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 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
8cdea7c0 BS |
17 | * This program is free software; you can redistribute it and/or modify |
18 | * it under the terms of the GNU General Public License as published by | |
19 | * the Free Software Foundation; either version 2 of the License, or | |
20 | * (at your option) any later version. | |
21 | * | |
22 | * This program is distributed in the hope that it will be useful, | |
23 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
24 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
25 | * GNU General Public License for more details. | |
26 | */ | |
27 | ||
28 | #include <linux/res_counter.h> | |
29 | #include <linux/memcontrol.h> | |
30 | #include <linux/cgroup.h> | |
78fb7466 | 31 | #include <linux/mm.h> |
4ffef5fe | 32 | #include <linux/hugetlb.h> |
d13d1443 | 33 | #include <linux/pagemap.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
f64c3f54 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 KS |
47 | #include <linux/eventfd.h> |
48 | #include <linux/sort.h> | |
66e1707b | 49 | #include <linux/fs.h> |
d2ceb9b7 | 50 | #include <linux/seq_file.h> |
33327948 | 51 | #include <linux/vmalloc.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
52d4b9ac | 54 | #include <linux/page_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
08e552c6 | 57 | #include "internal.h" |
d1a4c0b3 | 58 | #include <net/sock.h> |
4bd2c1ee | 59 | #include <net/ip.h> |
d1a4c0b3 | 60 | #include <net/tcp_memcontrol.h> |
8cdea7c0 | 61 | |
8697d331 BS |
62 | #include <asm/uaccess.h> |
63 | ||
cc8e970c KM |
64 | #include <trace/events/vmscan.h> |
65 | ||
a181b0e8 | 66 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
68ae564b DR |
67 | EXPORT_SYMBOL(mem_cgroup_subsys); |
68 | ||
a181b0e8 | 69 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
6bbda35c | 70 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 71 | |
c255a458 | 72 | #ifdef CONFIG_MEMCG_SWAP |
338c8431 | 73 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 74 | int do_swap_account __read_mostly; |
a42c390c MH |
75 | |
76 | /* for remember boot option*/ | |
c255a458 | 77 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
a42c390c MH |
78 | static int really_do_swap_account __initdata = 1; |
79 | #else | |
80 | static int really_do_swap_account __initdata = 0; | |
81 | #endif | |
82 | ||
c077719b | 83 | #else |
a0db00fc | 84 | #define do_swap_account 0 |
c077719b KH |
85 | #endif |
86 | ||
87 | ||
d52aa412 KH |
88 | /* |
89 | * Statistics for memory cgroup. | |
90 | */ | |
91 | enum mem_cgroup_stat_index { | |
92 | /* | |
93 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
94 | */ | |
95 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 96 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 97 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
bff6bb83 | 98 | MEM_CGROUP_STAT_SWAP, /* # of pages, swapped out */ |
d52aa412 KH |
99 | MEM_CGROUP_STAT_NSTATS, |
100 | }; | |
101 | ||
af7c4b0e JW |
102 | static const char * const mem_cgroup_stat_names[] = { |
103 | "cache", | |
104 | "rss", | |
105 | "mapped_file", | |
106 | "swap", | |
107 | }; | |
108 | ||
e9f8974f JW |
109 | enum mem_cgroup_events_index { |
110 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
111 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
456f998e YH |
112 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
113 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
114 | MEM_CGROUP_EVENTS_NSTATS, |
115 | }; | |
af7c4b0e JW |
116 | |
117 | static const char * const mem_cgroup_events_names[] = { | |
118 | "pgpgin", | |
119 | "pgpgout", | |
120 | "pgfault", | |
121 | "pgmajfault", | |
122 | }; | |
123 | ||
58cf188e SZ |
124 | static const char * const mem_cgroup_lru_names[] = { |
125 | "inactive_anon", | |
126 | "active_anon", | |
127 | "inactive_file", | |
128 | "active_file", | |
129 | "unevictable", | |
130 | }; | |
131 | ||
7a159cc9 JW |
132 | /* |
133 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
134 | * it will be incremated by the number of pages. This counter is used for | |
135 | * for trigger some periodic events. This is straightforward and better | |
136 | * than using jiffies etc. to handle periodic memcg event. | |
137 | */ | |
138 | enum mem_cgroup_events_target { | |
139 | MEM_CGROUP_TARGET_THRESH, | |
140 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 141 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
142 | MEM_CGROUP_NTARGETS, |
143 | }; | |
a0db00fc KS |
144 | #define THRESHOLDS_EVENTS_TARGET 128 |
145 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
146 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 147 | |
d52aa412 | 148 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 149 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 150 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
13114716 | 151 | unsigned long nr_page_events; |
7a159cc9 | 152 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
153 | }; |
154 | ||
527a5ec9 | 155 | struct mem_cgroup_reclaim_iter { |
5f578161 MH |
156 | /* |
157 | * last scanned hierarchy member. Valid only if last_dead_count | |
158 | * matches memcg->dead_count of the hierarchy root group. | |
159 | */ | |
542f85f9 | 160 | struct mem_cgroup *last_visited; |
5f578161 MH |
161 | unsigned long last_dead_count; |
162 | ||
527a5ec9 JW |
163 | /* scan generation, increased every round-trip */ |
164 | unsigned int generation; | |
165 | }; | |
166 | ||
6d12e2d8 KH |
167 | /* |
168 | * per-zone information in memory controller. | |
169 | */ | |
6d12e2d8 | 170 | struct mem_cgroup_per_zone { |
6290df54 | 171 | struct lruvec lruvec; |
1eb49272 | 172 | unsigned long lru_size[NR_LRU_LISTS]; |
3e2f41f1 | 173 | |
527a5ec9 JW |
174 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
175 | ||
f64c3f54 BS |
176 | struct rb_node tree_node; /* RB tree node */ |
177 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
178 | /* the soft limit is exceeded*/ | |
179 | bool on_tree; | |
d79154bb | 180 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
4e416953 | 181 | /* use container_of */ |
6d12e2d8 | 182 | }; |
6d12e2d8 KH |
183 | |
184 | struct mem_cgroup_per_node { | |
185 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
186 | }; | |
187 | ||
188 | struct mem_cgroup_lru_info { | |
45cf7ebd | 189 | struct mem_cgroup_per_node *nodeinfo[0]; |
6d12e2d8 KH |
190 | }; |
191 | ||
f64c3f54 BS |
192 | /* |
193 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
194 | * their hierarchy representation | |
195 | */ | |
196 | ||
197 | struct mem_cgroup_tree_per_zone { | |
198 | struct rb_root rb_root; | |
199 | spinlock_t lock; | |
200 | }; | |
201 | ||
202 | struct mem_cgroup_tree_per_node { | |
203 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
204 | }; | |
205 | ||
206 | struct mem_cgroup_tree { | |
207 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
208 | }; | |
209 | ||
210 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
211 | ||
2e72b634 KS |
212 | struct mem_cgroup_threshold { |
213 | struct eventfd_ctx *eventfd; | |
214 | u64 threshold; | |
215 | }; | |
216 | ||
9490ff27 | 217 | /* For threshold */ |
2e72b634 | 218 | struct mem_cgroup_threshold_ary { |
748dad36 | 219 | /* An array index points to threshold just below or equal to usage. */ |
5407a562 | 220 | int current_threshold; |
2e72b634 KS |
221 | /* Size of entries[] */ |
222 | unsigned int size; | |
223 | /* Array of thresholds */ | |
224 | struct mem_cgroup_threshold entries[0]; | |
225 | }; | |
2c488db2 KS |
226 | |
227 | struct mem_cgroup_thresholds { | |
228 | /* Primary thresholds array */ | |
229 | struct mem_cgroup_threshold_ary *primary; | |
230 | /* | |
231 | * Spare threshold array. | |
232 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
233 | * It must be able to store at least primary->size - 1 entries. | |
234 | */ | |
235 | struct mem_cgroup_threshold_ary *spare; | |
236 | }; | |
237 | ||
9490ff27 KH |
238 | /* for OOM */ |
239 | struct mem_cgroup_eventfd_list { | |
240 | struct list_head list; | |
241 | struct eventfd_ctx *eventfd; | |
242 | }; | |
2e72b634 | 243 | |
c0ff4b85 R |
244 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
245 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 246 | |
8cdea7c0 BS |
247 | /* |
248 | * The memory controller data structure. The memory controller controls both | |
249 | * page cache and RSS per cgroup. We would eventually like to provide | |
250 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
251 | * to help the administrator determine what knobs to tune. | |
252 | * | |
253 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
254 | * we hit the water mark. May be even add a low water mark, such that |
255 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
256 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
257 | */ |
258 | struct mem_cgroup { | |
259 | struct cgroup_subsys_state css; | |
260 | /* | |
261 | * the counter to account for memory usage | |
262 | */ | |
263 | struct res_counter res; | |
59927fb9 | 264 | |
70ddf637 AV |
265 | /* vmpressure notifications */ |
266 | struct vmpressure vmpressure; | |
267 | ||
59927fb9 HD |
268 | union { |
269 | /* | |
270 | * the counter to account for mem+swap usage. | |
271 | */ | |
272 | struct res_counter memsw; | |
273 | ||
274 | /* | |
275 | * rcu_freeing is used only when freeing struct mem_cgroup, | |
276 | * so put it into a union to avoid wasting more memory. | |
277 | * It must be disjoint from the css field. It could be | |
278 | * in a union with the res field, but res plays a much | |
279 | * larger part in mem_cgroup life than memsw, and might | |
280 | * be of interest, even at time of free, when debugging. | |
281 | * So share rcu_head with the less interesting memsw. | |
282 | */ | |
283 | struct rcu_head rcu_freeing; | |
284 | /* | |
3afe36b1 GC |
285 | * We also need some space for a worker in deferred freeing. |
286 | * By the time we call it, rcu_freeing is no longer in use. | |
59927fb9 HD |
287 | */ |
288 | struct work_struct work_freeing; | |
289 | }; | |
290 | ||
510fc4e1 GC |
291 | /* |
292 | * the counter to account for kernel memory usage. | |
293 | */ | |
294 | struct res_counter kmem; | |
18f59ea7 BS |
295 | /* |
296 | * Should the accounting and control be hierarchical, per subtree? | |
297 | */ | |
298 | bool use_hierarchy; | |
510fc4e1 | 299 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
79dfdacc MH |
300 | |
301 | bool oom_lock; | |
302 | atomic_t under_oom; | |
303 | ||
8c7c6e34 | 304 | atomic_t refcnt; |
14797e23 | 305 | |
1f4c025b | 306 | int swappiness; |
3c11ecf4 KH |
307 | /* OOM-Killer disable */ |
308 | int oom_kill_disable; | |
a7885eb8 | 309 | |
22a668d7 KH |
310 | /* set when res.limit == memsw.limit */ |
311 | bool memsw_is_minimum; | |
312 | ||
2e72b634 KS |
313 | /* protect arrays of thresholds */ |
314 | struct mutex thresholds_lock; | |
315 | ||
316 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 317 | struct mem_cgroup_thresholds thresholds; |
907860ed | 318 | |
2e72b634 | 319 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 320 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 321 | |
9490ff27 KH |
322 | /* For oom notifier event fd */ |
323 | struct list_head oom_notify; | |
185efc0f | 324 | |
7dc74be0 DN |
325 | /* |
326 | * Should we move charges of a task when a task is moved into this | |
327 | * mem_cgroup ? And what type of charges should we move ? | |
328 | */ | |
329 | unsigned long move_charge_at_immigrate; | |
619d094b KH |
330 | /* |
331 | * set > 0 if pages under this cgroup are moving to other cgroup. | |
332 | */ | |
333 | atomic_t moving_account; | |
312734c0 KH |
334 | /* taken only while moving_account > 0 */ |
335 | spinlock_t move_lock; | |
d52aa412 | 336 | /* |
c62b1a3b | 337 | * percpu counter. |
d52aa412 | 338 | */ |
3a7951b4 | 339 | struct mem_cgroup_stat_cpu __percpu *stat; |
711d3d2c KH |
340 | /* |
341 | * used when a cpu is offlined or other synchronizations | |
342 | * See mem_cgroup_read_stat(). | |
343 | */ | |
344 | struct mem_cgroup_stat_cpu nocpu_base; | |
345 | spinlock_t pcp_counter_lock; | |
d1a4c0b3 | 346 | |
5f578161 | 347 | atomic_t dead_count; |
4bd2c1ee | 348 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
d1a4c0b3 GC |
349 | struct tcp_memcontrol tcp_mem; |
350 | #endif | |
2633d7a0 GC |
351 | #if defined(CONFIG_MEMCG_KMEM) |
352 | /* analogous to slab_common's slab_caches list. per-memcg */ | |
353 | struct list_head memcg_slab_caches; | |
354 | /* Not a spinlock, we can take a lot of time walking the list */ | |
355 | struct mutex slab_caches_mutex; | |
356 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ | |
357 | int kmemcg_id; | |
358 | #endif | |
45cf7ebd GC |
359 | |
360 | int last_scanned_node; | |
361 | #if MAX_NUMNODES > 1 | |
362 | nodemask_t scan_nodes; | |
363 | atomic_t numainfo_events; | |
364 | atomic_t numainfo_updating; | |
365 | #endif | |
70ddf637 | 366 | |
45cf7ebd GC |
367 | /* |
368 | * Per cgroup active and inactive list, similar to the | |
369 | * per zone LRU lists. | |
370 | * | |
371 | * WARNING: This has to be the last element of the struct. Don't | |
372 | * add new fields after this point. | |
373 | */ | |
374 | struct mem_cgroup_lru_info info; | |
8cdea7c0 BS |
375 | }; |
376 | ||
45cf7ebd GC |
377 | static size_t memcg_size(void) |
378 | { | |
379 | return sizeof(struct mem_cgroup) + | |
380 | nr_node_ids * sizeof(struct mem_cgroup_per_node); | |
381 | } | |
382 | ||
510fc4e1 GC |
383 | /* internal only representation about the status of kmem accounting. */ |
384 | enum { | |
385 | KMEM_ACCOUNTED_ACTIVE = 0, /* accounted by this cgroup itself */ | |
a8964b9b | 386 | KMEM_ACCOUNTED_ACTIVATED, /* static key enabled. */ |
7de37682 | 387 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
510fc4e1 GC |
388 | }; |
389 | ||
a8964b9b GC |
390 | /* We account when limit is on, but only after call sites are patched */ |
391 | #define KMEM_ACCOUNTED_MASK \ | |
392 | ((1 << KMEM_ACCOUNTED_ACTIVE) | (1 << KMEM_ACCOUNTED_ACTIVATED)) | |
510fc4e1 GC |
393 | |
394 | #ifdef CONFIG_MEMCG_KMEM | |
395 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) | |
396 | { | |
397 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
398 | } | |
7de37682 GC |
399 | |
400 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) | |
401 | { | |
402 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); | |
403 | } | |
404 | ||
a8964b9b GC |
405 | static void memcg_kmem_set_activated(struct mem_cgroup *memcg) |
406 | { | |
407 | set_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
408 | } | |
409 | ||
55007d84 GC |
410 | static void memcg_kmem_clear_activated(struct mem_cgroup *memcg) |
411 | { | |
412 | clear_bit(KMEM_ACCOUNTED_ACTIVATED, &memcg->kmem_account_flags); | |
413 | } | |
414 | ||
7de37682 GC |
415 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
416 | { | |
417 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) | |
418 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); | |
419 | } | |
420 | ||
421 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) | |
422 | { | |
423 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, | |
424 | &memcg->kmem_account_flags); | |
425 | } | |
510fc4e1 GC |
426 | #endif |
427 | ||
7dc74be0 DN |
428 | /* Stuffs for move charges at task migration. */ |
429 | /* | |
ee5e8472 GC |
430 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
431 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. | |
7dc74be0 DN |
432 | */ |
433 | enum move_type { | |
4ffef5fe | 434 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 435 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
436 | NR_MOVE_TYPE, |
437 | }; | |
438 | ||
4ffef5fe DN |
439 | /* "mc" and its members are protected by cgroup_mutex */ |
440 | static struct move_charge_struct { | |
b1dd693e | 441 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
442 | struct mem_cgroup *from; |
443 | struct mem_cgroup *to; | |
ee5e8472 | 444 | unsigned long immigrate_flags; |
4ffef5fe | 445 | unsigned long precharge; |
854ffa8d | 446 | unsigned long moved_charge; |
483c30b5 | 447 | unsigned long moved_swap; |
8033b97c DN |
448 | struct task_struct *moving_task; /* a task moving charges */ |
449 | wait_queue_head_t waitq; /* a waitq for other context */ | |
450 | } mc = { | |
2bd9bb20 | 451 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
452 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
453 | }; | |
4ffef5fe | 454 | |
90254a65 DN |
455 | static bool move_anon(void) |
456 | { | |
ee5e8472 | 457 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
90254a65 DN |
458 | } |
459 | ||
87946a72 DN |
460 | static bool move_file(void) |
461 | { | |
ee5e8472 | 462 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
87946a72 DN |
463 | } |
464 | ||
4e416953 BS |
465 | /* |
466 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
467 | * limit reclaim to prevent infinite loops, if they ever occur. | |
468 | */ | |
a0db00fc KS |
469 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
470 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 | |
4e416953 | 471 | |
217bc319 KH |
472 | enum charge_type { |
473 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 474 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 475 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 476 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
477 | NR_CHARGE_TYPE, |
478 | }; | |
479 | ||
8c7c6e34 | 480 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
481 | enum res_type { |
482 | _MEM, | |
483 | _MEMSWAP, | |
484 | _OOM_TYPE, | |
510fc4e1 | 485 | _KMEM, |
86ae53e1 GC |
486 | }; |
487 | ||
a0db00fc KS |
488 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
489 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 490 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
491 | /* Used for OOM nofiier */ |
492 | #define OOM_CONTROL (0) | |
8c7c6e34 | 493 | |
75822b44 BS |
494 | /* |
495 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
496 | */ | |
497 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
498 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
499 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
500 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
501 | ||
0999821b GC |
502 | /* |
503 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
504 | * As a consequence, any change that needs to protect against new child cgroups | |
505 | * appearing has to hold it as well. | |
506 | */ | |
507 | static DEFINE_MUTEX(memcg_create_mutex); | |
508 | ||
c0ff4b85 R |
509 | static void mem_cgroup_get(struct mem_cgroup *memcg); |
510 | static void mem_cgroup_put(struct mem_cgroup *memcg); | |
e1aab161 | 511 | |
b2145145 WL |
512 | static inline |
513 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) | |
514 | { | |
515 | return container_of(s, struct mem_cgroup, css); | |
516 | } | |
517 | ||
70ddf637 AV |
518 | /* Some nice accessors for the vmpressure. */ |
519 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
520 | { | |
521 | if (!memcg) | |
522 | memcg = root_mem_cgroup; | |
523 | return &memcg->vmpressure; | |
524 | } | |
525 | ||
526 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
527 | { | |
528 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
529 | } | |
530 | ||
531 | struct vmpressure *css_to_vmpressure(struct cgroup_subsys_state *css) | |
532 | { | |
533 | return &mem_cgroup_from_css(css)->vmpressure; | |
534 | } | |
535 | ||
7ffc0edc MH |
536 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
537 | { | |
538 | return (memcg == root_mem_cgroup); | |
539 | } | |
540 | ||
e1aab161 | 541 | /* Writing them here to avoid exposing memcg's inner layout */ |
4bd2c1ee | 542 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
e1aab161 | 543 | |
e1aab161 GC |
544 | void sock_update_memcg(struct sock *sk) |
545 | { | |
376be5ff | 546 | if (mem_cgroup_sockets_enabled) { |
e1aab161 | 547 | struct mem_cgroup *memcg; |
3f134619 | 548 | struct cg_proto *cg_proto; |
e1aab161 GC |
549 | |
550 | BUG_ON(!sk->sk_prot->proto_cgroup); | |
551 | ||
f3f511e1 GC |
552 | /* Socket cloning can throw us here with sk_cgrp already |
553 | * filled. It won't however, necessarily happen from | |
554 | * process context. So the test for root memcg given | |
555 | * the current task's memcg won't help us in this case. | |
556 | * | |
557 | * Respecting the original socket's memcg is a better | |
558 | * decision in this case. | |
559 | */ | |
560 | if (sk->sk_cgrp) { | |
561 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); | |
562 | mem_cgroup_get(sk->sk_cgrp->memcg); | |
563 | return; | |
564 | } | |
565 | ||
e1aab161 GC |
566 | rcu_read_lock(); |
567 | memcg = mem_cgroup_from_task(current); | |
3f134619 GC |
568 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
569 | if (!mem_cgroup_is_root(memcg) && memcg_proto_active(cg_proto)) { | |
e1aab161 | 570 | mem_cgroup_get(memcg); |
3f134619 | 571 | sk->sk_cgrp = cg_proto; |
e1aab161 GC |
572 | } |
573 | rcu_read_unlock(); | |
574 | } | |
575 | } | |
576 | EXPORT_SYMBOL(sock_update_memcg); | |
577 | ||
578 | void sock_release_memcg(struct sock *sk) | |
579 | { | |
376be5ff | 580 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
e1aab161 GC |
581 | struct mem_cgroup *memcg; |
582 | WARN_ON(!sk->sk_cgrp->memcg); | |
583 | memcg = sk->sk_cgrp->memcg; | |
584 | mem_cgroup_put(memcg); | |
585 | } | |
586 | } | |
d1a4c0b3 GC |
587 | |
588 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) | |
589 | { | |
590 | if (!memcg || mem_cgroup_is_root(memcg)) | |
591 | return NULL; | |
592 | ||
593 | return &memcg->tcp_mem.cg_proto; | |
594 | } | |
595 | EXPORT_SYMBOL(tcp_proto_cgroup); | |
e1aab161 | 596 | |
3f134619 GC |
597 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
598 | { | |
599 | if (!memcg_proto_activated(&memcg->tcp_mem.cg_proto)) | |
600 | return; | |
601 | static_key_slow_dec(&memcg_socket_limit_enabled); | |
602 | } | |
603 | #else | |
604 | static void disarm_sock_keys(struct mem_cgroup *memcg) | |
605 | { | |
606 | } | |
607 | #endif | |
608 | ||
a8964b9b | 609 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 GC |
610 | /* |
611 | * This will be the memcg's index in each cache's ->memcg_params->memcg_caches. | |
612 | * There are two main reasons for not using the css_id for this: | |
613 | * 1) this works better in sparse environments, where we have a lot of memcgs, | |
614 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
615 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
616 | * 200 entry array for that. | |
617 | * | |
618 | * 2) In order not to violate the cgroup API, we would like to do all memory | |
619 | * allocation in ->create(). At that point, we haven't yet allocated the | |
620 | * css_id. Having a separate index prevents us from messing with the cgroup | |
621 | * core for this | |
622 | * | |
623 | * The current size of the caches array is stored in | |
624 | * memcg_limited_groups_array_size. It will double each time we have to | |
625 | * increase it. | |
626 | */ | |
627 | static DEFINE_IDA(kmem_limited_groups); | |
749c5415 GC |
628 | int memcg_limited_groups_array_size; |
629 | ||
55007d84 GC |
630 | /* |
631 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
632 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
633 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
634 | * tunable, but that is strictly not necessary. | |
635 | * | |
636 | * MAX_SIZE should be as large as the number of css_ids. Ideally, we could get | |
637 | * this constant directly from cgroup, but it is understandable that this is | |
638 | * better kept as an internal representation in cgroup.c. In any case, the | |
639 | * css_id space is not getting any smaller, and we don't have to necessarily | |
640 | * increase ours as well if it increases. | |
641 | */ | |
642 | #define MEMCG_CACHES_MIN_SIZE 4 | |
643 | #define MEMCG_CACHES_MAX_SIZE 65535 | |
644 | ||
d7f25f8a GC |
645 | /* |
646 | * A lot of the calls to the cache allocation functions are expected to be | |
647 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
648 | * conditional to this static branch, we'll have to allow modules that does | |
649 | * kmem_cache_alloc and the such to see this symbol as well | |
650 | */ | |
a8964b9b | 651 | struct static_key memcg_kmem_enabled_key; |
d7f25f8a | 652 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b GC |
653 | |
654 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
655 | { | |
55007d84 | 656 | if (memcg_kmem_is_active(memcg)) { |
a8964b9b | 657 | static_key_slow_dec(&memcg_kmem_enabled_key); |
55007d84 GC |
658 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
659 | } | |
bea207c8 GC |
660 | /* |
661 | * This check can't live in kmem destruction function, | |
662 | * since the charges will outlive the cgroup | |
663 | */ | |
664 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); | |
a8964b9b GC |
665 | } |
666 | #else | |
667 | static void disarm_kmem_keys(struct mem_cgroup *memcg) | |
668 | { | |
669 | } | |
670 | #endif /* CONFIG_MEMCG_KMEM */ | |
671 | ||
672 | static void disarm_static_keys(struct mem_cgroup *memcg) | |
673 | { | |
674 | disarm_sock_keys(memcg); | |
675 | disarm_kmem_keys(memcg); | |
676 | } | |
677 | ||
c0ff4b85 | 678 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
8c7c6e34 | 679 | |
f64c3f54 | 680 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 681 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
f64c3f54 | 682 | { |
45cf7ebd | 683 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
c0ff4b85 | 684 | return &memcg->info.nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
685 | } |
686 | ||
c0ff4b85 | 687 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
d324236b | 688 | { |
c0ff4b85 | 689 | return &memcg->css; |
d324236b WF |
690 | } |
691 | ||
f64c3f54 | 692 | static struct mem_cgroup_per_zone * |
c0ff4b85 | 693 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 694 | { |
97a6c37b JW |
695 | int nid = page_to_nid(page); |
696 | int zid = page_zonenum(page); | |
f64c3f54 | 697 | |
c0ff4b85 | 698 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
f64c3f54 BS |
699 | } |
700 | ||
701 | static struct mem_cgroup_tree_per_zone * | |
702 | soft_limit_tree_node_zone(int nid, int zid) | |
703 | { | |
704 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
705 | } | |
706 | ||
707 | static struct mem_cgroup_tree_per_zone * | |
708 | soft_limit_tree_from_page(struct page *page) | |
709 | { | |
710 | int nid = page_to_nid(page); | |
711 | int zid = page_zonenum(page); | |
712 | ||
713 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
714 | } | |
715 | ||
716 | static void | |
c0ff4b85 | 717 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
f64c3f54 | 718 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
719 | struct mem_cgroup_tree_per_zone *mctz, |
720 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
721 | { |
722 | struct rb_node **p = &mctz->rb_root.rb_node; | |
723 | struct rb_node *parent = NULL; | |
724 | struct mem_cgroup_per_zone *mz_node; | |
725 | ||
726 | if (mz->on_tree) | |
727 | return; | |
728 | ||
ef8745c1 KH |
729 | mz->usage_in_excess = new_usage_in_excess; |
730 | if (!mz->usage_in_excess) | |
731 | return; | |
f64c3f54 BS |
732 | while (*p) { |
733 | parent = *p; | |
734 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
735 | tree_node); | |
736 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
737 | p = &(*p)->rb_left; | |
738 | /* | |
739 | * We can't avoid mem cgroups that are over their soft | |
740 | * limit by the same amount | |
741 | */ | |
742 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
743 | p = &(*p)->rb_right; | |
744 | } | |
745 | rb_link_node(&mz->tree_node, parent, p); | |
746 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
747 | mz->on_tree = true; | |
4e416953 BS |
748 | } |
749 | ||
750 | static void | |
c0ff4b85 | 751 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
4e416953 BS |
752 | struct mem_cgroup_per_zone *mz, |
753 | struct mem_cgroup_tree_per_zone *mctz) | |
754 | { | |
755 | if (!mz->on_tree) | |
756 | return; | |
757 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
758 | mz->on_tree = false; | |
759 | } | |
760 | ||
f64c3f54 | 761 | static void |
c0ff4b85 | 762 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
f64c3f54 BS |
763 | struct mem_cgroup_per_zone *mz, |
764 | struct mem_cgroup_tree_per_zone *mctz) | |
765 | { | |
766 | spin_lock(&mctz->lock); | |
c0ff4b85 | 767 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
768 | spin_unlock(&mctz->lock); |
769 | } | |
770 | ||
f64c3f54 | 771 | |
c0ff4b85 | 772 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 773 | { |
ef8745c1 | 774 | unsigned long long excess; |
f64c3f54 BS |
775 | struct mem_cgroup_per_zone *mz; |
776 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
777 | int nid = page_to_nid(page); |
778 | int zid = page_zonenum(page); | |
f64c3f54 BS |
779 | mctz = soft_limit_tree_from_page(page); |
780 | ||
781 | /* | |
4e649152 KH |
782 | * Necessary to update all ancestors when hierarchy is used. |
783 | * because their event counter is not touched. | |
f64c3f54 | 784 | */ |
c0ff4b85 R |
785 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
786 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
787 | excess = res_counter_soft_limit_excess(&memcg->res); | |
4e649152 KH |
788 | /* |
789 | * We have to update the tree if mz is on RB-tree or | |
790 | * mem is over its softlimit. | |
791 | */ | |
ef8745c1 | 792 | if (excess || mz->on_tree) { |
4e649152 KH |
793 | spin_lock(&mctz->lock); |
794 | /* if on-tree, remove it */ | |
795 | if (mz->on_tree) | |
c0ff4b85 | 796 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
4e649152 | 797 | /* |
ef8745c1 KH |
798 | * Insert again. mz->usage_in_excess will be updated. |
799 | * If excess is 0, no tree ops. | |
4e649152 | 800 | */ |
c0ff4b85 | 801 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
4e649152 KH |
802 | spin_unlock(&mctz->lock); |
803 | } | |
f64c3f54 BS |
804 | } |
805 | } | |
806 | ||
c0ff4b85 | 807 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
f64c3f54 BS |
808 | { |
809 | int node, zone; | |
810 | struct mem_cgroup_per_zone *mz; | |
811 | struct mem_cgroup_tree_per_zone *mctz; | |
812 | ||
3ed28fa1 | 813 | for_each_node(node) { |
f64c3f54 | 814 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
c0ff4b85 | 815 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
f64c3f54 | 816 | mctz = soft_limit_tree_node_zone(node, zone); |
c0ff4b85 | 817 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
f64c3f54 BS |
818 | } |
819 | } | |
820 | } | |
821 | ||
4e416953 BS |
822 | static struct mem_cgroup_per_zone * |
823 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
824 | { | |
825 | struct rb_node *rightmost = NULL; | |
26251eaf | 826 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
827 | |
828 | retry: | |
26251eaf | 829 | mz = NULL; |
4e416953 BS |
830 | rightmost = rb_last(&mctz->rb_root); |
831 | if (!rightmost) | |
832 | goto done; /* Nothing to reclaim from */ | |
833 | ||
834 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
835 | /* | |
836 | * Remove the node now but someone else can add it back, | |
837 | * we will to add it back at the end of reclaim to its correct | |
838 | * position in the tree. | |
839 | */ | |
d79154bb HD |
840 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
841 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || | |
842 | !css_tryget(&mz->memcg->css)) | |
4e416953 BS |
843 | goto retry; |
844 | done: | |
845 | return mz; | |
846 | } | |
847 | ||
848 | static struct mem_cgroup_per_zone * | |
849 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
850 | { | |
851 | struct mem_cgroup_per_zone *mz; | |
852 | ||
853 | spin_lock(&mctz->lock); | |
854 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
855 | spin_unlock(&mctz->lock); | |
856 | return mz; | |
857 | } | |
858 | ||
711d3d2c KH |
859 | /* |
860 | * Implementation Note: reading percpu statistics for memcg. | |
861 | * | |
862 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
863 | * synchronization to implement "quick" read. There are trade-off between | |
864 | * reading cost and precision of value. Then, we may have a chance to implement | |
865 | * a periodic synchronizion of counter in memcg's counter. | |
866 | * | |
867 | * But this _read() function is used for user interface now. The user accounts | |
868 | * memory usage by memory cgroup and he _always_ requires exact value because | |
869 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
870 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
871 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
872 | * | |
873 | * If there are kernel internal actions which can make use of some not-exact | |
874 | * value, and reading all cpu value can be performance bottleneck in some | |
875 | * common workload, threashold and synchonization as vmstat[] should be | |
876 | * implemented. | |
877 | */ | |
c0ff4b85 | 878 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
7a159cc9 | 879 | enum mem_cgroup_stat_index idx) |
c62b1a3b | 880 | { |
7a159cc9 | 881 | long val = 0; |
c62b1a3b | 882 | int cpu; |
c62b1a3b | 883 | |
711d3d2c KH |
884 | get_online_cpus(); |
885 | for_each_online_cpu(cpu) | |
c0ff4b85 | 886 | val += per_cpu(memcg->stat->count[idx], cpu); |
711d3d2c | 887 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
888 | spin_lock(&memcg->pcp_counter_lock); |
889 | val += memcg->nocpu_base.count[idx]; | |
890 | spin_unlock(&memcg->pcp_counter_lock); | |
711d3d2c KH |
891 | #endif |
892 | put_online_cpus(); | |
c62b1a3b KH |
893 | return val; |
894 | } | |
895 | ||
c0ff4b85 | 896 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
0c3e73e8 BS |
897 | bool charge) |
898 | { | |
899 | int val = (charge) ? 1 : -1; | |
bff6bb83 | 900 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
0c3e73e8 BS |
901 | } |
902 | ||
c0ff4b85 | 903 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
904 | enum mem_cgroup_events_index idx) |
905 | { | |
906 | unsigned long val = 0; | |
907 | int cpu; | |
908 | ||
909 | for_each_online_cpu(cpu) | |
c0ff4b85 | 910 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f | 911 | #ifdef CONFIG_HOTPLUG_CPU |
c0ff4b85 R |
912 | spin_lock(&memcg->pcp_counter_lock); |
913 | val += memcg->nocpu_base.events[idx]; | |
914 | spin_unlock(&memcg->pcp_counter_lock); | |
e9f8974f JW |
915 | #endif |
916 | return val; | |
917 | } | |
918 | ||
c0ff4b85 | 919 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b2402857 | 920 | bool anon, int nr_pages) |
d52aa412 | 921 | { |
c62b1a3b KH |
922 | preempt_disable(); |
923 | ||
b2402857 KH |
924 | /* |
925 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
926 | * counted as CACHE even if it's on ANON LRU. | |
927 | */ | |
928 | if (anon) | |
929 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], | |
c0ff4b85 | 930 | nr_pages); |
d52aa412 | 931 | else |
b2402857 | 932 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 933 | nr_pages); |
55e462b0 | 934 | |
e401f176 KH |
935 | /* pagein of a big page is an event. So, ignore page size */ |
936 | if (nr_pages > 0) | |
c0ff4b85 | 937 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 938 | else { |
c0ff4b85 | 939 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
940 | nr_pages = -nr_pages; /* for event */ |
941 | } | |
e401f176 | 942 | |
13114716 | 943 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
2e72b634 | 944 | |
c62b1a3b | 945 | preempt_enable(); |
6d12e2d8 KH |
946 | } |
947 | ||
bb2a0de9 | 948 | unsigned long |
4d7dcca2 | 949 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
074291fe KK |
950 | { |
951 | struct mem_cgroup_per_zone *mz; | |
952 | ||
953 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); | |
954 | return mz->lru_size[lru]; | |
955 | } | |
956 | ||
957 | static unsigned long | |
c0ff4b85 | 958 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
bb2a0de9 | 959 | unsigned int lru_mask) |
889976db YH |
960 | { |
961 | struct mem_cgroup_per_zone *mz; | |
f156ab93 | 962 | enum lru_list lru; |
bb2a0de9 KH |
963 | unsigned long ret = 0; |
964 | ||
c0ff4b85 | 965 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
bb2a0de9 | 966 | |
f156ab93 HD |
967 | for_each_lru(lru) { |
968 | if (BIT(lru) & lru_mask) | |
969 | ret += mz->lru_size[lru]; | |
bb2a0de9 KH |
970 | } |
971 | return ret; | |
972 | } | |
973 | ||
974 | static unsigned long | |
c0ff4b85 | 975 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 KH |
976 | int nid, unsigned int lru_mask) |
977 | { | |
889976db YH |
978 | u64 total = 0; |
979 | int zid; | |
980 | ||
bb2a0de9 | 981 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
c0ff4b85 R |
982 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
983 | nid, zid, lru_mask); | |
bb2a0de9 | 984 | |
889976db YH |
985 | return total; |
986 | } | |
bb2a0de9 | 987 | |
c0ff4b85 | 988 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 989 | unsigned int lru_mask) |
6d12e2d8 | 990 | { |
889976db | 991 | int nid; |
6d12e2d8 KH |
992 | u64 total = 0; |
993 | ||
31aaea4a | 994 | for_each_node_state(nid, N_MEMORY) |
c0ff4b85 | 995 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
6d12e2d8 | 996 | return total; |
d52aa412 KH |
997 | } |
998 | ||
f53d7ce3 JW |
999 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
1000 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
1001 | { |
1002 | unsigned long val, next; | |
1003 | ||
13114716 | 1004 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 1005 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 1006 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
1007 | if ((long)next - (long)val < 0) { |
1008 | switch (target) { | |
1009 | case MEM_CGROUP_TARGET_THRESH: | |
1010 | next = val + THRESHOLDS_EVENTS_TARGET; | |
1011 | break; | |
1012 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
1013 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
1014 | break; | |
1015 | case MEM_CGROUP_TARGET_NUMAINFO: | |
1016 | next = val + NUMAINFO_EVENTS_TARGET; | |
1017 | break; | |
1018 | default: | |
1019 | break; | |
1020 | } | |
1021 | __this_cpu_write(memcg->stat->targets[target], next); | |
1022 | return true; | |
7a159cc9 | 1023 | } |
f53d7ce3 | 1024 | return false; |
d2265e6f KH |
1025 | } |
1026 | ||
1027 | /* | |
1028 | * Check events in order. | |
1029 | * | |
1030 | */ | |
c0ff4b85 | 1031 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f | 1032 | { |
4799401f | 1033 | preempt_disable(); |
d2265e6f | 1034 | /* threshold event is triggered in finer grain than soft limit */ |
f53d7ce3 JW |
1035 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
1036 | MEM_CGROUP_TARGET_THRESH))) { | |
82b3f2a7 AM |
1037 | bool do_softlimit; |
1038 | bool do_numainfo __maybe_unused; | |
f53d7ce3 JW |
1039 | |
1040 | do_softlimit = mem_cgroup_event_ratelimit(memcg, | |
1041 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
1042 | #if MAX_NUMNODES > 1 | |
1043 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
1044 | MEM_CGROUP_TARGET_NUMAINFO); | |
1045 | #endif | |
1046 | preempt_enable(); | |
1047 | ||
c0ff4b85 | 1048 | mem_cgroup_threshold(memcg); |
f53d7ce3 | 1049 | if (unlikely(do_softlimit)) |
c0ff4b85 | 1050 | mem_cgroup_update_tree(memcg, page); |
453a9bf3 | 1051 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 1052 | if (unlikely(do_numainfo)) |
c0ff4b85 | 1053 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 1054 | #endif |
f53d7ce3 JW |
1055 | } else |
1056 | preempt_enable(); | |
d2265e6f KH |
1057 | } |
1058 | ||
d1a4c0b3 | 1059 | struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 | 1060 | { |
b2145145 WL |
1061 | return mem_cgroup_from_css( |
1062 | cgroup_subsys_state(cont, mem_cgroup_subsys_id)); | |
8cdea7c0 BS |
1063 | } |
1064 | ||
cf475ad2 | 1065 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1066 | { |
31a78f23 BS |
1067 | /* |
1068 | * mm_update_next_owner() may clear mm->owner to NULL | |
1069 | * if it races with swapoff, page migration, etc. | |
1070 | * So this can be called with p == NULL. | |
1071 | */ | |
1072 | if (unlikely(!p)) | |
1073 | return NULL; | |
1074 | ||
b2145145 | 1075 | return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id)); |
78fb7466 PE |
1076 | } |
1077 | ||
a433658c | 1078 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 1079 | { |
c0ff4b85 | 1080 | struct mem_cgroup *memcg = NULL; |
0b7f569e KH |
1081 | |
1082 | if (!mm) | |
1083 | return NULL; | |
54595fe2 KH |
1084 | /* |
1085 | * Because we have no locks, mm->owner's may be being moved to other | |
1086 | * cgroup. We use css_tryget() here even if this looks | |
1087 | * pessimistic (rather than adding locks here). | |
1088 | */ | |
1089 | rcu_read_lock(); | |
1090 | do { | |
c0ff4b85 R |
1091 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1092 | if (unlikely(!memcg)) | |
54595fe2 | 1093 | break; |
c0ff4b85 | 1094 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1095 | rcu_read_unlock(); |
c0ff4b85 | 1096 | return memcg; |
54595fe2 KH |
1097 | } |
1098 | ||
16248d8f MH |
1099 | /* |
1100 | * Returns a next (in a pre-order walk) alive memcg (with elevated css | |
1101 | * ref. count) or NULL if the whole root's subtree has been visited. | |
1102 | * | |
1103 | * helper function to be used by mem_cgroup_iter | |
1104 | */ | |
1105 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, | |
1106 | struct mem_cgroup *last_visited) | |
1107 | { | |
1108 | struct cgroup *prev_cgroup, *next_cgroup; | |
1109 | ||
1110 | /* | |
1111 | * Root is not visited by cgroup iterators so it needs an | |
1112 | * explicit visit. | |
1113 | */ | |
1114 | if (!last_visited) | |
1115 | return root; | |
1116 | ||
1117 | prev_cgroup = (last_visited == root) ? NULL | |
1118 | : last_visited->css.cgroup; | |
1119 | skip_node: | |
1120 | next_cgroup = cgroup_next_descendant_pre( | |
1121 | prev_cgroup, root->css.cgroup); | |
1122 | ||
1123 | /* | |
1124 | * Even if we found a group we have to make sure it is | |
1125 | * alive. css && !memcg means that the groups should be | |
1126 | * skipped and we should continue the tree walk. | |
1127 | * last_visited css is safe to use because it is | |
1128 | * protected by css_get and the tree walk is rcu safe. | |
1129 | */ | |
1130 | if (next_cgroup) { | |
1131 | struct mem_cgroup *mem = mem_cgroup_from_cont( | |
1132 | next_cgroup); | |
1133 | if (css_tryget(&mem->css)) | |
1134 | return mem; | |
1135 | else { | |
1136 | prev_cgroup = next_cgroup; | |
1137 | goto skip_node; | |
1138 | } | |
1139 | } | |
1140 | ||
1141 | return NULL; | |
1142 | } | |
1143 | ||
5660048c JW |
1144 | /** |
1145 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1146 | * @root: hierarchy root | |
1147 | * @prev: previously returned memcg, NULL on first invocation | |
1148 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1149 | * | |
1150 | * Returns references to children of the hierarchy below @root, or | |
1151 | * @root itself, or %NULL after a full round-trip. | |
1152 | * | |
1153 | * Caller must pass the return value in @prev on subsequent | |
1154 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1155 | * to cancel a hierarchy walk before the round-trip is complete. | |
1156 | * | |
1157 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
1158 | * divide up the memcgs in the hierarchy among all concurrent | |
1159 | * reclaimers operating on the same zone and priority. | |
1160 | */ | |
1161 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, | |
1162 | struct mem_cgroup *prev, | |
1163 | struct mem_cgroup_reclaim_cookie *reclaim) | |
14067bb3 | 1164 | { |
9f3a0d09 | 1165 | struct mem_cgroup *memcg = NULL; |
542f85f9 | 1166 | struct mem_cgroup *last_visited = NULL; |
5f578161 | 1167 | unsigned long uninitialized_var(dead_count); |
711d3d2c | 1168 | |
5660048c JW |
1169 | if (mem_cgroup_disabled()) |
1170 | return NULL; | |
1171 | ||
9f3a0d09 JW |
1172 | if (!root) |
1173 | root = root_mem_cgroup; | |
7d74b06f | 1174 | |
9f3a0d09 | 1175 | if (prev && !reclaim) |
542f85f9 | 1176 | last_visited = prev; |
14067bb3 | 1177 | |
9f3a0d09 JW |
1178 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1179 | if (prev) | |
c40046f3 | 1180 | goto out_css_put; |
9f3a0d09 JW |
1181 | return root; |
1182 | } | |
14067bb3 | 1183 | |
542f85f9 | 1184 | rcu_read_lock(); |
9f3a0d09 | 1185 | while (!memcg) { |
527a5ec9 | 1186 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
711d3d2c | 1187 | |
527a5ec9 JW |
1188 | if (reclaim) { |
1189 | int nid = zone_to_nid(reclaim->zone); | |
1190 | int zid = zone_idx(reclaim->zone); | |
1191 | struct mem_cgroup_per_zone *mz; | |
1192 | ||
1193 | mz = mem_cgroup_zoneinfo(root, nid, zid); | |
1194 | iter = &mz->reclaim_iter[reclaim->priority]; | |
542f85f9 MH |
1195 | last_visited = iter->last_visited; |
1196 | if (prev && reclaim->generation != iter->generation) { | |
5f578161 | 1197 | iter->last_visited = NULL; |
542f85f9 MH |
1198 | goto out_unlock; |
1199 | } | |
5f578161 MH |
1200 | |
1201 | /* | |
1202 | * If the dead_count mismatches, a destruction | |
1203 | * has happened or is happening concurrently. | |
1204 | * If the dead_count matches, a destruction | |
1205 | * might still happen concurrently, but since | |
1206 | * we checked under RCU, that destruction | |
1207 | * won't free the object until we release the | |
1208 | * RCU reader lock. Thus, the dead_count | |
1209 | * check verifies the pointer is still valid, | |
1210 | * css_tryget() verifies the cgroup pointed to | |
1211 | * is alive. | |
1212 | */ | |
1213 | dead_count = atomic_read(&root->dead_count); | |
1214 | smp_rmb(); | |
1215 | last_visited = iter->last_visited; | |
1216 | if (last_visited) { | |
1217 | if ((dead_count != iter->last_dead_count) || | |
1218 | !css_tryget(&last_visited->css)) { | |
1219 | last_visited = NULL; | |
1220 | } | |
1221 | } | |
527a5ec9 | 1222 | } |
7d74b06f | 1223 | |
16248d8f | 1224 | memcg = __mem_cgroup_iter_next(root, last_visited); |
14067bb3 | 1225 | |
527a5ec9 | 1226 | if (reclaim) { |
542f85f9 MH |
1227 | if (last_visited) |
1228 | css_put(&last_visited->css); | |
1229 | ||
19f39402 | 1230 | iter->last_visited = memcg; |
5f578161 MH |
1231 | smp_wmb(); |
1232 | iter->last_dead_count = dead_count; | |
542f85f9 | 1233 | |
19f39402 | 1234 | if (!memcg) |
527a5ec9 JW |
1235 | iter->generation++; |
1236 | else if (!prev && memcg) | |
1237 | reclaim->generation = iter->generation; | |
1238 | } | |
9f3a0d09 | 1239 | |
19f39402 | 1240 | if (prev && !memcg) |
542f85f9 | 1241 | goto out_unlock; |
9f3a0d09 | 1242 | } |
542f85f9 MH |
1243 | out_unlock: |
1244 | rcu_read_unlock(); | |
c40046f3 MH |
1245 | out_css_put: |
1246 | if (prev && prev != root) | |
1247 | css_put(&prev->css); | |
1248 | ||
9f3a0d09 | 1249 | return memcg; |
14067bb3 | 1250 | } |
7d74b06f | 1251 | |
5660048c JW |
1252 | /** |
1253 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1254 | * @root: hierarchy root | |
1255 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1256 | */ | |
1257 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1258 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1259 | { |
1260 | if (!root) | |
1261 | root = root_mem_cgroup; | |
1262 | if (prev && prev != root) | |
1263 | css_put(&prev->css); | |
1264 | } | |
7d74b06f | 1265 | |
9f3a0d09 JW |
1266 | /* |
1267 | * Iteration constructs for visiting all cgroups (under a tree). If | |
1268 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
1269 | * be used for reference counting. | |
1270 | */ | |
1271 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 1272 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 1273 | iter != NULL; \ |
527a5ec9 | 1274 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 1275 | |
9f3a0d09 | 1276 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 1277 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 1278 | iter != NULL; \ |
527a5ec9 | 1279 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 1280 | |
68ae564b | 1281 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
456f998e | 1282 | { |
c0ff4b85 | 1283 | struct mem_cgroup *memcg; |
456f998e | 1284 | |
456f998e | 1285 | rcu_read_lock(); |
c0ff4b85 R |
1286 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1287 | if (unlikely(!memcg)) | |
456f998e YH |
1288 | goto out; |
1289 | ||
1290 | switch (idx) { | |
456f998e | 1291 | case PGFAULT: |
0e574a93 JW |
1292 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
1293 | break; | |
1294 | case PGMAJFAULT: | |
1295 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); | |
456f998e YH |
1296 | break; |
1297 | default: | |
1298 | BUG(); | |
1299 | } | |
1300 | out: | |
1301 | rcu_read_unlock(); | |
1302 | } | |
68ae564b | 1303 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
456f998e | 1304 | |
925b7673 JW |
1305 | /** |
1306 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
1307 | * @zone: zone of the wanted lruvec | |
fa9add64 | 1308 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
1309 | * |
1310 | * Returns the lru list vector holding pages for the given @zone and | |
1311 | * @mem. This can be the global zone lruvec, if the memory controller | |
1312 | * is disabled. | |
1313 | */ | |
1314 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1315 | struct mem_cgroup *memcg) | |
1316 | { | |
1317 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1318 | struct lruvec *lruvec; |
925b7673 | 1319 | |
bea8c150 HD |
1320 | if (mem_cgroup_disabled()) { |
1321 | lruvec = &zone->lruvec; | |
1322 | goto out; | |
1323 | } | |
925b7673 JW |
1324 | |
1325 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); | |
bea8c150 HD |
1326 | lruvec = &mz->lruvec; |
1327 | out: | |
1328 | /* | |
1329 | * Since a node can be onlined after the mem_cgroup was created, | |
1330 | * we have to be prepared to initialize lruvec->zone here; | |
1331 | * and if offlined then reonlined, we need to reinitialize it. | |
1332 | */ | |
1333 | if (unlikely(lruvec->zone != zone)) | |
1334 | lruvec->zone = zone; | |
1335 | return lruvec; | |
925b7673 JW |
1336 | } |
1337 | ||
08e552c6 KH |
1338 | /* |
1339 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
1340 | * Operations are called by routine of global LRU independently from memcg. | |
1341 | * What we have to take care of here is validness of pc->mem_cgroup. | |
1342 | * | |
1343 | * Changes to pc->mem_cgroup happens when | |
1344 | * 1. charge | |
1345 | * 2. moving account | |
1346 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
1347 | * It is added to LRU before charge. | |
1348 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
1349 | * When moving account, the page is not on LRU. It's isolated. | |
1350 | */ | |
4f98a2fe | 1351 | |
925b7673 | 1352 | /** |
fa9add64 | 1353 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
925b7673 | 1354 | * @page: the page |
fa9add64 | 1355 | * @zone: zone of the page |
925b7673 | 1356 | */ |
fa9add64 | 1357 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1358 | { |
08e552c6 | 1359 | struct mem_cgroup_per_zone *mz; |
925b7673 JW |
1360 | struct mem_cgroup *memcg; |
1361 | struct page_cgroup *pc; | |
bea8c150 | 1362 | struct lruvec *lruvec; |
6d12e2d8 | 1363 | |
bea8c150 HD |
1364 | if (mem_cgroup_disabled()) { |
1365 | lruvec = &zone->lruvec; | |
1366 | goto out; | |
1367 | } | |
925b7673 | 1368 | |
08e552c6 | 1369 | pc = lookup_page_cgroup(page); |
38c5d72f | 1370 | memcg = pc->mem_cgroup; |
7512102c HD |
1371 | |
1372 | /* | |
fa9add64 | 1373 | * Surreptitiously switch any uncharged offlist page to root: |
7512102c HD |
1374 | * an uncharged page off lru does nothing to secure |
1375 | * its former mem_cgroup from sudden removal. | |
1376 | * | |
1377 | * Our caller holds lru_lock, and PageCgroupUsed is updated | |
1378 | * under page_cgroup lock: between them, they make all uses | |
1379 | * of pc->mem_cgroup safe. | |
1380 | */ | |
fa9add64 | 1381 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
7512102c HD |
1382 | pc->mem_cgroup = memcg = root_mem_cgroup; |
1383 | ||
925b7673 | 1384 | mz = page_cgroup_zoneinfo(memcg, page); |
bea8c150 HD |
1385 | lruvec = &mz->lruvec; |
1386 | out: | |
1387 | /* | |
1388 | * Since a node can be onlined after the mem_cgroup was created, | |
1389 | * we have to be prepared to initialize lruvec->zone here; | |
1390 | * and if offlined then reonlined, we need to reinitialize it. | |
1391 | */ | |
1392 | if (unlikely(lruvec->zone != zone)) | |
1393 | lruvec->zone = zone; | |
1394 | return lruvec; | |
08e552c6 | 1395 | } |
b69408e8 | 1396 | |
925b7673 | 1397 | /** |
fa9add64 HD |
1398 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1399 | * @lruvec: mem_cgroup per zone lru vector | |
1400 | * @lru: index of lru list the page is sitting on | |
1401 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1402 | * |
fa9add64 HD |
1403 | * This function must be called when a page is added to or removed from an |
1404 | * lru list. | |
3f58a829 | 1405 | */ |
fa9add64 HD |
1406 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1407 | int nr_pages) | |
3f58a829 MK |
1408 | { |
1409 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1410 | unsigned long *lru_size; |
3f58a829 MK |
1411 | |
1412 | if (mem_cgroup_disabled()) | |
1413 | return; | |
1414 | ||
fa9add64 HD |
1415 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1416 | lru_size = mz->lru_size + lru; | |
1417 | *lru_size += nr_pages; | |
1418 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1419 | } |
544122e5 | 1420 | |
3e92041d | 1421 | /* |
c0ff4b85 | 1422 | * Checks whether given mem is same or in the root_mem_cgroup's |
3e92041d MH |
1423 | * hierarchy subtree |
1424 | */ | |
c3ac9a8a JW |
1425 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
1426 | struct mem_cgroup *memcg) | |
3e92041d | 1427 | { |
91c63734 JW |
1428 | if (root_memcg == memcg) |
1429 | return true; | |
3a981f48 | 1430 | if (!root_memcg->use_hierarchy || !memcg) |
91c63734 | 1431 | return false; |
c3ac9a8a JW |
1432 | return css_is_ancestor(&memcg->css, &root_memcg->css); |
1433 | } | |
1434 | ||
1435 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, | |
1436 | struct mem_cgroup *memcg) | |
1437 | { | |
1438 | bool ret; | |
1439 | ||
91c63734 | 1440 | rcu_read_lock(); |
c3ac9a8a | 1441 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
91c63734 JW |
1442 | rcu_read_unlock(); |
1443 | return ret; | |
3e92041d MH |
1444 | } |
1445 | ||
c0ff4b85 | 1446 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *memcg) |
4c4a2214 DR |
1447 | { |
1448 | int ret; | |
0b7f569e | 1449 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1450 | struct task_struct *p; |
4c4a2214 | 1451 | |
158e0a2d | 1452 | p = find_lock_task_mm(task); |
de077d22 DR |
1453 | if (p) { |
1454 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1455 | task_unlock(p); | |
1456 | } else { | |
1457 | /* | |
1458 | * All threads may have already detached their mm's, but the oom | |
1459 | * killer still needs to detect if they have already been oom | |
1460 | * killed to prevent needlessly killing additional tasks. | |
1461 | */ | |
1462 | task_lock(task); | |
1463 | curr = mem_cgroup_from_task(task); | |
1464 | if (curr) | |
1465 | css_get(&curr->css); | |
1466 | task_unlock(task); | |
1467 | } | |
0b7f569e KH |
1468 | if (!curr) |
1469 | return 0; | |
d31f56db | 1470 | /* |
c0ff4b85 | 1471 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
d31f56db | 1472 | * use_hierarchy of "curr" here make this function true if hierarchy is |
c0ff4b85 R |
1473 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
1474 | * hierarchy(even if use_hierarchy is disabled in "memcg"). | |
d31f56db | 1475 | */ |
c0ff4b85 | 1476 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
0b7f569e | 1477 | css_put(&curr->css); |
4c4a2214 DR |
1478 | return ret; |
1479 | } | |
1480 | ||
c56d5c7d | 1481 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1482 | { |
9b272977 | 1483 | unsigned long inactive_ratio; |
14797e23 | 1484 | unsigned long inactive; |
9b272977 | 1485 | unsigned long active; |
c772be93 | 1486 | unsigned long gb; |
14797e23 | 1487 | |
4d7dcca2 HD |
1488 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
1489 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); | |
14797e23 | 1490 | |
c772be93 KM |
1491 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1492 | if (gb) | |
1493 | inactive_ratio = int_sqrt(10 * gb); | |
1494 | else | |
1495 | inactive_ratio = 1; | |
1496 | ||
9b272977 | 1497 | return inactive * inactive_ratio < active; |
14797e23 KM |
1498 | } |
1499 | ||
6d61ef40 BS |
1500 | #define mem_cgroup_from_res_counter(counter, member) \ |
1501 | container_of(counter, struct mem_cgroup, member) | |
1502 | ||
19942822 | 1503 | /** |
9d11ea9f | 1504 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1505 | * @memcg: the memory cgroup |
19942822 | 1506 | * |
9d11ea9f | 1507 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1508 | * pages. |
19942822 | 1509 | */ |
c0ff4b85 | 1510 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1511 | { |
9d11ea9f JW |
1512 | unsigned long long margin; |
1513 | ||
c0ff4b85 | 1514 | margin = res_counter_margin(&memcg->res); |
9d11ea9f | 1515 | if (do_swap_account) |
c0ff4b85 | 1516 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
7ec99d62 | 1517 | return margin >> PAGE_SHIFT; |
19942822 JW |
1518 | } |
1519 | ||
1f4c025b | 1520 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1521 | { |
1522 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1523 | |
1524 | /* root ? */ | |
1525 | if (cgrp->parent == NULL) | |
1526 | return vm_swappiness; | |
1527 | ||
bf1ff263 | 1528 | return memcg->swappiness; |
a7885eb8 KM |
1529 | } |
1530 | ||
619d094b KH |
1531 | /* |
1532 | * memcg->moving_account is used for checking possibility that some thread is | |
1533 | * calling move_account(). When a thread on CPU-A starts moving pages under | |
1534 | * a memcg, other threads should check memcg->moving_account under | |
1535 | * rcu_read_lock(), like this: | |
1536 | * | |
1537 | * CPU-A CPU-B | |
1538 | * rcu_read_lock() | |
1539 | * memcg->moving_account+1 if (memcg->mocing_account) | |
1540 | * take heavy locks. | |
1541 | * synchronize_rcu() update something. | |
1542 | * rcu_read_unlock() | |
1543 | * start move here. | |
1544 | */ | |
4331f7d3 KH |
1545 | |
1546 | /* for quick checking without looking up memcg */ | |
1547 | atomic_t memcg_moving __read_mostly; | |
1548 | ||
c0ff4b85 | 1549 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
32047e2a | 1550 | { |
4331f7d3 | 1551 | atomic_inc(&memcg_moving); |
619d094b | 1552 | atomic_inc(&memcg->moving_account); |
32047e2a KH |
1553 | synchronize_rcu(); |
1554 | } | |
1555 | ||
c0ff4b85 | 1556 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
32047e2a | 1557 | { |
619d094b KH |
1558 | /* |
1559 | * Now, mem_cgroup_clear_mc() may call this function with NULL. | |
1560 | * We check NULL in callee rather than caller. | |
1561 | */ | |
4331f7d3 KH |
1562 | if (memcg) { |
1563 | atomic_dec(&memcg_moving); | |
619d094b | 1564 | atomic_dec(&memcg->moving_account); |
4331f7d3 | 1565 | } |
32047e2a | 1566 | } |
619d094b | 1567 | |
32047e2a KH |
1568 | /* |
1569 | * 2 routines for checking "mem" is under move_account() or not. | |
1570 | * | |
13fd1dd9 AM |
1571 | * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This |
1572 | * is used for avoiding races in accounting. If true, | |
32047e2a KH |
1573 | * pc->mem_cgroup may be overwritten. |
1574 | * | |
1575 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1576 | * under hierarchy of moving cgroups. This is for | |
1577 | * waiting at hith-memory prressure caused by "move". | |
1578 | */ | |
1579 | ||
13fd1dd9 | 1580 | static bool mem_cgroup_stolen(struct mem_cgroup *memcg) |
32047e2a KH |
1581 | { |
1582 | VM_BUG_ON(!rcu_read_lock_held()); | |
619d094b | 1583 | return atomic_read(&memcg->moving_account) > 0; |
32047e2a | 1584 | } |
4b534334 | 1585 | |
c0ff4b85 | 1586 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1587 | { |
2bd9bb20 KH |
1588 | struct mem_cgroup *from; |
1589 | struct mem_cgroup *to; | |
4b534334 | 1590 | bool ret = false; |
2bd9bb20 KH |
1591 | /* |
1592 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1593 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1594 | */ | |
1595 | spin_lock(&mc.lock); | |
1596 | from = mc.from; | |
1597 | to = mc.to; | |
1598 | if (!from) | |
1599 | goto unlock; | |
3e92041d | 1600 | |
c0ff4b85 R |
1601 | ret = mem_cgroup_same_or_subtree(memcg, from) |
1602 | || mem_cgroup_same_or_subtree(memcg, to); | |
2bd9bb20 KH |
1603 | unlock: |
1604 | spin_unlock(&mc.lock); | |
4b534334 KH |
1605 | return ret; |
1606 | } | |
1607 | ||
c0ff4b85 | 1608 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1609 | { |
1610 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1611 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1612 | DEFINE_WAIT(wait); |
1613 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1614 | /* moving charge context might have finished. */ | |
1615 | if (mc.moving_task) | |
1616 | schedule(); | |
1617 | finish_wait(&mc.waitq, &wait); | |
1618 | return true; | |
1619 | } | |
1620 | } | |
1621 | return false; | |
1622 | } | |
1623 | ||
312734c0 KH |
1624 | /* |
1625 | * Take this lock when | |
1626 | * - a code tries to modify page's memcg while it's USED. | |
1627 | * - a code tries to modify page state accounting in a memcg. | |
13fd1dd9 | 1628 | * see mem_cgroup_stolen(), too. |
312734c0 KH |
1629 | */ |
1630 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, | |
1631 | unsigned long *flags) | |
1632 | { | |
1633 | spin_lock_irqsave(&memcg->move_lock, *flags); | |
1634 | } | |
1635 | ||
1636 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, | |
1637 | unsigned long *flags) | |
1638 | { | |
1639 | spin_unlock_irqrestore(&memcg->move_lock, *flags); | |
1640 | } | |
1641 | ||
58cf188e | 1642 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1643 | /** |
58cf188e | 1644 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1645 | * @memcg: The memory cgroup that went over limit |
1646 | * @p: Task that is going to be killed | |
1647 | * | |
1648 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1649 | * enabled | |
1650 | */ | |
1651 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1652 | { | |
1653 | struct cgroup *task_cgrp; | |
1654 | struct cgroup *mem_cgrp; | |
1655 | /* | |
1656 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1657 | * on the assumption that OOM is serialized for memory controller. | |
1658 | * If this assumption is broken, revisit this code. | |
1659 | */ | |
1660 | static char memcg_name[PATH_MAX]; | |
1661 | int ret; | |
58cf188e SZ |
1662 | struct mem_cgroup *iter; |
1663 | unsigned int i; | |
e222432b | 1664 | |
58cf188e | 1665 | if (!p) |
e222432b BS |
1666 | return; |
1667 | ||
e222432b BS |
1668 | rcu_read_lock(); |
1669 | ||
1670 | mem_cgrp = memcg->css.cgroup; | |
1671 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1672 | ||
1673 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1674 | if (ret < 0) { | |
1675 | /* | |
1676 | * Unfortunately, we are unable to convert to a useful name | |
1677 | * But we'll still print out the usage information | |
1678 | */ | |
1679 | rcu_read_unlock(); | |
1680 | goto done; | |
1681 | } | |
1682 | rcu_read_unlock(); | |
1683 | ||
d045197f | 1684 | pr_info("Task in %s killed", memcg_name); |
e222432b BS |
1685 | |
1686 | rcu_read_lock(); | |
1687 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1688 | if (ret < 0) { | |
1689 | rcu_read_unlock(); | |
1690 | goto done; | |
1691 | } | |
1692 | rcu_read_unlock(); | |
1693 | ||
1694 | /* | |
1695 | * Continues from above, so we don't need an KERN_ level | |
1696 | */ | |
d045197f | 1697 | pr_cont(" as a result of limit of %s\n", memcg_name); |
e222432b BS |
1698 | done: |
1699 | ||
d045197f | 1700 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1701 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1702 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1703 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
d045197f | 1704 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
e222432b BS |
1705 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1706 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1707 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
d045197f | 1708 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
510fc4e1 GC |
1709 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
1710 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, | |
1711 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); | |
58cf188e SZ |
1712 | |
1713 | for_each_mem_cgroup_tree(iter, memcg) { | |
1714 | pr_info("Memory cgroup stats"); | |
1715 | ||
1716 | rcu_read_lock(); | |
1717 | ret = cgroup_path(iter->css.cgroup, memcg_name, PATH_MAX); | |
1718 | if (!ret) | |
1719 | pr_cont(" for %s", memcg_name); | |
1720 | rcu_read_unlock(); | |
1721 | pr_cont(":"); | |
1722 | ||
1723 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
1724 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) | |
1725 | continue; | |
1726 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], | |
1727 | K(mem_cgroup_read_stat(iter, i))); | |
1728 | } | |
1729 | ||
1730 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1731 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1732 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1733 | ||
1734 | pr_cont("\n"); | |
1735 | } | |
e222432b BS |
1736 | } |
1737 | ||
81d39c20 KH |
1738 | /* |
1739 | * This function returns the number of memcg under hierarchy tree. Returns | |
1740 | * 1(self count) if no children. | |
1741 | */ | |
c0ff4b85 | 1742 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1743 | { |
1744 | int num = 0; | |
7d74b06f KH |
1745 | struct mem_cgroup *iter; |
1746 | ||
c0ff4b85 | 1747 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1748 | num++; |
81d39c20 KH |
1749 | return num; |
1750 | } | |
1751 | ||
a63d83f4 DR |
1752 | /* |
1753 | * Return the memory (and swap, if configured) limit for a memcg. | |
1754 | */ | |
9cbb78bb | 1755 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 DR |
1756 | { |
1757 | u64 limit; | |
a63d83f4 | 1758 | |
f3e8eb70 | 1759 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
f3e8eb70 | 1760 | |
a63d83f4 | 1761 | /* |
9a5a8f19 | 1762 | * Do not consider swap space if we cannot swap due to swappiness |
a63d83f4 | 1763 | */ |
9a5a8f19 MH |
1764 | if (mem_cgroup_swappiness(memcg)) { |
1765 | u64 memsw; | |
1766 | ||
1767 | limit += total_swap_pages << PAGE_SHIFT; | |
1768 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1769 | ||
1770 | /* | |
1771 | * If memsw is finite and limits the amount of swap space | |
1772 | * available to this memcg, return that limit. | |
1773 | */ | |
1774 | limit = min(limit, memsw); | |
1775 | } | |
1776 | ||
1777 | return limit; | |
a63d83f4 DR |
1778 | } |
1779 | ||
19965460 DR |
1780 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1781 | int order) | |
9cbb78bb DR |
1782 | { |
1783 | struct mem_cgroup *iter; | |
1784 | unsigned long chosen_points = 0; | |
1785 | unsigned long totalpages; | |
1786 | unsigned int points = 0; | |
1787 | struct task_struct *chosen = NULL; | |
1788 | ||
876aafbf | 1789 | /* |
465adcf1 DR |
1790 | * If current has a pending SIGKILL or is exiting, then automatically |
1791 | * select it. The goal is to allow it to allocate so that it may | |
1792 | * quickly exit and free its memory. | |
876aafbf | 1793 | */ |
465adcf1 | 1794 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
876aafbf DR |
1795 | set_thread_flag(TIF_MEMDIE); |
1796 | return; | |
1797 | } | |
1798 | ||
1799 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); | |
9cbb78bb DR |
1800 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
1801 | for_each_mem_cgroup_tree(iter, memcg) { | |
1802 | struct cgroup *cgroup = iter->css.cgroup; | |
1803 | struct cgroup_iter it; | |
1804 | struct task_struct *task; | |
1805 | ||
1806 | cgroup_iter_start(cgroup, &it); | |
1807 | while ((task = cgroup_iter_next(cgroup, &it))) { | |
1808 | switch (oom_scan_process_thread(task, totalpages, NULL, | |
1809 | false)) { | |
1810 | case OOM_SCAN_SELECT: | |
1811 | if (chosen) | |
1812 | put_task_struct(chosen); | |
1813 | chosen = task; | |
1814 | chosen_points = ULONG_MAX; | |
1815 | get_task_struct(chosen); | |
1816 | /* fall through */ | |
1817 | case OOM_SCAN_CONTINUE: | |
1818 | continue; | |
1819 | case OOM_SCAN_ABORT: | |
1820 | cgroup_iter_end(cgroup, &it); | |
1821 | mem_cgroup_iter_break(memcg, iter); | |
1822 | if (chosen) | |
1823 | put_task_struct(chosen); | |
1824 | return; | |
1825 | case OOM_SCAN_OK: | |
1826 | break; | |
1827 | }; | |
1828 | points = oom_badness(task, memcg, NULL, totalpages); | |
1829 | if (points > chosen_points) { | |
1830 | if (chosen) | |
1831 | put_task_struct(chosen); | |
1832 | chosen = task; | |
1833 | chosen_points = points; | |
1834 | get_task_struct(chosen); | |
1835 | } | |
1836 | } | |
1837 | cgroup_iter_end(cgroup, &it); | |
1838 | } | |
1839 | ||
1840 | if (!chosen) | |
1841 | return; | |
1842 | points = chosen_points * 1000 / totalpages; | |
9cbb78bb DR |
1843 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
1844 | NULL, "Memory cgroup out of memory"); | |
9cbb78bb DR |
1845 | } |
1846 | ||
5660048c JW |
1847 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
1848 | gfp_t gfp_mask, | |
1849 | unsigned long flags) | |
1850 | { | |
1851 | unsigned long total = 0; | |
1852 | bool noswap = false; | |
1853 | int loop; | |
1854 | ||
1855 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) | |
1856 | noswap = true; | |
1857 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) | |
1858 | noswap = true; | |
1859 | ||
1860 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { | |
1861 | if (loop) | |
1862 | drain_all_stock_async(memcg); | |
1863 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); | |
1864 | /* | |
1865 | * Allow limit shrinkers, which are triggered directly | |
1866 | * by userspace, to catch signals and stop reclaim | |
1867 | * after minimal progress, regardless of the margin. | |
1868 | */ | |
1869 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) | |
1870 | break; | |
1871 | if (mem_cgroup_margin(memcg)) | |
1872 | break; | |
1873 | /* | |
1874 | * If nothing was reclaimed after two attempts, there | |
1875 | * may be no reclaimable pages in this hierarchy. | |
1876 | */ | |
1877 | if (loop && !total) | |
1878 | break; | |
1879 | } | |
1880 | return total; | |
1881 | } | |
1882 | ||
4d0c066d KH |
1883 | /** |
1884 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1885 | * @memcg: the target memcg |
4d0c066d KH |
1886 | * @nid: the node ID to be checked. |
1887 | * @noswap : specify true here if the user wants flle only information. | |
1888 | * | |
1889 | * This function returns whether the specified memcg contains any | |
1890 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1891 | * pages in the node. | |
1892 | */ | |
c0ff4b85 | 1893 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1894 | int nid, bool noswap) |
1895 | { | |
c0ff4b85 | 1896 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1897 | return true; |
1898 | if (noswap || !total_swap_pages) | |
1899 | return false; | |
c0ff4b85 | 1900 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1901 | return true; |
1902 | return false; | |
1903 | ||
1904 | } | |
889976db YH |
1905 | #if MAX_NUMNODES > 1 |
1906 | ||
1907 | /* | |
1908 | * Always updating the nodemask is not very good - even if we have an empty | |
1909 | * list or the wrong list here, we can start from some node and traverse all | |
1910 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1911 | * | |
1912 | */ | |
c0ff4b85 | 1913 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1914 | { |
1915 | int nid; | |
453a9bf3 KH |
1916 | /* |
1917 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1918 | * pagein/pageout changes since the last update. | |
1919 | */ | |
c0ff4b85 | 1920 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1921 | return; |
c0ff4b85 | 1922 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1923 | return; |
1924 | ||
889976db | 1925 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1926 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1927 | |
31aaea4a | 1928 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1929 | |
c0ff4b85 R |
1930 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1931 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1932 | } |
453a9bf3 | 1933 | |
c0ff4b85 R |
1934 | atomic_set(&memcg->numainfo_events, 0); |
1935 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1936 | } |
1937 | ||
1938 | /* | |
1939 | * Selecting a node where we start reclaim from. Because what we need is just | |
1940 | * reducing usage counter, start from anywhere is O,K. Considering | |
1941 | * memory reclaim from current node, there are pros. and cons. | |
1942 | * | |
1943 | * Freeing memory from current node means freeing memory from a node which | |
1944 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1945 | * hit limits, it will see a contention on a node. But freeing from remote | |
1946 | * node means more costs for memory reclaim because of memory latency. | |
1947 | * | |
1948 | * Now, we use round-robin. Better algorithm is welcomed. | |
1949 | */ | |
c0ff4b85 | 1950 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1951 | { |
1952 | int node; | |
1953 | ||
c0ff4b85 R |
1954 | mem_cgroup_may_update_nodemask(memcg); |
1955 | node = memcg->last_scanned_node; | |
889976db | 1956 | |
c0ff4b85 | 1957 | node = next_node(node, memcg->scan_nodes); |
889976db | 1958 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1959 | node = first_node(memcg->scan_nodes); |
889976db YH |
1960 | /* |
1961 | * We call this when we hit limit, not when pages are added to LRU. | |
1962 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1963 | * memcg is too small and all pages are not on LRU. In that case, | |
1964 | * we use curret node. | |
1965 | */ | |
1966 | if (unlikely(node == MAX_NUMNODES)) | |
1967 | node = numa_node_id(); | |
1968 | ||
c0ff4b85 | 1969 | memcg->last_scanned_node = node; |
889976db YH |
1970 | return node; |
1971 | } | |
1972 | ||
4d0c066d KH |
1973 | /* |
1974 | * Check all nodes whether it contains reclaimable pages or not. | |
1975 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1976 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1977 | * enough new information. We need to do double check. | |
1978 | */ | |
6bbda35c | 1979 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d KH |
1980 | { |
1981 | int nid; | |
1982 | ||
1983 | /* | |
1984 | * quick check...making use of scan_node. | |
1985 | * We can skip unused nodes. | |
1986 | */ | |
c0ff4b85 R |
1987 | if (!nodes_empty(memcg->scan_nodes)) { |
1988 | for (nid = first_node(memcg->scan_nodes); | |
4d0c066d | 1989 | nid < MAX_NUMNODES; |
c0ff4b85 | 1990 | nid = next_node(nid, memcg->scan_nodes)) { |
4d0c066d | 1991 | |
c0ff4b85 | 1992 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
1993 | return true; |
1994 | } | |
1995 | } | |
1996 | /* | |
1997 | * Check rest of nodes. | |
1998 | */ | |
31aaea4a | 1999 | for_each_node_state(nid, N_MEMORY) { |
c0ff4b85 | 2000 | if (node_isset(nid, memcg->scan_nodes)) |
4d0c066d | 2001 | continue; |
c0ff4b85 | 2002 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
4d0c066d KH |
2003 | return true; |
2004 | } | |
2005 | return false; | |
2006 | } | |
2007 | ||
889976db | 2008 | #else |
c0ff4b85 | 2009 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
2010 | { |
2011 | return 0; | |
2012 | } | |
4d0c066d | 2013 | |
6bbda35c | 2014 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
4d0c066d | 2015 | { |
c0ff4b85 | 2016 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
4d0c066d | 2017 | } |
889976db YH |
2018 | #endif |
2019 | ||
5660048c JW |
2020 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
2021 | struct zone *zone, | |
2022 | gfp_t gfp_mask, | |
2023 | unsigned long *total_scanned) | |
6d61ef40 | 2024 | { |
9f3a0d09 | 2025 | struct mem_cgroup *victim = NULL; |
5660048c | 2026 | int total = 0; |
04046e1a | 2027 | int loop = 0; |
9d11ea9f | 2028 | unsigned long excess; |
185efc0f | 2029 | unsigned long nr_scanned; |
527a5ec9 JW |
2030 | struct mem_cgroup_reclaim_cookie reclaim = { |
2031 | .zone = zone, | |
2032 | .priority = 0, | |
2033 | }; | |
9d11ea9f | 2034 | |
c0ff4b85 | 2035 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
04046e1a | 2036 | |
4e416953 | 2037 | while (1) { |
527a5ec9 | 2038 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
9f3a0d09 | 2039 | if (!victim) { |
04046e1a | 2040 | loop++; |
4e416953 BS |
2041 | if (loop >= 2) { |
2042 | /* | |
2043 | * If we have not been able to reclaim | |
2044 | * anything, it might because there are | |
2045 | * no reclaimable pages under this hierarchy | |
2046 | */ | |
5660048c | 2047 | if (!total) |
4e416953 | 2048 | break; |
4e416953 | 2049 | /* |
25985edc | 2050 | * We want to do more targeted reclaim. |
4e416953 BS |
2051 | * excess >> 2 is not to excessive so as to |
2052 | * reclaim too much, nor too less that we keep | |
2053 | * coming back to reclaim from this cgroup | |
2054 | */ | |
2055 | if (total >= (excess >> 2) || | |
9f3a0d09 | 2056 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
4e416953 | 2057 | break; |
4e416953 | 2058 | } |
9f3a0d09 | 2059 | continue; |
4e416953 | 2060 | } |
5660048c | 2061 | if (!mem_cgroup_reclaimable(victim, false)) |
6d61ef40 | 2062 | continue; |
5660048c JW |
2063 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
2064 | zone, &nr_scanned); | |
2065 | *total_scanned += nr_scanned; | |
2066 | if (!res_counter_soft_limit_excess(&root_memcg->res)) | |
9f3a0d09 | 2067 | break; |
6d61ef40 | 2068 | } |
9f3a0d09 | 2069 | mem_cgroup_iter_break(root_memcg, victim); |
04046e1a | 2070 | return total; |
6d61ef40 BS |
2071 | } |
2072 | ||
867578cb KH |
2073 | /* |
2074 | * Check OOM-Killer is already running under our hierarchy. | |
2075 | * If someone is running, return false. | |
1af8efe9 | 2076 | * Has to be called with memcg_oom_lock |
867578cb | 2077 | */ |
c0ff4b85 | 2078 | static bool mem_cgroup_oom_lock(struct mem_cgroup *memcg) |
867578cb | 2079 | { |
79dfdacc | 2080 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 2081 | |
9f3a0d09 | 2082 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 2083 | if (iter->oom_lock) { |
79dfdacc MH |
2084 | /* |
2085 | * this subtree of our hierarchy is already locked | |
2086 | * so we cannot give a lock. | |
2087 | */ | |
79dfdacc | 2088 | failed = iter; |
9f3a0d09 JW |
2089 | mem_cgroup_iter_break(memcg, iter); |
2090 | break; | |
23751be0 JW |
2091 | } else |
2092 | iter->oom_lock = true; | |
7d74b06f | 2093 | } |
867578cb | 2094 | |
79dfdacc | 2095 | if (!failed) |
23751be0 | 2096 | return true; |
79dfdacc MH |
2097 | |
2098 | /* | |
2099 | * OK, we failed to lock the whole subtree so we have to clean up | |
2100 | * what we set up to the failing subtree | |
2101 | */ | |
9f3a0d09 | 2102 | for_each_mem_cgroup_tree(iter, memcg) { |
79dfdacc | 2103 | if (iter == failed) { |
9f3a0d09 JW |
2104 | mem_cgroup_iter_break(memcg, iter); |
2105 | break; | |
79dfdacc MH |
2106 | } |
2107 | iter->oom_lock = false; | |
2108 | } | |
23751be0 | 2109 | return false; |
a636b327 | 2110 | } |
0b7f569e | 2111 | |
79dfdacc | 2112 | /* |
1af8efe9 | 2113 | * Has to be called with memcg_oom_lock |
79dfdacc | 2114 | */ |
c0ff4b85 | 2115 | static int mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 2116 | { |
7d74b06f KH |
2117 | struct mem_cgroup *iter; |
2118 | ||
c0ff4b85 | 2119 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2120 | iter->oom_lock = false; |
2121 | return 0; | |
2122 | } | |
2123 | ||
c0ff4b85 | 2124 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2125 | { |
2126 | struct mem_cgroup *iter; | |
2127 | ||
c0ff4b85 | 2128 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc MH |
2129 | atomic_inc(&iter->under_oom); |
2130 | } | |
2131 | ||
c0ff4b85 | 2132 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
2133 | { |
2134 | struct mem_cgroup *iter; | |
2135 | ||
867578cb KH |
2136 | /* |
2137 | * When a new child is created while the hierarchy is under oom, | |
2138 | * mem_cgroup_oom_lock() may not be called. We have to use | |
2139 | * atomic_add_unless() here. | |
2140 | */ | |
c0ff4b85 | 2141 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 2142 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
2143 | } |
2144 | ||
1af8efe9 | 2145 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
2146 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
2147 | ||
dc98df5a | 2148 | struct oom_wait_info { |
d79154bb | 2149 | struct mem_cgroup *memcg; |
dc98df5a KH |
2150 | wait_queue_t wait; |
2151 | }; | |
2152 | ||
2153 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
2154 | unsigned mode, int sync, void *arg) | |
2155 | { | |
d79154bb HD |
2156 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
2157 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
2158 | struct oom_wait_info *oom_wait_info; |
2159 | ||
2160 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 2161 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 2162 | |
dc98df5a | 2163 | /* |
d79154bb | 2164 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
dc98df5a KH |
2165 | * Then we can use css_is_ancestor without taking care of RCU. |
2166 | */ | |
c0ff4b85 R |
2167 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
2168 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) | |
dc98df5a | 2169 | return 0; |
dc98df5a KH |
2170 | return autoremove_wake_function(wait, mode, sync, arg); |
2171 | } | |
2172 | ||
c0ff4b85 | 2173 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
dc98df5a | 2174 | { |
c0ff4b85 R |
2175 | /* for filtering, pass "memcg" as argument. */ |
2176 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); | |
dc98df5a KH |
2177 | } |
2178 | ||
c0ff4b85 | 2179 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 2180 | { |
c0ff4b85 R |
2181 | if (memcg && atomic_read(&memcg->under_oom)) |
2182 | memcg_wakeup_oom(memcg); | |
3c11ecf4 KH |
2183 | } |
2184 | ||
867578cb KH |
2185 | /* |
2186 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
2187 | */ | |
6bbda35c KS |
2188 | static bool mem_cgroup_handle_oom(struct mem_cgroup *memcg, gfp_t mask, |
2189 | int order) | |
0b7f569e | 2190 | { |
dc98df5a | 2191 | struct oom_wait_info owait; |
3c11ecf4 | 2192 | bool locked, need_to_kill; |
867578cb | 2193 | |
d79154bb | 2194 | owait.memcg = memcg; |
dc98df5a KH |
2195 | owait.wait.flags = 0; |
2196 | owait.wait.func = memcg_oom_wake_function; | |
2197 | owait.wait.private = current; | |
2198 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 2199 | need_to_kill = true; |
c0ff4b85 | 2200 | mem_cgroup_mark_under_oom(memcg); |
79dfdacc | 2201 | |
c0ff4b85 | 2202 | /* At first, try to OOM lock hierarchy under memcg.*/ |
1af8efe9 | 2203 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 2204 | locked = mem_cgroup_oom_lock(memcg); |
867578cb KH |
2205 | /* |
2206 | * Even if signal_pending(), we can't quit charge() loop without | |
2207 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
2208 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
2209 | */ | |
3c11ecf4 | 2210 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
c0ff4b85 | 2211 | if (!locked || memcg->oom_kill_disable) |
3c11ecf4 KH |
2212 | need_to_kill = false; |
2213 | if (locked) | |
c0ff4b85 | 2214 | mem_cgroup_oom_notify(memcg); |
1af8efe9 | 2215 | spin_unlock(&memcg_oom_lock); |
867578cb | 2216 | |
3c11ecf4 KH |
2217 | if (need_to_kill) { |
2218 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
e845e199 | 2219 | mem_cgroup_out_of_memory(memcg, mask, order); |
3c11ecf4 | 2220 | } else { |
867578cb | 2221 | schedule(); |
dc98df5a | 2222 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 2223 | } |
1af8efe9 | 2224 | spin_lock(&memcg_oom_lock); |
79dfdacc | 2225 | if (locked) |
c0ff4b85 R |
2226 | mem_cgroup_oom_unlock(memcg); |
2227 | memcg_wakeup_oom(memcg); | |
1af8efe9 | 2228 | spin_unlock(&memcg_oom_lock); |
867578cb | 2229 | |
c0ff4b85 | 2230 | mem_cgroup_unmark_under_oom(memcg); |
79dfdacc | 2231 | |
867578cb KH |
2232 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
2233 | return false; | |
2234 | /* Give chance to dying process */ | |
715a5ee8 | 2235 | schedule_timeout_uninterruptible(1); |
867578cb | 2236 | return true; |
0b7f569e KH |
2237 | } |
2238 | ||
d69b042f BS |
2239 | /* |
2240 | * Currently used to update mapped file statistics, but the routine can be | |
2241 | * generalized to update other statistics as well. | |
32047e2a KH |
2242 | * |
2243 | * Notes: Race condition | |
2244 | * | |
2245 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
2246 | * it tends to be costly. But considering some conditions, we doesn't need | |
2247 | * to do so _always_. | |
2248 | * | |
2249 | * Considering "charge", lock_page_cgroup() is not required because all | |
2250 | * file-stat operations happen after a page is attached to radix-tree. There | |
2251 | * are no race with "charge". | |
2252 | * | |
2253 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
2254 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
2255 | * if there are race with "uncharge". Statistics itself is properly handled | |
2256 | * by flags. | |
2257 | * | |
2258 | * Considering "move", this is an only case we see a race. To make the race | |
619d094b KH |
2259 | * small, we check mm->moving_account and detect there are possibility of race |
2260 | * If there is, we take a lock. | |
d69b042f | 2261 | */ |
26174efd | 2262 | |
89c06bd5 KH |
2263 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
2264 | bool *locked, unsigned long *flags) | |
2265 | { | |
2266 | struct mem_cgroup *memcg; | |
2267 | struct page_cgroup *pc; | |
2268 | ||
2269 | pc = lookup_page_cgroup(page); | |
2270 | again: | |
2271 | memcg = pc->mem_cgroup; | |
2272 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
2273 | return; | |
2274 | /* | |
2275 | * If this memory cgroup is not under account moving, we don't | |
da92c47d | 2276 | * need to take move_lock_mem_cgroup(). Because we already hold |
89c06bd5 | 2277 | * rcu_read_lock(), any calls to move_account will be delayed until |
13fd1dd9 | 2278 | * rcu_read_unlock() if mem_cgroup_stolen() == true. |
89c06bd5 | 2279 | */ |
13fd1dd9 | 2280 | if (!mem_cgroup_stolen(memcg)) |
89c06bd5 KH |
2281 | return; |
2282 | ||
2283 | move_lock_mem_cgroup(memcg, flags); | |
2284 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { | |
2285 | move_unlock_mem_cgroup(memcg, flags); | |
2286 | goto again; | |
2287 | } | |
2288 | *locked = true; | |
2289 | } | |
2290 | ||
2291 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) | |
2292 | { | |
2293 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2294 | ||
2295 | /* | |
2296 | * It's guaranteed that pc->mem_cgroup never changes while | |
2297 | * lock is held because a routine modifies pc->mem_cgroup | |
da92c47d | 2298 | * should take move_lock_mem_cgroup(). |
89c06bd5 KH |
2299 | */ |
2300 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); | |
2301 | } | |
2302 | ||
2a7106f2 GT |
2303 | void mem_cgroup_update_page_stat(struct page *page, |
2304 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f | 2305 | { |
c0ff4b85 | 2306 | struct mem_cgroup *memcg; |
32047e2a | 2307 | struct page_cgroup *pc = lookup_page_cgroup(page); |
dbd4ea78 | 2308 | unsigned long uninitialized_var(flags); |
d69b042f | 2309 | |
cfa44946 | 2310 | if (mem_cgroup_disabled()) |
d69b042f | 2311 | return; |
89c06bd5 | 2312 | |
c0ff4b85 R |
2313 | memcg = pc->mem_cgroup; |
2314 | if (unlikely(!memcg || !PageCgroupUsed(pc))) | |
89c06bd5 | 2315 | return; |
26174efd | 2316 | |
26174efd | 2317 | switch (idx) { |
2a7106f2 | 2318 | case MEMCG_NR_FILE_MAPPED: |
2a7106f2 | 2319 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
2320 | break; |
2321 | default: | |
2322 | BUG(); | |
8725d541 | 2323 | } |
d69b042f | 2324 | |
c0ff4b85 | 2325 | this_cpu_add(memcg->stat->count[idx], val); |
d69b042f | 2326 | } |
26174efd | 2327 | |
cdec2e42 KH |
2328 | /* |
2329 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
2330 | * TODO: maybe necessary to use big numbers in big irons. | |
2331 | */ | |
7ec99d62 | 2332 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
2333 | struct memcg_stock_pcp { |
2334 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2335 | unsigned int nr_pages; |
cdec2e42 | 2336 | struct work_struct work; |
26fe6168 | 2337 | unsigned long flags; |
a0db00fc | 2338 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2339 | }; |
2340 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2341 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2342 | |
a0956d54 SS |
2343 | /** |
2344 | * consume_stock: Try to consume stocked charge on this cpu. | |
2345 | * @memcg: memcg to consume from. | |
2346 | * @nr_pages: how many pages to charge. | |
2347 | * | |
2348 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2349 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2350 | * service an allocation will refill the stock. | |
2351 | * | |
2352 | * returns true if successful, false otherwise. | |
cdec2e42 | 2353 | */ |
a0956d54 | 2354 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2355 | { |
2356 | struct memcg_stock_pcp *stock; | |
2357 | bool ret = true; | |
2358 | ||
a0956d54 SS |
2359 | if (nr_pages > CHARGE_BATCH) |
2360 | return false; | |
2361 | ||
cdec2e42 | 2362 | stock = &get_cpu_var(memcg_stock); |
a0956d54 SS |
2363 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
2364 | stock->nr_pages -= nr_pages; | |
cdec2e42 KH |
2365 | else /* need to call res_counter_charge */ |
2366 | ret = false; | |
2367 | put_cpu_var(memcg_stock); | |
2368 | return ret; | |
2369 | } | |
2370 | ||
2371 | /* | |
2372 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2373 | */ | |
2374 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2375 | { | |
2376 | struct mem_cgroup *old = stock->cached; | |
2377 | ||
11c9ea4e JW |
2378 | if (stock->nr_pages) { |
2379 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2380 | ||
2381 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2382 | if (do_swap_account) |
11c9ea4e JW |
2383 | res_counter_uncharge(&old->memsw, bytes); |
2384 | stock->nr_pages = 0; | |
cdec2e42 KH |
2385 | } |
2386 | stock->cached = NULL; | |
cdec2e42 KH |
2387 | } |
2388 | ||
2389 | /* | |
2390 | * This must be called under preempt disabled or must be called by | |
2391 | * a thread which is pinned to local cpu. | |
2392 | */ | |
2393 | static void drain_local_stock(struct work_struct *dummy) | |
2394 | { | |
2395 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2396 | drain_stock(stock); | |
26fe6168 | 2397 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2398 | } |
2399 | ||
e4777496 MH |
2400 | static void __init memcg_stock_init(void) |
2401 | { | |
2402 | int cpu; | |
2403 | ||
2404 | for_each_possible_cpu(cpu) { | |
2405 | struct memcg_stock_pcp *stock = | |
2406 | &per_cpu(memcg_stock, cpu); | |
2407 | INIT_WORK(&stock->work, drain_local_stock); | |
2408 | } | |
2409 | } | |
2410 | ||
cdec2e42 KH |
2411 | /* |
2412 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2413 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2414 | */ |
c0ff4b85 | 2415 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2416 | { |
2417 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2418 | ||
c0ff4b85 | 2419 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2420 | drain_stock(stock); |
c0ff4b85 | 2421 | stock->cached = memcg; |
cdec2e42 | 2422 | } |
11c9ea4e | 2423 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2424 | put_cpu_var(memcg_stock); |
2425 | } | |
2426 | ||
2427 | /* | |
c0ff4b85 | 2428 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
d38144b7 MH |
2429 | * of the hierarchy under it. sync flag says whether we should block |
2430 | * until the work is done. | |
cdec2e42 | 2431 | */ |
c0ff4b85 | 2432 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
cdec2e42 | 2433 | { |
26fe6168 | 2434 | int cpu, curcpu; |
d38144b7 | 2435 | |
cdec2e42 | 2436 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2437 | get_online_cpus(); |
5af12d0e | 2438 | curcpu = get_cpu(); |
cdec2e42 KH |
2439 | for_each_online_cpu(cpu) { |
2440 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2441 | struct mem_cgroup *memcg; |
26fe6168 | 2442 | |
c0ff4b85 R |
2443 | memcg = stock->cached; |
2444 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 2445 | continue; |
c0ff4b85 | 2446 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
3e92041d | 2447 | continue; |
d1a05b69 MH |
2448 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2449 | if (cpu == curcpu) | |
2450 | drain_local_stock(&stock->work); | |
2451 | else | |
2452 | schedule_work_on(cpu, &stock->work); | |
2453 | } | |
cdec2e42 | 2454 | } |
5af12d0e | 2455 | put_cpu(); |
d38144b7 MH |
2456 | |
2457 | if (!sync) | |
2458 | goto out; | |
2459 | ||
2460 | for_each_online_cpu(cpu) { | |
2461 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2462 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2463 | flush_work(&stock->work); |
2464 | } | |
2465 | out: | |
cdec2e42 | 2466 | put_online_cpus(); |
d38144b7 MH |
2467 | } |
2468 | ||
2469 | /* | |
2470 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2471 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2472 | * expects some charges will be back to res_counter later but cannot wait for | |
2473 | * it. | |
2474 | */ | |
c0ff4b85 | 2475 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
d38144b7 | 2476 | { |
9f50fad6 MH |
2477 | /* |
2478 | * If someone calls draining, avoid adding more kworker runs. | |
2479 | */ | |
2480 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2481 | return; | |
c0ff4b85 | 2482 | drain_all_stock(root_memcg, false); |
9f50fad6 | 2483 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2484 | } |
2485 | ||
2486 | /* This is a synchronous drain interface. */ | |
c0ff4b85 | 2487 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
cdec2e42 KH |
2488 | { |
2489 | /* called when force_empty is called */ | |
9f50fad6 | 2490 | mutex_lock(&percpu_charge_mutex); |
c0ff4b85 | 2491 | drain_all_stock(root_memcg, true); |
9f50fad6 | 2492 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2493 | } |
2494 | ||
711d3d2c KH |
2495 | /* |
2496 | * This function drains percpu counter value from DEAD cpu and | |
2497 | * move it to local cpu. Note that this function can be preempted. | |
2498 | */ | |
c0ff4b85 | 2499 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
711d3d2c KH |
2500 | { |
2501 | int i; | |
2502 | ||
c0ff4b85 | 2503 | spin_lock(&memcg->pcp_counter_lock); |
6104621d | 2504 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
c0ff4b85 | 2505 | long x = per_cpu(memcg->stat->count[i], cpu); |
711d3d2c | 2506 | |
c0ff4b85 R |
2507 | per_cpu(memcg->stat->count[i], cpu) = 0; |
2508 | memcg->nocpu_base.count[i] += x; | |
711d3d2c | 2509 | } |
e9f8974f | 2510 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
c0ff4b85 | 2511 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
e9f8974f | 2512 | |
c0ff4b85 R |
2513 | per_cpu(memcg->stat->events[i], cpu) = 0; |
2514 | memcg->nocpu_base.events[i] += x; | |
e9f8974f | 2515 | } |
c0ff4b85 | 2516 | spin_unlock(&memcg->pcp_counter_lock); |
711d3d2c KH |
2517 | } |
2518 | ||
2519 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2520 | unsigned long action, |
2521 | void *hcpu) | |
2522 | { | |
2523 | int cpu = (unsigned long)hcpu; | |
2524 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2525 | struct mem_cgroup *iter; |
cdec2e42 | 2526 | |
619d094b | 2527 | if (action == CPU_ONLINE) |
1489ebad | 2528 | return NOTIFY_OK; |
1489ebad | 2529 | |
d833049b | 2530 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 2531 | return NOTIFY_OK; |
711d3d2c | 2532 | |
9f3a0d09 | 2533 | for_each_mem_cgroup(iter) |
711d3d2c KH |
2534 | mem_cgroup_drain_pcp_counter(iter, cpu); |
2535 | ||
cdec2e42 KH |
2536 | stock = &per_cpu(memcg_stock, cpu); |
2537 | drain_stock(stock); | |
2538 | return NOTIFY_OK; | |
2539 | } | |
2540 | ||
4b534334 KH |
2541 | |
2542 | /* See __mem_cgroup_try_charge() for details */ | |
2543 | enum { | |
2544 | CHARGE_OK, /* success */ | |
2545 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2546 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2547 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2548 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2549 | }; | |
2550 | ||
c0ff4b85 | 2551 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
4c9c5359 SS |
2552 | unsigned int nr_pages, unsigned int min_pages, |
2553 | bool oom_check) | |
4b534334 | 2554 | { |
7ec99d62 | 2555 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2556 | struct mem_cgroup *mem_over_limit; |
2557 | struct res_counter *fail_res; | |
2558 | unsigned long flags = 0; | |
2559 | int ret; | |
2560 | ||
c0ff4b85 | 2561 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
4b534334 KH |
2562 | |
2563 | if (likely(!ret)) { | |
2564 | if (!do_swap_account) | |
2565 | return CHARGE_OK; | |
c0ff4b85 | 2566 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
4b534334 KH |
2567 | if (likely(!ret)) |
2568 | return CHARGE_OK; | |
2569 | ||
c0ff4b85 | 2570 | res_counter_uncharge(&memcg->res, csize); |
4b534334 KH |
2571 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2572 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2573 | } else | |
2574 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2575 | /* |
9221edb7 JW |
2576 | * Never reclaim on behalf of optional batching, retry with a |
2577 | * single page instead. | |
2578 | */ | |
4c9c5359 | 2579 | if (nr_pages > min_pages) |
4b534334 KH |
2580 | return CHARGE_RETRY; |
2581 | ||
2582 | if (!(gfp_mask & __GFP_WAIT)) | |
2583 | return CHARGE_WOULDBLOCK; | |
2584 | ||
4c9c5359 SS |
2585 | if (gfp_mask & __GFP_NORETRY) |
2586 | return CHARGE_NOMEM; | |
2587 | ||
5660048c | 2588 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
7ec99d62 | 2589 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2590 | return CHARGE_RETRY; |
4b534334 | 2591 | /* |
19942822 JW |
2592 | * Even though the limit is exceeded at this point, reclaim |
2593 | * may have been able to free some pages. Retry the charge | |
2594 | * before killing the task. | |
2595 | * | |
2596 | * Only for regular pages, though: huge pages are rather | |
2597 | * unlikely to succeed so close to the limit, and we fall back | |
2598 | * to regular pages anyway in case of failure. | |
4b534334 | 2599 | */ |
4c9c5359 | 2600 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
4b534334 KH |
2601 | return CHARGE_RETRY; |
2602 | ||
2603 | /* | |
2604 | * At task move, charge accounts can be doubly counted. So, it's | |
2605 | * better to wait until the end of task_move if something is going on. | |
2606 | */ | |
2607 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2608 | return CHARGE_RETRY; | |
2609 | ||
2610 | /* If we don't need to call oom-killer at el, return immediately */ | |
2611 | if (!oom_check) | |
2612 | return CHARGE_NOMEM; | |
2613 | /* check OOM */ | |
e845e199 | 2614 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask, get_order(csize))) |
4b534334 KH |
2615 | return CHARGE_OOM_DIE; |
2616 | ||
2617 | return CHARGE_RETRY; | |
2618 | } | |
2619 | ||
f817ed48 | 2620 | /* |
38c5d72f KH |
2621 | * __mem_cgroup_try_charge() does |
2622 | * 1. detect memcg to be charged against from passed *mm and *ptr, | |
2623 | * 2. update res_counter | |
2624 | * 3. call memory reclaim if necessary. | |
2625 | * | |
2626 | * In some special case, if the task is fatal, fatal_signal_pending() or | |
2627 | * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup | |
2628 | * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon | |
2629 | * as possible without any hazards. 2: all pages should have a valid | |
2630 | * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg | |
2631 | * pointer, that is treated as a charge to root_mem_cgroup. | |
2632 | * | |
2633 | * So __mem_cgroup_try_charge() will return | |
2634 | * 0 ... on success, filling *ptr with a valid memcg pointer. | |
2635 | * -ENOMEM ... charge failure because of resource limits. | |
2636 | * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup. | |
2637 | * | |
2638 | * Unlike the exported interface, an "oom" parameter is added. if oom==true, | |
2639 | * the oom-killer can be invoked. | |
8a9f3ccd | 2640 | */ |
f817ed48 | 2641 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2642 | gfp_t gfp_mask, |
7ec99d62 | 2643 | unsigned int nr_pages, |
c0ff4b85 | 2644 | struct mem_cgroup **ptr, |
7ec99d62 | 2645 | bool oom) |
8a9f3ccd | 2646 | { |
7ec99d62 | 2647 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 | 2648 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
c0ff4b85 | 2649 | struct mem_cgroup *memcg = NULL; |
4b534334 | 2650 | int ret; |
a636b327 | 2651 | |
867578cb KH |
2652 | /* |
2653 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2654 | * in system level. So, allow to go ahead dying process in addition to | |
2655 | * MEMDIE process. | |
2656 | */ | |
2657 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2658 | || fatal_signal_pending(current))) | |
2659 | goto bypass; | |
a636b327 | 2660 | |
8a9f3ccd | 2661 | /* |
3be91277 HD |
2662 | * We always charge the cgroup the mm_struct belongs to. |
2663 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd | 2664 | * thread group leader migrates. It's possible that mm is not |
24467cac | 2665 | * set, if so charge the root memcg (happens for pagecache usage). |
8a9f3ccd | 2666 | */ |
c0ff4b85 | 2667 | if (!*ptr && !mm) |
38c5d72f | 2668 | *ptr = root_mem_cgroup; |
f75ca962 | 2669 | again: |
c0ff4b85 R |
2670 | if (*ptr) { /* css should be a valid one */ |
2671 | memcg = *ptr; | |
c0ff4b85 | 2672 | if (mem_cgroup_is_root(memcg)) |
f75ca962 | 2673 | goto done; |
a0956d54 | 2674 | if (consume_stock(memcg, nr_pages)) |
f75ca962 | 2675 | goto done; |
c0ff4b85 | 2676 | css_get(&memcg->css); |
4b534334 | 2677 | } else { |
f75ca962 | 2678 | struct task_struct *p; |
54595fe2 | 2679 | |
f75ca962 KH |
2680 | rcu_read_lock(); |
2681 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2682 | /* |
ebb76ce1 | 2683 | * Because we don't have task_lock(), "p" can exit. |
c0ff4b85 | 2684 | * In that case, "memcg" can point to root or p can be NULL with |
ebb76ce1 KH |
2685 | * race with swapoff. Then, we have small risk of mis-accouning. |
2686 | * But such kind of mis-account by race always happens because | |
2687 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2688 | * small race, here. | |
2689 | * (*) swapoff at el will charge against mm-struct not against | |
2690 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 | 2691 | */ |
c0ff4b85 | 2692 | memcg = mem_cgroup_from_task(p); |
38c5d72f KH |
2693 | if (!memcg) |
2694 | memcg = root_mem_cgroup; | |
2695 | if (mem_cgroup_is_root(memcg)) { | |
f75ca962 KH |
2696 | rcu_read_unlock(); |
2697 | goto done; | |
2698 | } | |
a0956d54 | 2699 | if (consume_stock(memcg, nr_pages)) { |
f75ca962 KH |
2700 | /* |
2701 | * It seems dagerous to access memcg without css_get(). | |
2702 | * But considering how consume_stok works, it's not | |
2703 | * necessary. If consume_stock success, some charges | |
2704 | * from this memcg are cached on this cpu. So, we | |
2705 | * don't need to call css_get()/css_tryget() before | |
2706 | * calling consume_stock(). | |
2707 | */ | |
2708 | rcu_read_unlock(); | |
2709 | goto done; | |
2710 | } | |
2711 | /* after here, we may be blocked. we need to get refcnt */ | |
c0ff4b85 | 2712 | if (!css_tryget(&memcg->css)) { |
f75ca962 KH |
2713 | rcu_read_unlock(); |
2714 | goto again; | |
2715 | } | |
2716 | rcu_read_unlock(); | |
2717 | } | |
8a9f3ccd | 2718 | |
4b534334 KH |
2719 | do { |
2720 | bool oom_check; | |
7a81b88c | 2721 | |
4b534334 | 2722 | /* If killed, bypass charge */ |
f75ca962 | 2723 | if (fatal_signal_pending(current)) { |
c0ff4b85 | 2724 | css_put(&memcg->css); |
4b534334 | 2725 | goto bypass; |
f75ca962 | 2726 | } |
6d61ef40 | 2727 | |
4b534334 KH |
2728 | oom_check = false; |
2729 | if (oom && !nr_oom_retries) { | |
2730 | oom_check = true; | |
2731 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2732 | } |
66e1707b | 2733 | |
4c9c5359 SS |
2734 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages, |
2735 | oom_check); | |
4b534334 KH |
2736 | switch (ret) { |
2737 | case CHARGE_OK: | |
2738 | break; | |
2739 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2740 | batch = nr_pages; |
c0ff4b85 R |
2741 | css_put(&memcg->css); |
2742 | memcg = NULL; | |
f75ca962 | 2743 | goto again; |
4b534334 | 2744 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
c0ff4b85 | 2745 | css_put(&memcg->css); |
4b534334 KH |
2746 | goto nomem; |
2747 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 | 2748 | if (!oom) { |
c0ff4b85 | 2749 | css_put(&memcg->css); |
867578cb | 2750 | goto nomem; |
f75ca962 | 2751 | } |
4b534334 KH |
2752 | /* If oom, we never return -ENOMEM */ |
2753 | nr_oom_retries--; | |
2754 | break; | |
2755 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
c0ff4b85 | 2756 | css_put(&memcg->css); |
867578cb | 2757 | goto bypass; |
66e1707b | 2758 | } |
4b534334 KH |
2759 | } while (ret != CHARGE_OK); |
2760 | ||
7ec99d62 | 2761 | if (batch > nr_pages) |
c0ff4b85 R |
2762 | refill_stock(memcg, batch - nr_pages); |
2763 | css_put(&memcg->css); | |
0c3e73e8 | 2764 | done: |
c0ff4b85 | 2765 | *ptr = memcg; |
7a81b88c KH |
2766 | return 0; |
2767 | nomem: | |
c0ff4b85 | 2768 | *ptr = NULL; |
7a81b88c | 2769 | return -ENOMEM; |
867578cb | 2770 | bypass: |
38c5d72f KH |
2771 | *ptr = root_mem_cgroup; |
2772 | return -EINTR; | |
7a81b88c | 2773 | } |
8a9f3ccd | 2774 | |
a3032a2c DN |
2775 | /* |
2776 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2777 | * This function is for that and do uncharge, put css's refcnt. | |
2778 | * gotten by try_charge(). | |
2779 | */ | |
c0ff4b85 | 2780 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
e7018b8d | 2781 | unsigned int nr_pages) |
a3032a2c | 2782 | { |
c0ff4b85 | 2783 | if (!mem_cgroup_is_root(memcg)) { |
e7018b8d JW |
2784 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2785 | ||
c0ff4b85 | 2786 | res_counter_uncharge(&memcg->res, bytes); |
a3032a2c | 2787 | if (do_swap_account) |
c0ff4b85 | 2788 | res_counter_uncharge(&memcg->memsw, bytes); |
a3032a2c | 2789 | } |
854ffa8d DN |
2790 | } |
2791 | ||
d01dd17f KH |
2792 | /* |
2793 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. | |
2794 | * This is useful when moving usage to parent cgroup. | |
2795 | */ | |
2796 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, | |
2797 | unsigned int nr_pages) | |
2798 | { | |
2799 | unsigned long bytes = nr_pages * PAGE_SIZE; | |
2800 | ||
2801 | if (mem_cgroup_is_root(memcg)) | |
2802 | return; | |
2803 | ||
2804 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); | |
2805 | if (do_swap_account) | |
2806 | res_counter_uncharge_until(&memcg->memsw, | |
2807 | memcg->memsw.parent, bytes); | |
2808 | } | |
2809 | ||
a3b2d692 KH |
2810 | /* |
2811 | * A helper function to get mem_cgroup from ID. must be called under | |
e9316080 TH |
2812 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
2813 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be | |
2814 | * called against removed memcg.) | |
a3b2d692 KH |
2815 | */ |
2816 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2817 | { | |
2818 | struct cgroup_subsys_state *css; | |
2819 | ||
2820 | /* ID 0 is unused ID */ | |
2821 | if (!id) | |
2822 | return NULL; | |
2823 | css = css_lookup(&mem_cgroup_subsys, id); | |
2824 | if (!css) | |
2825 | return NULL; | |
b2145145 | 2826 | return mem_cgroup_from_css(css); |
a3b2d692 KH |
2827 | } |
2828 | ||
e42d9d5d | 2829 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2830 | { |
c0ff4b85 | 2831 | struct mem_cgroup *memcg = NULL; |
3c776e64 | 2832 | struct page_cgroup *pc; |
a3b2d692 | 2833 | unsigned short id; |
b5a84319 KH |
2834 | swp_entry_t ent; |
2835 | ||
3c776e64 DN |
2836 | VM_BUG_ON(!PageLocked(page)); |
2837 | ||
3c776e64 | 2838 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2839 | lock_page_cgroup(pc); |
a3b2d692 | 2840 | if (PageCgroupUsed(pc)) { |
c0ff4b85 R |
2841 | memcg = pc->mem_cgroup; |
2842 | if (memcg && !css_tryget(&memcg->css)) | |
2843 | memcg = NULL; | |
e42d9d5d | 2844 | } else if (PageSwapCache(page)) { |
3c776e64 | 2845 | ent.val = page_private(page); |
9fb4b7cc | 2846 | id = lookup_swap_cgroup_id(ent); |
a3b2d692 | 2847 | rcu_read_lock(); |
c0ff4b85 R |
2848 | memcg = mem_cgroup_lookup(id); |
2849 | if (memcg && !css_tryget(&memcg->css)) | |
2850 | memcg = NULL; | |
a3b2d692 | 2851 | rcu_read_unlock(); |
3c776e64 | 2852 | } |
c0bd3f63 | 2853 | unlock_page_cgroup(pc); |
c0ff4b85 | 2854 | return memcg; |
b5a84319 KH |
2855 | } |
2856 | ||
c0ff4b85 | 2857 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
5564e88b | 2858 | struct page *page, |
7ec99d62 | 2859 | unsigned int nr_pages, |
9ce70c02 HD |
2860 | enum charge_type ctype, |
2861 | bool lrucare) | |
7a81b88c | 2862 | { |
ce587e65 | 2863 | struct page_cgroup *pc = lookup_page_cgroup(page); |
9ce70c02 | 2864 | struct zone *uninitialized_var(zone); |
fa9add64 | 2865 | struct lruvec *lruvec; |
9ce70c02 | 2866 | bool was_on_lru = false; |
b2402857 | 2867 | bool anon; |
9ce70c02 | 2868 | |
ca3e0214 | 2869 | lock_page_cgroup(pc); |
90deb788 | 2870 | VM_BUG_ON(PageCgroupUsed(pc)); |
ca3e0214 KH |
2871 | /* |
2872 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2873 | * accessed by any other context at this point. | |
2874 | */ | |
9ce70c02 HD |
2875 | |
2876 | /* | |
2877 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2878 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2879 | */ | |
2880 | if (lrucare) { | |
2881 | zone = page_zone(page); | |
2882 | spin_lock_irq(&zone->lru_lock); | |
2883 | if (PageLRU(page)) { | |
fa9add64 | 2884 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 | 2885 | ClearPageLRU(page); |
fa9add64 | 2886 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2887 | was_on_lru = true; |
2888 | } | |
2889 | } | |
2890 | ||
c0ff4b85 | 2891 | pc->mem_cgroup = memcg; |
261fb61a KH |
2892 | /* |
2893 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2894 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2895 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2896 | * before USED bit, we need memory barrier here. | |
2897 | * See mem_cgroup_add_lru_list(), etc. | |
2898 | */ | |
08e552c6 | 2899 | smp_wmb(); |
b2402857 | 2900 | SetPageCgroupUsed(pc); |
3be91277 | 2901 | |
9ce70c02 HD |
2902 | if (lrucare) { |
2903 | if (was_on_lru) { | |
fa9add64 | 2904 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
9ce70c02 HD |
2905 | VM_BUG_ON(PageLRU(page)); |
2906 | SetPageLRU(page); | |
fa9add64 | 2907 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
9ce70c02 HD |
2908 | } |
2909 | spin_unlock_irq(&zone->lru_lock); | |
2910 | } | |
2911 | ||
41326c17 | 2912 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
b2402857 KH |
2913 | anon = true; |
2914 | else | |
2915 | anon = false; | |
2916 | ||
2917 | mem_cgroup_charge_statistics(memcg, anon, nr_pages); | |
52d4b9ac | 2918 | unlock_page_cgroup(pc); |
9ce70c02 | 2919 | |
430e4863 KH |
2920 | /* |
2921 | * "charge_statistics" updated event counter. Then, check it. | |
2922 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2923 | * if they exceeds softlimit. | |
2924 | */ | |
c0ff4b85 | 2925 | memcg_check_events(memcg, page); |
7a81b88c | 2926 | } |
66e1707b | 2927 | |
7cf27982 GC |
2928 | static DEFINE_MUTEX(set_limit_mutex); |
2929 | ||
7ae1e1d0 GC |
2930 | #ifdef CONFIG_MEMCG_KMEM |
2931 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) | |
2932 | { | |
2933 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && | |
2934 | (memcg->kmem_account_flags & KMEM_ACCOUNTED_MASK); | |
2935 | } | |
2936 | ||
1f458cbf GC |
2937 | /* |
2938 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer | |
2939 | * in the memcg_cache_params struct. | |
2940 | */ | |
2941 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) | |
2942 | { | |
2943 | struct kmem_cache *cachep; | |
2944 | ||
2945 | VM_BUG_ON(p->is_root_cache); | |
2946 | cachep = p->root_cache; | |
2947 | return cachep->memcg_params->memcg_caches[memcg_cache_id(p->memcg)]; | |
2948 | } | |
2949 | ||
749c5415 GC |
2950 | #ifdef CONFIG_SLABINFO |
2951 | static int mem_cgroup_slabinfo_read(struct cgroup *cont, struct cftype *cft, | |
2952 | struct seq_file *m) | |
2953 | { | |
2954 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); | |
2955 | struct memcg_cache_params *params; | |
2956 | ||
2957 | if (!memcg_can_account_kmem(memcg)) | |
2958 | return -EIO; | |
2959 | ||
2960 | print_slabinfo_header(m); | |
2961 | ||
2962 | mutex_lock(&memcg->slab_caches_mutex); | |
2963 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) | |
2964 | cache_show(memcg_params_to_cache(params), m); | |
2965 | mutex_unlock(&memcg->slab_caches_mutex); | |
2966 | ||
2967 | return 0; | |
2968 | } | |
2969 | #endif | |
2970 | ||
7ae1e1d0 GC |
2971 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
2972 | { | |
2973 | struct res_counter *fail_res; | |
2974 | struct mem_cgroup *_memcg; | |
2975 | int ret = 0; | |
2976 | bool may_oom; | |
2977 | ||
2978 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); | |
2979 | if (ret) | |
2980 | return ret; | |
2981 | ||
2982 | /* | |
2983 | * Conditions under which we can wait for the oom_killer. Those are | |
2984 | * the same conditions tested by the core page allocator | |
2985 | */ | |
2986 | may_oom = (gfp & __GFP_FS) && !(gfp & __GFP_NORETRY); | |
2987 | ||
2988 | _memcg = memcg; | |
2989 | ret = __mem_cgroup_try_charge(NULL, gfp, size >> PAGE_SHIFT, | |
2990 | &_memcg, may_oom); | |
2991 | ||
2992 | if (ret == -EINTR) { | |
2993 | /* | |
2994 | * __mem_cgroup_try_charge() chosed to bypass to root due to | |
2995 | * OOM kill or fatal signal. Since our only options are to | |
2996 | * either fail the allocation or charge it to this cgroup, do | |
2997 | * it as a temporary condition. But we can't fail. From a | |
2998 | * kmem/slab perspective, the cache has already been selected, | |
2999 | * by mem_cgroup_kmem_get_cache(), so it is too late to change | |
3000 | * our minds. | |
3001 | * | |
3002 | * This condition will only trigger if the task entered | |
3003 | * memcg_charge_kmem in a sane state, but was OOM-killed during | |
3004 | * __mem_cgroup_try_charge() above. Tasks that were already | |
3005 | * dying when the allocation triggers should have been already | |
3006 | * directed to the root cgroup in memcontrol.h | |
3007 | */ | |
3008 | res_counter_charge_nofail(&memcg->res, size, &fail_res); | |
3009 | if (do_swap_account) | |
3010 | res_counter_charge_nofail(&memcg->memsw, size, | |
3011 | &fail_res); | |
3012 | ret = 0; | |
3013 | } else if (ret) | |
3014 | res_counter_uncharge(&memcg->kmem, size); | |
3015 | ||
3016 | return ret; | |
3017 | } | |
3018 | ||
3019 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) | |
3020 | { | |
7ae1e1d0 GC |
3021 | res_counter_uncharge(&memcg->res, size); |
3022 | if (do_swap_account) | |
3023 | res_counter_uncharge(&memcg->memsw, size); | |
7de37682 GC |
3024 | |
3025 | /* Not down to 0 */ | |
3026 | if (res_counter_uncharge(&memcg->kmem, size)) | |
3027 | return; | |
3028 | ||
3029 | if (memcg_kmem_test_and_clear_dead(memcg)) | |
3030 | mem_cgroup_put(memcg); | |
7ae1e1d0 GC |
3031 | } |
3032 | ||
2633d7a0 GC |
3033 | void memcg_cache_list_add(struct mem_cgroup *memcg, struct kmem_cache *cachep) |
3034 | { | |
3035 | if (!memcg) | |
3036 | return; | |
3037 | ||
3038 | mutex_lock(&memcg->slab_caches_mutex); | |
3039 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); | |
3040 | mutex_unlock(&memcg->slab_caches_mutex); | |
3041 | } | |
3042 | ||
3043 | /* | |
3044 | * helper for acessing a memcg's index. It will be used as an index in the | |
3045 | * child cache array in kmem_cache, and also to derive its name. This function | |
3046 | * will return -1 when this is not a kmem-limited memcg. | |
3047 | */ | |
3048 | int memcg_cache_id(struct mem_cgroup *memcg) | |
3049 | { | |
3050 | return memcg ? memcg->kmemcg_id : -1; | |
3051 | } | |
3052 | ||
55007d84 GC |
3053 | /* |
3054 | * This ends up being protected by the set_limit mutex, during normal | |
3055 | * operation, because that is its main call site. | |
3056 | * | |
3057 | * But when we create a new cache, we can call this as well if its parent | |
3058 | * is kmem-limited. That will have to hold set_limit_mutex as well. | |
3059 | */ | |
3060 | int memcg_update_cache_sizes(struct mem_cgroup *memcg) | |
3061 | { | |
3062 | int num, ret; | |
3063 | ||
3064 | num = ida_simple_get(&kmem_limited_groups, | |
3065 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); | |
3066 | if (num < 0) | |
3067 | return num; | |
3068 | /* | |
3069 | * After this point, kmem_accounted (that we test atomically in | |
3070 | * the beginning of this conditional), is no longer 0. This | |
3071 | * guarantees only one process will set the following boolean | |
3072 | * to true. We don't need test_and_set because we're protected | |
3073 | * by the set_limit_mutex anyway. | |
3074 | */ | |
3075 | memcg_kmem_set_activated(memcg); | |
3076 | ||
3077 | ret = memcg_update_all_caches(num+1); | |
3078 | if (ret) { | |
3079 | ida_simple_remove(&kmem_limited_groups, num); | |
3080 | memcg_kmem_clear_activated(memcg); | |
3081 | return ret; | |
3082 | } | |
3083 | ||
3084 | memcg->kmemcg_id = num; | |
3085 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); | |
3086 | mutex_init(&memcg->slab_caches_mutex); | |
3087 | return 0; | |
3088 | } | |
3089 | ||
3090 | static size_t memcg_caches_array_size(int num_groups) | |
3091 | { | |
3092 | ssize_t size; | |
3093 | if (num_groups <= 0) | |
3094 | return 0; | |
3095 | ||
3096 | size = 2 * num_groups; | |
3097 | if (size < MEMCG_CACHES_MIN_SIZE) | |
3098 | size = MEMCG_CACHES_MIN_SIZE; | |
3099 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
3100 | size = MEMCG_CACHES_MAX_SIZE; | |
3101 | ||
3102 | return size; | |
3103 | } | |
3104 | ||
3105 | /* | |
3106 | * We should update the current array size iff all caches updates succeed. This | |
3107 | * can only be done from the slab side. The slab mutex needs to be held when | |
3108 | * calling this. | |
3109 | */ | |
3110 | void memcg_update_array_size(int num) | |
3111 | { | |
3112 | if (num > memcg_limited_groups_array_size) | |
3113 | memcg_limited_groups_array_size = memcg_caches_array_size(num); | |
3114 | } | |
3115 | ||
15cf17d2 KK |
3116 | static void kmem_cache_destroy_work_func(struct work_struct *w); |
3117 | ||
55007d84 GC |
3118 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
3119 | { | |
3120 | struct memcg_cache_params *cur_params = s->memcg_params; | |
3121 | ||
3122 | VM_BUG_ON(s->memcg_params && !s->memcg_params->is_root_cache); | |
3123 | ||
3124 | if (num_groups > memcg_limited_groups_array_size) { | |
3125 | int i; | |
3126 | ssize_t size = memcg_caches_array_size(num_groups); | |
3127 | ||
3128 | size *= sizeof(void *); | |
3129 | size += sizeof(struct memcg_cache_params); | |
3130 | ||
3131 | s->memcg_params = kzalloc(size, GFP_KERNEL); | |
3132 | if (!s->memcg_params) { | |
3133 | s->memcg_params = cur_params; | |
3134 | return -ENOMEM; | |
3135 | } | |
3136 | ||
15cf17d2 KK |
3137 | INIT_WORK(&s->memcg_params->destroy, |
3138 | kmem_cache_destroy_work_func); | |
55007d84 GC |
3139 | s->memcg_params->is_root_cache = true; |
3140 | ||
3141 | /* | |
3142 | * There is the chance it will be bigger than | |
3143 | * memcg_limited_groups_array_size, if we failed an allocation | |
3144 | * in a cache, in which case all caches updated before it, will | |
3145 | * have a bigger array. | |
3146 | * | |
3147 | * But if that is the case, the data after | |
3148 | * memcg_limited_groups_array_size is certainly unused | |
3149 | */ | |
3150 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3151 | if (!cur_params->memcg_caches[i]) | |
3152 | continue; | |
3153 | s->memcg_params->memcg_caches[i] = | |
3154 | cur_params->memcg_caches[i]; | |
3155 | } | |
3156 | ||
3157 | /* | |
3158 | * Ideally, we would wait until all caches succeed, and only | |
3159 | * then free the old one. But this is not worth the extra | |
3160 | * pointer per-cache we'd have to have for this. | |
3161 | * | |
3162 | * It is not a big deal if some caches are left with a size | |
3163 | * bigger than the others. And all updates will reset this | |
3164 | * anyway. | |
3165 | */ | |
3166 | kfree(cur_params); | |
3167 | } | |
3168 | return 0; | |
3169 | } | |
3170 | ||
943a451a GC |
3171 | int memcg_register_cache(struct mem_cgroup *memcg, struct kmem_cache *s, |
3172 | struct kmem_cache *root_cache) | |
2633d7a0 GC |
3173 | { |
3174 | size_t size = sizeof(struct memcg_cache_params); | |
3175 | ||
3176 | if (!memcg_kmem_enabled()) | |
3177 | return 0; | |
3178 | ||
55007d84 GC |
3179 | if (!memcg) |
3180 | size += memcg_limited_groups_array_size * sizeof(void *); | |
3181 | ||
2633d7a0 GC |
3182 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
3183 | if (!s->memcg_params) | |
3184 | return -ENOMEM; | |
3185 | ||
15cf17d2 KK |
3186 | INIT_WORK(&s->memcg_params->destroy, |
3187 | kmem_cache_destroy_work_func); | |
943a451a | 3188 | if (memcg) { |
2633d7a0 | 3189 | s->memcg_params->memcg = memcg; |
943a451a | 3190 | s->memcg_params->root_cache = root_cache; |
4ba902b5 GC |
3191 | } else |
3192 | s->memcg_params->is_root_cache = true; | |
3193 | ||
2633d7a0 GC |
3194 | return 0; |
3195 | } | |
3196 | ||
3197 | void memcg_release_cache(struct kmem_cache *s) | |
3198 | { | |
d7f25f8a GC |
3199 | struct kmem_cache *root; |
3200 | struct mem_cgroup *memcg; | |
3201 | int id; | |
3202 | ||
3203 | /* | |
3204 | * This happens, for instance, when a root cache goes away before we | |
3205 | * add any memcg. | |
3206 | */ | |
3207 | if (!s->memcg_params) | |
3208 | return; | |
3209 | ||
3210 | if (s->memcg_params->is_root_cache) | |
3211 | goto out; | |
3212 | ||
3213 | memcg = s->memcg_params->memcg; | |
3214 | id = memcg_cache_id(memcg); | |
3215 | ||
3216 | root = s->memcg_params->root_cache; | |
3217 | root->memcg_params->memcg_caches[id] = NULL; | |
d7f25f8a GC |
3218 | |
3219 | mutex_lock(&memcg->slab_caches_mutex); | |
3220 | list_del(&s->memcg_params->list); | |
3221 | mutex_unlock(&memcg->slab_caches_mutex); | |
3222 | ||
fd0ccaf2 | 3223 | mem_cgroup_put(memcg); |
d7f25f8a | 3224 | out: |
2633d7a0 GC |
3225 | kfree(s->memcg_params); |
3226 | } | |
3227 | ||
0e9d92f2 GC |
3228 | /* |
3229 | * During the creation a new cache, we need to disable our accounting mechanism | |
3230 | * altogether. This is true even if we are not creating, but rather just | |
3231 | * enqueing new caches to be created. | |
3232 | * | |
3233 | * This is because that process will trigger allocations; some visible, like | |
3234 | * explicit kmallocs to auxiliary data structures, name strings and internal | |
3235 | * cache structures; some well concealed, like INIT_WORK() that can allocate | |
3236 | * objects during debug. | |
3237 | * | |
3238 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back | |
3239 | * to it. This may not be a bounded recursion: since the first cache creation | |
3240 | * failed to complete (waiting on the allocation), we'll just try to create the | |
3241 | * cache again, failing at the same point. | |
3242 | * | |
3243 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of | |
3244 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory | |
3245 | * inside the following two functions. | |
3246 | */ | |
3247 | static inline void memcg_stop_kmem_account(void) | |
3248 | { | |
3249 | VM_BUG_ON(!current->mm); | |
3250 | current->memcg_kmem_skip_account++; | |
3251 | } | |
3252 | ||
3253 | static inline void memcg_resume_kmem_account(void) | |
3254 | { | |
3255 | VM_BUG_ON(!current->mm); | |
3256 | current->memcg_kmem_skip_account--; | |
3257 | } | |
3258 | ||
1f458cbf GC |
3259 | static void kmem_cache_destroy_work_func(struct work_struct *w) |
3260 | { | |
3261 | struct kmem_cache *cachep; | |
3262 | struct memcg_cache_params *p; | |
3263 | ||
3264 | p = container_of(w, struct memcg_cache_params, destroy); | |
3265 | ||
3266 | cachep = memcg_params_to_cache(p); | |
3267 | ||
22933152 GC |
3268 | /* |
3269 | * If we get down to 0 after shrink, we could delete right away. | |
3270 | * However, memcg_release_pages() already puts us back in the workqueue | |
3271 | * in that case. If we proceed deleting, we'll get a dangling | |
3272 | * reference, and removing the object from the workqueue in that case | |
3273 | * is unnecessary complication. We are not a fast path. | |
3274 | * | |
3275 | * Note that this case is fundamentally different from racing with | |
3276 | * shrink_slab(): if memcg_cgroup_destroy_cache() is called in | |
3277 | * kmem_cache_shrink, not only we would be reinserting a dead cache | |
3278 | * into the queue, but doing so from inside the worker racing to | |
3279 | * destroy it. | |
3280 | * | |
3281 | * So if we aren't down to zero, we'll just schedule a worker and try | |
3282 | * again | |
3283 | */ | |
3284 | if (atomic_read(&cachep->memcg_params->nr_pages) != 0) { | |
3285 | kmem_cache_shrink(cachep); | |
3286 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) | |
3287 | return; | |
3288 | } else | |
1f458cbf GC |
3289 | kmem_cache_destroy(cachep); |
3290 | } | |
3291 | ||
3292 | void mem_cgroup_destroy_cache(struct kmem_cache *cachep) | |
3293 | { | |
3294 | if (!cachep->memcg_params->dead) | |
3295 | return; | |
3296 | ||
22933152 GC |
3297 | /* |
3298 | * There are many ways in which we can get here. | |
3299 | * | |
3300 | * We can get to a memory-pressure situation while the delayed work is | |
3301 | * still pending to run. The vmscan shrinkers can then release all | |
3302 | * cache memory and get us to destruction. If this is the case, we'll | |
3303 | * be executed twice, which is a bug (the second time will execute over | |
3304 | * bogus data). In this case, cancelling the work should be fine. | |
3305 | * | |
3306 | * But we can also get here from the worker itself, if | |
3307 | * kmem_cache_shrink is enough to shake all the remaining objects and | |
3308 | * get the page count to 0. In this case, we'll deadlock if we try to | |
3309 | * cancel the work (the worker runs with an internal lock held, which | |
3310 | * is the same lock we would hold for cancel_work_sync().) | |
3311 | * | |
3312 | * Since we can't possibly know who got us here, just refrain from | |
3313 | * running if there is already work pending | |
3314 | */ | |
3315 | if (work_pending(&cachep->memcg_params->destroy)) | |
3316 | return; | |
1f458cbf GC |
3317 | /* |
3318 | * We have to defer the actual destroying to a workqueue, because | |
3319 | * we might currently be in a context that cannot sleep. | |
3320 | */ | |
3321 | schedule_work(&cachep->memcg_params->destroy); | |
3322 | } | |
3323 | ||
d9c10ddd MH |
3324 | /* |
3325 | * This lock protects updaters, not readers. We want readers to be as fast as | |
3326 | * they can, and they will either see NULL or a valid cache value. Our model | |
3327 | * allow them to see NULL, in which case the root memcg will be selected. | |
3328 | * | |
3329 | * We need this lock because multiple allocations to the same cache from a non | |
3330 | * will span more than one worker. Only one of them can create the cache. | |
3331 | */ | |
3332 | static DEFINE_MUTEX(memcg_cache_mutex); | |
d7f25f8a | 3333 | |
d9c10ddd MH |
3334 | /* |
3335 | * Called with memcg_cache_mutex held | |
3336 | */ | |
d7f25f8a GC |
3337 | static struct kmem_cache *kmem_cache_dup(struct mem_cgroup *memcg, |
3338 | struct kmem_cache *s) | |
3339 | { | |
d7f25f8a | 3340 | struct kmem_cache *new; |
d9c10ddd | 3341 | static char *tmp_name = NULL; |
d7f25f8a | 3342 | |
d9c10ddd MH |
3343 | lockdep_assert_held(&memcg_cache_mutex); |
3344 | ||
3345 | /* | |
3346 | * kmem_cache_create_memcg duplicates the given name and | |
3347 | * cgroup_name for this name requires RCU context. | |
3348 | * This static temporary buffer is used to prevent from | |
3349 | * pointless shortliving allocation. | |
3350 | */ | |
3351 | if (!tmp_name) { | |
3352 | tmp_name = kmalloc(PATH_MAX, GFP_KERNEL); | |
3353 | if (!tmp_name) | |
3354 | return NULL; | |
3355 | } | |
3356 | ||
3357 | rcu_read_lock(); | |
3358 | snprintf(tmp_name, PATH_MAX, "%s(%d:%s)", s->name, | |
3359 | memcg_cache_id(memcg), cgroup_name(memcg->css.cgroup)); | |
3360 | rcu_read_unlock(); | |
d7f25f8a | 3361 | |
d9c10ddd | 3362 | new = kmem_cache_create_memcg(memcg, tmp_name, s->object_size, s->align, |
943a451a | 3363 | (s->flags & ~SLAB_PANIC), s->ctor, s); |
d7f25f8a | 3364 | |
d79923fa GC |
3365 | if (new) |
3366 | new->allocflags |= __GFP_KMEMCG; | |
3367 | ||
d7f25f8a GC |
3368 | return new; |
3369 | } | |
3370 | ||
d7f25f8a GC |
3371 | static struct kmem_cache *memcg_create_kmem_cache(struct mem_cgroup *memcg, |
3372 | struct kmem_cache *cachep) | |
3373 | { | |
3374 | struct kmem_cache *new_cachep; | |
3375 | int idx; | |
3376 | ||
3377 | BUG_ON(!memcg_can_account_kmem(memcg)); | |
3378 | ||
3379 | idx = memcg_cache_id(memcg); | |
3380 | ||
3381 | mutex_lock(&memcg_cache_mutex); | |
3382 | new_cachep = cachep->memcg_params->memcg_caches[idx]; | |
3383 | if (new_cachep) | |
3384 | goto out; | |
3385 | ||
3386 | new_cachep = kmem_cache_dup(memcg, cachep); | |
d7f25f8a GC |
3387 | if (new_cachep == NULL) { |
3388 | new_cachep = cachep; | |
3389 | goto out; | |
3390 | } | |
3391 | ||
3392 | mem_cgroup_get(memcg); | |
1f458cbf | 3393 | atomic_set(&new_cachep->memcg_params->nr_pages , 0); |
d7f25f8a GC |
3394 | |
3395 | cachep->memcg_params->memcg_caches[idx] = new_cachep; | |
3396 | /* | |
3397 | * the readers won't lock, make sure everybody sees the updated value, | |
3398 | * so they won't put stuff in the queue again for no reason | |
3399 | */ | |
3400 | wmb(); | |
3401 | out: | |
3402 | mutex_unlock(&memcg_cache_mutex); | |
3403 | return new_cachep; | |
3404 | } | |
3405 | ||
7cf27982 GC |
3406 | void kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
3407 | { | |
3408 | struct kmem_cache *c; | |
3409 | int i; | |
3410 | ||
3411 | if (!s->memcg_params) | |
3412 | return; | |
3413 | if (!s->memcg_params->is_root_cache) | |
3414 | return; | |
3415 | ||
3416 | /* | |
3417 | * If the cache is being destroyed, we trust that there is no one else | |
3418 | * requesting objects from it. Even if there are, the sanity checks in | |
3419 | * kmem_cache_destroy should caught this ill-case. | |
3420 | * | |
3421 | * Still, we don't want anyone else freeing memcg_caches under our | |
3422 | * noses, which can happen if a new memcg comes to life. As usual, | |
3423 | * we'll take the set_limit_mutex to protect ourselves against this. | |
3424 | */ | |
3425 | mutex_lock(&set_limit_mutex); | |
3426 | for (i = 0; i < memcg_limited_groups_array_size; i++) { | |
3427 | c = s->memcg_params->memcg_caches[i]; | |
3428 | if (!c) | |
3429 | continue; | |
3430 | ||
3431 | /* | |
3432 | * We will now manually delete the caches, so to avoid races | |
3433 | * we need to cancel all pending destruction workers and | |
3434 | * proceed with destruction ourselves. | |
3435 | * | |
3436 | * kmem_cache_destroy() will call kmem_cache_shrink internally, | |
3437 | * and that could spawn the workers again: it is likely that | |
3438 | * the cache still have active pages until this very moment. | |
3439 | * This would lead us back to mem_cgroup_destroy_cache. | |
3440 | * | |
3441 | * But that will not execute at all if the "dead" flag is not | |
3442 | * set, so flip it down to guarantee we are in control. | |
3443 | */ | |
3444 | c->memcg_params->dead = false; | |
22933152 | 3445 | cancel_work_sync(&c->memcg_params->destroy); |
7cf27982 GC |
3446 | kmem_cache_destroy(c); |
3447 | } | |
3448 | mutex_unlock(&set_limit_mutex); | |
3449 | } | |
3450 | ||
d7f25f8a GC |
3451 | struct create_work { |
3452 | struct mem_cgroup *memcg; | |
3453 | struct kmem_cache *cachep; | |
3454 | struct work_struct work; | |
3455 | }; | |
3456 | ||
1f458cbf GC |
3457 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
3458 | { | |
3459 | struct kmem_cache *cachep; | |
3460 | struct memcg_cache_params *params; | |
3461 | ||
3462 | if (!memcg_kmem_is_active(memcg)) | |
3463 | return; | |
3464 | ||
3465 | mutex_lock(&memcg->slab_caches_mutex); | |
3466 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) { | |
3467 | cachep = memcg_params_to_cache(params); | |
3468 | cachep->memcg_params->dead = true; | |
1f458cbf GC |
3469 | schedule_work(&cachep->memcg_params->destroy); |
3470 | } | |
3471 | mutex_unlock(&memcg->slab_caches_mutex); | |
3472 | } | |
3473 | ||
d7f25f8a GC |
3474 | static void memcg_create_cache_work_func(struct work_struct *w) |
3475 | { | |
3476 | struct create_work *cw; | |
3477 | ||
3478 | cw = container_of(w, struct create_work, work); | |
3479 | memcg_create_kmem_cache(cw->memcg, cw->cachep); | |
3480 | /* Drop the reference gotten when we enqueued. */ | |
3481 | css_put(&cw->memcg->css); | |
3482 | kfree(cw); | |
3483 | } | |
3484 | ||
3485 | /* | |
3486 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 3487 | */ |
0e9d92f2 GC |
3488 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3489 | struct kmem_cache *cachep) | |
d7f25f8a GC |
3490 | { |
3491 | struct create_work *cw; | |
3492 | ||
3493 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); | |
ca0dde97 LZ |
3494 | if (cw == NULL) { |
3495 | css_put(&memcg->css); | |
d7f25f8a GC |
3496 | return; |
3497 | } | |
3498 | ||
3499 | cw->memcg = memcg; | |
3500 | cw->cachep = cachep; | |
3501 | ||
3502 | INIT_WORK(&cw->work, memcg_create_cache_work_func); | |
3503 | schedule_work(&cw->work); | |
3504 | } | |
3505 | ||
0e9d92f2 GC |
3506 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
3507 | struct kmem_cache *cachep) | |
3508 | { | |
3509 | /* | |
3510 | * We need to stop accounting when we kmalloc, because if the | |
3511 | * corresponding kmalloc cache is not yet created, the first allocation | |
3512 | * in __memcg_create_cache_enqueue will recurse. | |
3513 | * | |
3514 | * However, it is better to enclose the whole function. Depending on | |
3515 | * the debugging options enabled, INIT_WORK(), for instance, can | |
3516 | * trigger an allocation. This too, will make us recurse. Because at | |
3517 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
3518 | * the safest choice is to do it like this, wrapping the whole function. | |
3519 | */ | |
3520 | memcg_stop_kmem_account(); | |
3521 | __memcg_create_cache_enqueue(memcg, cachep); | |
3522 | memcg_resume_kmem_account(); | |
3523 | } | |
d7f25f8a GC |
3524 | /* |
3525 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
3526 | * We try to use the current memcg's version of the cache. | |
3527 | * | |
3528 | * If the cache does not exist yet, if we are the first user of it, | |
3529 | * we either create it immediately, if possible, or create it asynchronously | |
3530 | * in a workqueue. | |
3531 | * In the latter case, we will let the current allocation go through with | |
3532 | * the original cache. | |
3533 | * | |
3534 | * Can't be called in interrupt context or from kernel threads. | |
3535 | * This function needs to be called with rcu_read_lock() held. | |
3536 | */ | |
3537 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, | |
3538 | gfp_t gfp) | |
3539 | { | |
3540 | struct mem_cgroup *memcg; | |
3541 | int idx; | |
3542 | ||
3543 | VM_BUG_ON(!cachep->memcg_params); | |
3544 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); | |
3545 | ||
0e9d92f2 GC |
3546 | if (!current->mm || current->memcg_kmem_skip_account) |
3547 | return cachep; | |
3548 | ||
d7f25f8a GC |
3549 | rcu_read_lock(); |
3550 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); | |
d7f25f8a GC |
3551 | |
3552 | if (!memcg_can_account_kmem(memcg)) | |
ca0dde97 | 3553 | goto out; |
d7f25f8a GC |
3554 | |
3555 | idx = memcg_cache_id(memcg); | |
3556 | ||
3557 | /* | |
3558 | * barrier to mare sure we're always seeing the up to date value. The | |
3559 | * code updating memcg_caches will issue a write barrier to match this. | |
3560 | */ | |
3561 | read_barrier_depends(); | |
ca0dde97 LZ |
3562 | if (likely(cachep->memcg_params->memcg_caches[idx])) { |
3563 | cachep = cachep->memcg_params->memcg_caches[idx]; | |
3564 | goto out; | |
d7f25f8a GC |
3565 | } |
3566 | ||
ca0dde97 LZ |
3567 | /* The corresponding put will be done in the workqueue. */ |
3568 | if (!css_tryget(&memcg->css)) | |
3569 | goto out; | |
3570 | rcu_read_unlock(); | |
3571 | ||
3572 | /* | |
3573 | * If we are in a safe context (can wait, and not in interrupt | |
3574 | * context), we could be be predictable and return right away. | |
3575 | * This would guarantee that the allocation being performed | |
3576 | * already belongs in the new cache. | |
3577 | * | |
3578 | * However, there are some clashes that can arrive from locking. | |
3579 | * For instance, because we acquire the slab_mutex while doing | |
3580 | * kmem_cache_dup, this means no further allocation could happen | |
3581 | * with the slab_mutex held. | |
3582 | * | |
3583 | * Also, because cache creation issue get_online_cpus(), this | |
3584 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, | |
3585 | * that ends up reversed during cpu hotplug. (cpuset allocates | |
3586 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, | |
3587 | * better to defer everything. | |
3588 | */ | |
3589 | memcg_create_cache_enqueue(memcg, cachep); | |
3590 | return cachep; | |
3591 | out: | |
3592 | rcu_read_unlock(); | |
3593 | return cachep; | |
d7f25f8a GC |
3594 | } |
3595 | EXPORT_SYMBOL(__memcg_kmem_get_cache); | |
3596 | ||
7ae1e1d0 GC |
3597 | /* |
3598 | * We need to verify if the allocation against current->mm->owner's memcg is | |
3599 | * possible for the given order. But the page is not allocated yet, so we'll | |
3600 | * need a further commit step to do the final arrangements. | |
3601 | * | |
3602 | * It is possible for the task to switch cgroups in this mean time, so at | |
3603 | * commit time, we can't rely on task conversion any longer. We'll then use | |
3604 | * the handle argument to return to the caller which cgroup we should commit | |
3605 | * against. We could also return the memcg directly and avoid the pointer | |
3606 | * passing, but a boolean return value gives better semantics considering | |
3607 | * the compiled-out case as well. | |
3608 | * | |
3609 | * Returning true means the allocation is possible. | |
3610 | */ | |
3611 | bool | |
3612 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) | |
3613 | { | |
3614 | struct mem_cgroup *memcg; | |
3615 | int ret; | |
3616 | ||
3617 | *_memcg = NULL; | |
3618 | memcg = try_get_mem_cgroup_from_mm(current->mm); | |
3619 | ||
3620 | /* | |
3621 | * very rare case described in mem_cgroup_from_task. Unfortunately there | |
3622 | * isn't much we can do without complicating this too much, and it would | |
3623 | * be gfp-dependent anyway. Just let it go | |
3624 | */ | |
3625 | if (unlikely(!memcg)) | |
3626 | return true; | |
3627 | ||
3628 | if (!memcg_can_account_kmem(memcg)) { | |
3629 | css_put(&memcg->css); | |
3630 | return true; | |
3631 | } | |
3632 | ||
7ae1e1d0 GC |
3633 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
3634 | if (!ret) | |
3635 | *_memcg = memcg; | |
7ae1e1d0 GC |
3636 | |
3637 | css_put(&memcg->css); | |
3638 | return (ret == 0); | |
3639 | } | |
3640 | ||
3641 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
3642 | int order) | |
3643 | { | |
3644 | struct page_cgroup *pc; | |
3645 | ||
3646 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3647 | ||
3648 | /* The page allocation failed. Revert */ | |
3649 | if (!page) { | |
3650 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 GC |
3651 | return; |
3652 | } | |
3653 | ||
3654 | pc = lookup_page_cgroup(page); | |
3655 | lock_page_cgroup(pc); | |
3656 | pc->mem_cgroup = memcg; | |
3657 | SetPageCgroupUsed(pc); | |
3658 | unlock_page_cgroup(pc); | |
3659 | } | |
3660 | ||
3661 | void __memcg_kmem_uncharge_pages(struct page *page, int order) | |
3662 | { | |
3663 | struct mem_cgroup *memcg = NULL; | |
3664 | struct page_cgroup *pc; | |
3665 | ||
3666 | ||
3667 | pc = lookup_page_cgroup(page); | |
3668 | /* | |
3669 | * Fast unlocked return. Theoretically might have changed, have to | |
3670 | * check again after locking. | |
3671 | */ | |
3672 | if (!PageCgroupUsed(pc)) | |
3673 | return; | |
3674 | ||
3675 | lock_page_cgroup(pc); | |
3676 | if (PageCgroupUsed(pc)) { | |
3677 | memcg = pc->mem_cgroup; | |
3678 | ClearPageCgroupUsed(pc); | |
3679 | } | |
3680 | unlock_page_cgroup(pc); | |
3681 | ||
3682 | /* | |
3683 | * We trust that only if there is a memcg associated with the page, it | |
3684 | * is a valid allocation | |
3685 | */ | |
3686 | if (!memcg) | |
3687 | return; | |
3688 | ||
3689 | VM_BUG_ON(mem_cgroup_is_root(memcg)); | |
3690 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); | |
7ae1e1d0 | 3691 | } |
1f458cbf GC |
3692 | #else |
3693 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) | |
3694 | { | |
3695 | } | |
7ae1e1d0 GC |
3696 | #endif /* CONFIG_MEMCG_KMEM */ |
3697 | ||
ca3e0214 KH |
3698 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3699 | ||
a0db00fc | 3700 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
ca3e0214 KH |
3701 | /* |
3702 | * Because tail pages are not marked as "used", set it. We're under | |
e94c8a9c KH |
3703 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
3704 | * charge/uncharge will be never happen and move_account() is done under | |
3705 | * compound_lock(), so we don't have to take care of races. | |
ca3e0214 | 3706 | */ |
e94c8a9c | 3707 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 KH |
3708 | { |
3709 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
e94c8a9c KH |
3710 | struct page_cgroup *pc; |
3711 | int i; | |
ca3e0214 | 3712 | |
3d37c4a9 KH |
3713 | if (mem_cgroup_disabled()) |
3714 | return; | |
e94c8a9c KH |
3715 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3716 | pc = head_pc + i; | |
3717 | pc->mem_cgroup = head_pc->mem_cgroup; | |
3718 | smp_wmb();/* see __commit_charge() */ | |
e94c8a9c KH |
3719 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
3720 | } | |
ca3e0214 | 3721 | } |
12d27107 | 3722 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3723 | |
f817ed48 | 3724 | /** |
de3638d9 | 3725 | * mem_cgroup_move_account - move account of the page |
5564e88b | 3726 | * @page: the page |
7ec99d62 | 3727 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
3728 | * @pc: page_cgroup of the page. |
3729 | * @from: mem_cgroup which the page is moved from. | |
3730 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
3731 | * | |
3732 | * The caller must confirm following. | |
08e552c6 | 3733 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 3734 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 3735 | * |
2f3479b1 KH |
3736 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
3737 | * from old cgroup. | |
f817ed48 | 3738 | */ |
7ec99d62 JW |
3739 | static int mem_cgroup_move_account(struct page *page, |
3740 | unsigned int nr_pages, | |
3741 | struct page_cgroup *pc, | |
3742 | struct mem_cgroup *from, | |
2f3479b1 | 3743 | struct mem_cgroup *to) |
f817ed48 | 3744 | { |
de3638d9 JW |
3745 | unsigned long flags; |
3746 | int ret; | |
b2402857 | 3747 | bool anon = PageAnon(page); |
987eba66 | 3748 | |
f817ed48 | 3749 | VM_BUG_ON(from == to); |
5564e88b | 3750 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
3751 | /* |
3752 | * The page is isolated from LRU. So, collapse function | |
3753 | * will not handle this page. But page splitting can happen. | |
3754 | * Do this check under compound_page_lock(). The caller should | |
3755 | * hold it. | |
3756 | */ | |
3757 | ret = -EBUSY; | |
7ec99d62 | 3758 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
3759 | goto out; |
3760 | ||
3761 | lock_page_cgroup(pc); | |
3762 | ||
3763 | ret = -EINVAL; | |
3764 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
3765 | goto unlock; | |
3766 | ||
312734c0 | 3767 | move_lock_mem_cgroup(from, &flags); |
f817ed48 | 3768 | |
2ff76f11 | 3769 | if (!anon && page_mapped(page)) { |
c62b1a3b KH |
3770 | /* Update mapped_file data for mem_cgroup */ |
3771 | preempt_disable(); | |
3772 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3773 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
3774 | preempt_enable(); | |
d69b042f | 3775 | } |
b2402857 | 3776 | mem_cgroup_charge_statistics(from, anon, -nr_pages); |
d69b042f | 3777 | |
854ffa8d | 3778 | /* caller should have done css_get */ |
08e552c6 | 3779 | pc->mem_cgroup = to; |
b2402857 | 3780 | mem_cgroup_charge_statistics(to, anon, nr_pages); |
312734c0 | 3781 | move_unlock_mem_cgroup(from, &flags); |
de3638d9 JW |
3782 | ret = 0; |
3783 | unlock: | |
57f9fd7d | 3784 | unlock_page_cgroup(pc); |
d2265e6f KH |
3785 | /* |
3786 | * check events | |
3787 | */ | |
5564e88b JW |
3788 | memcg_check_events(to, page); |
3789 | memcg_check_events(from, page); | |
de3638d9 | 3790 | out: |
f817ed48 KH |
3791 | return ret; |
3792 | } | |
3793 | ||
2ef37d3f MH |
3794 | /** |
3795 | * mem_cgroup_move_parent - moves page to the parent group | |
3796 | * @page: the page to move | |
3797 | * @pc: page_cgroup of the page | |
3798 | * @child: page's cgroup | |
3799 | * | |
3800 | * move charges to its parent or the root cgroup if the group has no | |
3801 | * parent (aka use_hierarchy==0). | |
3802 | * Although this might fail (get_page_unless_zero, isolate_lru_page or | |
3803 | * mem_cgroup_move_account fails) the failure is always temporary and | |
3804 | * it signals a race with a page removal/uncharge or migration. In the | |
3805 | * first case the page is on the way out and it will vanish from the LRU | |
3806 | * on the next attempt and the call should be retried later. | |
3807 | * Isolation from the LRU fails only if page has been isolated from | |
3808 | * the LRU since we looked at it and that usually means either global | |
3809 | * reclaim or migration going on. The page will either get back to the | |
3810 | * LRU or vanish. | |
3811 | * Finaly mem_cgroup_move_account fails only if the page got uncharged | |
3812 | * (!PageCgroupUsed) or moved to a different group. The page will | |
3813 | * disappear in the next attempt. | |
f817ed48 | 3814 | */ |
5564e88b JW |
3815 | static int mem_cgroup_move_parent(struct page *page, |
3816 | struct page_cgroup *pc, | |
6068bf01 | 3817 | struct mem_cgroup *child) |
f817ed48 | 3818 | { |
f817ed48 | 3819 | struct mem_cgroup *parent; |
7ec99d62 | 3820 | unsigned int nr_pages; |
4be4489f | 3821 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
3822 | int ret; |
3823 | ||
d8423011 | 3824 | VM_BUG_ON(mem_cgroup_is_root(child)); |
f817ed48 | 3825 | |
57f9fd7d DN |
3826 | ret = -EBUSY; |
3827 | if (!get_page_unless_zero(page)) | |
3828 | goto out; | |
3829 | if (isolate_lru_page(page)) | |
3830 | goto put; | |
52dbb905 | 3831 | |
7ec99d62 | 3832 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 3833 | |
cc926f78 KH |
3834 | parent = parent_mem_cgroup(child); |
3835 | /* | |
3836 | * If no parent, move charges to root cgroup. | |
3837 | */ | |
3838 | if (!parent) | |
3839 | parent = root_mem_cgroup; | |
f817ed48 | 3840 | |
2ef37d3f MH |
3841 | if (nr_pages > 1) { |
3842 | VM_BUG_ON(!PageTransHuge(page)); | |
987eba66 | 3843 | flags = compound_lock_irqsave(page); |
2ef37d3f | 3844 | } |
987eba66 | 3845 | |
cc926f78 | 3846 | ret = mem_cgroup_move_account(page, nr_pages, |
2f3479b1 | 3847 | pc, child, parent); |
cc926f78 KH |
3848 | if (!ret) |
3849 | __mem_cgroup_cancel_local_charge(child, nr_pages); | |
8dba474f | 3850 | |
7ec99d62 | 3851 | if (nr_pages > 1) |
987eba66 | 3852 | compound_unlock_irqrestore(page, flags); |
08e552c6 | 3853 | putback_lru_page(page); |
57f9fd7d | 3854 | put: |
40d58138 | 3855 | put_page(page); |
57f9fd7d | 3856 | out: |
f817ed48 KH |
3857 | return ret; |
3858 | } | |
3859 | ||
7a81b88c KH |
3860 | /* |
3861 | * Charge the memory controller for page usage. | |
3862 | * Return | |
3863 | * 0 if the charge was successful | |
3864 | * < 0 if the cgroup is over its limit | |
3865 | */ | |
3866 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 3867 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 3868 | { |
c0ff4b85 | 3869 | struct mem_cgroup *memcg = NULL; |
7ec99d62 | 3870 | unsigned int nr_pages = 1; |
8493ae43 | 3871 | bool oom = true; |
7a81b88c | 3872 | int ret; |
ec168510 | 3873 | |
37c2ac78 | 3874 | if (PageTransHuge(page)) { |
7ec99d62 | 3875 | nr_pages <<= compound_order(page); |
37c2ac78 | 3876 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
3877 | /* |
3878 | * Never OOM-kill a process for a huge page. The | |
3879 | * fault handler will fall back to regular pages. | |
3880 | */ | |
3881 | oom = false; | |
37c2ac78 | 3882 | } |
7a81b88c | 3883 | |
c0ff4b85 | 3884 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &memcg, oom); |
38c5d72f | 3885 | if (ret == -ENOMEM) |
7a81b88c | 3886 | return ret; |
ce587e65 | 3887 | __mem_cgroup_commit_charge(memcg, page, nr_pages, ctype, false); |
8a9f3ccd | 3888 | return 0; |
8a9f3ccd BS |
3889 | } |
3890 | ||
7a81b88c KH |
3891 | int mem_cgroup_newpage_charge(struct page *page, |
3892 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 3893 | { |
f8d66542 | 3894 | if (mem_cgroup_disabled()) |
cede86ac | 3895 | return 0; |
7a0524cf JW |
3896 | VM_BUG_ON(page_mapped(page)); |
3897 | VM_BUG_ON(page->mapping && !PageAnon(page)); | |
3898 | VM_BUG_ON(!mm); | |
217bc319 | 3899 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
41326c17 | 3900 | MEM_CGROUP_CHARGE_TYPE_ANON); |
217bc319 KH |
3901 | } |
3902 | ||
54595fe2 KH |
3903 | /* |
3904 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
3905 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 3906 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
3907 | * "commit()" or removed by "cancel()" |
3908 | */ | |
0435a2fd JW |
3909 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
3910 | struct page *page, | |
3911 | gfp_t mask, | |
3912 | struct mem_cgroup **memcgp) | |
8c7c6e34 | 3913 | { |
c0ff4b85 | 3914 | struct mem_cgroup *memcg; |
90deb788 | 3915 | struct page_cgroup *pc; |
54595fe2 | 3916 | int ret; |
8c7c6e34 | 3917 | |
90deb788 JW |
3918 | pc = lookup_page_cgroup(page); |
3919 | /* | |
3920 | * Every swap fault against a single page tries to charge the | |
3921 | * page, bail as early as possible. shmem_unuse() encounters | |
3922 | * already charged pages, too. The USED bit is protected by | |
3923 | * the page lock, which serializes swap cache removal, which | |
3924 | * in turn serializes uncharging. | |
3925 | */ | |
3926 | if (PageCgroupUsed(pc)) | |
3927 | return 0; | |
8c7c6e34 KH |
3928 | if (!do_swap_account) |
3929 | goto charge_cur_mm; | |
c0ff4b85 R |
3930 | memcg = try_get_mem_cgroup_from_page(page); |
3931 | if (!memcg) | |
54595fe2 | 3932 | goto charge_cur_mm; |
72835c86 JW |
3933 | *memcgp = memcg; |
3934 | ret = __mem_cgroup_try_charge(NULL, mask, 1, memcgp, true); | |
c0ff4b85 | 3935 | css_put(&memcg->css); |
38c5d72f KH |
3936 | if (ret == -EINTR) |
3937 | ret = 0; | |
54595fe2 | 3938 | return ret; |
8c7c6e34 | 3939 | charge_cur_mm: |
38c5d72f KH |
3940 | ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true); |
3941 | if (ret == -EINTR) | |
3942 | ret = 0; | |
3943 | return ret; | |
8c7c6e34 KH |
3944 | } |
3945 | ||
0435a2fd JW |
3946 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
3947 | gfp_t gfp_mask, struct mem_cgroup **memcgp) | |
3948 | { | |
3949 | *memcgp = NULL; | |
3950 | if (mem_cgroup_disabled()) | |
3951 | return 0; | |
bdf4f4d2 JW |
3952 | /* |
3953 | * A racing thread's fault, or swapoff, may have already | |
3954 | * updated the pte, and even removed page from swap cache: in | |
3955 | * those cases unuse_pte()'s pte_same() test will fail; but | |
3956 | * there's also a KSM case which does need to charge the page. | |
3957 | */ | |
3958 | if (!PageSwapCache(page)) { | |
3959 | int ret; | |
3960 | ||
3961 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true); | |
3962 | if (ret == -EINTR) | |
3963 | ret = 0; | |
3964 | return ret; | |
3965 | } | |
0435a2fd JW |
3966 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
3967 | } | |
3968 | ||
827a03d2 JW |
3969 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
3970 | { | |
3971 | if (mem_cgroup_disabled()) | |
3972 | return; | |
3973 | if (!memcg) | |
3974 | return; | |
3975 | __mem_cgroup_cancel_charge(memcg, 1); | |
3976 | } | |
3977 | ||
83aae4c7 | 3978 | static void |
72835c86 | 3979 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
83aae4c7 | 3980 | enum charge_type ctype) |
7a81b88c | 3981 | { |
f8d66542 | 3982 | if (mem_cgroup_disabled()) |
7a81b88c | 3983 | return; |
72835c86 | 3984 | if (!memcg) |
7a81b88c | 3985 | return; |
5a6475a4 | 3986 | |
ce587e65 | 3987 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
8c7c6e34 KH |
3988 | /* |
3989 | * Now swap is on-memory. This means this page may be | |
3990 | * counted both as mem and swap....double count. | |
03f3c433 KH |
3991 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
3992 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
3993 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 3994 | */ |
03f3c433 | 3995 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 3996 | swp_entry_t ent = {.val = page_private(page)}; |
86493009 | 3997 | mem_cgroup_uncharge_swap(ent); |
8c7c6e34 | 3998 | } |
7a81b88c KH |
3999 | } |
4000 | ||
72835c86 JW |
4001 | void mem_cgroup_commit_charge_swapin(struct page *page, |
4002 | struct mem_cgroup *memcg) | |
83aae4c7 | 4003 | { |
72835c86 | 4004 | __mem_cgroup_commit_charge_swapin(page, memcg, |
41326c17 | 4005 | MEM_CGROUP_CHARGE_TYPE_ANON); |
83aae4c7 DN |
4006 | } |
4007 | ||
827a03d2 JW |
4008 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
4009 | gfp_t gfp_mask) | |
7a81b88c | 4010 | { |
827a03d2 JW |
4011 | struct mem_cgroup *memcg = NULL; |
4012 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
4013 | int ret; | |
4014 | ||
f8d66542 | 4015 | if (mem_cgroup_disabled()) |
827a03d2 JW |
4016 | return 0; |
4017 | if (PageCompound(page)) | |
4018 | return 0; | |
4019 | ||
827a03d2 JW |
4020 | if (!PageSwapCache(page)) |
4021 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, type); | |
4022 | else { /* page is swapcache/shmem */ | |
0435a2fd JW |
4023 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
4024 | gfp_mask, &memcg); | |
827a03d2 JW |
4025 | if (!ret) |
4026 | __mem_cgroup_commit_charge_swapin(page, memcg, type); | |
4027 | } | |
4028 | return ret; | |
7a81b88c KH |
4029 | } |
4030 | ||
c0ff4b85 | 4031 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
7ec99d62 JW |
4032 | unsigned int nr_pages, |
4033 | const enum charge_type ctype) | |
569b846d KH |
4034 | { |
4035 | struct memcg_batch_info *batch = NULL; | |
4036 | bool uncharge_memsw = true; | |
7ec99d62 | 4037 | |
569b846d KH |
4038 | /* If swapout, usage of swap doesn't decrease */ |
4039 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
4040 | uncharge_memsw = false; | |
569b846d KH |
4041 | |
4042 | batch = ¤t->memcg_batch; | |
4043 | /* | |
4044 | * In usual, we do css_get() when we remember memcg pointer. | |
4045 | * But in this case, we keep res->usage until end of a series of | |
4046 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
4047 | */ | |
4048 | if (!batch->memcg) | |
c0ff4b85 | 4049 | batch->memcg = memcg; |
3c11ecf4 KH |
4050 | /* |
4051 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 4052 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
4053 | * the same cgroup and we have chance to coalesce uncharges. |
4054 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
4055 | * because we want to do uncharge as soon as possible. | |
4056 | */ | |
4057 | ||
4058 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
4059 | goto direct_uncharge; | |
4060 | ||
7ec99d62 | 4061 | if (nr_pages > 1) |
ec168510 AA |
4062 | goto direct_uncharge; |
4063 | ||
569b846d KH |
4064 | /* |
4065 | * In typical case, batch->memcg == mem. This means we can | |
4066 | * merge a series of uncharges to an uncharge of res_counter. | |
4067 | * If not, we uncharge res_counter ony by one. | |
4068 | */ | |
c0ff4b85 | 4069 | if (batch->memcg != memcg) |
569b846d KH |
4070 | goto direct_uncharge; |
4071 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 4072 | batch->nr_pages++; |
569b846d | 4073 | if (uncharge_memsw) |
7ffd4ca7 | 4074 | batch->memsw_nr_pages++; |
569b846d KH |
4075 | return; |
4076 | direct_uncharge: | |
c0ff4b85 | 4077 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
569b846d | 4078 | if (uncharge_memsw) |
c0ff4b85 R |
4079 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
4080 | if (unlikely(batch->memcg != memcg)) | |
4081 | memcg_oom_recover(memcg); | |
569b846d | 4082 | } |
7a81b88c | 4083 | |
8a9f3ccd | 4084 | /* |
69029cd5 | 4085 | * uncharge if !page_mapped(page) |
8a9f3ccd | 4086 | */ |
8c7c6e34 | 4087 | static struct mem_cgroup * |
0030f535 JW |
4088 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
4089 | bool end_migration) | |
8a9f3ccd | 4090 | { |
c0ff4b85 | 4091 | struct mem_cgroup *memcg = NULL; |
7ec99d62 JW |
4092 | unsigned int nr_pages = 1; |
4093 | struct page_cgroup *pc; | |
b2402857 | 4094 | bool anon; |
8a9f3ccd | 4095 | |
f8d66542 | 4096 | if (mem_cgroup_disabled()) |
8c7c6e34 | 4097 | return NULL; |
4077960e | 4098 | |
0c59b89c | 4099 | VM_BUG_ON(PageSwapCache(page)); |
d13d1443 | 4100 | |
37c2ac78 | 4101 | if (PageTransHuge(page)) { |
7ec99d62 | 4102 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
4103 | VM_BUG_ON(!PageTransHuge(page)); |
4104 | } | |
8697d331 | 4105 | /* |
3c541e14 | 4106 | * Check if our page_cgroup is valid |
8697d331 | 4107 | */ |
52d4b9ac | 4108 | pc = lookup_page_cgroup(page); |
cfa44946 | 4109 | if (unlikely(!PageCgroupUsed(pc))) |
8c7c6e34 | 4110 | return NULL; |
b9c565d5 | 4111 | |
52d4b9ac | 4112 | lock_page_cgroup(pc); |
d13d1443 | 4113 | |
c0ff4b85 | 4114 | memcg = pc->mem_cgroup; |
8c7c6e34 | 4115 | |
d13d1443 KH |
4116 | if (!PageCgroupUsed(pc)) |
4117 | goto unlock_out; | |
4118 | ||
b2402857 KH |
4119 | anon = PageAnon(page); |
4120 | ||
d13d1443 | 4121 | switch (ctype) { |
41326c17 | 4122 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
2ff76f11 KH |
4123 | /* |
4124 | * Generally PageAnon tells if it's the anon statistics to be | |
4125 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is | |
4126 | * used before page reached the stage of being marked PageAnon. | |
4127 | */ | |
b2402857 KH |
4128 | anon = true; |
4129 | /* fallthrough */ | |
8a9478ca | 4130 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 4131 | /* See mem_cgroup_prepare_migration() */ |
0030f535 JW |
4132 | if (page_mapped(page)) |
4133 | goto unlock_out; | |
4134 | /* | |
4135 | * Pages under migration may not be uncharged. But | |
4136 | * end_migration() /must/ be the one uncharging the | |
4137 | * unused post-migration page and so it has to call | |
4138 | * here with the migration bit still set. See the | |
4139 | * res_counter handling below. | |
4140 | */ | |
4141 | if (!end_migration && PageCgroupMigration(pc)) | |
d13d1443 KH |
4142 | goto unlock_out; |
4143 | break; | |
4144 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
4145 | if (!PageAnon(page)) { /* Shared memory */ | |
4146 | if (page->mapping && !page_is_file_cache(page)) | |
4147 | goto unlock_out; | |
4148 | } else if (page_mapped(page)) /* Anon */ | |
4149 | goto unlock_out; | |
4150 | break; | |
4151 | default: | |
4152 | break; | |
52d4b9ac | 4153 | } |
d13d1443 | 4154 | |
b2402857 | 4155 | mem_cgroup_charge_statistics(memcg, anon, -nr_pages); |
04046e1a | 4156 | |
52d4b9ac | 4157 | ClearPageCgroupUsed(pc); |
544122e5 KH |
4158 | /* |
4159 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
4160 | * freed from LRU. This is safe because uncharged page is expected not | |
4161 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
4162 | * special functions. | |
4163 | */ | |
b9c565d5 | 4164 | |
52d4b9ac | 4165 | unlock_page_cgroup(pc); |
f75ca962 | 4166 | /* |
c0ff4b85 | 4167 | * even after unlock, we have memcg->res.usage here and this memcg |
f75ca962 KH |
4168 | * will never be freed. |
4169 | */ | |
c0ff4b85 | 4170 | memcg_check_events(memcg, page); |
f75ca962 | 4171 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
c0ff4b85 R |
4172 | mem_cgroup_swap_statistics(memcg, true); |
4173 | mem_cgroup_get(memcg); | |
f75ca962 | 4174 | } |
0030f535 JW |
4175 | /* |
4176 | * Migration does not charge the res_counter for the | |
4177 | * replacement page, so leave it alone when phasing out the | |
4178 | * page that is unused after the migration. | |
4179 | */ | |
4180 | if (!end_migration && !mem_cgroup_is_root(memcg)) | |
c0ff4b85 | 4181 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
6d12e2d8 | 4182 | |
c0ff4b85 | 4183 | return memcg; |
d13d1443 KH |
4184 | |
4185 | unlock_out: | |
4186 | unlock_page_cgroup(pc); | |
8c7c6e34 | 4187 | return NULL; |
3c541e14 BS |
4188 | } |
4189 | ||
69029cd5 KH |
4190 | void mem_cgroup_uncharge_page(struct page *page) |
4191 | { | |
52d4b9ac KH |
4192 | /* early check. */ |
4193 | if (page_mapped(page)) | |
4194 | return; | |
40f23a21 | 4195 | VM_BUG_ON(page->mapping && !PageAnon(page)); |
0c59b89c JW |
4196 | if (PageSwapCache(page)) |
4197 | return; | |
0030f535 | 4198 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
69029cd5 KH |
4199 | } |
4200 | ||
4201 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
4202 | { | |
4203 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 4204 | VM_BUG_ON(page->mapping); |
0030f535 | 4205 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
69029cd5 KH |
4206 | } |
4207 | ||
569b846d KH |
4208 | /* |
4209 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
4210 | * In that cases, pages are freed continuously and we can expect pages | |
4211 | * are in the same memcg. All these calls itself limits the number of | |
4212 | * pages freed at once, then uncharge_start/end() is called properly. | |
4213 | * This may be called prural(2) times in a context, | |
4214 | */ | |
4215 | ||
4216 | void mem_cgroup_uncharge_start(void) | |
4217 | { | |
4218 | current->memcg_batch.do_batch++; | |
4219 | /* We can do nest. */ | |
4220 | if (current->memcg_batch.do_batch == 1) { | |
4221 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
4222 | current->memcg_batch.nr_pages = 0; |
4223 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
4224 | } |
4225 | } | |
4226 | ||
4227 | void mem_cgroup_uncharge_end(void) | |
4228 | { | |
4229 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
4230 | ||
4231 | if (!batch->do_batch) | |
4232 | return; | |
4233 | ||
4234 | batch->do_batch--; | |
4235 | if (batch->do_batch) /* If stacked, do nothing. */ | |
4236 | return; | |
4237 | ||
4238 | if (!batch->memcg) | |
4239 | return; | |
4240 | /* | |
4241 | * This "batch->memcg" is valid without any css_get/put etc... | |
4242 | * bacause we hide charges behind us. | |
4243 | */ | |
7ffd4ca7 JW |
4244 | if (batch->nr_pages) |
4245 | res_counter_uncharge(&batch->memcg->res, | |
4246 | batch->nr_pages * PAGE_SIZE); | |
4247 | if (batch->memsw_nr_pages) | |
4248 | res_counter_uncharge(&batch->memcg->memsw, | |
4249 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 4250 | memcg_oom_recover(batch->memcg); |
569b846d KH |
4251 | /* forget this pointer (for sanity check) */ |
4252 | batch->memcg = NULL; | |
4253 | } | |
4254 | ||
e767e056 | 4255 | #ifdef CONFIG_SWAP |
8c7c6e34 | 4256 | /* |
e767e056 | 4257 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
4258 | * memcg information is recorded to swap_cgroup of "ent" |
4259 | */ | |
8a9478ca KH |
4260 | void |
4261 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
4262 | { |
4263 | struct mem_cgroup *memcg; | |
8a9478ca KH |
4264 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
4265 | ||
4266 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
4267 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
4268 | ||
0030f535 | 4269 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
8c7c6e34 | 4270 | |
f75ca962 KH |
4271 | /* |
4272 | * record memcg information, if swapout && memcg != NULL, | |
4273 | * mem_cgroup_get() was called in uncharge(). | |
4274 | */ | |
4275 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 4276 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 4277 | } |
e767e056 | 4278 | #endif |
8c7c6e34 | 4279 | |
c255a458 | 4280 | #ifdef CONFIG_MEMCG_SWAP |
8c7c6e34 KH |
4281 | /* |
4282 | * called from swap_entry_free(). remove record in swap_cgroup and | |
4283 | * uncharge "memsw" account. | |
4284 | */ | |
4285 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 4286 | { |
8c7c6e34 | 4287 | struct mem_cgroup *memcg; |
a3b2d692 | 4288 | unsigned short id; |
8c7c6e34 KH |
4289 | |
4290 | if (!do_swap_account) | |
4291 | return; | |
4292 | ||
a3b2d692 KH |
4293 | id = swap_cgroup_record(ent, 0); |
4294 | rcu_read_lock(); | |
4295 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 4296 | if (memcg) { |
a3b2d692 KH |
4297 | /* |
4298 | * We uncharge this because swap is freed. | |
4299 | * This memcg can be obsolete one. We avoid calling css_tryget | |
4300 | */ | |
0c3e73e8 | 4301 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 4302 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 4303 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
4304 | mem_cgroup_put(memcg); |
4305 | } | |
a3b2d692 | 4306 | rcu_read_unlock(); |
d13d1443 | 4307 | } |
02491447 DN |
4308 | |
4309 | /** | |
4310 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
4311 | * @entry: swap entry to be moved | |
4312 | * @from: mem_cgroup which the entry is moved from | |
4313 | * @to: mem_cgroup which the entry is moved to | |
4314 | * | |
4315 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
4316 | * as the mem_cgroup's id of @from. | |
4317 | * | |
4318 | * Returns 0 on success, -EINVAL on failure. | |
4319 | * | |
4320 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
4321 | * both res and memsw, and called css_get(). | |
4322 | */ | |
4323 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4324 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4325 | { |
4326 | unsigned short old_id, new_id; | |
4327 | ||
4328 | old_id = css_id(&from->css); | |
4329 | new_id = css_id(&to->css); | |
4330 | ||
4331 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 4332 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 4333 | mem_cgroup_swap_statistics(to, true); |
02491447 | 4334 | /* |
483c30b5 DN |
4335 | * This function is only called from task migration context now. |
4336 | * It postpones res_counter and refcount handling till the end | |
4337 | * of task migration(mem_cgroup_clear_mc()) for performance | |
4338 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
4339 | * because if the process that has been moved to @to does | |
4340 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 4341 | */ |
02491447 | 4342 | mem_cgroup_get(to); |
02491447 DN |
4343 | return 0; |
4344 | } | |
4345 | return -EINVAL; | |
4346 | } | |
4347 | #else | |
4348 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 4349 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
4350 | { |
4351 | return -EINVAL; | |
4352 | } | |
8c7c6e34 | 4353 | #endif |
d13d1443 | 4354 | |
ae41be37 | 4355 | /* |
01b1ae63 KH |
4356 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
4357 | * page belongs to. | |
ae41be37 | 4358 | */ |
0030f535 JW |
4359 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
4360 | struct mem_cgroup **memcgp) | |
ae41be37 | 4361 | { |
c0ff4b85 | 4362 | struct mem_cgroup *memcg = NULL; |
b32967ff | 4363 | unsigned int nr_pages = 1; |
7ec99d62 | 4364 | struct page_cgroup *pc; |
ac39cf8c | 4365 | enum charge_type ctype; |
8869b8f6 | 4366 | |
72835c86 | 4367 | *memcgp = NULL; |
56039efa | 4368 | |
f8d66542 | 4369 | if (mem_cgroup_disabled()) |
0030f535 | 4370 | return; |
4077960e | 4371 | |
b32967ff MG |
4372 | if (PageTransHuge(page)) |
4373 | nr_pages <<= compound_order(page); | |
4374 | ||
52d4b9ac KH |
4375 | pc = lookup_page_cgroup(page); |
4376 | lock_page_cgroup(pc); | |
4377 | if (PageCgroupUsed(pc)) { | |
c0ff4b85 R |
4378 | memcg = pc->mem_cgroup; |
4379 | css_get(&memcg->css); | |
ac39cf8c | 4380 | /* |
4381 | * At migrating an anonymous page, its mapcount goes down | |
4382 | * to 0 and uncharge() will be called. But, even if it's fully | |
4383 | * unmapped, migration may fail and this page has to be | |
4384 | * charged again. We set MIGRATION flag here and delay uncharge | |
4385 | * until end_migration() is called | |
4386 | * | |
4387 | * Corner Case Thinking | |
4388 | * A) | |
4389 | * When the old page was mapped as Anon and it's unmap-and-freed | |
4390 | * while migration was ongoing. | |
4391 | * If unmap finds the old page, uncharge() of it will be delayed | |
4392 | * until end_migration(). If unmap finds a new page, it's | |
4393 | * uncharged when it make mapcount to be 1->0. If unmap code | |
4394 | * finds swap_migration_entry, the new page will not be mapped | |
4395 | * and end_migration() will find it(mapcount==0). | |
4396 | * | |
4397 | * B) | |
4398 | * When the old page was mapped but migraion fails, the kernel | |
4399 | * remaps it. A charge for it is kept by MIGRATION flag even | |
4400 | * if mapcount goes down to 0. We can do remap successfully | |
4401 | * without charging it again. | |
4402 | * | |
4403 | * C) | |
4404 | * The "old" page is under lock_page() until the end of | |
4405 | * migration, so, the old page itself will not be swapped-out. | |
4406 | * If the new page is swapped out before end_migraton, our | |
4407 | * hook to usual swap-out path will catch the event. | |
4408 | */ | |
4409 | if (PageAnon(page)) | |
4410 | SetPageCgroupMigration(pc); | |
e8589cc1 | 4411 | } |
52d4b9ac | 4412 | unlock_page_cgroup(pc); |
ac39cf8c | 4413 | /* |
4414 | * If the page is not charged at this point, | |
4415 | * we return here. | |
4416 | */ | |
c0ff4b85 | 4417 | if (!memcg) |
0030f535 | 4418 | return; |
01b1ae63 | 4419 | |
72835c86 | 4420 | *memcgp = memcg; |
ac39cf8c | 4421 | /* |
4422 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
4423 | * is called before end_migration, we can catch all events on this new | |
4424 | * page. In the case new page is migrated but not remapped, new page's | |
4425 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
4426 | */ | |
ac39cf8c | 4427 | if (PageAnon(page)) |
41326c17 | 4428 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
ac39cf8c | 4429 | else |
62ba7442 | 4430 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
0030f535 JW |
4431 | /* |
4432 | * The page is committed to the memcg, but it's not actually | |
4433 | * charged to the res_counter since we plan on replacing the | |
4434 | * old one and only one page is going to be left afterwards. | |
4435 | */ | |
b32967ff | 4436 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
ae41be37 | 4437 | } |
8869b8f6 | 4438 | |
69029cd5 | 4439 | /* remove redundant charge if migration failed*/ |
c0ff4b85 | 4440 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
50de1dd9 | 4441 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 4442 | { |
ac39cf8c | 4443 | struct page *used, *unused; |
01b1ae63 | 4444 | struct page_cgroup *pc; |
b2402857 | 4445 | bool anon; |
01b1ae63 | 4446 | |
c0ff4b85 | 4447 | if (!memcg) |
01b1ae63 | 4448 | return; |
b25ed609 | 4449 | |
50de1dd9 | 4450 | if (!migration_ok) { |
ac39cf8c | 4451 | used = oldpage; |
4452 | unused = newpage; | |
01b1ae63 | 4453 | } else { |
ac39cf8c | 4454 | used = newpage; |
01b1ae63 KH |
4455 | unused = oldpage; |
4456 | } | |
0030f535 | 4457 | anon = PageAnon(used); |
7d188958 JW |
4458 | __mem_cgroup_uncharge_common(unused, |
4459 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON | |
4460 | : MEM_CGROUP_CHARGE_TYPE_CACHE, | |
4461 | true); | |
0030f535 | 4462 | css_put(&memcg->css); |
69029cd5 | 4463 | /* |
ac39cf8c | 4464 | * We disallowed uncharge of pages under migration because mapcount |
4465 | * of the page goes down to zero, temporarly. | |
4466 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 4467 | */ |
ac39cf8c | 4468 | pc = lookup_page_cgroup(oldpage); |
4469 | lock_page_cgroup(pc); | |
4470 | ClearPageCgroupMigration(pc); | |
4471 | unlock_page_cgroup(pc); | |
ac39cf8c | 4472 | |
01b1ae63 | 4473 | /* |
ac39cf8c | 4474 | * If a page is a file cache, radix-tree replacement is very atomic |
4475 | * and we can skip this check. When it was an Anon page, its mapcount | |
4476 | * goes down to 0. But because we added MIGRATION flage, it's not | |
4477 | * uncharged yet. There are several case but page->mapcount check | |
4478 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
4479 | * check. (see prepare_charge() also) | |
69029cd5 | 4480 | */ |
b2402857 | 4481 | if (anon) |
ac39cf8c | 4482 | mem_cgroup_uncharge_page(used); |
ae41be37 | 4483 | } |
78fb7466 | 4484 | |
ab936cbc KH |
4485 | /* |
4486 | * At replace page cache, newpage is not under any memcg but it's on | |
4487 | * LRU. So, this function doesn't touch res_counter but handles LRU | |
4488 | * in correct way. Both pages are locked so we cannot race with uncharge. | |
4489 | */ | |
4490 | void mem_cgroup_replace_page_cache(struct page *oldpage, | |
4491 | struct page *newpage) | |
4492 | { | |
bde05d1c | 4493 | struct mem_cgroup *memcg = NULL; |
ab936cbc | 4494 | struct page_cgroup *pc; |
ab936cbc | 4495 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
ab936cbc KH |
4496 | |
4497 | if (mem_cgroup_disabled()) | |
4498 | return; | |
4499 | ||
4500 | pc = lookup_page_cgroup(oldpage); | |
4501 | /* fix accounting on old pages */ | |
4502 | lock_page_cgroup(pc); | |
bde05d1c HD |
4503 | if (PageCgroupUsed(pc)) { |
4504 | memcg = pc->mem_cgroup; | |
4505 | mem_cgroup_charge_statistics(memcg, false, -1); | |
4506 | ClearPageCgroupUsed(pc); | |
4507 | } | |
ab936cbc KH |
4508 | unlock_page_cgroup(pc); |
4509 | ||
bde05d1c HD |
4510 | /* |
4511 | * When called from shmem_replace_page(), in some cases the | |
4512 | * oldpage has already been charged, and in some cases not. | |
4513 | */ | |
4514 | if (!memcg) | |
4515 | return; | |
ab936cbc KH |
4516 | /* |
4517 | * Even if newpage->mapping was NULL before starting replacement, | |
4518 | * the newpage may be on LRU(or pagevec for LRU) already. We lock | |
4519 | * LRU while we overwrite pc->mem_cgroup. | |
4520 | */ | |
ce587e65 | 4521 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
ab936cbc KH |
4522 | } |
4523 | ||
f212ad7c DN |
4524 | #ifdef CONFIG_DEBUG_VM |
4525 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
4526 | { | |
4527 | struct page_cgroup *pc; | |
4528 | ||
4529 | pc = lookup_page_cgroup(page); | |
cfa44946 JW |
4530 | /* |
4531 | * Can be NULL while feeding pages into the page allocator for | |
4532 | * the first time, i.e. during boot or memory hotplug; | |
4533 | * or when mem_cgroup_disabled(). | |
4534 | */ | |
f212ad7c DN |
4535 | if (likely(pc) && PageCgroupUsed(pc)) |
4536 | return pc; | |
4537 | return NULL; | |
4538 | } | |
4539 | ||
4540 | bool mem_cgroup_bad_page_check(struct page *page) | |
4541 | { | |
4542 | if (mem_cgroup_disabled()) | |
4543 | return false; | |
4544 | ||
4545 | return lookup_page_cgroup_used(page) != NULL; | |
4546 | } | |
4547 | ||
4548 | void mem_cgroup_print_bad_page(struct page *page) | |
4549 | { | |
4550 | struct page_cgroup *pc; | |
4551 | ||
4552 | pc = lookup_page_cgroup_used(page); | |
4553 | if (pc) { | |
d045197f AM |
4554 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
4555 | pc, pc->flags, pc->mem_cgroup); | |
f212ad7c DN |
4556 | } |
4557 | } | |
4558 | #endif | |
4559 | ||
d38d2a75 | 4560 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 4561 | unsigned long long val) |
628f4235 | 4562 | { |
81d39c20 | 4563 | int retry_count; |
3c11ecf4 | 4564 | u64 memswlimit, memlimit; |
628f4235 | 4565 | int ret = 0; |
81d39c20 KH |
4566 | int children = mem_cgroup_count_children(memcg); |
4567 | u64 curusage, oldusage; | |
3c11ecf4 | 4568 | int enlarge; |
81d39c20 KH |
4569 | |
4570 | /* | |
4571 | * For keeping hierarchical_reclaim simple, how long we should retry | |
4572 | * is depends on callers. We set our retry-count to be function | |
4573 | * of # of children which we should visit in this loop. | |
4574 | */ | |
4575 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
4576 | ||
4577 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 4578 | |
3c11ecf4 | 4579 | enlarge = 0; |
8c7c6e34 | 4580 | while (retry_count) { |
628f4235 KH |
4581 | if (signal_pending(current)) { |
4582 | ret = -EINTR; | |
4583 | break; | |
4584 | } | |
8c7c6e34 KH |
4585 | /* |
4586 | * Rather than hide all in some function, I do this in | |
4587 | * open coded manner. You see what this really does. | |
aaad153e | 4588 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4589 | */ |
4590 | mutex_lock(&set_limit_mutex); | |
4591 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
4592 | if (memswlimit < val) { | |
4593 | ret = -EINVAL; | |
4594 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
4595 | break; |
4596 | } | |
3c11ecf4 KH |
4597 | |
4598 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4599 | if (memlimit < val) | |
4600 | enlarge = 1; | |
4601 | ||
8c7c6e34 | 4602 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
4603 | if (!ret) { |
4604 | if (memswlimit == val) | |
4605 | memcg->memsw_is_minimum = true; | |
4606 | else | |
4607 | memcg->memsw_is_minimum = false; | |
4608 | } | |
8c7c6e34 KH |
4609 | mutex_unlock(&set_limit_mutex); |
4610 | ||
4611 | if (!ret) | |
4612 | break; | |
4613 | ||
5660048c JW |
4614 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4615 | MEM_CGROUP_RECLAIM_SHRINK); | |
81d39c20 KH |
4616 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
4617 | /* Usage is reduced ? */ | |
4618 | if (curusage >= oldusage) | |
4619 | retry_count--; | |
4620 | else | |
4621 | oldusage = curusage; | |
8c7c6e34 | 4622 | } |
3c11ecf4 KH |
4623 | if (!ret && enlarge) |
4624 | memcg_oom_recover(memcg); | |
14797e23 | 4625 | |
8c7c6e34 KH |
4626 | return ret; |
4627 | } | |
4628 | ||
338c8431 LZ |
4629 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
4630 | unsigned long long val) | |
8c7c6e34 | 4631 | { |
81d39c20 | 4632 | int retry_count; |
3c11ecf4 | 4633 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
4634 | int children = mem_cgroup_count_children(memcg); |
4635 | int ret = -EBUSY; | |
3c11ecf4 | 4636 | int enlarge = 0; |
8c7c6e34 | 4637 | |
81d39c20 KH |
4638 | /* see mem_cgroup_resize_res_limit */ |
4639 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
4640 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
4641 | while (retry_count) { |
4642 | if (signal_pending(current)) { | |
4643 | ret = -EINTR; | |
4644 | break; | |
4645 | } | |
4646 | /* | |
4647 | * Rather than hide all in some function, I do this in | |
4648 | * open coded manner. You see what this really does. | |
aaad153e | 4649 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
8c7c6e34 KH |
4650 | */ |
4651 | mutex_lock(&set_limit_mutex); | |
4652 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
4653 | if (memlimit > val) { | |
4654 | ret = -EINVAL; | |
4655 | mutex_unlock(&set_limit_mutex); | |
4656 | break; | |
4657 | } | |
3c11ecf4 KH |
4658 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
4659 | if (memswlimit < val) | |
4660 | enlarge = 1; | |
8c7c6e34 | 4661 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
4662 | if (!ret) { |
4663 | if (memlimit == val) | |
4664 | memcg->memsw_is_minimum = true; | |
4665 | else | |
4666 | memcg->memsw_is_minimum = false; | |
4667 | } | |
8c7c6e34 KH |
4668 | mutex_unlock(&set_limit_mutex); |
4669 | ||
4670 | if (!ret) | |
4671 | break; | |
4672 | ||
5660048c JW |
4673 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
4674 | MEM_CGROUP_RECLAIM_NOSWAP | | |
4675 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 4676 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 4677 | /* Usage is reduced ? */ |
8c7c6e34 | 4678 | if (curusage >= oldusage) |
628f4235 | 4679 | retry_count--; |
81d39c20 KH |
4680 | else |
4681 | oldusage = curusage; | |
628f4235 | 4682 | } |
3c11ecf4 KH |
4683 | if (!ret && enlarge) |
4684 | memcg_oom_recover(memcg); | |
628f4235 KH |
4685 | return ret; |
4686 | } | |
4687 | ||
4e416953 | 4688 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
4689 | gfp_t gfp_mask, |
4690 | unsigned long *total_scanned) | |
4e416953 BS |
4691 | { |
4692 | unsigned long nr_reclaimed = 0; | |
4693 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
4694 | unsigned long reclaimed; | |
4695 | int loop = 0; | |
4696 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 4697 | unsigned long long excess; |
0ae5e89c | 4698 | unsigned long nr_scanned; |
4e416953 BS |
4699 | |
4700 | if (order > 0) | |
4701 | return 0; | |
4702 | ||
00918b6a | 4703 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
4704 | /* |
4705 | * This loop can run a while, specially if mem_cgroup's continuously | |
4706 | * keep exceeding their soft limit and putting the system under | |
4707 | * pressure | |
4708 | */ | |
4709 | do { | |
4710 | if (next_mz) | |
4711 | mz = next_mz; | |
4712 | else | |
4713 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
4714 | if (!mz) | |
4715 | break; | |
4716 | ||
0ae5e89c | 4717 | nr_scanned = 0; |
d79154bb | 4718 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
5660048c | 4719 | gfp_mask, &nr_scanned); |
4e416953 | 4720 | nr_reclaimed += reclaimed; |
0ae5e89c | 4721 | *total_scanned += nr_scanned; |
4e416953 BS |
4722 | spin_lock(&mctz->lock); |
4723 | ||
4724 | /* | |
4725 | * If we failed to reclaim anything from this memory cgroup | |
4726 | * it is time to move on to the next cgroup | |
4727 | */ | |
4728 | next_mz = NULL; | |
4729 | if (!reclaimed) { | |
4730 | do { | |
4731 | /* | |
4732 | * Loop until we find yet another one. | |
4733 | * | |
4734 | * By the time we get the soft_limit lock | |
4735 | * again, someone might have aded the | |
4736 | * group back on the RB tree. Iterate to | |
4737 | * make sure we get a different mem. | |
4738 | * mem_cgroup_largest_soft_limit_node returns | |
4739 | * NULL if no other cgroup is present on | |
4740 | * the tree | |
4741 | */ | |
4742 | next_mz = | |
4743 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 4744 | if (next_mz == mz) |
d79154bb | 4745 | css_put(&next_mz->memcg->css); |
39cc98f1 | 4746 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
4747 | break; |
4748 | } while (1); | |
4749 | } | |
d79154bb HD |
4750 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
4751 | excess = res_counter_soft_limit_excess(&mz->memcg->res); | |
4e416953 BS |
4752 | /* |
4753 | * One school of thought says that we should not add | |
4754 | * back the node to the tree if reclaim returns 0. | |
4755 | * But our reclaim could return 0, simply because due | |
4756 | * to priority we are exposing a smaller subset of | |
4757 | * memory to reclaim from. Consider this as a longer | |
4758 | * term TODO. | |
4759 | */ | |
ef8745c1 | 4760 | /* If excess == 0, no tree ops */ |
d79154bb | 4761 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
4e416953 | 4762 | spin_unlock(&mctz->lock); |
d79154bb | 4763 | css_put(&mz->memcg->css); |
4e416953 BS |
4764 | loop++; |
4765 | /* | |
4766 | * Could not reclaim anything and there are no more | |
4767 | * mem cgroups to try or we seem to be looping without | |
4768 | * reclaiming anything. | |
4769 | */ | |
4770 | if (!nr_reclaimed && | |
4771 | (next_mz == NULL || | |
4772 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
4773 | break; | |
4774 | } while (!nr_reclaimed); | |
4775 | if (next_mz) | |
d79154bb | 4776 | css_put(&next_mz->memcg->css); |
4e416953 BS |
4777 | return nr_reclaimed; |
4778 | } | |
4779 | ||
2ef37d3f MH |
4780 | /** |
4781 | * mem_cgroup_force_empty_list - clears LRU of a group | |
4782 | * @memcg: group to clear | |
4783 | * @node: NUMA node | |
4784 | * @zid: zone id | |
4785 | * @lru: lru to to clear | |
4786 | * | |
3c935d18 | 4787 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
2ef37d3f MH |
4788 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
4789 | * group. | |
cc847582 | 4790 | */ |
2ef37d3f | 4791 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
08e552c6 | 4792 | int node, int zid, enum lru_list lru) |
cc847582 | 4793 | { |
bea8c150 | 4794 | struct lruvec *lruvec; |
2ef37d3f | 4795 | unsigned long flags; |
072c56c1 | 4796 | struct list_head *list; |
925b7673 JW |
4797 | struct page *busy; |
4798 | struct zone *zone; | |
072c56c1 | 4799 | |
08e552c6 | 4800 | zone = &NODE_DATA(node)->node_zones[zid]; |
bea8c150 HD |
4801 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
4802 | list = &lruvec->lists[lru]; | |
cc847582 | 4803 | |
f817ed48 | 4804 | busy = NULL; |
2ef37d3f | 4805 | do { |
925b7673 | 4806 | struct page_cgroup *pc; |
5564e88b JW |
4807 | struct page *page; |
4808 | ||
08e552c6 | 4809 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 4810 | if (list_empty(list)) { |
08e552c6 | 4811 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 4812 | break; |
f817ed48 | 4813 | } |
925b7673 JW |
4814 | page = list_entry(list->prev, struct page, lru); |
4815 | if (busy == page) { | |
4816 | list_move(&page->lru, list); | |
648bcc77 | 4817 | busy = NULL; |
08e552c6 | 4818 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
4819 | continue; |
4820 | } | |
08e552c6 | 4821 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 4822 | |
925b7673 | 4823 | pc = lookup_page_cgroup(page); |
5564e88b | 4824 | |
3c935d18 | 4825 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
f817ed48 | 4826 | /* found lock contention or "pc" is obsolete. */ |
925b7673 | 4827 | busy = page; |
f817ed48 KH |
4828 | cond_resched(); |
4829 | } else | |
4830 | busy = NULL; | |
2ef37d3f | 4831 | } while (!list_empty(list)); |
cc847582 KH |
4832 | } |
4833 | ||
4834 | /* | |
c26251f9 MH |
4835 | * make mem_cgroup's charge to be 0 if there is no task by moving |
4836 | * all the charges and pages to the parent. | |
cc847582 | 4837 | * This enables deleting this mem_cgroup. |
c26251f9 MH |
4838 | * |
4839 | * Caller is responsible for holding css reference on the memcg. | |
cc847582 | 4840 | */ |
ab5196c2 | 4841 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
cc847582 | 4842 | { |
c26251f9 | 4843 | int node, zid; |
bea207c8 | 4844 | u64 usage; |
f817ed48 | 4845 | |
fce66477 | 4846 | do { |
52d4b9ac KH |
4847 | /* This is for making all *used* pages to be on LRU. */ |
4848 | lru_add_drain_all(); | |
c0ff4b85 | 4849 | drain_all_stock_sync(memcg); |
c0ff4b85 | 4850 | mem_cgroup_start_move(memcg); |
31aaea4a | 4851 | for_each_node_state(node, N_MEMORY) { |
2ef37d3f | 4852 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
f156ab93 HD |
4853 | enum lru_list lru; |
4854 | for_each_lru(lru) { | |
2ef37d3f | 4855 | mem_cgroup_force_empty_list(memcg, |
f156ab93 | 4856 | node, zid, lru); |
f817ed48 | 4857 | } |
1ecaab2b | 4858 | } |
f817ed48 | 4859 | } |
c0ff4b85 R |
4860 | mem_cgroup_end_move(memcg); |
4861 | memcg_oom_recover(memcg); | |
52d4b9ac | 4862 | cond_resched(); |
f817ed48 | 4863 | |
2ef37d3f | 4864 | /* |
bea207c8 GC |
4865 | * Kernel memory may not necessarily be trackable to a specific |
4866 | * process. So they are not migrated, and therefore we can't | |
4867 | * expect their value to drop to 0 here. | |
4868 | * Having res filled up with kmem only is enough. | |
4869 | * | |
2ef37d3f MH |
4870 | * This is a safety check because mem_cgroup_force_empty_list |
4871 | * could have raced with mem_cgroup_replace_page_cache callers | |
4872 | * so the lru seemed empty but the page could have been added | |
4873 | * right after the check. RES_USAGE should be safe as we always | |
4874 | * charge before adding to the LRU. | |
4875 | */ | |
bea207c8 GC |
4876 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
4877 | res_counter_read_u64(&memcg->kmem, RES_USAGE); | |
4878 | } while (usage > 0); | |
c26251f9 MH |
4879 | } |
4880 | ||
b5f99b53 GC |
4881 | /* |
4882 | * This mainly exists for tests during the setting of set of use_hierarchy. | |
4883 | * Since this is the very setting we are changing, the current hierarchy value | |
4884 | * is meaningless | |
4885 | */ | |
4886 | static inline bool __memcg_has_children(struct mem_cgroup *memcg) | |
4887 | { | |
4888 | struct cgroup *pos; | |
4889 | ||
4890 | /* bounce at first found */ | |
4891 | cgroup_for_each_child(pos, memcg->css.cgroup) | |
4892 | return true; | |
4893 | return false; | |
4894 | } | |
4895 | ||
4896 | /* | |
0999821b GC |
4897 | * Must be called with memcg_create_mutex held, unless the cgroup is guaranteed |
4898 | * to be already dead (as in mem_cgroup_force_empty, for instance). This is | |
b5f99b53 GC |
4899 | * from mem_cgroup_count_children(), in the sense that we don't really care how |
4900 | * many children we have; we only need to know if we have any. It also counts | |
4901 | * any memcg without hierarchy as infertile. | |
4902 | */ | |
4903 | static inline bool memcg_has_children(struct mem_cgroup *memcg) | |
4904 | { | |
4905 | return memcg->use_hierarchy && __memcg_has_children(memcg); | |
4906 | } | |
4907 | ||
c26251f9 MH |
4908 | /* |
4909 | * Reclaims as many pages from the given memcg as possible and moves | |
4910 | * the rest to the parent. | |
4911 | * | |
4912 | * Caller is responsible for holding css reference for memcg. | |
4913 | */ | |
4914 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
4915 | { | |
4916 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
4917 | struct cgroup *cgrp = memcg->css.cgroup; | |
f817ed48 | 4918 | |
c1e862c1 | 4919 | /* returns EBUSY if there is a task or if we come here twice. */ |
c26251f9 MH |
4920 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
4921 | return -EBUSY; | |
4922 | ||
c1e862c1 KH |
4923 | /* we call try-to-free pages for make this cgroup empty */ |
4924 | lru_add_drain_all(); | |
f817ed48 | 4925 | /* try to free all pages in this cgroup */ |
569530fb | 4926 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
f817ed48 | 4927 | int progress; |
c1e862c1 | 4928 | |
c26251f9 MH |
4929 | if (signal_pending(current)) |
4930 | return -EINTR; | |
4931 | ||
c0ff4b85 | 4932 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
185efc0f | 4933 | false); |
c1e862c1 | 4934 | if (!progress) { |
f817ed48 | 4935 | nr_retries--; |
c1e862c1 | 4936 | /* maybe some writeback is necessary */ |
8aa7e847 | 4937 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 4938 | } |
f817ed48 KH |
4939 | |
4940 | } | |
08e552c6 | 4941 | lru_add_drain(); |
ab5196c2 MH |
4942 | mem_cgroup_reparent_charges(memcg); |
4943 | ||
4944 | return 0; | |
cc847582 KH |
4945 | } |
4946 | ||
6bbda35c | 4947 | static int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
c1e862c1 | 4948 | { |
c26251f9 MH |
4949 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
4950 | int ret; | |
4951 | ||
d8423011 MH |
4952 | if (mem_cgroup_is_root(memcg)) |
4953 | return -EINVAL; | |
c26251f9 MH |
4954 | css_get(&memcg->css); |
4955 | ret = mem_cgroup_force_empty(memcg); | |
4956 | css_put(&memcg->css); | |
4957 | ||
4958 | return ret; | |
c1e862c1 KH |
4959 | } |
4960 | ||
4961 | ||
18f59ea7 BS |
4962 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
4963 | { | |
4964 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
4965 | } | |
4966 | ||
4967 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
4968 | u64 val) | |
4969 | { | |
4970 | int retval = 0; | |
c0ff4b85 | 4971 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
18f59ea7 | 4972 | struct cgroup *parent = cont->parent; |
c0ff4b85 | 4973 | struct mem_cgroup *parent_memcg = NULL; |
18f59ea7 BS |
4974 | |
4975 | if (parent) | |
c0ff4b85 | 4976 | parent_memcg = mem_cgroup_from_cont(parent); |
18f59ea7 | 4977 | |
0999821b | 4978 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
4979 | |
4980 | if (memcg->use_hierarchy == val) | |
4981 | goto out; | |
4982 | ||
18f59ea7 | 4983 | /* |
af901ca1 | 4984 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
4985 | * in the child subtrees. If it is unset, then the change can |
4986 | * occur, provided the current cgroup has no children. | |
4987 | * | |
4988 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
4989 | * set if there are no children. | |
4990 | */ | |
c0ff4b85 | 4991 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 4992 | (val == 1 || val == 0)) { |
b5f99b53 | 4993 | if (!__memcg_has_children(memcg)) |
c0ff4b85 | 4994 | memcg->use_hierarchy = val; |
18f59ea7 BS |
4995 | else |
4996 | retval = -EBUSY; | |
4997 | } else | |
4998 | retval = -EINVAL; | |
567fb435 GC |
4999 | |
5000 | out: | |
0999821b | 5001 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
5002 | |
5003 | return retval; | |
5004 | } | |
5005 | ||
0c3e73e8 | 5006 | |
c0ff4b85 | 5007 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
7a159cc9 | 5008 | enum mem_cgroup_stat_index idx) |
0c3e73e8 | 5009 | { |
7d74b06f | 5010 | struct mem_cgroup *iter; |
7a159cc9 | 5011 | long val = 0; |
0c3e73e8 | 5012 | |
7a159cc9 | 5013 | /* Per-cpu values can be negative, use a signed accumulator */ |
c0ff4b85 | 5014 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f KH |
5015 | val += mem_cgroup_read_stat(iter, idx); |
5016 | ||
5017 | if (val < 0) /* race ? */ | |
5018 | val = 0; | |
5019 | return val; | |
0c3e73e8 BS |
5020 | } |
5021 | ||
c0ff4b85 | 5022 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
104f3928 | 5023 | { |
7d74b06f | 5024 | u64 val; |
104f3928 | 5025 | |
c0ff4b85 | 5026 | if (!mem_cgroup_is_root(memcg)) { |
104f3928 | 5027 | if (!swap) |
65c64ce8 | 5028 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
104f3928 | 5029 | else |
65c64ce8 | 5030 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
104f3928 KS |
5031 | } |
5032 | ||
c0ff4b85 R |
5033 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
5034 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); | |
104f3928 | 5035 | |
7d74b06f | 5036 | if (swap) |
bff6bb83 | 5037 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
104f3928 KS |
5038 | |
5039 | return val << PAGE_SHIFT; | |
5040 | } | |
5041 | ||
af36f906 TH |
5042 | static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft, |
5043 | struct file *file, char __user *buf, | |
5044 | size_t nbytes, loff_t *ppos) | |
8cdea7c0 | 5045 | { |
c0ff4b85 | 5046 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af36f906 | 5047 | char str[64]; |
104f3928 | 5048 | u64 val; |
86ae53e1 GC |
5049 | int name, len; |
5050 | enum res_type type; | |
8c7c6e34 KH |
5051 | |
5052 | type = MEMFILE_TYPE(cft->private); | |
5053 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5054 | |
8c7c6e34 KH |
5055 | switch (type) { |
5056 | case _MEM: | |
104f3928 | 5057 | if (name == RES_USAGE) |
c0ff4b85 | 5058 | val = mem_cgroup_usage(memcg, false); |
104f3928 | 5059 | else |
c0ff4b85 | 5060 | val = res_counter_read_u64(&memcg->res, name); |
8c7c6e34 KH |
5061 | break; |
5062 | case _MEMSWAP: | |
104f3928 | 5063 | if (name == RES_USAGE) |
c0ff4b85 | 5064 | val = mem_cgroup_usage(memcg, true); |
104f3928 | 5065 | else |
c0ff4b85 | 5066 | val = res_counter_read_u64(&memcg->memsw, name); |
8c7c6e34 | 5067 | break; |
510fc4e1 GC |
5068 | case _KMEM: |
5069 | val = res_counter_read_u64(&memcg->kmem, name); | |
5070 | break; | |
8c7c6e34 KH |
5071 | default: |
5072 | BUG(); | |
8c7c6e34 | 5073 | } |
af36f906 TH |
5074 | |
5075 | len = scnprintf(str, sizeof(str), "%llu\n", (unsigned long long)val); | |
5076 | return simple_read_from_buffer(buf, nbytes, ppos, str, len); | |
8cdea7c0 | 5077 | } |
510fc4e1 GC |
5078 | |
5079 | static int memcg_update_kmem_limit(struct cgroup *cont, u64 val) | |
5080 | { | |
5081 | int ret = -EINVAL; | |
5082 | #ifdef CONFIG_MEMCG_KMEM | |
5083 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); | |
5084 | /* | |
5085 | * For simplicity, we won't allow this to be disabled. It also can't | |
5086 | * be changed if the cgroup has children already, or if tasks had | |
5087 | * already joined. | |
5088 | * | |
5089 | * If tasks join before we set the limit, a person looking at | |
5090 | * kmem.usage_in_bytes will have no way to determine when it took | |
5091 | * place, which makes the value quite meaningless. | |
5092 | * | |
5093 | * After it first became limited, changes in the value of the limit are | |
5094 | * of course permitted. | |
510fc4e1 | 5095 | */ |
0999821b | 5096 | mutex_lock(&memcg_create_mutex); |
510fc4e1 GC |
5097 | mutex_lock(&set_limit_mutex); |
5098 | if (!memcg->kmem_account_flags && val != RESOURCE_MAX) { | |
b5f99b53 | 5099 | if (cgroup_task_count(cont) || memcg_has_children(memcg)) { |
510fc4e1 GC |
5100 | ret = -EBUSY; |
5101 | goto out; | |
5102 | } | |
5103 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5104 | VM_BUG_ON(ret); | |
5105 | ||
55007d84 GC |
5106 | ret = memcg_update_cache_sizes(memcg); |
5107 | if (ret) { | |
5108 | res_counter_set_limit(&memcg->kmem, RESOURCE_MAX); | |
5109 | goto out; | |
5110 | } | |
692e89ab GC |
5111 | static_key_slow_inc(&memcg_kmem_enabled_key); |
5112 | /* | |
5113 | * setting the active bit after the inc will guarantee no one | |
5114 | * starts accounting before all call sites are patched | |
5115 | */ | |
5116 | memcg_kmem_set_active(memcg); | |
5117 | ||
7de37682 GC |
5118 | /* |
5119 | * kmem charges can outlive the cgroup. In the case of slab | |
5120 | * pages, for instance, a page contain objects from various | |
5121 | * processes, so it is unfeasible to migrate them away. We | |
5122 | * need to reference count the memcg because of that. | |
5123 | */ | |
5124 | mem_cgroup_get(memcg); | |
510fc4e1 GC |
5125 | } else |
5126 | ret = res_counter_set_limit(&memcg->kmem, val); | |
5127 | out: | |
5128 | mutex_unlock(&set_limit_mutex); | |
0999821b | 5129 | mutex_unlock(&memcg_create_mutex); |
510fc4e1 GC |
5130 | #endif |
5131 | return ret; | |
5132 | } | |
5133 | ||
6d043990 | 5134 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 5135 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 5136 | { |
55007d84 | 5137 | int ret = 0; |
510fc4e1 GC |
5138 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
5139 | if (!parent) | |
55007d84 GC |
5140 | goto out; |
5141 | ||
510fc4e1 | 5142 | memcg->kmem_account_flags = parent->kmem_account_flags; |
a8964b9b GC |
5143 | /* |
5144 | * When that happen, we need to disable the static branch only on those | |
5145 | * memcgs that enabled it. To achieve this, we would be forced to | |
5146 | * complicate the code by keeping track of which memcgs were the ones | |
5147 | * that actually enabled limits, and which ones got it from its | |
5148 | * parents. | |
5149 | * | |
5150 | * It is a lot simpler just to do static_key_slow_inc() on every child | |
5151 | * that is accounted. | |
5152 | */ | |
55007d84 GC |
5153 | if (!memcg_kmem_is_active(memcg)) |
5154 | goto out; | |
5155 | ||
5156 | /* | |
5157 | * destroy(), called if we fail, will issue static_key_slow_inc() and | |
5158 | * mem_cgroup_put() if kmem is enabled. We have to either call them | |
5159 | * unconditionally, or clear the KMEM_ACTIVE flag. I personally find | |
5160 | * this more consistent, since it always leads to the same destroy path | |
5161 | */ | |
5162 | mem_cgroup_get(memcg); | |
5163 | static_key_slow_inc(&memcg_kmem_enabled_key); | |
5164 | ||
5165 | mutex_lock(&set_limit_mutex); | |
5166 | ret = memcg_update_cache_sizes(memcg); | |
5167 | mutex_unlock(&set_limit_mutex); | |
55007d84 GC |
5168 | out: |
5169 | return ret; | |
510fc4e1 | 5170 | } |
6d043990 | 5171 | #endif /* CONFIG_MEMCG_KMEM */ |
510fc4e1 | 5172 | |
628f4235 KH |
5173 | /* |
5174 | * The user of this function is... | |
5175 | * RES_LIMIT. | |
5176 | */ | |
856c13aa PM |
5177 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
5178 | const char *buffer) | |
8cdea7c0 | 5179 | { |
628f4235 | 5180 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
5181 | enum res_type type; |
5182 | int name; | |
628f4235 KH |
5183 | unsigned long long val; |
5184 | int ret; | |
5185 | ||
8c7c6e34 KH |
5186 | type = MEMFILE_TYPE(cft->private); |
5187 | name = MEMFILE_ATTR(cft->private); | |
af36f906 | 5188 | |
8c7c6e34 | 5189 | switch (name) { |
628f4235 | 5190 | case RES_LIMIT: |
4b3bde4c BS |
5191 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
5192 | ret = -EINVAL; | |
5193 | break; | |
5194 | } | |
628f4235 KH |
5195 | /* This function does all necessary parse...reuse it */ |
5196 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
5197 | if (ret) |
5198 | break; | |
5199 | if (type == _MEM) | |
628f4235 | 5200 | ret = mem_cgroup_resize_limit(memcg, val); |
510fc4e1 | 5201 | else if (type == _MEMSWAP) |
8c7c6e34 | 5202 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
510fc4e1 GC |
5203 | else if (type == _KMEM) |
5204 | ret = memcg_update_kmem_limit(cont, val); | |
5205 | else | |
5206 | return -EINVAL; | |
628f4235 | 5207 | break; |
296c81d8 BS |
5208 | case RES_SOFT_LIMIT: |
5209 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
5210 | if (ret) | |
5211 | break; | |
5212 | /* | |
5213 | * For memsw, soft limits are hard to implement in terms | |
5214 | * of semantics, for now, we support soft limits for | |
5215 | * control without swap | |
5216 | */ | |
5217 | if (type == _MEM) | |
5218 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
5219 | else | |
5220 | ret = -EINVAL; | |
5221 | break; | |
628f4235 KH |
5222 | default: |
5223 | ret = -EINVAL; /* should be BUG() ? */ | |
5224 | break; | |
5225 | } | |
5226 | return ret; | |
8cdea7c0 BS |
5227 | } |
5228 | ||
fee7b548 KH |
5229 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
5230 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
5231 | { | |
5232 | struct cgroup *cgroup; | |
5233 | unsigned long long min_limit, min_memsw_limit, tmp; | |
5234 | ||
5235 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5236 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5237 | cgroup = memcg->css.cgroup; | |
5238 | if (!memcg->use_hierarchy) | |
5239 | goto out; | |
5240 | ||
5241 | while (cgroup->parent) { | |
5242 | cgroup = cgroup->parent; | |
5243 | memcg = mem_cgroup_from_cont(cgroup); | |
5244 | if (!memcg->use_hierarchy) | |
5245 | break; | |
5246 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
5247 | min_limit = min(min_limit, tmp); | |
5248 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
5249 | min_memsw_limit = min(min_memsw_limit, tmp); | |
5250 | } | |
5251 | out: | |
5252 | *mem_limit = min_limit; | |
5253 | *memsw_limit = min_memsw_limit; | |
fee7b548 KH |
5254 | } |
5255 | ||
29f2a4da | 5256 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 | 5257 | { |
af36f906 | 5258 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
86ae53e1 GC |
5259 | int name; |
5260 | enum res_type type; | |
c84872e1 | 5261 | |
8c7c6e34 KH |
5262 | type = MEMFILE_TYPE(event); |
5263 | name = MEMFILE_ATTR(event); | |
af36f906 | 5264 | |
8c7c6e34 | 5265 | switch (name) { |
29f2a4da | 5266 | case RES_MAX_USAGE: |
8c7c6e34 | 5267 | if (type == _MEM) |
c0ff4b85 | 5268 | res_counter_reset_max(&memcg->res); |
510fc4e1 | 5269 | else if (type == _MEMSWAP) |
c0ff4b85 | 5270 | res_counter_reset_max(&memcg->memsw); |
510fc4e1 GC |
5271 | else if (type == _KMEM) |
5272 | res_counter_reset_max(&memcg->kmem); | |
5273 | else | |
5274 | return -EINVAL; | |
29f2a4da PE |
5275 | break; |
5276 | case RES_FAILCNT: | |
8c7c6e34 | 5277 | if (type == _MEM) |
c0ff4b85 | 5278 | res_counter_reset_failcnt(&memcg->res); |
510fc4e1 | 5279 | else if (type == _MEMSWAP) |
c0ff4b85 | 5280 | res_counter_reset_failcnt(&memcg->memsw); |
510fc4e1 GC |
5281 | else if (type == _KMEM) |
5282 | res_counter_reset_failcnt(&memcg->kmem); | |
5283 | else | |
5284 | return -EINVAL; | |
29f2a4da PE |
5285 | break; |
5286 | } | |
f64c3f54 | 5287 | |
85cc59db | 5288 | return 0; |
c84872e1 PE |
5289 | } |
5290 | ||
7dc74be0 DN |
5291 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
5292 | struct cftype *cft) | |
5293 | { | |
5294 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
5295 | } | |
5296 | ||
02491447 | 5297 | #ifdef CONFIG_MMU |
7dc74be0 DN |
5298 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
5299 | struct cftype *cft, u64 val) | |
5300 | { | |
c0ff4b85 | 5301 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
7dc74be0 DN |
5302 | |
5303 | if (val >= (1 << NR_MOVE_TYPE)) | |
5304 | return -EINVAL; | |
ee5e8472 | 5305 | |
7dc74be0 | 5306 | /* |
ee5e8472 GC |
5307 | * No kind of locking is needed in here, because ->can_attach() will |
5308 | * check this value once in the beginning of the process, and then carry | |
5309 | * on with stale data. This means that changes to this value will only | |
5310 | * affect task migrations starting after the change. | |
7dc74be0 | 5311 | */ |
c0ff4b85 | 5312 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
5313 | return 0; |
5314 | } | |
02491447 DN |
5315 | #else |
5316 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
5317 | struct cftype *cft, u64 val) | |
5318 | { | |
5319 | return -ENOSYS; | |
5320 | } | |
5321 | #endif | |
7dc74be0 | 5322 | |
406eb0c9 | 5323 | #ifdef CONFIG_NUMA |
ab215884 | 5324 | static int memcg_numa_stat_show(struct cgroup *cont, struct cftype *cft, |
fada52ca | 5325 | struct seq_file *m) |
406eb0c9 YH |
5326 | { |
5327 | int nid; | |
5328 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
5329 | unsigned long node_nr; | |
d79154bb | 5330 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
406eb0c9 | 5331 | |
d79154bb | 5332 | total_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL); |
406eb0c9 | 5333 | seq_printf(m, "total=%lu", total_nr); |
31aaea4a | 5334 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5335 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL); |
406eb0c9 YH |
5336 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5337 | } | |
5338 | seq_putc(m, '\n'); | |
5339 | ||
d79154bb | 5340 | file_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_FILE); |
406eb0c9 | 5341 | seq_printf(m, "file=%lu", file_nr); |
31aaea4a | 5342 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5343 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5344 | LRU_ALL_FILE); |
406eb0c9 YH |
5345 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5346 | } | |
5347 | seq_putc(m, '\n'); | |
5348 | ||
d79154bb | 5349 | anon_nr = mem_cgroup_nr_lru_pages(memcg, LRU_ALL_ANON); |
406eb0c9 | 5350 | seq_printf(m, "anon=%lu", anon_nr); |
31aaea4a | 5351 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5352 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5353 | LRU_ALL_ANON); |
406eb0c9 YH |
5354 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5355 | } | |
5356 | seq_putc(m, '\n'); | |
5357 | ||
d79154bb | 5358 | unevictable_nr = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_UNEVICTABLE)); |
406eb0c9 | 5359 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
31aaea4a | 5360 | for_each_node_state(nid, N_MEMORY) { |
d79154bb | 5361 | node_nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
bb2a0de9 | 5362 | BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
5363 | seq_printf(m, " N%d=%lu", nid, node_nr); |
5364 | } | |
5365 | seq_putc(m, '\n'); | |
5366 | return 0; | |
5367 | } | |
5368 | #endif /* CONFIG_NUMA */ | |
5369 | ||
af7c4b0e JW |
5370 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
5371 | { | |
5372 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); | |
5373 | } | |
5374 | ||
ab215884 | 5375 | static int memcg_stat_show(struct cgroup *cont, struct cftype *cft, |
78ccf5b5 | 5376 | struct seq_file *m) |
d2ceb9b7 | 5377 | { |
d79154bb | 5378 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
af7c4b0e JW |
5379 | struct mem_cgroup *mi; |
5380 | unsigned int i; | |
406eb0c9 | 5381 | |
af7c4b0e | 5382 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
bff6bb83 | 5383 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5384 | continue; |
af7c4b0e JW |
5385 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
5386 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); | |
1dd3a273 | 5387 | } |
7b854121 | 5388 | |
af7c4b0e JW |
5389 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
5390 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
5391 | mem_cgroup_read_events(memcg, i)); | |
5392 | ||
5393 | for (i = 0; i < NR_LRU_LISTS; i++) | |
5394 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
5395 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
5396 | ||
14067bb3 | 5397 | /* Hierarchical information */ |
fee7b548 KH |
5398 | { |
5399 | unsigned long long limit, memsw_limit; | |
d79154bb | 5400 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
78ccf5b5 | 5401 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
fee7b548 | 5402 | if (do_swap_account) |
78ccf5b5 JW |
5403 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
5404 | memsw_limit); | |
fee7b548 | 5405 | } |
7f016ee8 | 5406 | |
af7c4b0e JW |
5407 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
5408 | long long val = 0; | |
5409 | ||
bff6bb83 | 5410 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
1dd3a273 | 5411 | continue; |
af7c4b0e JW |
5412 | for_each_mem_cgroup_tree(mi, memcg) |
5413 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
5414 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); | |
5415 | } | |
5416 | ||
5417 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
5418 | unsigned long long val = 0; | |
5419 | ||
5420 | for_each_mem_cgroup_tree(mi, memcg) | |
5421 | val += mem_cgroup_read_events(mi, i); | |
5422 | seq_printf(m, "total_%s %llu\n", | |
5423 | mem_cgroup_events_names[i], val); | |
5424 | } | |
5425 | ||
5426 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5427 | unsigned long long val = 0; | |
5428 | ||
5429 | for_each_mem_cgroup_tree(mi, memcg) | |
5430 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
5431 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 5432 | } |
14067bb3 | 5433 | |
7f016ee8 | 5434 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
5435 | { |
5436 | int nid, zid; | |
5437 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 5438 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
5439 | unsigned long recent_rotated[2] = {0, 0}; |
5440 | unsigned long recent_scanned[2] = {0, 0}; | |
5441 | ||
5442 | for_each_online_node(nid) | |
5443 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
d79154bb | 5444 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
89abfab1 | 5445 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 5446 | |
89abfab1 HD |
5447 | recent_rotated[0] += rstat->recent_rotated[0]; |
5448 | recent_rotated[1] += rstat->recent_rotated[1]; | |
5449 | recent_scanned[0] += rstat->recent_scanned[0]; | |
5450 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 5451 | } |
78ccf5b5 JW |
5452 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
5453 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
5454 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
5455 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
5456 | } |
5457 | #endif | |
5458 | ||
d2ceb9b7 KH |
5459 | return 0; |
5460 | } | |
5461 | ||
a7885eb8 KM |
5462 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
5463 | { | |
5464 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
5465 | ||
1f4c025b | 5466 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
5467 | } |
5468 | ||
5469 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
5470 | u64 val) | |
5471 | { | |
5472 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
5473 | struct mem_cgroup *parent; | |
068b38c1 | 5474 | |
a7885eb8 KM |
5475 | if (val > 100) |
5476 | return -EINVAL; | |
5477 | ||
5478 | if (cgrp->parent == NULL) | |
5479 | return -EINVAL; | |
5480 | ||
5481 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 | 5482 | |
0999821b | 5483 | mutex_lock(&memcg_create_mutex); |
068b38c1 | 5484 | |
a7885eb8 | 5485 | /* If under hierarchy, only empty-root can set this value */ |
b5f99b53 | 5486 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5487 | mutex_unlock(&memcg_create_mutex); |
a7885eb8 | 5488 | return -EINVAL; |
068b38c1 | 5489 | } |
a7885eb8 | 5490 | |
a7885eb8 | 5491 | memcg->swappiness = val; |
a7885eb8 | 5492 | |
0999821b | 5493 | mutex_unlock(&memcg_create_mutex); |
068b38c1 | 5494 | |
a7885eb8 KM |
5495 | return 0; |
5496 | } | |
5497 | ||
2e72b634 KS |
5498 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
5499 | { | |
5500 | struct mem_cgroup_threshold_ary *t; | |
5501 | u64 usage; | |
5502 | int i; | |
5503 | ||
5504 | rcu_read_lock(); | |
5505 | if (!swap) | |
2c488db2 | 5506 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 5507 | else |
2c488db2 | 5508 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
5509 | |
5510 | if (!t) | |
5511 | goto unlock; | |
5512 | ||
5513 | usage = mem_cgroup_usage(memcg, swap); | |
5514 | ||
5515 | /* | |
748dad36 | 5516 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
5517 | * If it's not true, a threshold was crossed after last |
5518 | * call of __mem_cgroup_threshold(). | |
5519 | */ | |
5407a562 | 5520 | i = t->current_threshold; |
2e72b634 KS |
5521 | |
5522 | /* | |
5523 | * Iterate backward over array of thresholds starting from | |
5524 | * current_threshold and check if a threshold is crossed. | |
5525 | * If none of thresholds below usage is crossed, we read | |
5526 | * only one element of the array here. | |
5527 | */ | |
5528 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
5529 | eventfd_signal(t->entries[i].eventfd, 1); | |
5530 | ||
5531 | /* i = current_threshold + 1 */ | |
5532 | i++; | |
5533 | ||
5534 | /* | |
5535 | * Iterate forward over array of thresholds starting from | |
5536 | * current_threshold+1 and check if a threshold is crossed. | |
5537 | * If none of thresholds above usage is crossed, we read | |
5538 | * only one element of the array here. | |
5539 | */ | |
5540 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
5541 | eventfd_signal(t->entries[i].eventfd, 1); | |
5542 | ||
5543 | /* Update current_threshold */ | |
5407a562 | 5544 | t->current_threshold = i - 1; |
2e72b634 KS |
5545 | unlock: |
5546 | rcu_read_unlock(); | |
5547 | } | |
5548 | ||
5549 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
5550 | { | |
ad4ca5f4 KS |
5551 | while (memcg) { |
5552 | __mem_cgroup_threshold(memcg, false); | |
5553 | if (do_swap_account) | |
5554 | __mem_cgroup_threshold(memcg, true); | |
5555 | ||
5556 | memcg = parent_mem_cgroup(memcg); | |
5557 | } | |
2e72b634 KS |
5558 | } |
5559 | ||
5560 | static int compare_thresholds(const void *a, const void *b) | |
5561 | { | |
5562 | const struct mem_cgroup_threshold *_a = a; | |
5563 | const struct mem_cgroup_threshold *_b = b; | |
5564 | ||
5565 | return _a->threshold - _b->threshold; | |
5566 | } | |
5567 | ||
c0ff4b85 | 5568 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
5569 | { |
5570 | struct mem_cgroup_eventfd_list *ev; | |
5571 | ||
c0ff4b85 | 5572 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 KH |
5573 | eventfd_signal(ev->eventfd, 1); |
5574 | return 0; | |
5575 | } | |
5576 | ||
c0ff4b85 | 5577 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 5578 | { |
7d74b06f KH |
5579 | struct mem_cgroup *iter; |
5580 | ||
c0ff4b85 | 5581 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 5582 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
5583 | } |
5584 | ||
5585 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
5586 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
5587 | { |
5588 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
5589 | struct mem_cgroup_thresholds *thresholds; |
5590 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5591 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5592 | u64 threshold, usage; |
2c488db2 | 5593 | int i, size, ret; |
2e72b634 KS |
5594 | |
5595 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
5596 | if (ret) | |
5597 | return ret; | |
5598 | ||
5599 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 5600 | |
2e72b634 | 5601 | if (type == _MEM) |
2c488db2 | 5602 | thresholds = &memcg->thresholds; |
2e72b634 | 5603 | else if (type == _MEMSWAP) |
2c488db2 | 5604 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5605 | else |
5606 | BUG(); | |
5607 | ||
5608 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
5609 | ||
5610 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 5611 | if (thresholds->primary) |
2e72b634 KS |
5612 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
5613 | ||
2c488db2 | 5614 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
5615 | |
5616 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 5617 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 5618 | GFP_KERNEL); |
2c488db2 | 5619 | if (!new) { |
2e72b634 KS |
5620 | ret = -ENOMEM; |
5621 | goto unlock; | |
5622 | } | |
2c488db2 | 5623 | new->size = size; |
2e72b634 KS |
5624 | |
5625 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
5626 | if (thresholds->primary) { |
5627 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 5628 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
5629 | } |
5630 | ||
2e72b634 | 5631 | /* Add new threshold */ |
2c488db2 KS |
5632 | new->entries[size - 1].eventfd = eventfd; |
5633 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
5634 | |
5635 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 5636 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
5637 | compare_thresholds, NULL); |
5638 | ||
5639 | /* Find current threshold */ | |
2c488db2 | 5640 | new->current_threshold = -1; |
2e72b634 | 5641 | for (i = 0; i < size; i++) { |
748dad36 | 5642 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 5643 | /* |
2c488db2 KS |
5644 | * new->current_threshold will not be used until |
5645 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
5646 | * it here. |
5647 | */ | |
2c488db2 | 5648 | ++new->current_threshold; |
748dad36 SZ |
5649 | } else |
5650 | break; | |
2e72b634 KS |
5651 | } |
5652 | ||
2c488db2 KS |
5653 | /* Free old spare buffer and save old primary buffer as spare */ |
5654 | kfree(thresholds->spare); | |
5655 | thresholds->spare = thresholds->primary; | |
5656 | ||
5657 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 5658 | |
907860ed | 5659 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
5660 | synchronize_rcu(); |
5661 | ||
2e72b634 KS |
5662 | unlock: |
5663 | mutex_unlock(&memcg->thresholds_lock); | |
5664 | ||
5665 | return ret; | |
5666 | } | |
5667 | ||
907860ed | 5668 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 5669 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
5670 | { |
5671 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
5672 | struct mem_cgroup_thresholds *thresholds; |
5673 | struct mem_cgroup_threshold_ary *new; | |
86ae53e1 | 5674 | enum res_type type = MEMFILE_TYPE(cft->private); |
2e72b634 | 5675 | u64 usage; |
2c488db2 | 5676 | int i, j, size; |
2e72b634 KS |
5677 | |
5678 | mutex_lock(&memcg->thresholds_lock); | |
5679 | if (type == _MEM) | |
2c488db2 | 5680 | thresholds = &memcg->thresholds; |
2e72b634 | 5681 | else if (type == _MEMSWAP) |
2c488db2 | 5682 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
5683 | else |
5684 | BUG(); | |
5685 | ||
371528ca AV |
5686 | if (!thresholds->primary) |
5687 | goto unlock; | |
5688 | ||
2e72b634 KS |
5689 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
5690 | ||
5691 | /* Check if a threshold crossed before removing */ | |
5692 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
5693 | ||
5694 | /* Calculate new number of threshold */ | |
2c488db2 KS |
5695 | size = 0; |
5696 | for (i = 0; i < thresholds->primary->size; i++) { | |
5697 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
5698 | size++; |
5699 | } | |
5700 | ||
2c488db2 | 5701 | new = thresholds->spare; |
907860ed | 5702 | |
2e72b634 KS |
5703 | /* Set thresholds array to NULL if we don't have thresholds */ |
5704 | if (!size) { | |
2c488db2 KS |
5705 | kfree(new); |
5706 | new = NULL; | |
907860ed | 5707 | goto swap_buffers; |
2e72b634 KS |
5708 | } |
5709 | ||
2c488db2 | 5710 | new->size = size; |
2e72b634 KS |
5711 | |
5712 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
5713 | new->current_threshold = -1; |
5714 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
5715 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
5716 | continue; |
5717 | ||
2c488db2 | 5718 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 5719 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 5720 | /* |
2c488db2 | 5721 | * new->current_threshold will not be used |
2e72b634 KS |
5722 | * until rcu_assign_pointer(), so it's safe to increment |
5723 | * it here. | |
5724 | */ | |
2c488db2 | 5725 | ++new->current_threshold; |
2e72b634 KS |
5726 | } |
5727 | j++; | |
5728 | } | |
5729 | ||
907860ed | 5730 | swap_buffers: |
2c488db2 KS |
5731 | /* Swap primary and spare array */ |
5732 | thresholds->spare = thresholds->primary; | |
8c757763 SZ |
5733 | /* If all events are unregistered, free the spare array */ |
5734 | if (!new) { | |
5735 | kfree(thresholds->spare); | |
5736 | thresholds->spare = NULL; | |
5737 | } | |
5738 | ||
2c488db2 | 5739 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 5740 | |
907860ed | 5741 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 5742 | synchronize_rcu(); |
371528ca | 5743 | unlock: |
2e72b634 | 5744 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 5745 | } |
c1e862c1 | 5746 | |
9490ff27 KH |
5747 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
5748 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
5749 | { | |
5750 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
5751 | struct mem_cgroup_eventfd_list *event; | |
86ae53e1 | 5752 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5753 | |
5754 | BUG_ON(type != _OOM_TYPE); | |
5755 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
5756 | if (!event) | |
5757 | return -ENOMEM; | |
5758 | ||
1af8efe9 | 5759 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
5760 | |
5761 | event->eventfd = eventfd; | |
5762 | list_add(&event->list, &memcg->oom_notify); | |
5763 | ||
5764 | /* already in OOM ? */ | |
79dfdacc | 5765 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 5766 | eventfd_signal(eventfd, 1); |
1af8efe9 | 5767 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5768 | |
5769 | return 0; | |
5770 | } | |
5771 | ||
907860ed | 5772 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
5773 | struct cftype *cft, struct eventfd_ctx *eventfd) |
5774 | { | |
c0ff4b85 | 5775 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
9490ff27 | 5776 | struct mem_cgroup_eventfd_list *ev, *tmp; |
86ae53e1 | 5777 | enum res_type type = MEMFILE_TYPE(cft->private); |
9490ff27 KH |
5778 | |
5779 | BUG_ON(type != _OOM_TYPE); | |
5780 | ||
1af8efe9 | 5781 | spin_lock(&memcg_oom_lock); |
9490ff27 | 5782 | |
c0ff4b85 | 5783 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
5784 | if (ev->eventfd == eventfd) { |
5785 | list_del(&ev->list); | |
5786 | kfree(ev); | |
5787 | } | |
5788 | } | |
5789 | ||
1af8efe9 | 5790 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
5791 | } |
5792 | ||
3c11ecf4 KH |
5793 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
5794 | struct cftype *cft, struct cgroup_map_cb *cb) | |
5795 | { | |
c0ff4b85 | 5796 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 | 5797 | |
c0ff4b85 | 5798 | cb->fill(cb, "oom_kill_disable", memcg->oom_kill_disable); |
3c11ecf4 | 5799 | |
c0ff4b85 | 5800 | if (atomic_read(&memcg->under_oom)) |
3c11ecf4 KH |
5801 | cb->fill(cb, "under_oom", 1); |
5802 | else | |
5803 | cb->fill(cb, "under_oom", 0); | |
5804 | return 0; | |
5805 | } | |
5806 | ||
3c11ecf4 KH |
5807 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
5808 | struct cftype *cft, u64 val) | |
5809 | { | |
c0ff4b85 | 5810 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
3c11ecf4 KH |
5811 | struct mem_cgroup *parent; |
5812 | ||
5813 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
5814 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
5815 | return -EINVAL; | |
5816 | ||
5817 | parent = mem_cgroup_from_cont(cgrp->parent); | |
5818 | ||
0999821b | 5819 | mutex_lock(&memcg_create_mutex); |
3c11ecf4 | 5820 | /* oom-kill-disable is a flag for subhierarchy. */ |
b5f99b53 | 5821 | if ((parent->use_hierarchy) || memcg_has_children(memcg)) { |
0999821b | 5822 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5823 | return -EINVAL; |
5824 | } | |
c0ff4b85 | 5825 | memcg->oom_kill_disable = val; |
4d845ebf | 5826 | if (!val) |
c0ff4b85 | 5827 | memcg_oom_recover(memcg); |
0999821b | 5828 | mutex_unlock(&memcg_create_mutex); |
3c11ecf4 KH |
5829 | return 0; |
5830 | } | |
5831 | ||
c255a458 | 5832 | #ifdef CONFIG_MEMCG_KMEM |
cbe128e3 | 5833 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa | 5834 | { |
55007d84 GC |
5835 | int ret; |
5836 | ||
2633d7a0 | 5837 | memcg->kmemcg_id = -1; |
55007d84 GC |
5838 | ret = memcg_propagate_kmem(memcg); |
5839 | if (ret) | |
5840 | return ret; | |
2633d7a0 | 5841 | |
1d62e436 | 5842 | return mem_cgroup_sockets_init(memcg, ss); |
573b400d | 5843 | } |
e5671dfa | 5844 | |
1d62e436 | 5845 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 | 5846 | { |
1d62e436 | 5847 | mem_cgroup_sockets_destroy(memcg); |
7de37682 GC |
5848 | |
5849 | memcg_kmem_mark_dead(memcg); | |
5850 | ||
5851 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) | |
5852 | return; | |
5853 | ||
5854 | /* | |
5855 | * Charges already down to 0, undo mem_cgroup_get() done in the charge | |
5856 | * path here, being careful not to race with memcg_uncharge_kmem: it is | |
5857 | * possible that the charges went down to 0 between mark_dead and the | |
5858 | * res_counter read, so in that case, we don't need the put | |
5859 | */ | |
5860 | if (memcg_kmem_test_and_clear_dead(memcg)) | |
5861 | mem_cgroup_put(memcg); | |
d1a4c0b3 | 5862 | } |
e5671dfa | 5863 | #else |
cbe128e3 | 5864 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
e5671dfa GC |
5865 | { |
5866 | return 0; | |
5867 | } | |
d1a4c0b3 | 5868 | |
1d62e436 | 5869 | static void kmem_cgroup_destroy(struct mem_cgroup *memcg) |
d1a4c0b3 GC |
5870 | { |
5871 | } | |
e5671dfa GC |
5872 | #endif |
5873 | ||
8cdea7c0 BS |
5874 | static struct cftype mem_cgroup_files[] = { |
5875 | { | |
0eea1030 | 5876 | .name = "usage_in_bytes", |
8c7c6e34 | 5877 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
af36f906 | 5878 | .read = mem_cgroup_read, |
9490ff27 KH |
5879 | .register_event = mem_cgroup_usage_register_event, |
5880 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 5881 | }, |
c84872e1 PE |
5882 | { |
5883 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 5884 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 5885 | .trigger = mem_cgroup_reset, |
af36f906 | 5886 | .read = mem_cgroup_read, |
c84872e1 | 5887 | }, |
8cdea7c0 | 5888 | { |
0eea1030 | 5889 | .name = "limit_in_bytes", |
8c7c6e34 | 5890 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 5891 | .write_string = mem_cgroup_write, |
af36f906 | 5892 | .read = mem_cgroup_read, |
8cdea7c0 | 5893 | }, |
296c81d8 BS |
5894 | { |
5895 | .name = "soft_limit_in_bytes", | |
5896 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
5897 | .write_string = mem_cgroup_write, | |
af36f906 | 5898 | .read = mem_cgroup_read, |
296c81d8 | 5899 | }, |
8cdea7c0 BS |
5900 | { |
5901 | .name = "failcnt", | |
8c7c6e34 | 5902 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 5903 | .trigger = mem_cgroup_reset, |
af36f906 | 5904 | .read = mem_cgroup_read, |
8cdea7c0 | 5905 | }, |
d2ceb9b7 KH |
5906 | { |
5907 | .name = "stat", | |
ab215884 | 5908 | .read_seq_string = memcg_stat_show, |
d2ceb9b7 | 5909 | }, |
c1e862c1 KH |
5910 | { |
5911 | .name = "force_empty", | |
5912 | .trigger = mem_cgroup_force_empty_write, | |
5913 | }, | |
18f59ea7 BS |
5914 | { |
5915 | .name = "use_hierarchy", | |
f00baae7 | 5916 | .flags = CFTYPE_INSANE, |
18f59ea7 BS |
5917 | .write_u64 = mem_cgroup_hierarchy_write, |
5918 | .read_u64 = mem_cgroup_hierarchy_read, | |
5919 | }, | |
a7885eb8 KM |
5920 | { |
5921 | .name = "swappiness", | |
5922 | .read_u64 = mem_cgroup_swappiness_read, | |
5923 | .write_u64 = mem_cgroup_swappiness_write, | |
5924 | }, | |
7dc74be0 DN |
5925 | { |
5926 | .name = "move_charge_at_immigrate", | |
5927 | .read_u64 = mem_cgroup_move_charge_read, | |
5928 | .write_u64 = mem_cgroup_move_charge_write, | |
5929 | }, | |
9490ff27 KH |
5930 | { |
5931 | .name = "oom_control", | |
3c11ecf4 KH |
5932 | .read_map = mem_cgroup_oom_control_read, |
5933 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
5934 | .register_event = mem_cgroup_oom_register_event, |
5935 | .unregister_event = mem_cgroup_oom_unregister_event, | |
5936 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
5937 | }, | |
70ddf637 AV |
5938 | { |
5939 | .name = "pressure_level", | |
5940 | .register_event = vmpressure_register_event, | |
5941 | .unregister_event = vmpressure_unregister_event, | |
5942 | }, | |
406eb0c9 YH |
5943 | #ifdef CONFIG_NUMA |
5944 | { | |
5945 | .name = "numa_stat", | |
ab215884 | 5946 | .read_seq_string = memcg_numa_stat_show, |
406eb0c9 YH |
5947 | }, |
5948 | #endif | |
510fc4e1 GC |
5949 | #ifdef CONFIG_MEMCG_KMEM |
5950 | { | |
5951 | .name = "kmem.limit_in_bytes", | |
5952 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
5953 | .write_string = mem_cgroup_write, | |
5954 | .read = mem_cgroup_read, | |
5955 | }, | |
5956 | { | |
5957 | .name = "kmem.usage_in_bytes", | |
5958 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
5959 | .read = mem_cgroup_read, | |
5960 | }, | |
5961 | { | |
5962 | .name = "kmem.failcnt", | |
5963 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
5964 | .trigger = mem_cgroup_reset, | |
5965 | .read = mem_cgroup_read, | |
5966 | }, | |
5967 | { | |
5968 | .name = "kmem.max_usage_in_bytes", | |
5969 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
5970 | .trigger = mem_cgroup_reset, | |
5971 | .read = mem_cgroup_read, | |
5972 | }, | |
749c5415 GC |
5973 | #ifdef CONFIG_SLABINFO |
5974 | { | |
5975 | .name = "kmem.slabinfo", | |
5976 | .read_seq_string = mem_cgroup_slabinfo_read, | |
5977 | }, | |
5978 | #endif | |
8c7c6e34 | 5979 | #endif |
6bc10349 | 5980 | { }, /* terminate */ |
af36f906 | 5981 | }; |
8c7c6e34 | 5982 | |
2d11085e MH |
5983 | #ifdef CONFIG_MEMCG_SWAP |
5984 | static struct cftype memsw_cgroup_files[] = { | |
5985 | { | |
5986 | .name = "memsw.usage_in_bytes", | |
5987 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5988 | .read = mem_cgroup_read, | |
5989 | .register_event = mem_cgroup_usage_register_event, | |
5990 | .unregister_event = mem_cgroup_usage_unregister_event, | |
5991 | }, | |
5992 | { | |
5993 | .name = "memsw.max_usage_in_bytes", | |
5994 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5995 | .trigger = mem_cgroup_reset, | |
5996 | .read = mem_cgroup_read, | |
5997 | }, | |
5998 | { | |
5999 | .name = "memsw.limit_in_bytes", | |
6000 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6001 | .write_string = mem_cgroup_write, | |
6002 | .read = mem_cgroup_read, | |
6003 | }, | |
6004 | { | |
6005 | .name = "memsw.failcnt", | |
6006 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6007 | .trigger = mem_cgroup_reset, | |
6008 | .read = mem_cgroup_read, | |
6009 | }, | |
6010 | { }, /* terminate */ | |
6011 | }; | |
6012 | #endif | |
c0ff4b85 | 6013 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
6014 | { |
6015 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 6016 | struct mem_cgroup_per_zone *mz; |
41e3355d | 6017 | int zone, tmp = node; |
1ecaab2b KH |
6018 | /* |
6019 | * This routine is called against possible nodes. | |
6020 | * But it's BUG to call kmalloc() against offline node. | |
6021 | * | |
6022 | * TODO: this routine can waste much memory for nodes which will | |
6023 | * never be onlined. It's better to use memory hotplug callback | |
6024 | * function. | |
6025 | */ | |
41e3355d KH |
6026 | if (!node_state(node, N_NORMAL_MEMORY)) |
6027 | tmp = -1; | |
17295c88 | 6028 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
6029 | if (!pn) |
6030 | return 1; | |
1ecaab2b | 6031 | |
1ecaab2b KH |
6032 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
6033 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 6034 | lruvec_init(&mz->lruvec); |
f64c3f54 | 6035 | mz->usage_in_excess = 0; |
4e416953 | 6036 | mz->on_tree = false; |
d79154bb | 6037 | mz->memcg = memcg; |
1ecaab2b | 6038 | } |
0a619e58 | 6039 | memcg->info.nodeinfo[node] = pn; |
6d12e2d8 KH |
6040 | return 0; |
6041 | } | |
6042 | ||
c0ff4b85 | 6043 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 6044 | { |
c0ff4b85 | 6045 | kfree(memcg->info.nodeinfo[node]); |
1ecaab2b KH |
6046 | } |
6047 | ||
33327948 KH |
6048 | static struct mem_cgroup *mem_cgroup_alloc(void) |
6049 | { | |
d79154bb | 6050 | struct mem_cgroup *memcg; |
45cf7ebd | 6051 | size_t size = memcg_size(); |
33327948 | 6052 | |
45cf7ebd | 6053 | /* Can be very big if nr_node_ids is very big */ |
c8dad2bb | 6054 | if (size < PAGE_SIZE) |
d79154bb | 6055 | memcg = kzalloc(size, GFP_KERNEL); |
33327948 | 6056 | else |
d79154bb | 6057 | memcg = vzalloc(size); |
33327948 | 6058 | |
d79154bb | 6059 | if (!memcg) |
e7bbcdf3 DC |
6060 | return NULL; |
6061 | ||
d79154bb HD |
6062 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
6063 | if (!memcg->stat) | |
d2e61b8d | 6064 | goto out_free; |
d79154bb HD |
6065 | spin_lock_init(&memcg->pcp_counter_lock); |
6066 | return memcg; | |
d2e61b8d DC |
6067 | |
6068 | out_free: | |
6069 | if (size < PAGE_SIZE) | |
d79154bb | 6070 | kfree(memcg); |
d2e61b8d | 6071 | else |
d79154bb | 6072 | vfree(memcg); |
d2e61b8d | 6073 | return NULL; |
33327948 KH |
6074 | } |
6075 | ||
59927fb9 | 6076 | /* |
c8b2a36f GC |
6077 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
6078 | * (scanning all at force_empty is too costly...) | |
6079 | * | |
6080 | * Instead of clearing all references at force_empty, we remember | |
6081 | * the number of reference from swap_cgroup and free mem_cgroup when | |
6082 | * it goes down to 0. | |
6083 | * | |
6084 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 6085 | */ |
c8b2a36f GC |
6086 | |
6087 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 6088 | { |
c8b2a36f | 6089 | int node; |
45cf7ebd | 6090 | size_t size = memcg_size(); |
59927fb9 | 6091 | |
c8b2a36f GC |
6092 | mem_cgroup_remove_from_trees(memcg); |
6093 | free_css_id(&mem_cgroup_subsys, &memcg->css); | |
6094 | ||
6095 | for_each_node(node) | |
6096 | free_mem_cgroup_per_zone_info(memcg, node); | |
6097 | ||
6098 | free_percpu(memcg->stat); | |
6099 | ||
3f134619 GC |
6100 | /* |
6101 | * We need to make sure that (at least for now), the jump label | |
6102 | * destruction code runs outside of the cgroup lock. This is because | |
6103 | * get_online_cpus(), which is called from the static_branch update, | |
6104 | * can't be called inside the cgroup_lock. cpusets are the ones | |
6105 | * enforcing this dependency, so if they ever change, we might as well. | |
6106 | * | |
6107 | * schedule_work() will guarantee this happens. Be careful if you need | |
6108 | * to move this code around, and make sure it is outside | |
6109 | * the cgroup_lock. | |
6110 | */ | |
a8964b9b | 6111 | disarm_static_keys(memcg); |
3afe36b1 GC |
6112 | if (size < PAGE_SIZE) |
6113 | kfree(memcg); | |
6114 | else | |
6115 | vfree(memcg); | |
59927fb9 | 6116 | } |
3afe36b1 | 6117 | |
59927fb9 | 6118 | |
8c7c6e34 | 6119 | /* |
c8b2a36f GC |
6120 | * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU, |
6121 | * but in process context. The work_freeing structure is overlaid | |
6122 | * on the rcu_freeing structure, which itself is overlaid on memsw. | |
8c7c6e34 | 6123 | */ |
c8b2a36f | 6124 | static void free_work(struct work_struct *work) |
33327948 | 6125 | { |
c8b2a36f | 6126 | struct mem_cgroup *memcg; |
08e552c6 | 6127 | |
c8b2a36f GC |
6128 | memcg = container_of(work, struct mem_cgroup, work_freeing); |
6129 | __mem_cgroup_free(memcg); | |
6130 | } | |
04046e1a | 6131 | |
c8b2a36f GC |
6132 | static void free_rcu(struct rcu_head *rcu_head) |
6133 | { | |
6134 | struct mem_cgroup *memcg; | |
08e552c6 | 6135 | |
c8b2a36f GC |
6136 | memcg = container_of(rcu_head, struct mem_cgroup, rcu_freeing); |
6137 | INIT_WORK(&memcg->work_freeing, free_work); | |
6138 | schedule_work(&memcg->work_freeing); | |
33327948 KH |
6139 | } |
6140 | ||
c0ff4b85 | 6141 | static void mem_cgroup_get(struct mem_cgroup *memcg) |
8c7c6e34 | 6142 | { |
c0ff4b85 | 6143 | atomic_inc(&memcg->refcnt); |
8c7c6e34 KH |
6144 | } |
6145 | ||
c0ff4b85 | 6146 | static void __mem_cgroup_put(struct mem_cgroup *memcg, int count) |
8c7c6e34 | 6147 | { |
c0ff4b85 R |
6148 | if (atomic_sub_and_test(count, &memcg->refcnt)) { |
6149 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
c8b2a36f | 6150 | call_rcu(&memcg->rcu_freeing, free_rcu); |
7bcc1bb1 DN |
6151 | if (parent) |
6152 | mem_cgroup_put(parent); | |
6153 | } | |
8c7c6e34 KH |
6154 | } |
6155 | ||
c0ff4b85 | 6156 | static void mem_cgroup_put(struct mem_cgroup *memcg) |
483c30b5 | 6157 | { |
c0ff4b85 | 6158 | __mem_cgroup_put(memcg, 1); |
483c30b5 DN |
6159 | } |
6160 | ||
7bcc1bb1 DN |
6161 | /* |
6162 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
6163 | */ | |
e1aab161 | 6164 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
7bcc1bb1 | 6165 | { |
c0ff4b85 | 6166 | if (!memcg->res.parent) |
7bcc1bb1 | 6167 | return NULL; |
c0ff4b85 | 6168 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
7bcc1bb1 | 6169 | } |
e1aab161 | 6170 | EXPORT_SYMBOL(parent_mem_cgroup); |
33327948 | 6171 | |
8787a1df | 6172 | static void __init mem_cgroup_soft_limit_tree_init(void) |
f64c3f54 BS |
6173 | { |
6174 | struct mem_cgroup_tree_per_node *rtpn; | |
6175 | struct mem_cgroup_tree_per_zone *rtpz; | |
6176 | int tmp, node, zone; | |
6177 | ||
3ed28fa1 | 6178 | for_each_node(node) { |
f64c3f54 BS |
6179 | tmp = node; |
6180 | if (!node_state(node, N_NORMAL_MEMORY)) | |
6181 | tmp = -1; | |
6182 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
8787a1df | 6183 | BUG_ON(!rtpn); |
f64c3f54 BS |
6184 | |
6185 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
6186 | ||
6187 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
6188 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
6189 | rtpz->rb_root = RB_ROOT; | |
6190 | spin_lock_init(&rtpz->lock); | |
6191 | } | |
6192 | } | |
f64c3f54 BS |
6193 | } |
6194 | ||
0eb253e2 | 6195 | static struct cgroup_subsys_state * __ref |
92fb9748 | 6196 | mem_cgroup_css_alloc(struct cgroup *cont) |
8cdea7c0 | 6197 | { |
d142e3e6 | 6198 | struct mem_cgroup *memcg; |
04046e1a | 6199 | long error = -ENOMEM; |
6d12e2d8 | 6200 | int node; |
8cdea7c0 | 6201 | |
c0ff4b85 R |
6202 | memcg = mem_cgroup_alloc(); |
6203 | if (!memcg) | |
04046e1a | 6204 | return ERR_PTR(error); |
78fb7466 | 6205 | |
3ed28fa1 | 6206 | for_each_node(node) |
c0ff4b85 | 6207 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 6208 | goto free_out; |
f64c3f54 | 6209 | |
c077719b | 6210 | /* root ? */ |
28dbc4b6 | 6211 | if (cont->parent == NULL) { |
a41c58a6 | 6212 | root_mem_cgroup = memcg; |
d142e3e6 GC |
6213 | res_counter_init(&memcg->res, NULL); |
6214 | res_counter_init(&memcg->memsw, NULL); | |
6215 | res_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 6216 | } |
28dbc4b6 | 6217 | |
d142e3e6 GC |
6218 | memcg->last_scanned_node = MAX_NUMNODES; |
6219 | INIT_LIST_HEAD(&memcg->oom_notify); | |
6220 | atomic_set(&memcg->refcnt, 1); | |
6221 | memcg->move_charge_at_immigrate = 0; | |
6222 | mutex_init(&memcg->thresholds_lock); | |
6223 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 6224 | vmpressure_init(&memcg->vmpressure); |
d142e3e6 GC |
6225 | |
6226 | return &memcg->css; | |
6227 | ||
6228 | free_out: | |
6229 | __mem_cgroup_free(memcg); | |
6230 | return ERR_PTR(error); | |
6231 | } | |
6232 | ||
6233 | static int | |
6234 | mem_cgroup_css_online(struct cgroup *cont) | |
6235 | { | |
6236 | struct mem_cgroup *memcg, *parent; | |
6237 | int error = 0; | |
6238 | ||
6239 | if (!cont->parent) | |
6240 | return 0; | |
6241 | ||
0999821b | 6242 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
6243 | memcg = mem_cgroup_from_cont(cont); |
6244 | parent = mem_cgroup_from_cont(cont->parent); | |
6245 | ||
6246 | memcg->use_hierarchy = parent->use_hierarchy; | |
6247 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
6248 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
6249 | ||
6250 | if (parent->use_hierarchy) { | |
c0ff4b85 R |
6251 | res_counter_init(&memcg->res, &parent->res); |
6252 | res_counter_init(&memcg->memsw, &parent->memsw); | |
510fc4e1 | 6253 | res_counter_init(&memcg->kmem, &parent->kmem); |
55007d84 | 6254 | |
7bcc1bb1 DN |
6255 | /* |
6256 | * We increment refcnt of the parent to ensure that we can | |
6257 | * safely access it on res_counter_charge/uncharge. | |
6258 | * This refcnt will be decremented when freeing this | |
6259 | * mem_cgroup(see mem_cgroup_put). | |
6260 | */ | |
6261 | mem_cgroup_get(parent); | |
18f59ea7 | 6262 | } else { |
c0ff4b85 R |
6263 | res_counter_init(&memcg->res, NULL); |
6264 | res_counter_init(&memcg->memsw, NULL); | |
510fc4e1 | 6265 | res_counter_init(&memcg->kmem, NULL); |
8c7f6edb TH |
6266 | /* |
6267 | * Deeper hierachy with use_hierarchy == false doesn't make | |
6268 | * much sense so let cgroup subsystem know about this | |
6269 | * unfortunate state in our controller. | |
6270 | */ | |
d142e3e6 | 6271 | if (parent != root_mem_cgroup) |
8c7f6edb | 6272 | mem_cgroup_subsys.broken_hierarchy = true; |
18f59ea7 | 6273 | } |
cbe128e3 GC |
6274 | |
6275 | error = memcg_init_kmem(memcg, &mem_cgroup_subsys); | |
0999821b | 6276 | mutex_unlock(&memcg_create_mutex); |
cbe128e3 GC |
6277 | if (error) { |
6278 | /* | |
6279 | * We call put now because our (and parent's) refcnts | |
6280 | * are already in place. mem_cgroup_put() will internally | |
6281 | * call __mem_cgroup_free, so return directly | |
6282 | */ | |
6283 | mem_cgroup_put(memcg); | |
e4715f01 GC |
6284 | if (parent->use_hierarchy) |
6285 | mem_cgroup_put(parent); | |
cbe128e3 | 6286 | } |
d142e3e6 | 6287 | return error; |
8cdea7c0 BS |
6288 | } |
6289 | ||
5f578161 MH |
6290 | /* |
6291 | * Announce all parents that a group from their hierarchy is gone. | |
6292 | */ | |
6293 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) | |
6294 | { | |
6295 | struct mem_cgroup *parent = memcg; | |
6296 | ||
6297 | while ((parent = parent_mem_cgroup(parent))) | |
6298 | atomic_inc(&parent->dead_count); | |
6299 | ||
6300 | /* | |
6301 | * if the root memcg is not hierarchical we have to check it | |
6302 | * explicitely. | |
6303 | */ | |
6304 | if (!root_mem_cgroup->use_hierarchy) | |
6305 | atomic_inc(&root_mem_cgroup->dead_count); | |
6306 | } | |
6307 | ||
92fb9748 | 6308 | static void mem_cgroup_css_offline(struct cgroup *cont) |
df878fb0 | 6309 | { |
c0ff4b85 | 6310 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
ec64f515 | 6311 | |
5f578161 | 6312 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
ab5196c2 | 6313 | mem_cgroup_reparent_charges(memcg); |
1f458cbf | 6314 | mem_cgroup_destroy_all_caches(memcg); |
df878fb0 KH |
6315 | } |
6316 | ||
92fb9748 | 6317 | static void mem_cgroup_css_free(struct cgroup *cont) |
8cdea7c0 | 6318 | { |
c0ff4b85 | 6319 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
c268e994 | 6320 | |
1d62e436 | 6321 | kmem_cgroup_destroy(memcg); |
d1a4c0b3 | 6322 | |
c0ff4b85 | 6323 | mem_cgroup_put(memcg); |
8cdea7c0 BS |
6324 | } |
6325 | ||
02491447 | 6326 | #ifdef CONFIG_MMU |
7dc74be0 | 6327 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
6328 | #define PRECHARGE_COUNT_AT_ONCE 256 |
6329 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 6330 | { |
854ffa8d DN |
6331 | int ret = 0; |
6332 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
c0ff4b85 | 6333 | struct mem_cgroup *memcg = mc.to; |
4ffef5fe | 6334 | |
c0ff4b85 | 6335 | if (mem_cgroup_is_root(memcg)) { |
854ffa8d DN |
6336 | mc.precharge += count; |
6337 | /* we don't need css_get for root */ | |
6338 | return ret; | |
6339 | } | |
6340 | /* try to charge at once */ | |
6341 | if (count > 1) { | |
6342 | struct res_counter *dummy; | |
6343 | /* | |
c0ff4b85 | 6344 | * "memcg" cannot be under rmdir() because we've already checked |
854ffa8d DN |
6345 | * by cgroup_lock_live_cgroup() that it is not removed and we |
6346 | * are still under the same cgroup_mutex. So we can postpone | |
6347 | * css_get(). | |
6348 | */ | |
c0ff4b85 | 6349 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
854ffa8d | 6350 | goto one_by_one; |
c0ff4b85 | 6351 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
854ffa8d | 6352 | PAGE_SIZE * count, &dummy)) { |
c0ff4b85 | 6353 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
854ffa8d DN |
6354 | goto one_by_one; |
6355 | } | |
6356 | mc.precharge += count; | |
854ffa8d DN |
6357 | return ret; |
6358 | } | |
6359 | one_by_one: | |
6360 | /* fall back to one by one charge */ | |
6361 | while (count--) { | |
6362 | if (signal_pending(current)) { | |
6363 | ret = -EINTR; | |
6364 | break; | |
6365 | } | |
6366 | if (!batch_count--) { | |
6367 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
6368 | cond_resched(); | |
6369 | } | |
c0ff4b85 R |
6370 | ret = __mem_cgroup_try_charge(NULL, |
6371 | GFP_KERNEL, 1, &memcg, false); | |
38c5d72f | 6372 | if (ret) |
854ffa8d | 6373 | /* mem_cgroup_clear_mc() will do uncharge later */ |
38c5d72f | 6374 | return ret; |
854ffa8d DN |
6375 | mc.precharge++; |
6376 | } | |
4ffef5fe DN |
6377 | return ret; |
6378 | } | |
6379 | ||
6380 | /** | |
8d32ff84 | 6381 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
6382 | * @vma: the vma the pte to be checked belongs |
6383 | * @addr: the address corresponding to the pte to be checked | |
6384 | * @ptent: the pte to be checked | |
02491447 | 6385 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
6386 | * |
6387 | * Returns | |
6388 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
6389 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
6390 | * move charge. if @target is not NULL, the page is stored in target->page | |
6391 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
6392 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
6393 | * target for charge migration. if @target is not NULL, the entry is stored | |
6394 | * in target->ent. | |
4ffef5fe DN |
6395 | * |
6396 | * Called with pte lock held. | |
6397 | */ | |
4ffef5fe DN |
6398 | union mc_target { |
6399 | struct page *page; | |
02491447 | 6400 | swp_entry_t ent; |
4ffef5fe DN |
6401 | }; |
6402 | ||
4ffef5fe | 6403 | enum mc_target_type { |
8d32ff84 | 6404 | MC_TARGET_NONE = 0, |
4ffef5fe | 6405 | MC_TARGET_PAGE, |
02491447 | 6406 | MC_TARGET_SWAP, |
4ffef5fe DN |
6407 | }; |
6408 | ||
90254a65 DN |
6409 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
6410 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 6411 | { |
90254a65 | 6412 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 6413 | |
90254a65 DN |
6414 | if (!page || !page_mapped(page)) |
6415 | return NULL; | |
6416 | if (PageAnon(page)) { | |
6417 | /* we don't move shared anon */ | |
4b91355e | 6418 | if (!move_anon()) |
90254a65 | 6419 | return NULL; |
87946a72 DN |
6420 | } else if (!move_file()) |
6421 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
6422 | return NULL; |
6423 | if (!get_page_unless_zero(page)) | |
6424 | return NULL; | |
6425 | ||
6426 | return page; | |
6427 | } | |
6428 | ||
4b91355e | 6429 | #ifdef CONFIG_SWAP |
90254a65 DN |
6430 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
6431 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6432 | { | |
90254a65 DN |
6433 | struct page *page = NULL; |
6434 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
6435 | ||
6436 | if (!move_anon() || non_swap_entry(ent)) | |
6437 | return NULL; | |
4b91355e KH |
6438 | /* |
6439 | * Because lookup_swap_cache() updates some statistics counter, | |
6440 | * we call find_get_page() with swapper_space directly. | |
6441 | */ | |
33806f06 | 6442 | page = find_get_page(swap_address_space(ent), ent.val); |
90254a65 DN |
6443 | if (do_swap_account) |
6444 | entry->val = ent.val; | |
6445 | ||
6446 | return page; | |
6447 | } | |
4b91355e KH |
6448 | #else |
6449 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
6450 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6451 | { | |
6452 | return NULL; | |
6453 | } | |
6454 | #endif | |
90254a65 | 6455 | |
87946a72 DN |
6456 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
6457 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
6458 | { | |
6459 | struct page *page = NULL; | |
87946a72 DN |
6460 | struct address_space *mapping; |
6461 | pgoff_t pgoff; | |
6462 | ||
6463 | if (!vma->vm_file) /* anonymous vma */ | |
6464 | return NULL; | |
6465 | if (!move_file()) | |
6466 | return NULL; | |
6467 | ||
87946a72 DN |
6468 | mapping = vma->vm_file->f_mapping; |
6469 | if (pte_none(ptent)) | |
6470 | pgoff = linear_page_index(vma, addr); | |
6471 | else /* pte_file(ptent) is true */ | |
6472 | pgoff = pte_to_pgoff(ptent); | |
6473 | ||
6474 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
6475 | page = find_get_page(mapping, pgoff); |
6476 | ||
6477 | #ifdef CONFIG_SWAP | |
6478 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
6479 | if (radix_tree_exceptional_entry(page)) { | |
6480 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 6481 | if (do_swap_account) |
aa3b1895 | 6482 | *entry = swap; |
33806f06 | 6483 | page = find_get_page(swap_address_space(swap), swap.val); |
87946a72 | 6484 | } |
aa3b1895 | 6485 | #endif |
87946a72 DN |
6486 | return page; |
6487 | } | |
6488 | ||
8d32ff84 | 6489 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
6490 | unsigned long addr, pte_t ptent, union mc_target *target) |
6491 | { | |
6492 | struct page *page = NULL; | |
6493 | struct page_cgroup *pc; | |
8d32ff84 | 6494 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
6495 | swp_entry_t ent = { .val = 0 }; |
6496 | ||
6497 | if (pte_present(ptent)) | |
6498 | page = mc_handle_present_pte(vma, addr, ptent); | |
6499 | else if (is_swap_pte(ptent)) | |
6500 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
6501 | else if (pte_none(ptent) || pte_file(ptent)) |
6502 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
6503 | |
6504 | if (!page && !ent.val) | |
8d32ff84 | 6505 | return ret; |
02491447 DN |
6506 | if (page) { |
6507 | pc = lookup_page_cgroup(page); | |
6508 | /* | |
6509 | * Do only loose check w/o page_cgroup lock. | |
6510 | * mem_cgroup_move_account() checks the pc is valid or not under | |
6511 | * the lock. | |
6512 | */ | |
6513 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6514 | ret = MC_TARGET_PAGE; | |
6515 | if (target) | |
6516 | target->page = page; | |
6517 | } | |
6518 | if (!ret || !target) | |
6519 | put_page(page); | |
6520 | } | |
90254a65 DN |
6521 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
6522 | if (ent.val && !ret && | |
9fb4b7cc | 6523 | css_id(&mc.from->css) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
6524 | ret = MC_TARGET_SWAP; |
6525 | if (target) | |
6526 | target->ent = ent; | |
4ffef5fe | 6527 | } |
4ffef5fe DN |
6528 | return ret; |
6529 | } | |
6530 | ||
12724850 NH |
6531 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
6532 | /* | |
6533 | * We don't consider swapping or file mapped pages because THP does not | |
6534 | * support them for now. | |
6535 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
6536 | */ | |
6537 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6538 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6539 | { | |
6540 | struct page *page = NULL; | |
6541 | struct page_cgroup *pc; | |
6542 | enum mc_target_type ret = MC_TARGET_NONE; | |
6543 | ||
6544 | page = pmd_page(pmd); | |
6545 | VM_BUG_ON(!page || !PageHead(page)); | |
6546 | if (!move_anon()) | |
6547 | return ret; | |
6548 | pc = lookup_page_cgroup(page); | |
6549 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
6550 | ret = MC_TARGET_PAGE; | |
6551 | if (target) { | |
6552 | get_page(page); | |
6553 | target->page = page; | |
6554 | } | |
6555 | } | |
6556 | return ret; | |
6557 | } | |
6558 | #else | |
6559 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
6560 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
6561 | { | |
6562 | return MC_TARGET_NONE; | |
6563 | } | |
6564 | #endif | |
6565 | ||
4ffef5fe DN |
6566 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
6567 | unsigned long addr, unsigned long end, | |
6568 | struct mm_walk *walk) | |
6569 | { | |
6570 | struct vm_area_struct *vma = walk->private; | |
6571 | pte_t *pte; | |
6572 | spinlock_t *ptl; | |
6573 | ||
12724850 NH |
6574 | if (pmd_trans_huge_lock(pmd, vma) == 1) { |
6575 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) | |
6576 | mc.precharge += HPAGE_PMD_NR; | |
6577 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6578 | return 0; |
12724850 | 6579 | } |
03319327 | 6580 | |
45f83cef AA |
6581 | if (pmd_trans_unstable(pmd)) |
6582 | return 0; | |
4ffef5fe DN |
6583 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
6584 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 6585 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
6586 | mc.precharge++; /* increment precharge temporarily */ |
6587 | pte_unmap_unlock(pte - 1, ptl); | |
6588 | cond_resched(); | |
6589 | ||
7dc74be0 DN |
6590 | return 0; |
6591 | } | |
6592 | ||
4ffef5fe DN |
6593 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
6594 | { | |
6595 | unsigned long precharge; | |
6596 | struct vm_area_struct *vma; | |
6597 | ||
dfe076b0 | 6598 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
6599 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6600 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
6601 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
6602 | .mm = mm, | |
6603 | .private = vma, | |
6604 | }; | |
6605 | if (is_vm_hugetlb_page(vma)) | |
6606 | continue; | |
4ffef5fe DN |
6607 | walk_page_range(vma->vm_start, vma->vm_end, |
6608 | &mem_cgroup_count_precharge_walk); | |
6609 | } | |
dfe076b0 | 6610 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
6611 | |
6612 | precharge = mc.precharge; | |
6613 | mc.precharge = 0; | |
6614 | ||
6615 | return precharge; | |
6616 | } | |
6617 | ||
4ffef5fe DN |
6618 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
6619 | { | |
dfe076b0 DN |
6620 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
6621 | ||
6622 | VM_BUG_ON(mc.moving_task); | |
6623 | mc.moving_task = current; | |
6624 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
6625 | } |
6626 | ||
dfe076b0 DN |
6627 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
6628 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 6629 | { |
2bd9bb20 KH |
6630 | struct mem_cgroup *from = mc.from; |
6631 | struct mem_cgroup *to = mc.to; | |
6632 | ||
4ffef5fe | 6633 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
6634 | if (mc.precharge) { |
6635 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
6636 | mc.precharge = 0; | |
6637 | } | |
6638 | /* | |
6639 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
6640 | * we must uncharge here. | |
6641 | */ | |
6642 | if (mc.moved_charge) { | |
6643 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
6644 | mc.moved_charge = 0; | |
4ffef5fe | 6645 | } |
483c30b5 DN |
6646 | /* we must fixup refcnts and charges */ |
6647 | if (mc.moved_swap) { | |
483c30b5 DN |
6648 | /* uncharge swap account from the old cgroup */ |
6649 | if (!mem_cgroup_is_root(mc.from)) | |
6650 | res_counter_uncharge(&mc.from->memsw, | |
6651 | PAGE_SIZE * mc.moved_swap); | |
6652 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
6653 | ||
6654 | if (!mem_cgroup_is_root(mc.to)) { | |
6655 | /* | |
6656 | * we charged both to->res and to->memsw, so we should | |
6657 | * uncharge to->res. | |
6658 | */ | |
6659 | res_counter_uncharge(&mc.to->res, | |
6660 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
6661 | } |
6662 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
6663 | mc.moved_swap = 0; |
6664 | } | |
dfe076b0 DN |
6665 | memcg_oom_recover(from); |
6666 | memcg_oom_recover(to); | |
6667 | wake_up_all(&mc.waitq); | |
6668 | } | |
6669 | ||
6670 | static void mem_cgroup_clear_mc(void) | |
6671 | { | |
6672 | struct mem_cgroup *from = mc.from; | |
6673 | ||
6674 | /* | |
6675 | * we must clear moving_task before waking up waiters at the end of | |
6676 | * task migration. | |
6677 | */ | |
6678 | mc.moving_task = NULL; | |
6679 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 6680 | spin_lock(&mc.lock); |
4ffef5fe DN |
6681 | mc.from = NULL; |
6682 | mc.to = NULL; | |
2bd9bb20 | 6683 | spin_unlock(&mc.lock); |
32047e2a | 6684 | mem_cgroup_end_move(from); |
4ffef5fe DN |
6685 | } |
6686 | ||
761b3ef5 LZ |
6687 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
6688 | struct cgroup_taskset *tset) | |
7dc74be0 | 6689 | { |
2f7ee569 | 6690 | struct task_struct *p = cgroup_taskset_first(tset); |
7dc74be0 | 6691 | int ret = 0; |
c0ff4b85 | 6692 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgroup); |
ee5e8472 | 6693 | unsigned long move_charge_at_immigrate; |
7dc74be0 | 6694 | |
ee5e8472 GC |
6695 | /* |
6696 | * We are now commited to this value whatever it is. Changes in this | |
6697 | * tunable will only affect upcoming migrations, not the current one. | |
6698 | * So we need to save it, and keep it going. | |
6699 | */ | |
6700 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; | |
6701 | if (move_charge_at_immigrate) { | |
7dc74be0 DN |
6702 | struct mm_struct *mm; |
6703 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
6704 | ||
c0ff4b85 | 6705 | VM_BUG_ON(from == memcg); |
7dc74be0 DN |
6706 | |
6707 | mm = get_task_mm(p); | |
6708 | if (!mm) | |
6709 | return 0; | |
7dc74be0 | 6710 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
6711 | if (mm->owner == p) { |
6712 | VM_BUG_ON(mc.from); | |
6713 | VM_BUG_ON(mc.to); | |
6714 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 6715 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 6716 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 6717 | mem_cgroup_start_move(from); |
2bd9bb20 | 6718 | spin_lock(&mc.lock); |
4ffef5fe | 6719 | mc.from = from; |
c0ff4b85 | 6720 | mc.to = memcg; |
ee5e8472 | 6721 | mc.immigrate_flags = move_charge_at_immigrate; |
2bd9bb20 | 6722 | spin_unlock(&mc.lock); |
dfe076b0 | 6723 | /* We set mc.moving_task later */ |
4ffef5fe DN |
6724 | |
6725 | ret = mem_cgroup_precharge_mc(mm); | |
6726 | if (ret) | |
6727 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
6728 | } |
6729 | mmput(mm); | |
7dc74be0 DN |
6730 | } |
6731 | return ret; | |
6732 | } | |
6733 | ||
761b3ef5 LZ |
6734 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
6735 | struct cgroup_taskset *tset) | |
7dc74be0 | 6736 | { |
4ffef5fe | 6737 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
6738 | } |
6739 | ||
4ffef5fe DN |
6740 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
6741 | unsigned long addr, unsigned long end, | |
6742 | struct mm_walk *walk) | |
7dc74be0 | 6743 | { |
4ffef5fe DN |
6744 | int ret = 0; |
6745 | struct vm_area_struct *vma = walk->private; | |
6746 | pte_t *pte; | |
6747 | spinlock_t *ptl; | |
12724850 NH |
6748 | enum mc_target_type target_type; |
6749 | union mc_target target; | |
6750 | struct page *page; | |
6751 | struct page_cgroup *pc; | |
4ffef5fe | 6752 | |
12724850 NH |
6753 | /* |
6754 | * We don't take compound_lock() here but no race with splitting thp | |
6755 | * happens because: | |
6756 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not | |
6757 | * under splitting, which means there's no concurrent thp split, | |
6758 | * - if another thread runs into split_huge_page() just after we | |
6759 | * entered this if-block, the thread must wait for page table lock | |
6760 | * to be unlocked in __split_huge_page_splitting(), where the main | |
6761 | * part of thp split is not executed yet. | |
6762 | */ | |
6763 | if (pmd_trans_huge_lock(pmd, vma) == 1) { | |
62ade86a | 6764 | if (mc.precharge < HPAGE_PMD_NR) { |
12724850 NH |
6765 | spin_unlock(&vma->vm_mm->page_table_lock); |
6766 | return 0; | |
6767 | } | |
6768 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
6769 | if (target_type == MC_TARGET_PAGE) { | |
6770 | page = target.page; | |
6771 | if (!isolate_lru_page(page)) { | |
6772 | pc = lookup_page_cgroup(page); | |
6773 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, | |
2f3479b1 | 6774 | pc, mc.from, mc.to)) { |
12724850 NH |
6775 | mc.precharge -= HPAGE_PMD_NR; |
6776 | mc.moved_charge += HPAGE_PMD_NR; | |
6777 | } | |
6778 | putback_lru_page(page); | |
6779 | } | |
6780 | put_page(page); | |
6781 | } | |
6782 | spin_unlock(&vma->vm_mm->page_table_lock); | |
1a5a9906 | 6783 | return 0; |
12724850 NH |
6784 | } |
6785 | ||
45f83cef AA |
6786 | if (pmd_trans_unstable(pmd)) |
6787 | return 0; | |
4ffef5fe DN |
6788 | retry: |
6789 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
6790 | for (; addr != end; addr += PAGE_SIZE) { | |
6791 | pte_t ptent = *(pte++); | |
02491447 | 6792 | swp_entry_t ent; |
4ffef5fe DN |
6793 | |
6794 | if (!mc.precharge) | |
6795 | break; | |
6796 | ||
8d32ff84 | 6797 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
6798 | case MC_TARGET_PAGE: |
6799 | page = target.page; | |
6800 | if (isolate_lru_page(page)) | |
6801 | goto put; | |
6802 | pc = lookup_page_cgroup(page); | |
7ec99d62 | 6803 | if (!mem_cgroup_move_account(page, 1, pc, |
2f3479b1 | 6804 | mc.from, mc.to)) { |
4ffef5fe | 6805 | mc.precharge--; |
854ffa8d DN |
6806 | /* we uncharge from mc.from later. */ |
6807 | mc.moved_charge++; | |
4ffef5fe DN |
6808 | } |
6809 | putback_lru_page(page); | |
8d32ff84 | 6810 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6811 | put_page(page); |
6812 | break; | |
02491447 DN |
6813 | case MC_TARGET_SWAP: |
6814 | ent = target.ent; | |
e91cbb42 | 6815 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6816 | mc.precharge--; |
483c30b5 DN |
6817 | /* we fixup refcnts and charges later. */ |
6818 | mc.moved_swap++; | |
6819 | } | |
02491447 | 6820 | break; |
4ffef5fe DN |
6821 | default: |
6822 | break; | |
6823 | } | |
6824 | } | |
6825 | pte_unmap_unlock(pte - 1, ptl); | |
6826 | cond_resched(); | |
6827 | ||
6828 | if (addr != end) { | |
6829 | /* | |
6830 | * We have consumed all precharges we got in can_attach(). | |
6831 | * We try charge one by one, but don't do any additional | |
6832 | * charges to mc.to if we have failed in charge once in attach() | |
6833 | * phase. | |
6834 | */ | |
854ffa8d | 6835 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6836 | if (!ret) |
6837 | goto retry; | |
6838 | } | |
6839 | ||
6840 | return ret; | |
6841 | } | |
6842 | ||
6843 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
6844 | { | |
6845 | struct vm_area_struct *vma; | |
6846 | ||
6847 | lru_add_drain_all(); | |
dfe076b0 DN |
6848 | retry: |
6849 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
6850 | /* | |
6851 | * Someone who are holding the mmap_sem might be waiting in | |
6852 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6853 | * and retry. Because we cancel precharges, we might not be able | |
6854 | * to move enough charges, but moving charge is a best-effort | |
6855 | * feature anyway, so it wouldn't be a big problem. | |
6856 | */ | |
6857 | __mem_cgroup_clear_mc(); | |
6858 | cond_resched(); | |
6859 | goto retry; | |
6860 | } | |
4ffef5fe DN |
6861 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
6862 | int ret; | |
6863 | struct mm_walk mem_cgroup_move_charge_walk = { | |
6864 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6865 | .mm = mm, | |
6866 | .private = vma, | |
6867 | }; | |
6868 | if (is_vm_hugetlb_page(vma)) | |
6869 | continue; | |
4ffef5fe DN |
6870 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
6871 | &mem_cgroup_move_charge_walk); | |
6872 | if (ret) | |
6873 | /* | |
6874 | * means we have consumed all precharges and failed in | |
6875 | * doing additional charge. Just abandon here. | |
6876 | */ | |
6877 | break; | |
6878 | } | |
dfe076b0 | 6879 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
6880 | } |
6881 | ||
761b3ef5 LZ |
6882 | static void mem_cgroup_move_task(struct cgroup *cont, |
6883 | struct cgroup_taskset *tset) | |
67e465a7 | 6884 | { |
2f7ee569 | 6885 | struct task_struct *p = cgroup_taskset_first(tset); |
a433658c | 6886 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 6887 | |
dfe076b0 | 6888 | if (mm) { |
a433658c KM |
6889 | if (mc.to) |
6890 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
6891 | mmput(mm); |
6892 | } | |
a433658c KM |
6893 | if (mc.to) |
6894 | mem_cgroup_clear_mc(); | |
67e465a7 | 6895 | } |
5cfb80a7 | 6896 | #else /* !CONFIG_MMU */ |
761b3ef5 LZ |
6897 | static int mem_cgroup_can_attach(struct cgroup *cgroup, |
6898 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6899 | { |
6900 | return 0; | |
6901 | } | |
761b3ef5 LZ |
6902 | static void mem_cgroup_cancel_attach(struct cgroup *cgroup, |
6903 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6904 | { |
6905 | } | |
761b3ef5 LZ |
6906 | static void mem_cgroup_move_task(struct cgroup *cont, |
6907 | struct cgroup_taskset *tset) | |
5cfb80a7 DN |
6908 | { |
6909 | } | |
6910 | #endif | |
67e465a7 | 6911 | |
f00baae7 TH |
6912 | /* |
6913 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
6914 | * to verify sane_behavior flag on each mount attempt. | |
6915 | */ | |
6916 | static void mem_cgroup_bind(struct cgroup *root) | |
6917 | { | |
6918 | /* | |
6919 | * use_hierarchy is forced with sane_behavior. cgroup core | |
6920 | * guarantees that @root doesn't have any children, so turning it | |
6921 | * on for the root memcg is enough. | |
6922 | */ | |
6923 | if (cgroup_sane_behavior(root)) | |
6924 | mem_cgroup_from_cont(root)->use_hierarchy = true; | |
6925 | } | |
6926 | ||
8cdea7c0 BS |
6927 | struct cgroup_subsys mem_cgroup_subsys = { |
6928 | .name = "memory", | |
6929 | .subsys_id = mem_cgroup_subsys_id, | |
92fb9748 | 6930 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6931 | .css_online = mem_cgroup_css_online, |
92fb9748 TH |
6932 | .css_offline = mem_cgroup_css_offline, |
6933 | .css_free = mem_cgroup_css_free, | |
7dc74be0 DN |
6934 | .can_attach = mem_cgroup_can_attach, |
6935 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 6936 | .attach = mem_cgroup_move_task, |
f00baae7 | 6937 | .bind = mem_cgroup_bind, |
6bc10349 | 6938 | .base_cftypes = mem_cgroup_files, |
6d12e2d8 | 6939 | .early_init = 0, |
04046e1a | 6940 | .use_id = 1, |
8cdea7c0 | 6941 | }; |
c077719b | 6942 | |
c255a458 | 6943 | #ifdef CONFIG_MEMCG_SWAP |
a42c390c MH |
6944 | static int __init enable_swap_account(char *s) |
6945 | { | |
6946 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 6947 | if (!strcmp(s, "1")) |
a42c390c | 6948 | really_do_swap_account = 1; |
a2c8990a | 6949 | else if (!strcmp(s, "0")) |
a42c390c MH |
6950 | really_do_swap_account = 0; |
6951 | return 1; | |
6952 | } | |
a2c8990a | 6953 | __setup("swapaccount=", enable_swap_account); |
c077719b | 6954 | |
2d11085e MH |
6955 | static void __init memsw_file_init(void) |
6956 | { | |
6acc8b02 MH |
6957 | WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, memsw_cgroup_files)); |
6958 | } | |
6959 | ||
6960 | static void __init enable_swap_cgroup(void) | |
6961 | { | |
6962 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6963 | do_swap_account = 1; | |
6964 | memsw_file_init(); | |
6965 | } | |
2d11085e | 6966 | } |
6acc8b02 | 6967 | |
2d11085e | 6968 | #else |
6acc8b02 | 6969 | static void __init enable_swap_cgroup(void) |
2d11085e MH |
6970 | { |
6971 | } | |
c077719b | 6972 | #endif |
2d11085e MH |
6973 | |
6974 | /* | |
1081312f MH |
6975 | * subsys_initcall() for memory controller. |
6976 | * | |
6977 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
6978 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
6979 | * everything that doesn't depend on a specific mem_cgroup structure should | |
6980 | * be initialized from here. | |
2d11085e MH |
6981 | */ |
6982 | static int __init mem_cgroup_init(void) | |
6983 | { | |
6984 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); | |
6acc8b02 | 6985 | enable_swap_cgroup(); |
8787a1df | 6986 | mem_cgroup_soft_limit_tree_init(); |
e4777496 | 6987 | memcg_stock_init(); |
2d11085e MH |
6988 | return 0; |
6989 | } | |
6990 | subsys_initcall(mem_cgroup_init); |