)); | |
4301 | ||
4302 | for (i = 0; i < NR_LRU_LISTS; i++) | |
4303 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
4304 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
4305 | ||
14067bb3 | 4306 | /* Hierarchical information */ |
fee7b548 KH |
4307 | { |
4308 | unsigned long long limit, memsw_limit; | |
d79154bb | 4309 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 4310 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 4311 | if (do_swap_account) |
78ccf5b5 JW |
4312 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4313 | memsw_limit); | |
fee7b548 | 4314 | } |
7f016ee8 | 4315 | |
af7c4b0e JW |
4316 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
4317 | long long val = 0; | |
4318 | ||
bff6bb83 | 4319 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 4320 | continue; |
af7c4b0e JW |
4321 | for_each_mem_cgroup_tree(mi, memcg) |
4322 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
4323 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
4324 | } | |
4325 | ||
4326 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
4327 | unsigned long long val = 0; | |
4328 | ||
4329 | for_each_mem_cgroup_tree(mi, memcg) | |
4330 | val += mem_cgroup_read_events(mi, i); | |
4331 | seq_printf(m, "total_%s %llu\n", | |
4332 | mem_cgroup_events_names[i], val); | |
4333 | } | |
4334 | ||
4335 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
4336 | unsigned long long val = 0; | |
4337 | ||
4338 | for_each_mem_cgroup_tree(mi, memcg) | |
4339 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
4340 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 4341 | } |
14067bb3 | 4342 | |
7f016ee8 | 4343 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
4344 | { |
4345 | int nid, zid; | |
4346 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 4347 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
4348 | unsigned long recent_rotated[2] = {0, 0}; |
4349 | unsigned long recent_scanned[2] = {0, 0}; | |
4350 | ||
4351 | for_each_online_node(nid) | |
4352 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 4353 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 4354 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 4355 | |
89abfab1 HD |
4356 | recent_rotated[0] += rstat->recent_rotated[0]; |
4357 | recent_rotated[1] += rstat->recent_rotated[1]; | |
4358 | recent_scanned[0] += rstat->recent_scanned[0]; | |
4359 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 4360 | } |
78ccf5b5 JW |
4361 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
4362 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
4363 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
4364 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
4365 | } |
4366 | #endif | |
4367 | ||
d2ceb9b7 KH |
4368 | return 0; |
4369 | } | |
4370 | ||
a7885eb8 KM |
4371 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4372 | { | |
4373 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4374 | ||
1f4c025b | 4375 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4376 | } |
4377 | ||
4378 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4379 | u64 val) | |
4380 | { | |
4381 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4382 | struct mem_cgroup *parent; | |
068b38c1 | 4383 | |
a7885eb8 KM |
4384 | if (val > 100) |
4385 | return -EINVAL; | |
4386 | ||
4387 | if (cgrp->parent == NULL) | |
4388 | return -EINVAL; | |
4389 | ||
4390 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4391 | |
4392 | cgroup_lock(); | |
4393 | ||
a7885eb8 KM |
4394 | /* If under hierarchy, only empty-root can set this value */ |
4395 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4396 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4397 | cgroup_unlock(); | |
a7885eb8 | 4398 | return -EINVAL; |
068b38c1 | 4399 | } |
a7885eb8 | 4400 | |
a7885eb8 | 4401 | memcg->swappiness = val; |
a7885eb8 | 4402 | |
068b38c1 LZ |
4403 | cgroup_unlock(); |
4404 | ||
a7885eb8 KM |
4405 | return 0; |
4406 | } | |
4407 | ||
2e72b634 KS |
4408 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4409 | { | |
4410 | struct mem_cgroup_threshold_ary *t; | |
4411 | u64 usage; | |
4412 | int i; | |
4413 | ||
4414 | rcu_read_lock(); | |
4415 | if (!swap) | |
2c488db2 | 4416 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4417 | else |
2c488db2 | 4418 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4419 | |
4420 | if (!t) | |
4421 | goto unlock; | |
4422 | ||
4423 | usage = mem_cgroup_usage(memcg, swap); | |
4424 | ||
4425 | /* | |
748dad36 | 4426 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4427 | * If it's not true, a threshold was crossed after last |
4428 | * call of __mem_cgroup_threshold(). | |
4429 | */ | |
5407a562 | 4430 | i = t->current_threshold; |
2e72b634 KS |
4431 | |
4432 | /* | |
4433 | * Iterate backward over array of thresholds starting from | |
4434 | * current_threshold and check if a threshold is crossed. | |
4435 | * If none of thresholds below usage is crossed, we read | |
4436 | * only one element of the array here. | |
4437 | */ | |
4438 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4439 | eventfd_signal(t->entries[i].eventfd, 1); | |
4440 | ||
4441 | /* i = current_threshold + 1 */ | |
4442 | i++; | |
4443 | ||
4444 | /* | |
4445 | * Iterate forward over array of thresholds starting from | |
4446 | * current_threshold+1 and check if a threshold is crossed. | |
4447 | * If none of thresholds above usage is crossed, we read | |
4448 | * only one element of the array here. | |
4449 | */ | |
4450 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4451 | eventfd_signal(t->entries[i].eventfd, 1); | |
4452 | ||
4453 | /* Update current_threshold */ | |
5407a562 | 4454 | t->current_threshold = i - 1; |
2e72b634 KS |
4455 | unlock: |
4456 | rcu_read_unlock(); | |
4457 | } | |
4458 | ||
4459 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4460 | { | |
ad4ca5f4 KS |
4461 | while (memcg) { |
4462 | __mem_cgroup_threshold(memcg, false); | |
4463 | if (do_swap_account) | |
4464 | __mem_cgroup_threshold(memcg, true); | |
4465 | ||
4466 | memcg = parent_mem_cgroup(memcg); | |
4467 | } | |
2e72b634 KS |
4468 | } |
4469 | ||
4470 | static int compare_thresholds(const void *a, const void *b) | |
4471 | { | |
4472 | const struct mem_cgroup_threshold *_a = a; | |
4473 | const struct mem_cgroup_threshold *_b = b; | |
4474 | ||
4475 | return _a->threshold - _b->threshold; | |
4476 | } | |
4477 | ||
c0ff4b85 | 4478 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4479 | { |
4480 | struct mem_cgroup_eventfd_list *ev; | |
4481 | ||
c0ff4b85 | 4482 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
4483 | eventfd_signal(ev->eventfd, 1); |
4484 | return 0; | |
4485 | } | |
4486 | ||
c0ff4b85 | 4487 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4488 | { |
7d74b06f KH |
4489 | struct mem_cgroup *iter; |
4490 | ||
c0ff4b85 | 4491 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4492 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4493 | } |
4494 | ||
4495 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4496 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4497 | { |
4498 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4499 | struct mem_cgroup_thresholds *thresholds; |
4500 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 4501 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 4502 | u64 threshold, usage; |
2c488db2 | 4503 | int i, size, ret; |
2e72b634 KS |
4504 | |
4505 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4506 | if (ret) | |
4507 | return ret; | |
4508 | ||
4509 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4510 | |
2e72b634 | 4511 | if (type == _MEM) |
2c488db2 | 4512 | thresholds = &memcg->thresholds; |
2e72b634 | 4513 | else if (type == _MEMSWAP) |
2c488db2 | 4514 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4515 | else |
4516 | BUG(); | |
4517 | ||
4518 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4519 | ||
4520 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4521 | if (thresholds->primary) |
2e72b634 KS |
4522 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4523 | ||
2c488db2 | 4524 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4525 | |
4526 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4527 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4528 | GFP_KERNEL); |
2c488db2 | 4529 | if (!new) { |
2e72b634 KS |
4530 | ret = -ENOMEM; |
4531 | goto unlock; | |
4532 | } | |
2c488db2 | 4533 | new->size = size; |
2e72b634 KS |
4534 | |
4535 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4536 | if (thresholds->primary) { |
4537 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4538 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4539 | } |
4540 | ||
2e72b634 | 4541 | /* Add new threshold */ |
2c488db2 KS |
4542 | new->entries[size - 1].eventfd = eventfd; |
4543 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4544 | |
4545 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4546 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4547 | compare_thresholds, NULL); |
4548 | ||
4549 | /* Find current threshold */ | |
2c488db2 | 4550 | new->current_threshold = -1; |
2e72b634 | 4551 | for (i = 0; i < size; i++) { |
748dad36 | 4552 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4553 | /* |
2c488db2 KS |
4554 | * new->current_threshold will not be used until |
4555 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4556 | * it here. |
4557 | */ | |
2c488db2 | 4558 | ++new->current_threshold; |
748dad36 SZ |
4559 | } else |
4560 | break; | |
2e72b634 KS |
4561 | } |
4562 | ||
2c488db2 KS |
4563 | /* Free old spare buffer and save old primary buffer as spare */ |
4564 | kfree(thresholds->spare); | |
4565 | thresholds->spare = thresholds->primary; | |
4566 | ||
4567 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4568 | |
907860ed | 4569 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4570 | synchronize_rcu(); |
4571 | ||
2e72b634 KS |
4572 | unlock: |
4573 | mutex_unlock(&memcg->thresholds_lock); | |
4574 | ||
4575 | return ret; | |
4576 | } | |
4577 | ||
907860ed | 4578 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4579 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4580 | { |
4581 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4582 | struct mem_cgroup_thresholds *thresholds; |
4583 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 4584 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 4585 | u64 usage; |
2c488db2 | 4586 | int i, j, size; |
2e72b634 KS |
4587 | |
4588 | mutex_lock(&memcg->thresholds_lock); | |
4589 | if (type == _MEM) | |
2c488db2 | 4590 | thresholds = &memcg->thresholds; |
2e72b634 | 4591 | else if (type == _MEMSWAP) |
2c488db2 | 4592 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4593 | else |
4594 | BUG(); | |
4595 | ||
371528ca AV |
4596 | if (!thresholds->primary) |
4597 | goto unlock; | |
4598 | ||
2e72b634 KS |
4599 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
4600 | ||
4601 | /* Check if a threshold crossed before removing */ | |
4602 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4603 | ||
4604 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4605 | size = 0; |
4606 | for (i = 0; i < thresholds->primary->size; i++) { | |
4607 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4608 | size++; |
4609 | } | |
4610 | ||
2c488db2 | 4611 | new = thresholds->spare; |
907860ed | 4612 | |
2e72b634 KS |
4613 | /* Set thresholds array to NULL if we don't have thresholds */ |
4614 | if (!size) { | |
2c488db2 KS |
4615 | kfree(new); |
4616 | new = NULL; | |
907860ed | 4617 | goto swap_buffers; |
2e72b634 KS |
4618 | } |
4619 | ||
2c488db2 | 4620 | new->size = size; |
2e72b634 KS |
4621 | |
4622 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4623 | new->current_threshold = -1; |
4624 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4625 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4626 | continue; |
4627 | ||
2c488db2 | 4628 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4629 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4630 | /* |
2c488db2 | 4631 | * new->current_threshold will not be used |
2e72b634 KS |
4632 | * until rcu_assign_pointer(), so it's safe to increment |
4633 | * it here. | |
4634 | */ | |
2c488db2 | 4635 | ++new->current_threshold; |
2e72b634 KS |
4636 | } |
4637 | j++; | |
4638 | } | |
4639 | ||
907860ed | 4640 | swap_buffers: |
2c488db2 KS |
4641 | /* Swap primary and spare array */ |
4642 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
4643 | /* If all events are unregistered, free the spare array */ |
4644 | if (!new) { | |
4645 | kfree(thresholds->spare); | |
4646 | thresholds->spare = NULL; | |
4647 | } | |
4648 | ||
2c488db2 | 4649 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4650 | |
907860ed | 4651 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4652 | synchronize_rcu(); |
371528ca | 4653 | unlock: |
2e72b634 | 4654 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4655 | } |
c1e862c1 | 4656 | |
9490ff27 KH |
4657 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4658 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4659 | { | |
4660 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4661 | struct mem_cgroup_eventfd_list *event; | |
86ae53e1 | 4662 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
4663 | |
4664 | BUG_ON(type != _OOM_TYPE); | |
4665 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4666 | if (!event) | |
4667 | return -ENOMEM; | |
4668 | ||
1af8efe9 | 4669 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4670 | |
4671 | event->eventfd = eventfd; | |
4672 | list_add(&event->list, &memcg->oom_notify); | |
4673 | ||
4674 | /* already in OOM ? */ | |
79dfdacc | 4675 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4676 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4677 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4678 | |
4679 | return 0; | |
4680 | } | |
4681 | ||
907860ed | 4682 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4683 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4684 | { | |
c0ff4b85 | 4685 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 | 4686 | struct mem_cgroup_eventfd_list *ev, *tmp; |
86ae53e1 | 4687 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
4688 | |
4689 | BUG_ON(type != _OOM_TYPE); | |
4690 | ||
1af8efe9 | 4691 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4692 | |
c0ff4b85 | 4693 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4694 | if (ev->eventfd == eventfd) { |
4695 | list_del(&ev->list); | |
4696 | kfree(ev); | |
4697 | } | |
4698 | } | |
4699 | ||
1af8efe9 | 4700 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4701 | } |
4702 | ||
3c11ecf4 KH |
4703 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4704 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4705 | { | |
c0ff4b85 | 4706 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 4707 | |
c0ff4b85 | 4708 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 4709 | |
c0ff4b85 | 4710 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
4711 | cb->fill(cb, "under_oom", 1); |
4712 | else | |
4713 | cb->fill(cb, "under_oom", 0); | |
4714 | return 0; | |
4715 | } | |
4716 | ||
3c11ecf4 KH |
4717 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4718 | struct cftype *cft, u64 val) | |
4719 | { | |
c0ff4b85 | 4720 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
4721 | struct mem_cgroup *parent; |
4722 | ||
4723 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4724 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4725 | return -EINVAL; | |
4726 | ||
4727 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4728 | ||
4729 | cgroup_lock(); | |
4730 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4731 | if ((parent->use_hierarchy) || | |
c0ff4b85 | 4732 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3c11ecf4 KH |
4733 | cgroup_unlock(); |
4734 | return -EINVAL; | |
4735 | } | |
c0ff4b85 | 4736 | memcg->oom_kill_disable = val; |
4d845ebf | 4737 | if (!val) |
c0ff4b85 | 4738 | memcg_oom_recover(memcg); |
3c11ecf4 KH |
4739 | cgroup_unlock(); |
4740 | return 0; | |
4741 | } | |
4742 | ||
c255a458 | 4743 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 4744 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 4745 | { |
510fc4e1 | 4746 | memcg_propagate_kmem(memcg); |
1d62e436 | 4747 | return mem_cgroup_sockets_init(memcg, ss); |
e5671dfa GC |
4748 | }; |
4749 | ||
1d62e436 | 4750 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 | 4751 | { |
1d62e436 | 4752 | mem_cgroup_sockets_destroy(memcg); |
d1a4c0b3 | 4753 | } |
e5671dfa | 4754 | #else |
cbe128e3 | 4755 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
4756 | { |
4757 | return 0; | |
4758 | } | |
d1a4c0b3 | 4759 | |
1d62e436 | 4760 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 GC |
4761 | { |
4762 | } | |
e5671dfa GC |
4763 | #endif |
4764 | ||
8cdea7c0 BS |
4765 | static struct cftype mem_cgroup_files[] = { |
4766 | { | |
0eea1030 | 4767 | .name = "usage_in_bytes", |
8c7c6e34 | 4768 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 4769 | .read = mem_cgroup_read, |
9490ff27 KH |
4770 | .register_event = mem_cgroup_usage_register_event, |
4771 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4772 | }, |
c84872e1 PE |
4773 | { |
4774 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4775 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4776 | .trigger = mem_cgroup_reset, |
af36f906 | 4777 | .read = mem_cgroup_read, |
c84872e1 | 4778 | }, |
8cdea7c0 | 4779 | { |
0eea1030 | 4780 | .name = "limit_in_bytes", |
8c7c6e34 | 4781 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4782 | .write_string = mem_cgroup_write, |
af36f906 | 4783 | .read = mem_cgroup_read, |
8cdea7c0 | 4784 | }, |
296c81d8 BS |
4785 | { |
4786 | .name = "soft_limit_in_bytes", | |
4787 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4788 | .write_string = mem_cgroup_write, | |
af36f906 | 4789 | .read = mem_cgroup_read, |
296c81d8 | 4790 | }, |
8cdea7c0 BS |
4791 | { |
4792 | .name = "failcnt", | |
8c7c6e34 | 4793 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4794 | .trigger = mem_cgroup_reset, |
af36f906 | 4795 | .read = mem_cgroup_read, |
8cdea7c0 | 4796 | }, |
d2ceb9b7 KH |
4797 | { |
4798 | .name = "stat", | |
ab215884 | 4799 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 4800 | }, |
c1e862c1 KH |
4801 | { |
4802 | .name = "force_empty", | |
4803 | .trigger = mem_cgroup_force_empty_write, | |
4804 | }, | |
18f59ea7 BS |
4805 | { |
4806 | .name = "use_hierarchy", | |
4807 | .write_u64 = mem_cgroup_hierarchy_write, | |
4808 | .read_u64 = mem_cgroup_hierarchy_read, | |
4809 | }, | |
a7885eb8 KM |
4810 | { |
4811 | .name = "swappiness", | |
4812 | .read_u64 = mem_cgroup_swappiness_read, | |
4813 | .write_u64 = mem_cgroup_swappiness_write, | |
4814 | }, | |
7dc74be0 DN |
4815 | { |
4816 | .name = "move_charge_at_immigrate", | |
4817 | .read_u64 = mem_cgroup_move_charge_read, | |
4818 | .write_u64 = mem_cgroup_move_charge_write, | |
4819 | }, | |
9490ff27 KH |
4820 | { |
4821 | .name = "oom_control", | |
3c11ecf4 KH |
4822 | .read_map = mem_cgroup_oom_control_read, |
4823 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4824 | .register_event = mem_cgroup_oom_register_event, |
4825 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4826 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4827 | }, | |
406eb0c9 YH |
4828 | #ifdef CONFIG_NUMA |
4829 | { | |
4830 | .name = "numa_stat", | |
ab215884 | 4831 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
4832 | }, |
4833 | #endif | |
c255a458 | 4834 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4835 | { |
4836 | .name = "memsw.usage_in_bytes", | |
4837 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
af36f906 | 4838 | .read = mem_cgroup_read, |
9490ff27 KH |
4839 | .register_event = mem_cgroup_usage_register_event, |
4840 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4841 | }, |
4842 | { | |
4843 | .name = "memsw.max_usage_in_bytes", | |
4844 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4845 | .trigger = mem_cgroup_reset, | |
af36f906 | 4846 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4847 | }, |
4848 | { | |
4849 | .name = "memsw.limit_in_bytes", | |
4850 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4851 | .write_string = mem_cgroup_write, | |
af36f906 | 4852 | .read = mem_cgroup_read, |
8c7c6e34 KH |
4853 | }, |
4854 | { | |
4855 | .name = "memsw.failcnt", | |
4856 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4857 | .trigger = mem_cgroup_reset, | |
af36f906 | 4858 | .read = mem_cgroup_read, |
8c7c6e34 | 4859 | }, |
510fc4e1 GC |
4860 | #endif |
4861 | #ifdef CONFIG_MEMCG_KMEM | |
4862 | { | |
4863 | .name = "kmem.limit_in_bytes", | |
4864 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
4865 | .write_string = mem_cgroup_write, | |
4866 | .read = mem_cgroup_read, | |
4867 | }, | |
4868 | { | |
4869 | .name = "kmem.usage_in_bytes", | |
4870 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
4871 | .read = mem_cgroup_read, | |
4872 | }, | |
4873 | { | |
4874 | .name = "kmem.failcnt", | |
4875 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
4876 | .trigger = mem_cgroup_reset, | |
4877 | .read = mem_cgroup_read, | |
4878 | }, | |
4879 | { | |
4880 | .name = "kmem.max_usage_in_bytes", | |
4881 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
4882 | .trigger = mem_cgroup_reset, | |
4883 | .read = mem_cgroup_read, | |
4884 | }, | |
8c7c6e34 | 4885 | #endif |
6bc10349 | 4886 | { }, /* terminate */ |
af36f906 | 4887 | }; |
8c7c6e34 | 4888 | |
c0ff4b85 | 4889 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4890 | { |
4891 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4892 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4893 | int zone, tmp = node; |
1ecaab2b KH |
4894 | /* |
4895 | * This routine is called against possible nodes. | |
4896 | * But it's BUG to call kmalloc() against offline node. | |
4897 | * | |
4898 | * TODO: this routine can waste much memory for nodes which will | |
4899 | * never be onlined. It's better to use memory hotplug callback | |
4900 | * function. | |
4901 | */ | |
41e3355d KH |
4902 | if (!node_state(node, N_NORMAL_MEMORY)) |
4903 | tmp = -1; | |
17295c88 | 4904 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4905 | if (!pn) |
4906 | return 1; | |
1ecaab2b | 4907 | |
1ecaab2b KH |
4908 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4909 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4910 | lruvec_init(&mz->lruvec); |
f64c3f54 | 4911 | mz->usage_in_excess = 0; |
4e416953 | 4912 | mz->on_tree = false; |
d79154bb | 4913 | mz->memcg = memcg; |
1ecaab2b | 4914 | } |
0a619e58 | 4915 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
4916 | return 0; |
4917 | } | |
4918 | ||
c0ff4b85 | 4919 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4920 | { |
c0ff4b85 | 4921 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
4922 | } |
4923 | ||
33327948 KH |
4924 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4925 | { | |
d79154bb | 4926 | struct mem_cgroup *memcg; |
c62b1a3b | 4927 | int size = sizeof(struct mem_cgroup); |
33327948 | 4928 | |
c62b1a3b | 4929 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4930 | if (size < PAGE_SIZE) |
d79154bb | 4931 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 4932 | else |
d79154bb | 4933 | memcg = vzalloc(size); |
33327948 | 4934 | |
d79154bb | 4935 | if (!memcg) |
e7bbcdf3 DC |
4936 | return NULL; |
4937 | ||
d79154bb HD |
4938 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4939 | if (!memcg->stat) | |
d2e61b8d | 4940 | goto out_free; |
d79154bb HD |
4941 | spin_lock_init(&memcg->pcp_counter_lock); |
4942 | return memcg; | |
d2e61b8d DC |
4943 | |
4944 | out_free: | |
4945 | if (size < PAGE_SIZE) | |
d79154bb | 4946 | kfree(memcg); |
d2e61b8d | 4947 | else |
d79154bb | 4948 | vfree(memcg); |
d2e61b8d | 4949 | return NULL; |
33327948 KH |
4950 | } |
4951 | ||
59927fb9 | 4952 | /* |
3afe36b1 | 4953 | * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU, |
59927fb9 HD |
4954 | * but in process context. The work_freeing structure is overlaid |
4955 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
4956 | */ | |
3afe36b1 | 4957 | static void free_work(struct work_struct *work) |
59927fb9 HD |
4958 | { |
4959 | struct mem_cgroup *memcg; | |
3afe36b1 | 4960 | int size = sizeof(struct mem_cgroup); |
59927fb9 HD |
4961 | |
4962 | memcg = container_of(work, struct mem_cgroup, work_freeing); | |
3f134619 GC |
4963 | /* |
4964 | * We need to make sure that (at least for now), the jump label | |
4965 | * destruction code runs outside of the cgroup lock. This is because | |
4966 | * get_online_cpus(), which is called from the static_branch update, | |
4967 | * can't be called inside the cgroup_lock. cpusets are the ones | |
4968 | * enforcing this dependency, so if they ever change, we might as well. | |
4969 | * | |
4970 | * schedule_work() will guarantee this happens. Be careful if you need | |
4971 | * to move this code around, and make sure it is outside | |
4972 | * the cgroup_lock. | |
4973 | */ | |
4974 | disarm_sock_keys(memcg); | |
3afe36b1 GC |
4975 | if (size < PAGE_SIZE) |
4976 | kfree(memcg); | |
4977 | else | |
4978 | vfree(memcg); | |
59927fb9 | 4979 | } |
3afe36b1 GC |
4980 | |
4981 | static void free_rcu(struct rcu_head *rcu_head) | |
59927fb9 HD |
4982 | { |
4983 | struct mem_cgroup *memcg; | |
4984 | ||
4985 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); | |
3afe36b1 | 4986 | INIT_WORK(&memcg->work_freeing, free_work); |
59927fb9 HD |
4987 | schedule_work(&memcg->work_freeing); |
4988 | } | |
4989 | ||
8c7c6e34 KH |
4990 | /* |
4991 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4992 | * (scanning all at force_empty is too costly...) | |
4993 | * | |
4994 | * Instead of clearing all references at force_empty, we remember | |
4995 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4996 | * it goes down to 0. | |
4997 | * | |
8c7c6e34 KH |
4998 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4999 | */ | |
5000 | ||
c0ff4b85 | 5001 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
33327948 | 5002 | { |
08e552c6 KH |
5003 | int node; |
5004 | ||
c0ff4b85 R |
5005 | mem_cgroup_remove_from_trees(memcg); |
5006 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
04046e1a | 5007 | |
3ed28fa1 | 5008 | for_each_node(node) |
c0ff4b85 | 5009 | free_mem_cgroup_per_zone_info(memcg, node); |
08e552c6 | 5010 | |
c0ff4b85 | 5011 | free_percpu(memcg->stat); |
3afe36b1 | 5012 | call_rcu(&memcg->rcu_freeing, free_rcu); |
33327948 KH |
5013 | } |
5014 | ||
c0ff4b85 | 5015 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 5016 | { |
c0ff4b85 | 5017 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
5018 | } |
5019 | ||
c0ff4b85 | 5020 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 5021 | { |
c0ff4b85 R |
5022 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
5023 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
5024 | __mem_cgroup_free(memcg); | |
7bcc1bb1 DN |
5025 | if (parent) |
5026 | mem_cgroup_put(parent); | |
5027 | } | |
8c7c6e34 KH |
5028 | } |
5029 | ||
c0ff4b85 | 5030 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 5031 | { |
c0ff4b85 | 5032 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
5033 | } |
5034 | ||
7bcc1bb1 DN |
5035 | /* |
5036 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
5037 | */ | |
e1aab161 | 5038 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 5039 | { |
c0ff4b85 | 5040 | if (!memcg->res.parent) |
7bcc1bb1 | 5041 | return NULL; |
c0ff4b85 | 5042 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 5043 | } |
e1aab161 | 5044 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 5045 | |
c255a458 | 5046 | #ifdef CONFIG_MEMCG_SWAP |
c077719b KH |
5047 | static void __init enable_swap_cgroup(void) |
5048 | { | |
f8d66542 | 5049 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
5050 | do_swap_account = 1; |
5051 | } | |
5052 | #else | |
5053 | static void __init enable_swap_cgroup(void) | |
5054 | { | |
5055 | } | |
5056 | #endif | |
5057 | ||
f64c3f54 BS |
5058 | static int mem_cgroup_soft_limit_tree_init(void) |
5059 | { | |
5060 | struct mem_cgroup_tree_per_node *rtpn; | |
5061 | struct mem_cgroup_tree_per_zone *rtpz; | |
5062 | int tmp, node, zone; | |
5063 | ||
3ed28fa1 | 5064 | for_each_node(node) { |
f64c3f54 BS |
5065 | tmp = node; |
5066 | if (!node_state(node, N_NORMAL_MEMORY)) | |
5067 | tmp = -1; | |
5068 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
5069 | if (!rtpn) | |
c3cecc68 | 5070 | goto err_cleanup; |
f64c3f54 BS |
5071 | |
5072 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5073 | ||
5074 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5075 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5076 | rtpz->rb_root = RB_ROOT; | |
5077 | spin_lock_init(&rtpz->lock); | |
5078 | } | |
5079 | } | |
5080 | return 0; | |
c3cecc68 MH |
5081 | |
5082 | err_cleanup: | |
3ed28fa1 | 5083 | for_each_node(node) { |
c3cecc68 MH |
5084 | if (!soft_limit_tree.rb_tree_per_node[node]) |
5085 | break; | |
5086 | kfree(soft_limit_tree.rb_tree_per_node[node]); | |
5087 | soft_limit_tree.rb_tree_per_node[node] = NULL; | |
5088 | } | |
5089 | return 1; | |
5090 | ||
f64c3f54 BS |
5091 | } |
5092 | ||
0eb253e2 | 5093 | static struct cgroup_subsys_state * __ref |
92fb9748 | 5094 | mem_cgroup_css_alloc(struct cgroup *cont) |
8cdea7c0 | 5095 | { |
c0ff4b85 | 5096 | struct mem_cgroup *memcg, *parent; |
04046e1a | 5097 | long error = -ENOMEM; |
6d12e2d8 | 5098 | int node; |
8cdea7c0 | 5099 | |
c0ff4b85 R |
5100 | memcg = mem_cgroup_alloc(); |
5101 | if (!memcg) | |
04046e1a | 5102 | return ERR_PTR(error); |
78fb7466 | 5103 | |
3ed28fa1 | 5104 | for_each_node(node) |
c0ff4b85 | 5105 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 5106 | goto free_out; |
f64c3f54 | 5107 | |
c077719b | 5108 | /* root ? */ |
28dbc4b6 | 5109 | if (cont->parent == NULL) { |
cdec2e42 | 5110 | int cpu; |
c077719b | 5111 | enable_swap_cgroup(); |
28dbc4b6 | 5112 | parent = NULL; |
f64c3f54 BS |
5113 | if (mem_cgroup_soft_limit_tree_init()) |
5114 | goto free_out; | |
a41c58a6 | 5115 | root_mem_cgroup = memcg; |
cdec2e42 KH |
5116 | for_each_possible_cpu(cpu) { |
5117 | struct memcg_stock_pcp *stock = | |
5118 | &per_cpu(memcg_stock, cpu); | |
5119 | INIT_WORK(&stock->work, drain_local_stock); | |
5120 | } | |
711d3d2c | 5121 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 5122 | } else { |
28dbc4b6 | 5123 | parent = mem_cgroup_from_cont(cont->parent); |
c0ff4b85 R |
5124 | memcg->use_hierarchy = parent->use_hierarchy; |
5125 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
18f59ea7 | 5126 | } |
28dbc4b6 | 5127 | |
18f59ea7 | 5128 | if (parent && parent->use_hierarchy) { |
c0ff4b85 R |
5129 | res_counter_init(&memcg->res, &parent->res); |
5130 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 5131 | res_counter_init(&memcg->kmem, &parent->kmem); |
7bcc1bb1 DN |
5132 | /* |
5133 | * We increment refcnt of the parent to ensure that we can | |
5134 | * safely access it on res_counter_charge/uncharge. | |
5135 | * This refcnt will be decremented when freeing this | |
5136 | * mem_cgroup(see mem_cgroup_put). | |
5137 | */ | |
5138 | mem_cgroup_get(parent); | |
18f59ea7 | 5139 | } else { |
c0ff4b85 R |
5140 | res_counter_init(&memcg->res, NULL); |
5141 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 5142 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
5143 | /* |
5144 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5145 | * much sense so let cgroup subsystem know about this | |
5146 | * unfortunate state in our controller. | |
5147 | */ | |
5148 | if (parent && parent != root_mem_cgroup) | |
5149 | mem_cgroup_subsys.broken_hierarchy = true; | |
18f59ea7 | 5150 | } |
c0ff4b85 R |
5151 | memcg->last_scanned_node = MAX_NUMNODES; |
5152 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6d61ef40 | 5153 | |
a7885eb8 | 5154 | if (parent) |
c0ff4b85 R |
5155 | memcg->swappiness = mem_cgroup_swappiness(parent); |
5156 | atomic_set(&memcg->refcnt, 1); | |
5157 | memcg->move_charge_at_immigrate = 0; | |
5158 | mutex_init(&memcg->thresholds_lock); | |
312734c0 | 5159 | spin_lock_init(&memcg->move_lock); |
cbe128e3 GC |
5160 | |
5161 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
5162 | if (error) { | |
5163 | /* | |
5164 | * We call put now because our (and parent's) refcnts | |
5165 | * are already in place. mem_cgroup_put() will internally | |
5166 | * call __mem_cgroup_free, so return directly | |
5167 | */ | |
5168 | mem_cgroup_put(memcg); | |
5169 | return ERR_PTR(error); | |
5170 | } | |
c0ff4b85 | 5171 | return &memcg->css; |
6d12e2d8 | 5172 | free_out: |
c0ff4b85 | 5173 | __mem_cgroup_free(memcg); |
04046e1a | 5174 | return ERR_PTR(error); |
8cdea7c0 BS |
5175 | } |
5176 | ||
92fb9748 | 5177 | static void mem_cgroup_css_offline(struct cgroup *cont) |
df878fb0 | 5178 | { |
c0ff4b85 | 5179 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 5180 | |
ab5196c2 | 5181 | mem_cgroup_reparent_charges(memcg); |
df878fb0 KH |
5182 | } |
5183 | ||
92fb9748 | 5184 | static void mem_cgroup_css_free(struct cgroup *cont) |
8cdea7c0 | 5185 | { |
c0ff4b85 | 5186 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 5187 | |
1d62e436 | 5188 | kmem_cgroup_destroy(memcg); |
d1a4c0b3 | 5189 | |
c0ff4b85 | 5190 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
5191 | } |
5192 | ||
02491447 | 5193 | #ifdef CONFIG_MMU |
7dc74be0 | 5194 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
5195 | #define PRECHARGE_COUNT_AT_ONCE 256 |
5196 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 5197 | { |
854ffa8d DN |
5198 | int ret = 0; |
5199 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 5200 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 5201 | |
c0ff4b85 | 5202 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
5203 | mc.precharge += count; |
5204 | /* we don't need css_get for root */ | |
5205 | return ret; | |
5206 | } | |
5207 | /* try to charge at once */ | |
5208 | if (count > 1) { | |
5209 | struct res_counter *dummy; | |
5210 | /* | |
c0ff4b85 | 5211 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
5212 | * by cgroup_lock_live_cgroup() that it is not removed and we |
5213 | * are still under the same cgroup_mutex. So we can postpone | |
5214 | * css_get(). | |
5215 | */ | |
c0ff4b85 | 5216 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 5217 | goto one_by_one; |
c0ff4b85 | 5218 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 5219 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 5220 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
5221 | goto one_by_one; |
5222 | } | |
5223 | mc.precharge += count; | |
854ffa8d DN |
5224 | return ret; |
5225 | } | |
5226 | one_by_one: | |
5227 | /* fall back to one by one charge */ | |
5228 | while (count--) { | |
5229 | if (signal_pending(current)) { | |
5230 | ret = -EINTR; | |
5231 | break; | |
5232 | } | |
5233 | if (!batch_count--) { | |
5234 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5235 | cond_resched(); | |
5236 | } | |
c0ff4b85 R |
5237 | ret = __mem_cgroup_try_charge(NULL, |
5238 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 5239 | if (ret) |
854ffa8d | 5240 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 5241 | return ret; |
854ffa8d DN |
5242 | mc.precharge++; |
5243 | } | |
4ffef5fe DN |
5244 | return ret; |
5245 | } | |
5246 | ||
5247 | /** | |
8d32ff84 | 5248 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
5249 | * @vma: the vma the pte to be checked belongs |
5250 | * @addr: the address corresponding to the pte to be checked | |
5251 | * @ptent: the pte to be checked | |
02491447 | 5252 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5253 | * |
5254 | * Returns | |
5255 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5256 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5257 | * move charge. if @target is not NULL, the page is stored in target->page | |
5258 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5259 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5260 | * target for charge migration. if @target is not NULL, the entry is stored | |
5261 | * in target->ent. | |
4ffef5fe DN |
5262 | * |
5263 | * Called with pte lock held. | |
5264 | */ | |
4ffef5fe DN |
5265 | union mc_target { |
5266 | struct page *page; | |
02491447 | 5267 | swp_entry_t ent; |
4ffef5fe DN |
5268 | }; |
5269 | ||
4ffef5fe | 5270 | enum mc_target_type { |
8d32ff84 | 5271 | MC_TARGET_NONE = 0, |
4ffef5fe | 5272 | MC_TARGET_PAGE, |
02491447 | 5273 | MC_TARGET_SWAP, |
4ffef5fe DN |
5274 | }; |
5275 | ||
90254a65 DN |
5276 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5277 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5278 | { |
90254a65 | 5279 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5280 | |
90254a65 DN |
5281 | if (!page || !page_mapped(page)) |
5282 | return NULL; | |
5283 | if (PageAnon(page)) { | |
5284 | /* we don't move shared anon */ | |
4b91355e | 5285 | if (!move_anon()) |
90254a65 | 5286 | return NULL; |
87946a72 DN |
5287 | } else if (!move_file()) |
5288 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5289 | return NULL; |
5290 | if (!get_page_unless_zero(page)) | |
5291 | return NULL; | |
5292 | ||
5293 | return page; | |
5294 | } | |
5295 | ||
4b91355e | 5296 | #ifdef CONFIG_SWAP |
90254a65 DN |
5297 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
5298 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5299 | { | |
90254a65 DN |
5300 | struct page *page = NULL; |
5301 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5302 | ||
5303 | if (!move_anon() || non_swap_entry(ent)) | |
5304 | return NULL; | |
4b91355e KH |
5305 | /* |
5306 | * Because lookup_swap_cache() updates some statistics counter, | |
5307 | * we call find_get_page() with swapper_space directly. | |
5308 | */ | |
5309 | page = find_get_page(&swapper_space, ent.val); | |
90254a65 DN |
5310 | if (do_swap_account) |
5311 | entry->val = ent.val; | |
5312 | ||
5313 | return page; | |
5314 | } | |
4b91355e KH |
5315 | #else |
5316 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5317 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5318 | { | |
5319 | return NULL; | |
5320 | } | |
5321 | #endif | |
90254a65 | 5322 | |
87946a72 DN |
5323 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5324 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5325 | { | |
5326 | struct page *page = NULL; | |
87946a72 DN |
5327 | struct address_space *mapping; |
5328 | pgoff_t pgoff; | |
5329 | ||
5330 | if (!vma->vm_file) /* anonymous vma */ | |
5331 | return NULL; | |
5332 | if (!move_file()) | |
5333 | return NULL; | |
5334 | ||
87946a72 DN |
5335 | mapping = vma->vm_file->f_mapping; |
5336 | if (pte_none(ptent)) | |
5337 | pgoff = linear_page_index(vma, addr); | |
5338 | else /* pte_file(ptent) is true */ | |
5339 | pgoff = pte_to_pgoff(ptent); | |
5340 | ||
5341 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5342 | page = find_get_page(mapping, pgoff); |
5343 | ||
5344 | #ifdef CONFIG_SWAP | |
5345 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5346 | if (radix_tree_exceptional_entry(page)) { | |
5347 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5348 | if (do_swap_account) |
aa3b1895 HD |
5349 | *entry = swap; |
5350 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5351 | } |
aa3b1895 | 5352 | #endif |
87946a72 DN |
5353 | return page; |
5354 | } | |
5355 | ||
8d32ff84 | 5356 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5357 | unsigned long addr, pte_t ptent, union mc_target *target) |
5358 | { | |
5359 | struct page *page = NULL; | |
5360 | struct page_cgroup *pc; | |
8d32ff84 | 5361 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5362 | swp_entry_t ent = { .val = 0 }; |
5363 | ||
5364 | if (pte_present(ptent)) | |
5365 | page = mc_handle_present_pte(vma, addr, ptent); | |
5366 | else if (is_swap_pte(ptent)) | |
5367 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5368 | else if (pte_none(ptent) || pte_file(ptent)) |
5369 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5370 | |
5371 | if (!page && !ent.val) | |
8d32ff84 | 5372 | return ret; |
02491447 DN |
5373 | if (page) { |
5374 | pc = lookup_page_cgroup(page); | |
5375 | /* | |
5376 | * Do only loose check w/o page_cgroup lock. | |
5377 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5378 | * the lock. | |
5379 | */ | |
5380 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5381 | ret = MC_TARGET_PAGE; | |
5382 | if (target) | |
5383 | target->page = page; | |
5384 | } | |
5385 | if (!ret || !target) | |
5386 | put_page(page); | |
5387 | } | |
90254a65 DN |
5388 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5389 | if (ent.val && !ret && | |
9fb4b7cc | 5390 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5391 | ret = MC_TARGET_SWAP; |
5392 | if (target) | |
5393 | target->ent = ent; | |
4ffef5fe | 5394 | } |
4ffef5fe DN |
5395 | return ret; |
5396 | } | |
5397 | ||
12724850 NH |
5398 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5399 | /* | |
5400 | * We don't consider swapping or file mapped pages because THP does not | |
5401 | * support them for now. | |
5402 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
5403 | */ | |
5404 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5405 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5406 | { | |
5407 | struct page *page = NULL; | |
5408 | struct page_cgroup *pc; | |
5409 | enum mc_target_type ret = MC_TARGET_NONE; | |
5410 | ||
5411 | page = pmd_page(pmd); | |
5412 | VM_BUG_ON(!page || !PageHead(page)); | |
5413 | if (!move_anon()) | |
5414 | return ret; | |
5415 | pc = lookup_page_cgroup(page); | |
5416 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5417 | ret = MC_TARGET_PAGE; | |
5418 | if (target) { | |
5419 | get_page(page); | |
5420 | target->page = page; | |
5421 | } | |
5422 | } | |
5423 | return ret; | |
5424 | } | |
5425 | #else | |
5426 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5427 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5428 | { | |
5429 | return MC_TARGET_NONE; | |
5430 | } | |
5431 | #endif | |
5432 | ||
4ffef5fe DN |
5433 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5434 | unsigned long addr, unsigned long end, | |
5435 | struct mm_walk *walk) | |
5436 | { | |
5437 | struct vm_area_struct *vma = walk->private; | |
5438 | pte_t *pte; | |
5439 | spinlock_t *ptl; | |
5440 | ||
12724850 NH |
5441 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
5442 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
5443 | mc.precharge += HPAGE_PMD_NR; | |
5444 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5445 | return 0; |
12724850 | 5446 | } |
03319327 | 5447 | |
45f83cef AA |
5448 | if (pmd_trans_unstable(pmd)) |
5449 | return 0; | |
4ffef5fe DN |
5450 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5451 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5452 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5453 | mc.precharge++; /* increment precharge temporarily */ |
5454 | pte_unmap_unlock(pte - 1, ptl); | |
5455 | cond_resched(); | |
5456 | ||
7dc74be0 DN |
5457 | return 0; |
5458 | } | |
5459 | ||
4ffef5fe DN |
5460 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5461 | { | |
5462 | unsigned long precharge; | |
5463 | struct vm_area_struct *vma; | |
5464 | ||
dfe076b0 | 5465 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5466 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5467 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5468 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5469 | .mm = mm, | |
5470 | .private = vma, | |
5471 | }; | |
5472 | if (is_vm_hugetlb_page(vma)) | |
5473 | continue; | |
4ffef5fe DN |
5474 | walk_page_range(vma->vm_start, vma->vm_end, |
5475 | &mem_cgroup_count_precharge_walk); | |
5476 | } | |
dfe076b0 | 5477 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5478 | |
5479 | precharge = mc.precharge; | |
5480 | mc.precharge = 0; | |
5481 | ||
5482 | return precharge; | |
5483 | } | |
5484 | ||
4ffef5fe DN |
5485 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5486 | { | |
dfe076b0 DN |
5487 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5488 | ||
5489 | VM_BUG_ON(mc.moving_task); | |
5490 | mc.moving_task = current; | |
5491 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5492 | } |
5493 | ||
dfe076b0 DN |
5494 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5495 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5496 | { |
2bd9bb20 KH |
5497 | struct mem_cgroup *from = mc.from; |
5498 | struct mem_cgroup *to = mc.to; | |
5499 | ||
4ffef5fe | 5500 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5501 | if (mc.precharge) { |
5502 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5503 | mc.precharge = 0; | |
5504 | } | |
5505 | /* | |
5506 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5507 | * we must uncharge here. | |
5508 | */ | |
5509 | if (mc.moved_charge) { | |
5510 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5511 | mc.moved_charge = 0; | |
4ffef5fe | 5512 | } |
483c30b5 DN |
5513 | /* we must fixup refcnts and charges */ |
5514 | if (mc.moved_swap) { | |
483c30b5 DN |
5515 | /* uncharge swap account from the old cgroup */ |
5516 | if (!mem_cgroup_is_root(mc.from)) | |
5517 | res_counter_uncharge(&mc.from->memsw, | |
5518 | PAGE_SIZE * mc.moved_swap); | |
5519 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5520 | ||
5521 | if (!mem_cgroup_is_root(mc.to)) { | |
5522 | /* | |
5523 | * we charged both to->res and to->memsw, so we should | |
5524 | * uncharge to->res. | |
5525 | */ | |
5526 | res_counter_uncharge(&mc.to->res, | |
5527 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5528 | } |
5529 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5530 | mc.moved_swap = 0; |
5531 | } | |
dfe076b0 DN |
5532 | memcg_oom_recover(from); |
5533 | memcg_oom_recover(to); | |
5534 | wake_up_all(&mc.waitq); | |
5535 | } | |
5536 | ||
5537 | static void mem_cgroup_clear_mc(void) | |
5538 | { | |
5539 | struct mem_cgroup *from = mc.from; | |
5540 | ||
5541 | /* | |
5542 | * we must clear moving_task before waking up waiters at the end of | |
5543 | * task migration. | |
5544 | */ | |
5545 | mc.moving_task = NULL; | |
5546 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5547 | spin_lock(&mc.lock); |
4ffef5fe DN |
5548 | mc.from = NULL; |
5549 | mc.to = NULL; | |
2bd9bb20 | 5550 | spin_unlock(&mc.lock); |
32047e2a | 5551 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5552 | } |
5553 | ||
761b3ef5 LZ |
5554 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5555 | struct cgroup_taskset *tset) | |
7dc74be0 | 5556 | { |
2f7ee569 | 5557 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 5558 | int ret = 0; |
c0ff4b85 | 5559 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
7dc74be0 | 5560 | |
c0ff4b85 | 5561 | if (memcg->move_charge_at_immigrate) { |
7dc74be0 DN |
5562 | struct mm_struct *mm; |
5563 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5564 | ||
c0ff4b85 | 5565 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
5566 | |
5567 | mm = get_task_mm(p); | |
5568 | if (!mm) | |
5569 | return 0; | |
7dc74be0 | 5570 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5571 | if (mm->owner == p) { |
5572 | VM_BUG_ON(mc.from); | |
5573 | VM_BUG_ON(mc.to); | |
5574 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5575 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5576 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5577 | mem_cgroup_start_move(from); |
2bd9bb20 | 5578 | spin_lock(&mc.lock); |
4ffef5fe | 5579 | mc.from = from; |
c0ff4b85 | 5580 | mc.to = memcg; |
2bd9bb20 | 5581 | spin_unlock(&mc.lock); |
dfe076b0 | 5582 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5583 | |
5584 | ret = mem_cgroup_precharge_mc(mm); | |
5585 | if (ret) | |
5586 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5587 | } |
5588 | mmput(mm); | |
7dc74be0 DN |
5589 | } |
5590 | return ret; | |
5591 | } | |
5592 | ||
761b3ef5 LZ |
5593 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5594 | struct cgroup_taskset *tset) | |
7dc74be0 | 5595 | { |
4ffef5fe | 5596 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5597 | } |
5598 | ||
4ffef5fe DN |
5599 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5600 | unsigned long addr, unsigned long end, | |
5601 | struct mm_walk *walk) | |
7dc74be0 | 5602 | { |
4ffef5fe DN |
5603 | int ret = 0; |
5604 | struct vm_area_struct *vma = walk->private; | |
5605 | pte_t *pte; | |
5606 | spinlock_t *ptl; | |
12724850 NH |
5607 | enum mc_target_type target_type; |
5608 | union mc_target target; | |
5609 | struct page *page; | |
5610 | struct page_cgroup *pc; | |
4ffef5fe | 5611 | |
12724850 NH |
5612 | /* |
5613 | * We don't take compound_lock() here but no race with splitting thp | |
5614 | * happens because: | |
5615 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
5616 | * under splitting, which means there's no concurrent thp split, | |
5617 | * - if another thread runs into split_huge_page() just after we | |
5618 | * entered this if-block, the thread must wait for page table lock | |
5619 | * to be unlocked in __split_huge_page_splitting(), where the main | |
5620 | * part of thp split is not executed yet. | |
5621 | */ | |
5622 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 5623 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
5624 | spin_unlock(&vma->vm_mm->page_table_lock); |
5625 | return 0; | |
5626 | } | |
5627 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5628 | if (target_type == MC_TARGET_PAGE) { | |
5629 | page = target.page; | |
5630 | if (!isolate_lru_page(page)) { | |
5631 | pc = lookup_page_cgroup(page); | |
5632 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 5633 | pc, mc.from, mc.to)) { |
12724850 NH |
5634 | mc.precharge -= HPAGE_PMD_NR; |
5635 | mc.moved_charge += HPAGE_PMD_NR; | |
5636 | } | |
5637 | putback_lru_page(page); | |
5638 | } | |
5639 | put_page(page); | |
5640 | } | |
5641 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 5642 | return 0; |
12724850 NH |
5643 | } |
5644 | ||
45f83cef AA |
5645 | if (pmd_trans_unstable(pmd)) |
5646 | return 0; | |
4ffef5fe DN |
5647 | retry: |
5648 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5649 | for (; addr != end; addr += PAGE_SIZE) { | |
5650 | pte_t ptent = *(pte++); | |
02491447 | 5651 | swp_entry_t ent; |
4ffef5fe DN |
5652 | |
5653 | if (!mc.precharge) | |
5654 | break; | |
5655 | ||
8d32ff84 | 5656 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
5657 | case MC_TARGET_PAGE: |
5658 | page = target.page; | |
5659 | if (isolate_lru_page(page)) | |
5660 | goto put; | |
5661 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 5662 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 5663 | mc.from, mc.to)) { |
4ffef5fe | 5664 | mc.precharge--; |
854ffa8d DN |
5665 | /* we uncharge from mc.from later. */ |
5666 | mc.moved_charge++; | |
4ffef5fe DN |
5667 | } |
5668 | putback_lru_page(page); | |
8d32ff84 | 5669 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5670 | put_page(page); |
5671 | break; | |
02491447 DN |
5672 | case MC_TARGET_SWAP: |
5673 | ent = target.ent; | |
e91cbb42 | 5674 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5675 | mc.precharge--; |
483c30b5 DN |
5676 | /* we fixup refcnts and charges later. */ |
5677 | mc.moved_swap++; | |
5678 | } | |
02491447 | 5679 | break; |
4ffef5fe DN |
5680 | default: |
5681 | break; | |
5682 | } | |
5683 | } | |
5684 | pte_unmap_unlock(pte - 1, ptl); | |
5685 | cond_resched(); | |
5686 | ||
5687 | if (addr != end) { | |
5688 | /* | |
5689 | * We have consumed all precharges we got in can_attach(). | |
5690 | * We try charge one by one, but don't do any additional | |
5691 | * charges to mc.to if we have failed in charge once in attach() | |
5692 | * phase. | |
5693 | */ | |
854ffa8d | 5694 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5695 | if (!ret) |
5696 | goto retry; | |
5697 | } | |
5698 | ||
5699 | return ret; | |
5700 | } | |
5701 | ||
5702 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5703 | { | |
5704 | struct vm_area_struct *vma; | |
5705 | ||
5706 | lru_add_drain_all(); | |
dfe076b0 DN |
5707 | retry: |
5708 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5709 | /* | |
5710 | * Someone who are holding the mmap_sem might be waiting in | |
5711 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5712 | * and retry. Because we cancel precharges, we might not be able | |
5713 | * to move enough charges, but moving charge is a best-effort | |
5714 | * feature anyway, so it wouldn't be a big problem. | |
5715 | */ | |
5716 | __mem_cgroup_clear_mc(); | |
5717 | cond_resched(); | |
5718 | goto retry; | |
5719 | } | |
4ffef5fe DN |
5720 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5721 | int ret; | |
5722 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5723 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5724 | .mm = mm, | |
5725 | .private = vma, | |
5726 | }; | |
5727 | if (is_vm_hugetlb_page(vma)) | |
5728 | continue; | |
4ffef5fe DN |
5729 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5730 | &mem_cgroup_move_charge_walk); | |
5731 | if (ret) | |
5732 | /* | |
5733 | * means we have consumed all precharges and failed in | |
5734 | * doing additional charge. Just abandon here. | |
5735 | */ | |
5736 | break; | |
5737 | } | |
dfe076b0 | 5738 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5739 | } |
5740 | ||
761b3ef5 LZ |
5741 | static void mem_cgroup_move_task(struct cgroup *cont, |
5742 | struct cgroup_taskset *tset) | |
67e465a7 | 5743 | { |
2f7ee569 | 5744 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 5745 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5746 | |
dfe076b0 | 5747 | if (mm) { |
a433658c KM |
5748 | if (mc.to) |
5749 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5750 | mmput(mm); |
5751 | } | |
a433658c KM |
5752 | if (mc.to) |
5753 | mem_cgroup_clear_mc(); | |
67e465a7 | 5754 | } |
5cfb80a7 | 5755 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
5756 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
5757 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5758 | { |
5759 | return 0; | |
5760 | } | |
761b3ef5 LZ |
5761 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
5762 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5763 | { |
5764 | } | |
761b3ef5 LZ |
5765 | static void mem_cgroup_move_task(struct cgroup *cont, |
5766 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
5767 | { |
5768 | } | |
5769 | #endif | |
67e465a7 | 5770 | |
8cdea7c0 BS |
5771 | struct cgroup_subsys mem_cgroup_subsys = { |
5772 | .name = "memory", | |
5773 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 TH |
5774 | .css_alloc = mem_cgroup_css_alloc, |
5775 | .css_offline = mem_cgroup_css_offline, | |
5776 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
5777 | .can_attach = mem_cgroup_can_attach, |
5778 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5779 | .attach = mem_cgroup_move_task, |
6bc10349 | 5780 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 5781 | .early_init = 0, |
04046e1a | 5782 | .use_id = 1, |
8cdea7c0 | 5783 | }; |
c077719b | 5784 | |
c255a458 | 5785 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
5786 | static int __init enable_swap_account(char *s) |
5787 | { | |
5788 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5789 | if (!strcmp(s, "1")) |
a42c390c | 5790 | really_do_swap_account = 1; |
a2c8990a | 5791 | else if (!strcmp(s, "0")) |
a42c390c MH |
5792 | really_do_swap_account = 0; |
5793 | return 1; | |
5794 | } | |
a2c8990a | 5795 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5796 | |
c077719b | 5797 | #endif |